Einsatz von digitalen Entscheidungshilfen für das Management von heterogenen Pflanzenbeständen im ökologischen Landbau

Tobias Reuter Use of Digital Decision Support Tools for Managing Heterogeneous Fields in Organic Farming Einsatz von digitalen Entscheidungshilfen für das Management von heterogenen Pflanzenbeständen im ökologischen Landbau PhD Thesis Osnabrück University 2025 Use of Digital Decision Support Tools for Managing Heterogeneous Fields in Organic Farming Einsatz von digitalen Entscheidungshilfen für das Management von heterogenen Pflanzenbeständen im ökologischen Landbau Dissertation zur Erlangung des Doktorgrades Doktor der Naturwissenschaften (Dr. rer. nat.) des Fachbereichs Kulturund Sozialwissenschaften der Universität Osnabrück in Kooperation mit der Hochschule Osnabrück Fakultät Agrarwissenschaften und Landschaftsarchitektur vorgelegt von Tobias Reuter Aus Bad Godesberg am 05.09.2025, Osnabrück Index i Index Index ........................................................................................................................................................ i Abstract .................................................................................................................................................. ii Abbreviations ........................................................................................................................................ iii Figure index .......................................................................................................................................... iv Table index ............................................................................................................................................. v Chapter 1 General introduction ....................................................................................................... 6 1.1 Background and objective .......................................................................................................... 7 1.2 Structure of the thesis ................................................................................................................. 8 1.3 Study area ................................................................................................................................ 10 1.4 Organic agriculture in Germany and European Union ............................................................. 11 1.5 Heterogeneous field conditions ................................................................................................ 13 1.6 Precision Farming and decision support .................................................................................. 16 1.6.1 Data sensing and analysis........................................................................................................ 17 1.6.2 Decision making ....................................................................................................................... 20 1.6.3 Resource applications .............................................................................................................. 22 1.6.4 Crop mapping and evaluation .................................................................................................. 23 1.7 Research questions and hypotheses ....................................................................................... 24 Chapter 2 Scientific publications within the context of this work.............................................. 26 2.1 Site-specific mechanical weed management in maize (Zea mays) in North-West Germany .. 27 2.2 Delineation of management zones in clover-grass for site-specific management of subsequent crops .................................................................................................................................................. 45 2.3 Effects of mixed intercropping on the agronomic parameters of two organically grown malting barley cultivars (Hordeum vulgare) in Northwest Germany .................................................................. 66 Chapter 3 General discussion ........................................................................................................ 67 3.1 Integration of Precision Farming and organic agriculture: potential and challenges ............... 68 3.2 Variation of agricultural parameters between years and sites ................................................. 71 3.3 Outlook: Synergies between production and ecology .............................................................. 73 Chapter 4 Conclusions ................................................................................................................... 77 Summary .............................................................................................................................................. 78 Zusammenfassung (German summary) ............................................................................................ 80 References ........................................................................................................................................... 82 Acknowledgements ........................................................................................................................... 103 Appendix ............................................................................................................................................ 105 List of publications ............................................................................................................................... 105 Abstract ii Abstract The global agricultural sector faces the dual challenge of ensuring food security for a growing population while contending with a steady decline in biodiversity. Although organic farming offers significant environmental benefits—in terms of improved soil health and increased biodiversity—its yields are, on average, 20 % lower than those achieved with conventional practices. Precision Farming (PF) offers a promising solution to mitigate this yield gap by optimizing input applications (e.g., fertilizers) based on spatial variability within fields, which most fields exhibit. This dissertation presents possible solutions for enhancing the productivity and sustainability of organic farming. PF technologies, such as Unmanned aerial vehicles (UAVs) and Decision Support Systems (DSS), were integrated with traditional organic management practices including mechanical weeding, crop rotation and intercropping. To test this approaches a series of trials were conducted in northwestern Germany, reported in three peerreviewed papers. Site-specific mechanical weeding reduced the treated area by 58 % to 83 % in maize (depending on the year). Weeds were detected based on image recognition of UAVmultispectral images. The Relative Weed Cover was a successful decision support, as it helped to limit the weeding areas by considering both weed and maize cover. UAVs were also effectively employed for delineating of three clover–grass fields into distinct management zones. The used NDRE (Normalized Difference Red Edge)-Maps and fuzzy C-means clustering algorithms were appropriated for this task. These zones exhibited significant differences in clover proportion, biomass and the yield of subsequent cereal crops—driven primarily by variations in soil properties and topography. Such delineation supports the tailored management of sub-fields for optimal fertilization and crop selection. In addition, intercropping trials demonstrated that mixing malting barley with peas (Pisum sativum) enhanced protein content and land-use efficiency, whereas intercropping with linseed (Linum usitatissimum) resulted in reduced protein levels; nevertheless, sole stands of barley generally performed comparably to mixed cropping systems. Intercropping can therefore be used to influence the quality by partner crop to mitigate the spatial heterogeneity within the field. Collectively, these approaches advance sustainable agriculture by reducing unnecessary weed control and enabling more efficient resource use. This thesis contributes to the few research projects that combines PF-technology with organic farming practice. The results add to more efficient resource use and sustainable agriculture. The continued development of field robots and advanced DSS holds the potential to further modernize agriculture toward a biodiversity-based system. Future practices may involve subdividing fields into smaller management units that are integrated with landscape elements to promote biodiversity. However, additional long-term, user-focused research is necessary to fully understand the complex interactions within environmentally friendly agricultural landscapes. Abbreviations iii Abbreviations AI: Artificial Intelligence AMSL: Above mean sea-level ANN: Artificial Neural Network BBCH: Biological Federal Institute for Agriculture and Forestry (regarding plant growth stages) CCCI: Canopy Chlorophyll Content Index CGZ: Clover-grass zones CNN: Convolutional Neural Network Con: Control treatment CU: Cereal units DA: Digital Agriculture DAS: Days after sowing DSS: Decision Support Systems EU: European Union GNDVI: Green Normalized Difference Vegetation Index GNSS: Global Navigation Satellite System GS: Growth stages LCCI: Leaf Chlorophyll Content Index LER: Land Equivalent Ratio LiDAR: Light Detection and Ranging LLM: Large Language Models MCARI: Modified Chlorophyll Absorption Ratio Index Mg: Megagram ML: Machine Learning MZ: Management zone N: Nitrogen NDRE: Normalized Difference Red Edge NDVI: Normalized Difference Vegetation Index NUE: Nutrient use efficiency OA: Overall accuracy PA: Precision Agriculture PF: Precision Farming PSRI: Plant Senescence Reflectance Index RF: Random forest RGB: Red, Green, Blue RWC: Relative weed cover RYT: Relative yield total SMN: Soil mineral nitrogen SNC: Soil nitrogen content SOC: Soil organic carbon SOM: Soil organic matter SSWM: Site-specific weed management TKW: Thousand kernel weight UAV: Unmanned aerial vehicles VI: Vegetation Index VRA: Variable rate application VSWC: Volumetric soil water content WC: Weed cover XAI: eXplainable Artificial Intelligence Figure index iv Figure index Fig 1. Overview of three peer-reviewed papers on temporal and spatial scales. ................................... 9 Fig 2. Geographical distribution of experimental sites .......................................................................... 10 Fig 3. Total (point, solid line) and shared (triangle, dashed) area under organic management ........... 13 Fig 4. Overview of sources of spatial heterogeneity. ............................................................................ 14 Fig 5. Workflow of Precision Farming with examples in each step. ...................................................... 17 Fig 6. (Sub)plot arrangement and placement of measuring areas. ...................................................... 31 Fig 7. Mean air temperature [°C] (lines) and sum of precipitation [mm] (bars) ..................................... 31 Fig 8. Linear regression between volumetric soil water content [%] and weed cover [%]. ................... 37 Fig 9. Weed cover [%] estimated by image recognition before first site-specific weed management .. 38 Fig 10. Comparison of treated area [m²] due to second hoeing between treatment ............................ 39 Fig 11. Comparison of maize dry matter yield [g/m-2] at top and weed biomass [g m-2] at bottom ...... 39 Fig 12. Geographical Distribution of Experimental Sites ...................................................................... 48 Fig 13. Terrain based on digital elevation model of the trial fields. ....................................................... 49 Fig 14. Mean air temperature [°C] (lines) and sum of precipitation [mm] (bars) ................................... 51 Fig 15. NDRE-Maps (normalized difference red edge) of clover-grass for the different fields and selected dates. ...................................................................................................................................... 54 Fig 16. Maps for the delineated of clover-grass zones (CGZ) for each field. ....................................... 55 Fig 17. Yield of clover, grass, and weeds are compared between the two clover-grass zones ........... 55 Fig 18. Yield proportions of clover, grass, and weeds are compared between the two clover-grass zones .................................................................................................................................................. 56 Fig 19. Cereal plant height [cm] in comparison between the two clover-grass zones (CGZ) ............... 57 Fig 20. Cereal kernel yield [g m-2] in comparison between the two clover-grass zones (CGZ) at harvest. .................................................................................................................................................. 57 Fig 21. Development of volumetric soil water content in comparison between two clover-grass zones .. ................................................................................................................................................ 58 Fig S 1. Indexes for finding best number of clusters ............................................................................. 65 Fig S 2. Maps for the delineated of clover-grass zones (CGZ) for the single observed dates ............. 65 Picture 1: Sampling of winter wheat (Triticum aestivum) trial, 17th May 2023. ..................................... 6 Picture 2: Summer spelt (Triticum aestivum subsp. Spelta) trial, 7th July 2021 .................................. 26 Picture 3: Summer spelt (Triticum aestivum subsp. Spelta) trial, 19th July 2022 ................................ 67 Table index v Table index Table 1: Soil nutrient content. ............................................................................................................... 30 Table 2: Detailed field history and trial management for all experimental seasons. ............................ 32 Table 3: Parameters of the weed recognition for the different sampling dates. ................................... 33 Table 4: Comparison of volumetric water content ................................................................................ 35 Table 5: Comparison of maize plant height [cm], weed cover [%] and number of weed species ........ 36 Table 6: Accuracies for UAV-based image classification of weeds, maize and soil. ............................ 37 Table 7: Field description. ..................................................................................................................... 48 Table 8: Field management and soil sampling dates ........................................................................... 49 Table 13: Tested hypothesises in the three peer-reviewed paper. ....................................................... 69 Table S 1: Sampling dates field data and UAC campaigns. ................................................................. 63 General introduction 6 Chapter 1 General introduction Picture 1: Sampling of winter wheat (Triticum aestivum) trial, 17th May 2023. General introduction 7 1.1 Background and objective The global agricultural sector faces the challenge of feeding an anticipated population of nearly 10.000.000.000 people by 2050 (Willett et al., 2019), while simultaneously contending with decreasing agricultural land per capita (FAO, 2024). This situation is compounded by significant declines in both plant and animal wildlife populations (Geiger et al., 2010; Willett et al., 2019). Intensive use of fertilizers and pesticides has led to a promotion of plant and animal species (Francksen et al., 2022; Millard et al., 2021), which are often more harmful, than a diverse community (Esposito et al., 2023). Additionally, homogeneous agricultural landscapes—both in terms of crop rotation and landscape elements—fail to support diverse wildlife, as each species has specific requirements such as nutrient supply, tillage time and hosts (Meyer et al., 2019; Tscharntke et al., 2021). Diverse landscapes contribute to varied wildlife. This has both temporal and spatial scales. The temporal scale refers to varying and extended elements of crop rotation (Davis et al., 2012). The spatial scale includes landscape elements (flower strips, trees and hedgerows), habitat connectivity, crop mixtures, smaller field sizes and higher field edge density (Raatz et al., 2019). The concept of planetary boundaries summaries the global challenge for a stable earth system. It delineates critical parameters essential for maintaining earth's system stability and resilience. Out of these nine processes, six currently exceed their boundaries. Agriculture significantly impacts many of these processes, particularly biogeochemical flows, novel entities like pesticides, land-system change due to agricultural management and biosphere integrity (Richardson et al., 2023). Organic agriculture holds the potential to reduce the agricultural impact. It aims to produce food in an environmentally friendly manner, promoting closed nutrient cycles and minimizing the use of pesticides and fertilizers. This practice leads to average of 30 % higher species richness, especially plants (Tuck et al., 2014) and pollinators like bees profit from organic farming (Walker et al., 2024). However, this approach often results in yields that are 17 % to 20 % lower than conventional farming practices (de la Cruz et al., 2023; De Ponti et al., 2012), due to limitation of N fertiliser (Döring and Neuhoff, 2021). The prohibition of chemical pesticides necessitates the use of mechanical weeding, which is less efficient in weed control and can lead to soil erosion and plant injuries (Machleb et al., 2020; Melander et al., 2015). Agricultural fields, both conventional or organic managed, often exhibit a variety of conditions, including differences in soil texture, organic matter content, topography, nutrient availability and water accessibility (Aksakal et al., 2019; Zhu et al., 2013). These variations, influenced by field history (Schulp and Verburg, 2009) and landscape elements, such as flowering strips, hedgerows or trees (Raatz et al., 2019), result in diverse plant growth patterns that impact both productivity and wildlife (Pätzold et al., 2020). Precision agriculture addresses these issues by tailoring management practices to the specific conditions of sub-fields (EliChukwu, 2019). Modern agricultural practices often utilize sensors to monitor plant and soil conditions, with the collected data being analysed to provide management recommendations through digital Decision Support Systems (DSS; Balasundram et al., 2023). These systems leverage agricultural knowledge alongside statistical models or Artificial Intelligence to optimize decision-making processes (Eli-Chukwu, 2019). General introduction 14 Fig 4. Overview of sources of spatial heterogeneity. The grey area represents parent material, the brown area represents soil, and the green area represents the surface. Soil texture (a) and soil organic matter (SOM) (b) influence water and nutrient retention. Soil depth (c) limits rooting space and availability of water and nutrients. Topography (d) affects water flow and exposure to sunlight; typically, the lower slope is more fertile due to the accumulation of fine textured soil and SOM. Land management practices (e) impact soil fertility and nutrient levels due to tillage and fertilisation. Generally, organic farming and no-till management increase SOM and earthworm abundance. Past land use (f) affects yield and SOM content in the current period. Landscape elements like trees and hedgerows compete with crops for light, nutrients, and water. Landscape elements (g), such as field borders, restrict fertilisation and pesticide use. Weeds (h) occur in heterogeneous patches and compete with crops for nutrients, water, and light. Some pest species (i) are distributing from an infection point and building patches. Soil condition Soil texture is one of the most important soil parameters because it significantly affects soil chemical properties and crop yield (Bölenius et al., 2017; Franzluebbers et al., 2025). In particular, the clay content of soil influences the retention of water and nutrients due to its particle structure and negative charge (Boenecke et al., 2018; Girz and Mattila, 2024; Habib‑ur‑Rahman et al., 2022). Thus, a higher clay content can promote crop yields—for example, in grass leys (da Silva et al., 2022), as well as in rice (Oryza sativa), corn (Zea mays), cotton (Gossypium hirsutum) (Ouazaa et al., 2022) and winter wheat (Triticum aestivum; Groß et al., 2023; Usowicz and Lipiec, 2017). Moreover, because biological nitrogen fixation is correlated with the biomass production of legumes (Hoekstra et al., 2015), soil texture indirectly influences nitrogen content. It should be noted that crops respond differently to soil texture; for instance, soybeans (Glycine max) are significantly more affected by soil texture than pasture crops (Oldoni et al., 2025). In addition, SOM has a strong positive effect on soil fertility and crop yield because it supplies nutrients and can retain water and nitrogen (Groß et al., 2023). General introduction 15 Soil depth constrains the rooting space of plants and thus influences the availability of nutrients and water. In areas with deeper soils, more resources are available, which is generally associated with higher yields (Odone et al., 2024; Peralta et al., 2015; Sadras and Calviño, 2001). Notably, the response to soil depth is stronger in maize (Zea mays) than in wheat (Triticum aestivum) (Sadras and Calviño, 2001). Many fields feature a distinct relief. In many cases, the lower part of a slope is more productive than the upper slope (Bölenius et al., 2017; Peralta et al., 2015; Rodriguez Miranda et al., 2021) due to the accumulation of SOM and a finer soil texture that results in deeper soils downslope, thereby enhancing soil fertility. Additionally, downslope areas often exhibit higher soil water content because water naturally flows to lower elevations (Peralta et al., 2015; Rodriguez Miranda et al., 2021). In contrast, Habib‑ur‑Rahman et al. (2022) reported that yields at the top of the slope were 53 – 88 % higher than those downslope—where elevations were 2 to 5 m lower. This discrepancy was explained by the accumulation of SOM in a small depression within a convex slope. Moreover, topography influences exposure to sunlight (Peralta et al., 2015). Management influence Human activities, including historical land use, also influence soil properties and fertility. For example, the regular application of fertilizers is a common practice (Franzluebbers et al., 2025). Conversely, the use of heavy machinery can compact the soil, thus limiting rooting depth and reducing its capacity to store water—negatively affecting yields (Girz and Mattila, 2024). The avoidance of ploughing (also known as no-till management) can improve soil bulk density and stability (Franzluebbers et al., 2025), as well as increase SOM by approximately 21 %. But the effect on SOM varies between crops, with no effect in maize (Zea Maize) to strong increase in barley (Hordeum vulgare) (Bai et al., 2018). Other study showed a significant higher SOM content only in the top soil layer (0 – 15 cm), at deeper soil layers, no effect occurred (Jakab et al., 2023; Wulanningtyas et al., 2021). Such practices additionally stimulate soil biological activity, with earthworm abundance increasing by up to 50 % when pesticides are not applied (Bai et al., 2018). Organic farming is considered to have a positive effect on soil fertility as mentioned in section 1.4. Part of this benefit is attributable to the use of organic fertilizers, which can increase SOM by 21 – 47 % (Bai et al., 2018). A more diverse crop rotation has been shown to increase earthworm abundance by 60 % and SOM by 20 % (Bai et al., 2018). Moreover, permanent grasslands typically exhibit higher soil aggregate stability, enhanced biological activity and lower soil bulk density compared to arable lands (Franzluebbers et al., 2025). The effects on SOM and soil nitrogen can persist for decades; for example, changes in land use to arable land can still be detected 50 years later and continue to influence yields (Schuster et al., 2024). Similarly, land use changes from forest and plaggen agriculture in the Netherlands have produced measurable effects even after more than 100 years (Schulp and Verburg, 2009). Field borders have also evolved historically as fields were split or combined. Land-use changes—for example, converting permanent grassland to arable land—were sometimes implemented at the sub-field level (Schuster et al., 2024). Consequently, historical land-use decisions and crop management practices contribute to increased spatial variability. All fields are characterized by certain bordering elements, including field borders, agricultural roads, neighbouring fields, trees and hedgerows, all of which affect crop development on site (Raatz et al., 2019). Generally, border areas yield less (within 6 – 7 m of General introduction 16 the edge) compared to mid-field areas due to restrictions in pesticide and fertilizer applications (Raatz et al., 2019). Natural landscape elements such as trees and hedgerows can lead to yield reductions of 17.5 % and 7.9 %, respectively, over ranges of approximately 18 m—effects that are attributed to shading and competition for water and nutrients (Jose et al., 2000; Raatz et al., 2019). It should be noted that trees can also positively affect yields outside the immediate shading zone (beyond 20 m; Raatz et al., 2019). Beyond their influence on light, trees and hedgerows function as wind barriers and help reduce evapotranspiration (Jacobs et al., 2022). These effects vary from year to year, in dry years, shading and reduced evapotranspiration can be beneficial, whereas in wet years the opposite may occur (Raatz et al., 2019). This is especially important under future conditions of climate change with increasing heat. Trees can further provide protection against crop damage of extreme precipitation (Jacobs et al., 2022). The heterogeneous conditions influence not only crops but also weeds (Usowicz and Lipiec, 2017), resulting in a patchy distribution of wild plants (Metcalfe et al., 2019). Each species has its own specific environmental requirements (Pätzold et al., 2020) and the variability in conditions influences plant competitiveness. This further contributes to the heterogeneous distribution of both crops and weeds (Mhlanga et al., 2016). Many pathogens like Fusarium graminearum (Fusarium head blight) or Puccinia striiformis f. sp. tritici (Pst) (Yellow rust) are spreading from initial infection point. Centre around this infection point, nests are building, leading to a patchy distribution (Gao et al., 2023; A. Guo et al., 2021). 1.6 Precision Farming and decision support Precision Farming (PF), also known as Precision Agriculture, involves using sensors to gain insights into crop development, nutrient supply and growing conditions. Resources are then applied according to these conditions (Moran et al., 1997). The interpretation of this information can be either manual or automatic, leading to management decisions that adapt to the spatial and temporal heterogeneity within a field or landscape. This approach aims to use resources more efficiently (Balasundram et al., 2023; FAO, 2022; Karunathilake et al., 2023). The framework for PF is based on three key agricultural steps: (1) diagnosis, (2) decision making and (3) performing (FAO, 2022). The workflow includes: (i) data sensing, (ii) data analysis, (iii) decision making, (iv) resource application, (v) crop mapping and (vi) evaluation (Balasundram et al., 2023). Fig 5 illustrates the main workflow and provides examples. PF encompasses various topics, including crop farming, agronomics, livestock, aquaculture and agroforestry (Karunathilake et al., 2023). Smart farming, digital agriculture or Agriculture 4.0 is the further development of PF and sometimes used synonymously (Leddin et al., 2023). This concept provides a digital “ecosystem”, with the aim to connect and analyse data across the whole agricultural production chain. Digital agriculture combines a wide variety of sensors, autonomous sensing and application, big data analysis (AI and DSS) with Internet of things (IoT: a network of interconnected items and technologies; Basso and Antle, 2020; Goumagias et al., 2021). Unmanned aerial vehicle (UAV) and Unmanned ground vehicles (UGV), such as field robots, are commonly used in this concept (Balasundram et al., 2023). This study focuses on the management of heterogeneous crop fields using digital DSS. In this context, the term "Precision Farming" (PF) pertains specifically to crop farming, excluding other potential areas. The subsequent sections will provide a detailed examination of data sensing and analysis, decision making and resource application. General introduction 17 Fig 5. Workflow of Precision Farming with examples in each step. UAV = unmanned aerial vehicle. 1.6.1 Data sensing and analysis The sensing and analysis of data regarding crop development and soil conditions form the basis of PF applications. A wide range of tools and data sources is available—each with a distinct scope and application. In the following a selection of state of the art methods are described. Sensors are commonly employed, many based on optical principles measuring reflectance. For example, chlorophyll meters and green seekers assess the chlorophyll content of leaves, which serves as an indicator of plant nutrient levels. These sensors are available as handheld devices or can be mounted on tractors (Sharma and Bali, 2018). Remote sensing involves acquiring information about an object without physical contact (Fussell et al., 1986). Many remote sensing applications deploy sensors on satellites or UAVs, each with its own benefits. Satellites are often low-cost or even free and provide global coverage (Vidican et al., 2023). In contrast, UAVs can yield images with pixel resolutions under 1 cm (Nahrstedt et al., 2024), compared to satellites such as PlanetScope (3 m), Sentinel-2 (10 m), or Landsat (30 m). Additionally, UAVs offer greater flexibility—especially under cloudy conditions (Bareth et al., 2019). In research, satellite imagery is the most common, accounting for approximately 65 % of studies, followed by aerial vehicles at 22 % (Vidican et al., 2023). General introduction 18 Vegetation Indices Many remote sensing approaches utilize Vegetation Indices (VIs), which are spectral transformations combining two or more bands to identify vegetation and quantify biomass. The most commonly used index is the Normalized Difference Vegetation Index (NDVI) because of its suitability for a wide range of applications. This index correlates well with green biomass (Breunig et al., 2020; Munnaf et al., 2022; Sapkota et al., 2024; Vidican et al., 2023). However, NDVI is known to oversaturate at higher biomass values (Karunaratne et al., 2020; Xu et al., 2020). In addition, indices such as Green Normalized Difference Vegetation Index (GNDVI), NDVI, Modified Chlorophyll Absorption Ratio Index (MCARI) and Plant Senescence Reflectance Index (PSRI) can serve a broad range of objectives, while others like the Canopy Chlorophyll Content Index (CCCI) and Leaf Chlorophyll Content Index (LCCI) are more specific to applications such as crop classification and chlorophyll estimation (Vidican et al., 2023). VIs have been employed for a wide range of applications. They have been used to distinguish among crops—such as beans (Phaseolus vulgaris), beetroot (Beta vulgaris), grass (Poaceae), maize (Zea mays), potato (Solanum tuberosum) and wheat (Triticum aestivum)— at a landscape level (Sonobe et al., 2018) and to detect stress (Skendži et al., 2023). Because nutrient content and other quality parameters influence the spectral response, VIs can be used to assess these aspects, as demonstrated by Zeng and Chen (2018) for forages and by Barzin et al. (2020) for maize. Yield estimation and prediction represent a broad research area, having been applied to many crops including maize (Zea mays; Barzin et al., 2020), wheat (Triticum aestivum; Cheng et al., 2022), potato (Solanum tuberosum), sugar beet (Beta vulgaris) (Vannoppen and Gobin, 2022) and forages (Zeng and Chen, 2018). Combining VIs with other measurements (e.g., plant height) can improve biomass estimation in grasslands (Viljanen et al., 2018). RGB indices have also been successfully used for estimating forage biomass (Lussem et al., 2019). Although RGB cameras are more affordable, better results are typically achieved when combining RGB data with near-infrared (NIR) imagery, given strong correlation of NIR with chlorophyll content (Viljanen et al., 2018). Despite their broad utility, VIs face several challenges, including: (1) variability in crop growth stages (e.g., flowering) due to differences in leaf structure; (2) overlapping leaves; (3) variability within and between fields; (4) cloud cover interference and (5) similar spectral responses across different plant species. These factors can complicate the interpretation of VI data (Vidican et al., 2023). Image Analysis and Machine Learning Image analysis methods that use the original spectral and textural information can help mitigate some of the challenges associated with VI transformations, which may lead to information loss. While many analyses have traditionally relied on statistical models, machine learning (ML) approaches—including Random Forest (RF), Support Vector Machines (SVM) and Artificial Neural Networks (ANN)—are gaining prominence. These methods can handle complex, multidimensional data by independently correlating, weighting and extrapolating multiple features (Gerhards et al., 2020; Júnior et al., 2020; Lee et al., 2017). For instance, Azimi et al. (2021) compared Convolutional Neural Networks (CNN) with classical Machine learning methods and achieved higher accuracy when measuring stress levels due to nutrient deficiency. Machine learning methods such as Support Vector Machines and Random Forest have also been used to estimate wheat yield, with R² values around 0.9 General introduction 19 (Cheng et al., 2022). In grass sward studies, Random Forest was used to successfully estimate dry matter yield and nitrogen uptake, achieving R² values of 0.85 and 0.89, respectively (Oliveira et al., 2020). Viljanen et al. (2018) achieved an R² of 0.97 for dry matter estimation using a combination of sward height, RGB-spectral values and VIs, while Random Forest has also been deployed to identify crop proportions in grasslands using UAV imagery (Nahrstedt et al., 2024). CNNs provide accuracy advantages over classical ML approaches for weed detection, although they typically require longer training times (Júnior et al., 2020). For example, weed recognition in maize achieved an accuracy of 88 % using a CNN applied to high-resolution RGB images captured near ground level (Hasan et al., 2024). Comparable results have been obtained during the seedling stage (V1) of maize with UAV-captured RGB images, despite lower pixel resolution (Pei et al., 2022). In grassland studies, UAV multispectral images and CNN models have been used to map wild plant species, achieving an overall accuracy of 88 % (Pöttker et al., 2023). Multispectral UAV imagery has also been used to identify Cercospora Leaf Spot (caused by Cercospora beticola) in sugar beet (Beta vulgaris), where partial least squares discriminant analysis achieved an overall accuracy of 86 % (Barreto et al., 2023). For detecting Fusarium head blight (caused by Fusarium graminearum) in wheat (Triticum aestivum), a combination of VIs and texture indices achieved 93 % accuracy with UAV-mounted multispectral images (Gao et al., 2023). A similar approach was used for the detection of yellow rust (caused by Puccinia striiformis f. sp. tritici) in wheat (Triticum aestivum), where hyperspectral UAV images yielded R² values ranging from 0.55 to 0.88 depending on the infection stage, with lower accuracy observed during early infections (A. Guo et al., 2021). Overview of soil analysis methods Soil organic carbon has been estimated with R² values of 0.89 using hyperspectral images and 0.69 using time series of multispectral images. These results, achieved via a Neural Network model (Extreme Learning Machine), outperformed those obtained from partial least squares regression (Guo et al., 2021). Traditional soil sampling remains a valuable method for producing soil maps and often underpins remote sensing approaches (Ouazaa et al., 2022); however, it is limited by high costs (Buladaco II et al., 2024) and low spatial resolution (Ouazaa et al., 2022). An alternative is the use of proximal sensors, which operate either on an optical or an electrochemical principle. An example of an electrochemical sensor is the device developed by Smolka et al. (2017). This measurement technique is based on capillary electrophoresis, in which ions in a liquid sample are separated by an electric field, allowing them to pass sequentially past a detector. Evaluations demonstrate a strong correlation for nitrate (NO₃-) and potassium (K+) measurements; however, for ammonium (NH₄+) and phosphate (PO43-), the concentrations were too low to be reliably detected (Smolka et al., 2017). Farmlab (Stenon, Germany), a handheld device that utilizes impedance measurements and spectral analysis (ranging from ultraviolet to NIR) to estimate various nitrogen forms in the soil. While effective at detecting field heterogeneity, this device tends to overestimate mineral nitrogen (Nmin) by 38 kg N ha−1 in 75 % of cases compared to laboratory analyses (Vikuk et al., 2024). Additionally, electrical conductivity meters (e.g., the EM38 by Geonics Ltd., Mississauga, Canada) measure soil electrical conductivity via electromagnetic induction, providing a good estimate of soil fertility since the values correlate strongly with soil water content and clay content (Cimpoiaşua et al., 2020; Fortes et al., 2015). General introduction 20 Soil water content serves as an excellent indicator of soil heterogeneity and provides insight into the water supply available in a field. Time Domain Reflectometry (TDR) is commonly used for these measurements (Engels et al., 2025). One key advantage of TDR is its high accuracy—typically within 1 – 2 % of volumetric content—along with minimal calibration requirements. Additionally, TDR offers excellent spatial and temporal resolution, with options for automation and continuous monitoring. The devices are easy to use, fast and non-destructive (Skierucha et al., 2012). Soil compaction is a crucial factor affecting root growth, as well as soil water and air availability. Penetrometer can measure soil penetrability or penetration resistance on a routine basis and are used to estimate soil compaction (Herrick and Jones, 2002). Historical Yield Data Historical yield data from combined harvesters can offer valuable insights into field variability, as yield often correlates with soil fertility. Key to this analysis are the global navigation satellite system (GNSS) coordinates, which precisely locate yield data and multiyear records that help mitigate the effects of annual weather variability (Speranza et al., 2023). 1.6.2 Decision making After data are collected and analysed, they have to be interpreted, leading to a management decision. The data from yield maps or other technology mentioned above can already give valuable insight into the spatial variability and help farmers make informed decisions about the optimal timing of harvest (Ghazal et al., 2024). In general, decisions are categorized into three categories: (1) strategic decisions are on a high-level and complex like business decisions; (2) tactical decision or management decision are done weekly or monthly, like fertilizer use; and (3) operational decision, this is on a daily basis like worker distribution (Leddin et al., 2023). Farmers have to do several decisions each day and they cannot be an expert for every topic (Saikai et al., 2020). Management decisions are complex due to the need of broad knowledge and interaction in the agricultural field. Furthermore they are characterized by uncertainty and risk due to response of crops to weather differences within and between seasons. An additionally uncertainty is the price development of resources like fuel, fertilizer and product prices (Leddin et al., 2023). Decision Support Systems (DSS) and Artificial Intelligence (AI) can help with decision making due to the availability of handling highly complex and dynamic conditions in agriculture (Karunathilake et al., 2023). The challenge of DSS is to represent the reality with unique conditions at each farm (Saikai et al., 2020). Methods for Decision Support Systems Various methods for DSS are employed, including both statistical and AI-based approaches (Jha et al., 2019). Statistical modelling involves the development of mathematical models to represent relationships between variables using data. These models are utilized to comprehend patterns, make predictions and test hypotheses. Examples include regression analysis, k-means clustering and classification (Bockstaller and Girardin, 2003). Fuzzy logic, on the other hand, is a computational theory based on "gradual truths," as opposed to the conventional Boolean logic that relies solely on true or false (1 or 0) constructs, forming the foundational structure for all computational operations (Mamdani, 1974). Machine learning is described as the scientific field that enables machines to learn without explicit programming (Samuel, 2000). It is often used for computer vision but also for decision making (Jha et al., 2019; Saikai et al., 2020). Artificial Neural Networks (ANN) are also self-learning algorithms. General introduction 21 The architecture of ANN typically includes three or more layers: 1. Input layer 2. At least one hidden (middle) layer 3. Output layer. This is roughly inspired by the connectivity of the human brain. They are especially beneficial in the prediction of highly complex and dynamic settings (Song and He, 2005). Neuro-fuzzy logic is the combination of ANN with fuzzy logic (Wierman and Dobransky, 1993), the advantage is the implementation of human experience and knowledge (Papageorgiou et al., 2011). Expert systems are one the oldest approaches of AI. This system has two components: (1) the knowledge base contains the symbolic knowledge of the expert in the form of rules and heuristics. (2) An Inference engine that solves problems by interpreting the rules of the knowledge base. An expert system can be flexible, adjusted with new expert knowledge. As it is based on human knowledge, it is easier to explain (McKinion and Lemmon, 1985), especially compared to ANNs. Selected DSS applications in agriculture DSS were developed for a wide range of goals and application and have different levels of complexity. The DSS “Fruchtfolge” (German for crop rotation) was developed for crop recommendation and coarse manure application in Germany. The system uses customer reference number for field geometry, previous crops and soil data. Based on a crop model, regional yield and crop process, it suggests what crop to plant and how much manure to apply. For this an integer linear programming model is used. The case study farm shows a strong increase in profits (Pahmeyer et al., 2021). The Integrated Farm System Model (IFSM) is a research model for the dairy forage farm system. Unlike most farm models, IFSM simulates all major farm components on a process level and can therefore simulate economic performance. It simulates the growth of different crops including alfalfa (Medicago sativa), grass, maize (Zea mays), soybean (Glycine max) and small grain crops based on weather data. Nutrient supply, tillage and harvesting are considered. Based on available feeds and the nutrient requirements, the animal responses are computed with a cost-minimizing linear programming approach, to use the feed in an optimal way. Another component is the nutrient flow due to manure production, biological N-fixation and losses (denitrification and leaching). This model is foremost for research purposes but can help with decision support (Rotz et al., 2018). A model, developed to support farmers decision for harvest in Ireland is ”PastureBase”. This calculates the yield of each paddock based on a growth model using pastor cover (visual assessment or platemeter) and the use of paddock (grazed or silage). The yield was predicted with R² of 0.84 for grazed paddocks. The DSS achieved a R² of only 0.58 for silage used paddock due to lower number of observations. Nonetheless, this helps the farmer to find the right harvest time, as the yield is an important predictor (Hanrahan et al., 2017). Another expert system for pasture harvest management was developed by Reuter et al. (2023). It integrates weather forecasts, pasture height, growth stage, clover proportion and crude fibre content (derived from a growth model) to determine the optimal harvest date and form—either hay or silage. A harvest date is recommended only if a dry period of at least five days (for silage) or seven days (for hay) is forecasted. Validation of this rule set on 26 fields produced an overall R² of 0.75, while intensively managed fields for silage achieving an R² of 0.9. A Machine learning-model was established for site-specific applications in general with the approach that each farmer can construct a unique model for the own farm. The inputs can be drawn from historical data. The algorithm utilizes Bayesian optimization and can handle a wide range of management and environmental variables, adapting to unforeseen interaction effects. General introduction 22 This DSS was test in simulation and yielded positive results for learnability of complex sitespecific management and the higher profitability compared to uniform management (Saikai et al., 2020). Site-specific fertilisation is often based on yield potential maps, soil properties, or satelliteand UAV-images, as described in section 1.6.1. Most commonly, the fertilizer is distributed to even out differences or improve the yield in specific areas (Hagn et al., 2025; Morari et al., 2018). A DSS, based on soil data, was developed for the distribution of manure with the aim to increase the stable pool of SOM (Corti et al., 2023). The nutrient use efficiency and yield depend on the varying weather conditions during the growing season. One approach uses mid-term weather forecasts combined with crop forecast to plan the siteand season-specific N-management, leading to less used N-fertilizer without economic losses (Corti et al., 2023; Palka and Manschadi, 2024). Corti et al. (2023) showed that the effect of variable rate manure application varied between the years, but overall can improve the nitrogen use efficiency (NUE). Another complex decision topic is the management of weeds, as timing can be crucial, as well as the right herbicide choice. Much research focus on the site-specific application to reduce herbicides or weeding. This is often based on a simple weed threshold systems either for whole weed cover (Nikolić et al., 2021) or species specific threshold (Hamouz et al., 2014). Some studies follow even a zero-tolerance strategy (Allmendinger et al., 2024; Spaeth et al., 2024). Thresholds can either base on expert knowledge or experimentation and estimation of yield losses by weeds (Longchamps et al., 2014). The weed cover or number is mostly estimated with UAV or tractor mounted cameras as described in section 1.6.1. Some approaches are more complex. For the harrowing intensity a non-linear model was developed. The optimum angle of the weed harrow tines was predicted based on three variables: (1) the pre-harrow measured weed cover, (2) the draft force measured on the tines and (3) the target weed control efficacy required to lower the weed damage to below the biological threshold (Berge et al., 2024). The right timing of image acquisition for site-specific weeding can be estimated by a weed mergence prediction model with soil temperature and precipitation as inputs (Nikolić et al., 2021). Niemeyer et al. (2024) developed a reasoning system based on an expert system. It categorizes weeds into the classes “harmful” and “not harmful”, the classification depends on the distance between crop-plant and weed-plant and relative germinations time. Weeds classified as “non harmful” can be spared or tolerated in higher coverage than “harmful” species. In each grid cell the number of each species is counted and compared with the specific threshold. Zingsheim and Döring (2024) suggest that weeding robots should have the ability to measure weed cover and identify individual weed species to optimize crop yield and biodiversity. 1.6.3 Resource applications PF is the practice of applying resources based on the specific conditions within each field, resulting in more efficient resource management. These resources include soil, space, fertilizer and water (Ghazal et al., 2024). Adjusting resource applications to account for field variability is known as site-specific management or variable rate application (VRA). An important technology in this context is the global navigation satellite system (GNSS), which enables precise location mapping for targeted resource delivery (Karunathilake et al., 2023). One widely used practice is site-specific nutrient management, often implemented as variable rate fertilization. By applying fertilizers in accordance with crop needs and soil nutrientsupplying capacity, nitrogen use efficiency (NUE) can be significantly increased, while General introduction 23 reducing nutrient losses and greenhouse gas emissions (Balasundram et al., 2023). For instance, research in Germany demonstrated that sensor-based VRA could reduce nitrogen application by 38 kg ha-1, resulting in a 15 % higher NUE (Hagn et al., 2025). However, the effectiveness of VRA is not consistent across all sites and years. Trials in Austria have shown that the positive effects of site-specific nitrogen application—regarding yield, protein content and economic profit—vary by location (Palka and Manschadi, 2024). Similarly, site-specific manure application resulted in higher NUE, though outcomes fluctuated from year to year (Corti et al., 2023). Site-specific weeding reduces the use of chemicals, thereby enhancing environmental sustainability and reducing herbicide stress (Ghazal et al., 2024; Spaeth et al., 2024). When mechanical weeding is applied, it can also mitigate negative impacts such as plant injuries and soil degradation (Seitz et al., 2019; Woźniak, 2020). The reduction in treated area can vary between 10 % and 90 %, with many studies reporting savings between 40 % and 60 % (Allmendinger et al., 2024; Castaldi et al., 2017; Nikolić et al., 2021; Peña et al., 2013). These variations depend on factors such as weather conditions (López-Granados et al., 2016), weed infestation levels (Spaeth et al., 2024) and the timing and thresholds applied (Nikolić et al., 2021) and grid size (López-Granados et al., 2016). Li et al. (2022) tested an in-field, real-time spraying system that used industrial-grade RGB cameras to detect field thistle (Cirsium arvense var. integrifolium) plants and immediately apply herbicide. The system’s detection accuracy and relative hit rate were affected by driving speed; increasing the speed from 2 km ha-1 dropped the accuracy from 91 % to 80 % and the hit rate from 95 % to 93 %. Economic benefits from reduced herbicide application have been estimated to range from 16 € ha-1 to 33 € ha-1 (Nikolić et al., 2021) and up to 150 € ha-1 (Rajmis et al., 2022). As noted in section 1.6.1, pest detection also lays the foundation for site-specific management. For example, potential fungicide reductions of around 18 % have been observed (Rajmis et al., 2022). Another trial conducted in grassland achieved fungicide reductions between 51 % and 65 % (Booth et al., 2021). Seeding rates can be adjusted to local field conditions, as demonstrated with maize using NDVI and soil maps. Site-specific seeding resulted in 3 % to 7 % higher yield compared with uniform seeding, thereby improving the gross margin by 3 % to 9 % (Munnaf et al., 2022). For potatoes, yields increased to 17 Mg ha-1 with variable rate seeding compared to 14 Mg ha-1 with uniform seeding. Seeding maps based on either electrical conductivity or NDVI each resulted in higher gross margins than uniform seeding (Munnaf et al., 2020a). Agriculture is one of the largest users of freshwater; however, water-use efficiency remains low. Precision irrigation helps reduce water use by preventing overwatering and waterlogging (Neupane and Guo, 2019). For site-specific irrigation, crop models can predict water requirements at different growth stages. Using this approach, yields for cotton and ryegrass increased by 4.9 % and 8.5 %, respectively, while water application was reduced by approximately 5.5 % (Mccarthy et al., 2023). 1.6.4 Crop mapping and evaluation Crop mapping and the success of PF can be evaluated using the technology described in section 1.6.1, such as yield maps from combine harvesters (Hagn et al., 2025). Another approach involves estimating crop biomass through remote sensing data obtained from satellites and UAVs, as demonstrated for various crops including wheat (Triticum aestivum; Cheng et al., 2022), potato (Solanum tuberosum) and sugar beet (Beta vulgaris; Vannoppen Scientific publications within the context of this work 30 Table 1: Soil nutrient content. Mg and pH-value estimated with CaCL2-Method. P2O5 and K with Calcium-AcetatLactat-Method. Year P2O5 [mg/100 g soil] K2O [mg/100 g soil] Mg [mg/100 g soil] NO3 + NH4 [kg ha-1] pHvalue Corg [%] 2021 21.8 17.4 4.0 30.0 5.7 1.4 The predomital weed species in both years was Chenopodium album. Stellaria media, Poa trivialis (L.), Polygonum convolvulus (L.) and Galinsoga ciliata (RAF.) were common as well. The species Amaranthus, Equisetum arvense (L.), Spergula arvensis (L.), Capsella bursapastoris (L.), Veronica agrestis (L.), Echinochloa crus-galli (L.), Lamium purpureum (L.) occured only at some places and not at all times. Two different weed control thresholds were compared with a uniform weed management as control treatment (Con). This treatment is equivalent to conventional weeding, where every subplot is treated during all applications regardless of the weed pressure. The second treatment was a decision-making system based on a Weed Cover Threshold (WC) with the thresholds of 0.25 %, 0.5 % and 1.0 % weed cover. If the threshold of a subplot was exceeded, weed management was conducted. A weed cover of 0.06 to 0.31 % is the economic threshold for herbicide application in maize (Longchamps et al., 2014). As mechanical weeding is more costly than herbicide application, a threshold of 0.5 % was chosen, complemented by a more conservative threshold of 0.25 % and a less conservative threshold of 1.0 % weed cover. The third treatment considered for decision-making the Relative Weed Cover (RWC, Ngouajio et al. (1999). The RWC is a dimensionless value and was calculated as follows: 𝑅𝑊𝐶 = 𝑊𝑒𝑒𝑑 𝑐𝑜𝑣𝑒𝑟 (%) 𝑊𝑒𝑒𝑑 𝑐𝑜𝑣𝑒𝑟 (%)+𝐶𝑟𝑜𝑝 𝑐𝑜𝑣𝑒𝑟 (%) Equation 1 The treatment RWC had three thresholds: 0.1, 0.2 and 0.4 RWC. If the threshold of a subplot was exceeded, weed management was conducted. A RWC of 0.2 is equal to yield losses of around 10 % in comparison with a weed free plot (Ali et al., 2015; Ngouajio et al., 1999). In order to be able to evaluate the effects of different tolerance thresholds, a more conservative (RWC = 0.1) and a more tolerant (RWC = 0.4) treatment were also tested here. The experiment was structured as a randomized block design. The seven treatments were applied across four replicates, resulting in a total of 28 plots. Each plot measured 30 m in length and 3 m in width. To mitigate the influence of prior site-specific weeding, the trial’s location within the study site was altered in the subsequent year, and the plots were rerandomized. For site-specific management, each plot was subdivided into three subplots, each extending 10 m in length and maintaining the width of 3 m. Within each subplot, a 1 m² measuring area was designated for the assessment of plant traits. This area was positioned at the centre, spanning the two central rows of the subplot. Each measuring area including two 0.1 m2 areas, one covers the intra-row space, the other the inter-row space as shown in Fig 6. Their location was permanent for each trial year. Scientific publications within the context of this work 31 Fig 6. (Sub)plot arrangement and placement of measuring areas. Selection of each measurement area was meticulously conducted to ensure representation of the entire plant community within each subplot. This approach yielded a total of 168 samples on each designated measurement date. Weather conditions The weather data was recorded by a weather station of the Hochschule Osnabrueck University of Applied Science at 60 m AMSL about 1 km away from the field. Air temperature was measured at a height of 2 m. Fig 7 shows the weekly weather conditions for 2021 and 2022. Fig 7. Mean air temperature [°C] (lines) and sum of precipitation [mm] (bars) for each week of trial period (April – October) in comparison to the long-term period of 1996 – 2022 in grey in the trial years 2021 and 2022. Vertical dotted lines indicate selected management. With an annual rain amount of 646 mm in 2021 and 631 mm in 2022, the precipitation was much lower than the average of 872 mm for the period of 1996 to 2022. In the period of April to October, the year 2021 shows more rain than 2022 with 385 mm in comparison to 324 mm. In the first trial year, 56.8 mm more precipitation occurred in the sowing period (week 14 to Scientific publications within the context of this work 32 22). The mean temperature of 2021 is comparable to the long-term period average. However, with a temperature of 11 °C, 2021 was warmer than 2022 and the long-term average of 10 °C. Trial management To test the site-specific management under conditions of high weed pressure, maize was cultivated as previous crop. Before the pre-crop, five years of Phacelia tanacetifolia was grown. Maize, variety “Severeen” (FAO class 230), was seeded with the sowing machine Pracea (Amazone) and real time kinematic autosteering system. This variety is usable both for forage and grain harvest (BSA, 2021). 8 kernels m-2 were sown at 6 cm depth and 75 cm rowing space. Seeding dates and other management measures is shown in Table 2. Table 2: Detailed field history and trial management for all experimental seasons. Days after sowing = DAS. Growth stages (GS) marked: V0 – sowing, V5 – germination, V13/15/16 – third/fifth/sixth leave, R4 –doughty kernels. Management 2021 2022 Date [MM-DD] DAS GS Date [MM-DD] DAS GS Ploughing 05-10 -4 03-27 -40 Seedbed preparation 05-14 0 04-19 -17 Sowing 05-14 0 V0 05-06 0 V0 Pre-emergence uniform weeding 05-20 6 V5 05-11 5 V5 First site-specific mechanical weeding 06-09 26 V13 05-31 25 V13 Second site-specific mechanical weeding 06-25 42 V15 06-15 40 V16 Harvest 06-10 144 R4 09-14 131 R4 In 2022, 10 t ha-1 of dry chicken manure were applied before seeding. Weed detection UAV-based multispectral cameras were used for weed and maize plant detection. Different camera systems and flight altitudes were tested for image data acquisition to develop a flexible and system-independent concept for spatial differentiation of maize and weeds. A MicaSense Altum camera mounted on a DJI Matrice 210 as well as a DJI P4 Multispectral with a permanently installed multispectral camera were used to acquire UAV-based multispectral images. The MicaSense Altum records the surface reflectance in the visible (475 ± 32, 560 ± 27 and 668 ± 14 nm) and near infrared (717 ± 12 and 840 ± 57 nm) wavelengths. Spectral information is recorded in comparable wavelengths by the P4 multispectral camera (visible: 450 ± 16, 560 ± 16 and 650 ± 16 nm; near infrared: 730 ± 16 and 840 ± 26 nm). For observation dates in 2021, a flight altitude of 10 m was defined which led to a spatial resolution of 0.4 cm. In the following year, the flight altitude was increased to 25 m to be able to simulate larger area outputs and faster image acquisition, regarding a practice-oriented application context. The resulting images were processed using Agisoft Metashape (Version 1.7.2). Various weather conditions during data acquisition resulted in differences in exposure, which were corrected using an illumination sensor on the drones. Afterwards individual images were aligned to an orthophoto. For geometric correction of the image data eight reference panels distributed over the test area were measured as ground control points using a RTK GNSS receiver (Stonex S9III). The orthophotos of each recording date were classified using machine learning methods, to determine weed infestation in maize. The used methods and systems for both years are shown in Table 3. Scientific publications within the context of this work 33 Table 3: Parameters of the weed recognition for the different sampling dates. Date Flight altitude Pixel resolution Camera system Machine learning methods [YY-MM-DD] [m] [cm] 2021-06-03 10 0.4 MicaSense Altum Convolutional neural network 2021-06-17 10 0.4 MicaSense Altum Convolutional neural network 2022-05-19 25 1.0 MicaSense Altum Random forest 2022-06-10 25 1.0 DJI P4 multispectral Random forest In the first trial year a simple 2D convolutional neural network was implemented, this algorithm is inspired by the biological brain and uses multiple convolutional layers to extract features, such as texture, from images to identify objects (Zhu et al., 2017). The CNN architecture consisted of three convolutional layers, each employing a 3 x 3 pixel kernel to capture important image features. As input, the model received image patches measuring 32 x 32 pixels, extracted from the field trial plots showcasing differences in weed pressure. The training of the CNN encompassed 25 epochs and was executed using the Adam optimizer, a stochastic gradient descent variant. The training process featured a batch size of 16 and initiated with an initial learning rate set to 0.0001. The second method is RF classifier, which is based on a number of uncorrelated decision trees arrange in an ensemble. Within this ensemble, a classification is carried out by a majority decision on all decision trees (Breiman, 2001). The RF implementation was done using the scikit-learn software library (Pedregosa et al., 2011). The default settings were maintained for the classifier parameters, as the RF has been shown to achieve good results with them (Fernández-Delgado et al., 2014). The spectral information of trial plots was used for the classification. In contrast to the CNN, the RF does not consider texture features. For image pixel classification, distinction was made between the classes “maize”, “weeds” and “soil”. The resulting classification maps should enable the localization and quantification of weed pressure. The sampling of training and validation data was spatially independent of each other. To compare classification results of the considered dates for each of the classes, precision and recall, as well as the resulting F1-score and overall accuracy were calculated. The sampling of the validation data was randomly drafted from the recorded images. 𝑃 𝑟=𝑇𝑃 𝑇𝑃+𝐹𝑃 Equation 2 𝑅𝑒=𝑇𝑃 𝑇𝑃+𝐹𝑁 Equation 3 𝐹1 = 2∗ 𝑃𝑟∗ 𝑃𝑒 𝑃𝑟+𝑃𝑒 Equation 4 𝑂𝐴 = 𝑇𝑃+𝑇𝑁 𝑇𝑃+𝑇𝑁+𝐹𝑃+𝐹𝑃 Equation 5 Pr represents the precision, Re the recall, F1 the F1-score and OA the overall accuracy. TP (true positive) indicates the number of positive samples correctly classified and TN (true negative) the number of negative samples correctly classified. The number of incorrect classified positive or negative samples is represent as FP (false positive) and FN (false negative) respectively. Based on the classification maps, the number of pixels of each class were calculated for each subplot with the function “Zonal statistic” (Quantum GIS, Version 3.22). Subsequently WC and RWC were calculated for each subplot with the “Field Calculator” function of Quantum GIS using the Equation 6 – Equation 8. Scientific publications within the context of this work 34 𝑀𝐶 =∑ 𝑃𝑀 ∑𝑃𝑆+𝑃𝑊+𝑃𝑀 ∗100 Equation 6 𝑊𝐶 =∑𝑃𝑊 ∑𝑃𝑆+𝑃𝑊+𝑃𝑀 ∗100 Equation 7 𝑅𝑊𝐶 = 𝑊𝐶 𝑊𝐶+𝑀𝐶 Equation 8 where MC represents the maize cover, WC the weed cover, RWC the relative weed cover. PS, PM and PW correspond to the classified pixels of the classes Soil, Maize or Weed respectively. The WC and RWC were compared to the threshold of the corresponding treatment level for each subplot. If the value was higher than the threshold for a subplot, weed regulation was done as described in section Weed management, if not, no harrowing was conducted for this subplot and timestamp. Due to the time required to processing and analysing the UAV images, as well as varying weather conditions, weed regulation was applied seven to ten days after the flights. The classified maps were further used to show the weed distribution within the field. Weed management Between seeding and maize seed emerging, all plots were treated with a precision tine harrow (Treffler) at a speed of 10 to 12 km h-1, as common in organic practice to reduce the competition of weeds in the early growing period. This treatment was uniform for each variant as every subplot was managed. Two site-specific weed regulations with a hoe followed, matching the GS V13 and V16. The selection of the areas to be treated was based on weed detection. Subplots were hoed only if their weed cover per subplot exceeded the specific threshold for each treatment. An exception was the Con-treatment, where every plot was treated uniformly. Hoeing was done with a Kombi-PP (Schmotzer) in 2021 and a Chopstar (Einböck) in 2022. The change of equipment was done due to availability at the research station. Both machines were equipped with duck foot sweeps, finger weeder and automatic camera steering system. The devices were set up in the same way. Driving speed was 5 to 8 km h-1. With this equal equipment, comparable weeding success were archived with both machines. No chemical plant protection was applied to simulate organic farming conditions. Soil and plant analyses Visual assessments were performed five to seven days before and after each weed management intervention to estimate weed coverage and species diversity. Concurrently, the mean plant height within each subplot was ascertained by averaging the heights of four representative maize plants. Soil volumetric water content was gauged at a depth of 12 cm using a TDR-150 (Field Scout) device. For each measurement area, two readings were taken, one in the intra-row and one in the inter-row space, to calculate the average value. These soil moisture readings were taken alongside other measurements to characterize plant water availability and field heterogeneity. Additionally, the phenological stage of the maize was recorded. Maize yield was approximated by manually harvesting a 1 m² area at a height of approximately 15 cm above the ground within each subplot. The count of maize cobs was recorded. The harvested plants were then partitioned into cob and leaf-stem fractions and processed using a Schliesing 220 ZX wood chopper. The fresh weight of each fraction was measured, followed by a drying process in an oven at 105 °C. The dry weight was determined Scientific publications within the context of this work 35 post a 48-hour period. Weed biomass was treated similarly, being weighed fresh and subsequently dried. Calculations & statistics The first step was the calculation of arithmetic means for each plot for the parameter (volumetric water content, maize pant height, weed cover, number of weed species, treated area, whole plant yield, cob yield, number of cobs and weed biomass at harvest). In case of maize yield, cob yield and weed biomass, the dry matter was used. All parameters (maize yield, cob yield, weed biomass, weed cover, maize height, treated area, number of weed species, volumetric soil water content) were tested for normal distribution with the ShapiroWilk test (α = 0.05; Royston, 1995) and for homogeneity of variance with Levene’s test (α = 0.05; Fox, 2008). All parameters showed a homogeneity of variance. An analysis of variance (ANOVA; Chambers et al., 1992) followed with an alpha of 0.05. If the alternative hypothesis was accepted, a subsequent Tukey HSD-post hoc tests (α = 0.05; Holland, 1988) followed. For the variables maize yield, cob yield and weed biomass the fixed factors were the treatment and year, block and plot were random factors. For the variable maize height, number of weed species, volumetric soil water content the fixed factors were the treatment and date, block and plot were random factors. Coefficient of determination (R²) and a linear model are computed between volumetric soil water content and weed cover. All statistical analyses were carried out in R (Version 4.2.2; R Core Team, 2020) using the packages “nlme”, “emmeans”, “multcomp”, “rstatix” and “multcomp”. Results Plant development The volumetric soil water content (VSWC) was with 11.8 % (± 3.57 (standard deviation)) in 2021, overall higher than 7.3 % (± 4.90) in 2022. Between the treatments no differences occurred but the timestamps showed alterations (ANOVA, F-value = 6.38E+01, p-value = 8.55E-14). No interaction between year and treatment was observed. After the first site-specific weeding at DAS 26/25 to 30, the VSWC was higher in 2021 than in 2022 as shown in Table 4. In the time period after the second site-specific hoeing (DAS 40 – 50), the soil moisture conditions were wetter in 2021 than in 2022. Table 4: Comparison of volumetric water content [%] between timestamp. DAS = Days after sowing. SD = Standard deviation. Different letters indicate significant (Tukey HSD-post hoc test, alpha = 0.05) differences between timestamps. N = 28. Year DAS Volumetric soil water content Mean [%] SD Group 2021 30 8.65 1.12 B 2021 44 9.99 2.52 B 2021 46 9.85 2.52 B 2021 60 11.73 1.89 C 2022 34 11.56 3.01 C 2022 47 3.04 1.46 A The maize plant height developed faster in 2021 than in 2022, as shown in Table 5. In the first trial year the plant height was 11 cm larger at DAS 34 than in the second year and 64 cm larger at DAS 48. Between the treatments, no significant differences were observed. In 2021, the weed growth was slow and only increased significantly 49 days after sowing (Table 5). The hoeing at DAS 26 and 42 did not decrease the weed cover, but the growth Scientific publications within the context of this work 36 stagnated. The year 2022 showed a strong growth of weed between DAS 17 and 24. Before the first hoeing at DAS 25, the weed cover was higher than in 2021. After the weeding it decreased significantly till DAS 34. The weed cover increased by 5 % points from DAS 34 to 47, but without significant difference. The single treatments did not differ in weed cover. Table 5: Comparison of maize plant height [cm], weed cover [%] and number of weed species between time stamps. DAS = Days after sowing. SD = Standard deviation. Different lowercase letters indicate significant (Tukey HSDpost hoc test, alpha = 0.05) differences between timestamps. N = 28. Year DAS Maize plant height Weed cover Number of weed species Mean [cm] SD Group Mean [%] SD Group Mean SD Group 2021 20 11.63 0.67 A 2.59 1.46 A 3.66 0.56 E 2021 34 40.90 3.20 C 3.50 2.94 A 3.21 0.98 CDE 2021 49 114.26 5.88 E 11.41 10.13 BC 2.41 1.12 B 2021 60 192.45 7.92 F 14.00 11.05 CD 3.00 0.92 BCD 2022 17 12.57 1.23 A 4.80 3.52 AB 1.54 0.41 A 2022 24 12.57 1.23 A 19.89 14.78 D 3.51 0.61 DE 2022 34 29.28 3.73 B 10.44 4.10 BC 2.64 0.75 BC 2022 47 63.67 7.38 D 15.41 6.28 CD 1.49 0.37 A In both years the mean number of weed species Table 5 decreased after the first and second site-specific weeding (ANOVA, F-Value = 3.87E+01, p-value = 2.65E-09, Table 5). In 2021 the mean weed species count declined from 3.5 to 3.0 species, whereas in 2022, a decrease from 3.5 over 2.7 to 1.7 was observed. The predominant weed species across the years was Chenopodium album (L.) with 196 (2021) to 576 (2022) counts over all plots and timestamps. In addition, Stellaria media (L.) (150/144 counts) and Poa trivialis (141/113 counts) were present in large numbers. Polygonum convolvulus (L.) occurred manly in 2022 (138 in 2022 and only 44 in 2021). Galinsoga ciliate (RAF.) were found 43 to 70 times in 2021 and 2022 respectively. Few individuals of Amarathus (two times in both years), Capsella bursa-pastoris (L.) (only in 2021, six times), Echinochloa crus-galli (L.) (two and 15 times), Equisetum arvense (L.) (two times in 2022), Lamium purpureum (L.) (eleven and 17 times), Spergula arvensis (L.) (three times in 2021) and Veronica agrestis (L.) were found at four to seven spots per year. In 2021, it was observed that weed cover increased with higher volumetric soil water content, as indicated by a positive slope of 3.06 (Fig 8). The regression analysis yielded a significant result with an R² of 0.526 (p-value = 2.2E-16). However, this trend was not found in 2022. In that year, the relationship between weed cover and soil water content remained significant (p-value = 7.28E-3), but the strength of association was relatively weak (R² = 0.118). Unlike previous years, the relationship was negative, with a slope of -0.405. Notably, the regression analysis for both years remained significant (R² = 0.034, p-value = 1.99E-2). Overall, the slope of 0.421 suggests that weed cover tends to increase with rising soil water content. Scientific publications within the context of this work 37 Fig 8. Linear regression between volumetric soil water content [%] and weed cover [%]. Lines are the regression line with confidence interval of 95 %. Colour of points indicates the years. N = 170, n 2021 = 110, n 2022 = 58. During the validation of the resulting classification maps, an overall accuracy of at least 85.1 % was achieved across all timestamps (Table 6). In addition, F1-scores of at least 81.6 % and 82.3 % were achieved for the target classes “maize” and “weeds”. Both machine learning algorithms showed similar accuracy, with CNN achieving an OA of 85.1 % and 86.7 % and RF obtaining an OA of 88.8 % and 92.0 %. Neither the flight altitude nor the camera system affected the results. Table 6: Accuracies for UAV-based image classification of weeds, maize and soil. OA = Overall accuracy. Timestamp of acquisition Camera system F1-score [%] OA [%] Machine learning methods Maize Weeds Soil 2021-06-03 Micasense 81.6 86.1 87.9 85.1 Convolutional neural network 2021-06-17 Micasense 85.2 87.0 87.9 86.7 Convolutional neural network 2022-05-19 Micasense 85.6 82.3 98.3 88.8 Random forest 2022-06-10 Phantom MS 88.4 88.3 99.3 92.0 Random forest Fig 9 depicts the weed cover distribution derived from image classification before the first site-specific weeding intervention in each year. The mean weed cover was 2.24 % (± 1.60) in 2021 and 1.73 % (± 0.86) in 2022. The western section of the field was covered consistently higher with weed (2021: 2.7 %; 2022: 2.2 %) than the eastern section (2021: 1.58 %; 2022: 1.47 %). The western section also had the part with the highest weed cover with 11 % (2021) and 7.35 % (2022), as well as higher standard deviations of 1.78 and 0.97, versus 0.99 and 0.62 in the eastern section. This indicates the presence of some weed clusters with higher intensity in the western section, which were more dispersed in 2022 than in 2021. Before the second hoeing the situation was comparable but on a higher level. The western part showed higher weed cover with 3.43 % (± 3.1) versus 1.51 % (± 2.23) and 13.6 % (± 7.95) versus 8.09 % (± 9.82) in 2021 and 2022 respectively. Scientific publications within the context of this work 38 Fig 9. Weed cover [%] estimated by image recognition before first site-specific weed management in both trial years. Darker colours indicate higher weed cover. The whole field is displayed. In 2021, at the northern part of the field, a cover crop mixture was sown, in this part no weed detection was conducted and therefore not displayed. The position of the trial field varies in each year to reduce the effect of SSWM on the following year. As all threshold values were exceeded in both years before the first hoeing, all subplots were initially treated equally. For the second hoeing, significant differences between the variants were observed. In 2021, RWC0.4 showed significant less treated area than Con, WC0.25, WC0.5 and WC1.0 (Fig 10), none of the plots of RWC0.4 were hoed. RWC0.1, RWC0.2, WC0.5 and WC1.0 took an intermediate position. In 2022, the WC-Treatments were weeded completely, as well as Con and RWC0.1. Only RWC0.2 showed some spared plots, but without significant differences to the control treatment. Less plots of RWC0.4 were hoed than the plots of the other treatments, expect RWC0.2 in 2022. Overall, with RWC0.4 less area was treated. In average, more plots were hoed (ANOVA, F-Value = 2.15E+01, p-value = 3.79E-05) in 2022 than in 2021. Scientific publications within the context of this work 39 Fig 10. Comparison of treated area [m²] due to second hoeing between treatment (as shown as colours): Control (Con), Relative Weed Cover (RWC), Weed Cover (WC) and trial years. Whiskers show the minimum/maximum values. Bars indicate the interquartile range and columns the sum. Line inside the box symbolises the median. Single values are shown as points. Lowercase letters indicate significant (Tukey HSD-post hoc test, alpha = 0.05) differences between treatments and years. N=4. Maize yield and weed biomass at harvest The taller growth in 2021 led to 3.5 times higher maize yield in 2021 than in 2022 with significant differences in yield (ANOVA, F-value = 1.02E+03, p-value = 4.02E-30) (Fig 11). Fig 11. Comparison of maize dry matter yield [g/m-2] at top and weed biomass [g m-2] at bottom at harvest date between treatment (as shown as colours): Control (Con), Relative Weed Cover (RWC), Weed Cover (WC) and the trial years. Whiskers show the minimum/maximum values. Bars indicate the interquartile range. Line inside the box symbolises the median. Single values are shown as points. Letters indicate significant (Tukey HSD-post hoc test, alpha=0.05) differences between treatments and years, lowercase letters: maize dry matter yield, uppercase letter: weed biomass. N=4. Scientific publications within the context of this work 46 Introduction Modern agriculture faces the challenge of feeding a world population of about 10 billion people by 2050 and maintaining food security (Willett et al., 2019). At the same time, the intensive food production has significant negative effects on the environment, including a decline in plant and animal species due to land use changes and intensive use of nitrogen (N) fertilizers and pesticides (Geiger et al., 2010; Willett et al., 2019). Organic agriculture can enhance wildlife by promoting soil fertility and reducing the use of N fertilizers and pesticides. Organic farming, however, typically achieves only 80 % to 83 % of the yields of conventional farming, with a wide range of 40 – 130 % (de la Cruz et al., 2023; De Ponti et al., 2012). This yield gap is mainly due to lower inputs of N fertilizers (Döring and Neuhoff, 2021). In organic farming, N fertilizer sources are limited to organic options such as biological N fixation, green manure, compost, etc. The limited supply necessitates higher N use efficiency and better distribution within the field and over time. Most agricultural land is characterized by heterogeneous soil conditions, including soil texture, organic matter, relief, nutrient, and water supply. Consequently, plant development is influenced by spatial differences (Usowicz and Lipiec, 2017). While soil texture and soil organic matter (SOM) are relatively stable over time (Behera et al., 2018), nutrient and water supply can vary greatly within years (Aksakal et al., 2019; Zhu et al., 2013). Additionally, land use history can influence current crop development (Schulp and Verburg, 2009). The main N source in organic agriculture is the symbiotic N fixation by legumes, mainly clover-grass (Oberson et al., 2024). The amount of fixed N depends strongly on biomass production and varies with the weather conditions (Hoekstra et al., 2015). Thus, the development of clover affects the N supply for current and subsequent crops. Subfield adapted management can address issues of heterogeneous plant development. Fields can be delineated into management zones (MZ), where each subfield has similar plant and/or soil properties (Diacono et al., 2014; Peralta et al., 2015). This division in subfields can be done using various algorithms, with the fuzzy C-means clustering algorithm being widely used recently (Damian et al., 2020; Oldoni et al., 2025; Rodriguez Miranda et al., 2021). In conventional agriculture, spatial differences can be mitigated with fast-dissolving artificial fertilizers, but in organic agriculture, the legal fertilizer like green manure or cow manure work slowly. Therefore, field differences must be balanced in advance with organic fertilizers (Corti et al., 2023). Another approach is to adjust seeding density or use crop mixtures to even out spatial heterogeneity (Munnaf et al., 2022). There is a discussion about whether lowerperforming areas should receive resources like seeds or fertilizer, the "Robin Hood" approach, or if high-potential areas should receive more resources to maximize their potential the so called "King" approach. Typically, heterogeneous conditions are estimated by soil sampling, but this can be uneconomical due to labour and analysis costs (Buladaco II et al., 2024; Ouazaa et al., 2022). As an alternative, remote sensors can be employed to identify spatial differences within plant communities (Breunig et al., 2020) and can substitute soil sampling (Speranza et al., 2023), offering a less labour-intensive solution. Vegetation Indices (VI), such as the Normalized Difference Red Edge Index (NDRE), are particularly effective for this purpose and are widely utilized (Damian et al., 2020; Girz and Mattila, 2024; Leo et al., 2023). The NDRE considers the reflectance information from NIR and RedEdge to highlight differences in chlorophyll concentration within plant stands (Raymond Hunt et al., 2011). Unmanned aerial vehicles (UAVs) play a crucial role in this process, as they can swiftly and flexibly capture images Scientific publications within the context of this work 47 needed for NDRE calculation, making it a practical substitute for traditional soil sampling methods (Rasmussen et al. 2021). Although UAV images have been successfully used to delineate fields of cover and cereal crops, only a few studies have analysed management zones for crop rotation (Oldoni et al., 2025; Ouazaa et al., 2022). To our knowledge, none have conducted trials with clover-grass followed by a subsequent cereal crop under organic farming conditions. Consequently, this study aims to use UAV-based NDRE maps of clover-grass fields to delineate management zones and evaluate their impact on subsequent crop productivity. The findings might provide insights into optimizing N use efficiency in organic farming systems. This study aligns with the United Nations' Sustainable Development Goals (SDGs), particularly Goal 2: Zero Hunger, and Goal 15: Life on Land. By promoting sustainable agricultural practices, we aim to contribute to these global objectives. The following hypotheses were evaluated: • Hypothesis I: High productivity zones during the clover-grass period are also highly productive for the subsequent crop. • Hypothesis II: NDRE images are useful for delineating management zones in clovergrass fields. • Hypothesis III: Higher clover yields lead to higher fixed N and, therefore, higher soil N content (SNC) during both the clover-grass period and the subsequent crop. To test these hypotheses, three trials were conducted in northwestern Germany under organic on-farming conditions. Scientific publications within the context of this work 48 Materials & Methods Trial design and study site The study was conducted near Osnabrück in Lower Saxony, in the northwest of Germany, from 2020 to 2023. Three organically managed fields were selected, each approximately 1 hectare in size. The distances between them ranged between 500 and 4,100 m (Fig 12). Fig 12. Geographical Distribution of Experimental Sites in Northwestern Germany near Osnabrück (Background Source: Google Maps, 2024) The soil value is a German indicator of yield potential, with 0 indicating the lowest and 100 the highest fertility (Arbeitsgruppe Boden (Soil workingGroup) 2024 Table 7). Table 7: Field description. Soil types and soil values are based on NIBIS® Map Server (2023). The Soil value is a German indicator yield potential, there 0 indicates the lowest and 100 the highest fertility (Arbeitsgruppe Boden (Soil working Group), 2024). Data source of soil value and soil type: (LBEG, 2023). AMSL = above mean sea level. Field name Year Geographical coordinates Soil type Soil texture AMSL [m] Soil value PowerWeg 2020, 2021 52°18'59.3"N, 8°06'18.7"E Cambisol Loamy sand / sandy loam 113 – 115 52 – 55 Kiesschacht 2021, 2022 52°19'25.8"N, 8°09'28.9"E Cambisol Loamy sand / sandy loam 88 – 92 30 – 40 BremerStraße 2022, 2023 52°19'34.3"N, 8°09'46.7"E Regosol Loamy silt / sandy loam 121 – 123 42 – 62 Power Weg has a low slope gradient and is nearly flat with an above mean sea level (AMSL) of 113 to 115 m (Fig 13). The Kiesschacht field has the steepest relief, with elevations ranging from 88 to 92 m AMSL. Especially the western part has a steep slope. BremerStraße field has differences in slope comparable to the PowerWeg. Scientific publications within the context of this work 49 Fig 13. Terrain based on digital elevation model of the trial fields. AMSL (above mean sea level) [m] is indicated by colour: blue, lower values; yellow, higher values. Rectangles show the plots. Each map has a separate scale for easier identification of differences. The trial was conducted as an on-farm experiment, where the primary management decisions (time of harvest, crop choice, etc.) were made by the farmer. All fields have been managed organically since 1995. Each field was observed during two seasons: the clovergrass period and the subsequent cereal crop season. The seedbed for clover-grass was prepared using a plough and disk harrow after the harvest of the pre-crop in summer. A seed mixture of 32 kg ha−1 was sown, consisting of the following percentages: 11.11 % red clover (Trifolium pratense) 'Lucrum bio', 11.11 % red clover 'Titus bio', 11.11 % alsike clover (Trifolium hybridum L.) 'Aurora', 11.1 1 % white clover (Trifolium repens) 'Liflex', 7.78 % perennial ryegrass (Lolium perenne) 'Melfrost', and 47.78 % perennial ryegrass 'Valerio bio'. The clovergrass sward was harvested two to three times per year, with all biomasses removed from the field. The subsequent crop was a cereal. PowerWeg and Kiesschacht were tilled with a plough and disk harrow. Summer spelt (Triticum aestivum subsp. Spelta) 'Flauder' (180 kg ha−1) was sown in March 2020 and 2021. BremerStraße was tilled in the same way in October 2022, followed by the sowing of 195 kg ha−1 of winter wheat (Triticum aestivum L.) 'Grannosos'. Torsion harrowing was applied at PowerWeg at growth stage (GS) 15. On the field PowerWeg 10 Mg ha-1 of cow manure was applied and ploughing of clover-grass was performed. At BremerStraße, 13 Mg ha-1 of cow manure was applied during GS 37. A detailed list of the management practices can be found in Table 8. Table 8: Field management and soil sampling dates. Date format YYYY-MM-DD. Field name Clover-grass period Cereal period Tillage Seeding Harvest Tillage Seeding Harvest PowerWeg 2019-08-01 2019-08-01 2020-04-26 2020-07-11 2020-09-20 2021-03-24 2021-03-24 2021-08-11 Kiesschacht 2020-08-20 2020-08-25 2021-05-14 2021-07-16 2022-03-09 2022-03-14 2022-07-25 BremerStraße 2021-07-27 2021-08-01 2022-06-01 2022-08-03 2022-10-13 2022-10-17 2023-06-18 At the PowerWeg field, the predominant weed species during the clover-grass period were Papaver rhoeas (common poppy), Centaurea cyanus (cornflower), and Matricaria chamomilla (chamomile). During the summer spelt cropping, additional weed species such as Stellaria Scientific publications within the context of this work 50 media (chickweed), Trifolium spp. (clover), and Convolvulus arvensis (field bindweed) were present. One application of tine harrow was conducted at GS 59. In the Kiesschacht field, the most prevalent weeds were Lamium purpureum (red deadnettle), Centaurea cyanus (cornflower), Vicia spp. (vetch), and Rumex spp. (dock). During the subsequent cropping, Vicia spp. (vetch), Centaurea cyanus (cornflower), Papaver rhoeas (common poppy), Lolium ssp. (grass), and Stellaria media (chickweed) were found. In the clover-grass field BremerStraße, species such as Cirsium spp. (thistle), Lamium purpureum (red deadnettle), Brassica napus (rapeseed), Rumex spp. (dock), Veronica spp. (speedwell), Alopecurus spp. (foxtail), and Matricaria chamomilla (chamomile) occurred. During the winter wheat period, the field mostly featured species such as Trifolium spp. (clover), Veronica spp. (speedwell), Stellaria media (chickweed), and Galium aparine (cleavers). Plant and soil sampling On each field, 48 plots were established in a grid, with each plot being a square with a side length of 4 meters (Fig 13). A RTK GNSS error caused estimation-based placement of plots at Kiesschacht field, with exact positions recorded later. Within each plot, a sampling area of 0.25 m² was designated for clover-grass, and 1 m² was designated for the subsequent cereal crop. The position of the clover-grass sampling area was slightly shifted after each sampling to ensure an unharvested sampling area before each observation. In contrast, the cereal measurements area remained constant for the entire season, as the sampling was nondestructive with the exception of the harvest. The GPS position for each sampling area was recorded using the RTK GNSS receiver Helix M7 (Helix Geospace, United Kingdom). Right before each clover-grass harvest and every ten to 14 days afterwards, each sampling area was observed. The timestamp immediately preceding the harvest will be referred to as “harvest”, while all other timestamps will be designated as “sampling dates”. Plant height was calculated as the average of four measurements with a folding rule. Volumetric soil water content (VSWC) was measured at two points using a Field Scout TDR150 (SPECTRUM TECHNOLOGIES, USA) at a depth of 120 mm. Due to lack of access of measuring device, this was only used in trials starting in 2021. Visual assessments were conducted to estimate the yield proportion (Klapp and Stählin, 1936) and GS (Moore et al., 1991) of grass and clover species. Additionally, the sward was cut and separated into the fractions "Grass", "Clover", and "Weeds". For each sampling area, the total dry biomass and the dry mass of the fractions were determined by drying at 85 °C for 48 hours. During the cereal period, at relevant GS (determined according to BBCH (Maier, 2018)), plant height was determined for each sampling area using the same procedure as for clovergrass. At harvest time, the length of ears was measured in the same manner. The number of plants per meter was counted and then converted to a per-square-meter basis by multiplying the count by seven. VSWC was measured with in the same way as during the clover-grass period. Additionally, weed species and weed cover were estimated by visual assessment for an area of 0.1 m², as visual assessments are more precise for smaller areas. Yield was determined by hand-harvesting of all plants. Subsequently, cereals and weeds were separated, the fresh weight of each fraction was measured, then dried in an oven at 85 °C for 48 hours, after which the dry weight was determined. Kernels were separated from straw using a HALDRUP LT-35 laboratory thresher (Haldrup, Germany). A Contador seed counter (Pfeuffer, Germany) was used to determine thousand-kernel weight (TKW). Kernel and straw samples were then milled to 0.5 mm with a ZM 200 (Retsch, Germany), followed by drying for 48 hours at 105 °C. N estimation was performed with the Elemental Analysis Leco FP 628 (Leco, Netherlands). Finally, protein content was calculated by multiplying the N content by Scientific publications within the context of this work 51 6.25, based on the assumption that protein contains an average of 16 % N (ISO 16634-2, 2016). During the clover-grass and cereal periods, soil sampling was conducted at relevant timestamps; at least one sampling before tillage of clover-grass and during the vegetative growing of cereal. Three soil samples within each plot were collected from depths of 0 to 0.3, 0.3 to 0.6, and 0.6 to 0.9 m. Due to severe dry soil conditions, it was not always possible to sample the deeper layers. Each layer was homogenized and cooled until analysis. A commercial laboratory analysed the samples for soil mineral N (NH4+ and NO3−) based on VDLUFA I, A 6.1.4.1:2002 (VDLUFA, 2007). Soil texture and clay content were estimated by the same laboratory using VDLUFA I, D 2.1:1997 (VDLUFA, 2007). The exact days of data sampling and UAV campaigns can be found in Table S 1 (Supplementary Material). Weather conditions Weather data was recorded by a Germany's National Meteorological Service (Deutscher Wetterdienst) weather station at 103 meters above mean sea level, located approximately 1.0 to 4.3 km away from the fields (52°19'01.2"N, 8°10'09.8"E). Air temperature was measured at a height of two meters. Fig 14 illustrates the weekly weather conditions for the trial years in comparison to the long term period. Fig 14. Mean air temperature [°C] (lines) and sum of precipitation [mm] (bars) for each week of trial period in comparison to the long-term period of 2011 – 2019 in grey. Data source: (DWD, 2024). During the growing period from April to September, the year 2020 had an average temperature of 15.5 °C, which was 0.3 °C warmer than the long-term average of 15.2 °C. In 2021, the temperature was 14.3 °C, making it 1 °C cooler than the long-term average. The year 2022 experienced an average temperature of 15.7 °C, which was 0.5 °C warmer than the long-term average. The final trial period was comparable to the average, with a temperature of 15.1 °C. Peak temperatures were observed in all years, particularly during the summer months. The long-term average precipitation during the growing period was 461 mm. The years 2020 and 2022 experienced water deficits, with precipitation levels of 313 mm and 408 mm, respectively. Conversely, precipitation in 2021 and 2023 was higher than the long-term average, at 575 mm and 654 mm, respectively. Nevertheless, periods of severe drought occurred in all years. Scientific publications within the context of this work 52 Image acquisition and calculation of vegetation indices Parallel to the sampling in clover-grass, multispectral images were acquired via UAV. In 2020, an Altum camera (MicaSense, USA) mounted on a Matrice 210 (DJI, China) was used in flight, with attitude between 25 m and 46 m. In 2021 and 2022, a DJI P4 Multispectral (DJI, China) with a permanently-installed multispectral camera was used in a flight with an attitude of 10 m to 14 m. The MicaSense Altum captures surface reflectance in the visible wavelength range (475 nm ± 32 nm, 560 nm ± 27 nm, and 668 nm ± 14 nm) as well as in the near-infrared (717 nm ± 12 nm and 840 nm ± 57 nm). Similarly, the P4 multispectral camera records spectral data in comparable wavelengths (visible: 450 nm ± 16 nm, 560 nm ± 16 nm, and 650 nm ± 16 nm; near-infrared: 730 nm ± 16 nm and 840 nm ± 26 nm). The varying weather conditions during data collection led to differences in exposure, which were adjusted using an illumination sensor mounted on both UAVs. Subsequently, the individual images were aligned to an orthophoto. To correct the geometry of the image data, ten reference panels scattered across the test area were used as ground control points and measured with an RTK GNSS receiver Stonex S9III (Stonex, Germany). For this, the software Agisoft Metashape (Version 1.7.2) was used. The images were post-processed using Quantum GIS (Version 3.34). Based on the orthophotos, NDRE images (Tucker, 1979) were calculated using Equation 9 with the raster calculator function. 𝑁𝐷𝑅𝐸 = 𝑁𝐼𝑅−𝑅𝑒𝑑𝐸𝑑𝑔𝑒 𝑁𝐼𝑅+𝑅𝑒𝑑𝐸𝑑𝑔𝑒 Equation 9 In the NDRE calculation, NIR represents the reflectance in the near-infrared (840 nm), and RedEdge denotes to the reflectance in the red edge region (720 nm). All NDRE images were resampled to a pixel resolution of 1 meter using the r.resample function. This ensures uniform resolution, matching the lowest resolution of images collected at a flight altitude of 46 meters. A 1-meter pixel resolution requires less storage space and flight time, making it more practical. Additionally, the images were cropped to the field extent using the mask layer function to minimize the influence of borders, trails, and trees. Delineation of management zones The delineation of management zones was conducted using a spatial generalized fuzzy C-means algorithm within the R software environment (Version 4.2.2; R Core Team 2020). This analysis utilized the "geocmeans," "ggubr," "future," "tmap," and "terra" packages. This approach is an unsupervised machine learning algorithm that groups data points based on shared attributes. All points within a cluster are similar to each other, whereas the clusters themselves are distinct. Unlike K-means clustering, fuzzy C-means employs soft clustering, involving probabilistic cluster memberships. This method can effectively handle complex and overlapping clusters. The spatial generalized version of the algorithm converges more rapidly and is less sensitive to noise due to its incorporation of spatial information (Zhao et al. 2013). For each field, two timestamps of NDRE images were selected. These timestamps represent a compromise among timing just before the harvest date, image quality, and the combination of vegetative and reproductive stages of grass and clover. Two timestamps were chosen to balance temporal heterogeneity without the risk of redundancy. For PowerWeg, the chosen dates were June 12, 2020 (29 days before the second harvest, all species in reproductive stages) and September 10, 2020 (directly before the third harvest, grass in the vegetative stage, clover species in the reproductive stage). Data of the first harvest were not Scientific publications within the context of this work 53 collected, due to time delay in trial start. For Kiesschacht, May 14, 2021 was selected as it was directly before the first harvest, with all species except white clover in the reproductive stage. Additionally, August 26, 2021 was chosen: 11 days before the second harvest, with grass and white clover in the vegetative stage, and red clover and alsike clover in the flowering stage. On BremerStraße, the timestamps 29 days after the first harvest were selected (June 16, 2022), with red clover flowering and the other species in the vegetative stage. The second date was July 28, 2022, directly before the second harvest, where the grass plants were in elongation, red clover in the vegetative stage, and both alsike clover and white clover were flowering. The number of clusters (k) was estimated using the elbow method with the CalinskiHarabasz index, partition entropy, and the silhouette index for fuzzy C-means. CalinskiHarabasz index and silhouette index measure the cohesion of cluster, where higher values indicate dense clusters with better differentiation to other clusters. Partition entropy indicates the organisation or uncertainty, where higher values are a sign of higher organisation, and lower values represent a fuzzier result (Brito Da Silva et al., 2020). A significant difference between k = 2 and k = 3 was observed for all fields, with a flatter slope for higher values of k (Fig S 1, Supplementary Material); thus, k = 2 was chosen for all fields. Additional, k = 3 was tested, to evaluate the next higher number of clusters. Other parameters - m (1.5), alpha (1.3), beta (0.5), and window size (5 x 5) - were determined by computing the silhouette index and explained inertia for the combinations of m (1 – 2), alpha (0.5 – 2), beta (0.1 – 1) and window size (3 x 3, 5 x 5, 7 x 7). Additionally, at each field the clustering was conducted for each timestamp separately, for observing temporal difference within the clover-grass zone. The cluster assignment for each plot was extracted from the cluster image by majority decision and will be referred to as "Clover grass zone" (CGZ) from now on, where CGZ 1 is associated with lower and CGZ 2 with higher yield potential. In the case of three CGZs, the low-yield cluster is called CGZ 1, the high-yield cluster CGZ 3, while CGZ 2 has a position in between both. The term “Management Zone” (MZ) is used in discussion about sub-fields in general and includes CGZ. Mean comparison of zones The goal is to delineate the fields into zones with significantly different agricultural properties. For this, mean comparisons between the zones were conducted. Given the variation in soil conditions and weather across the fields and years, the CGZs were compared separately for each field. For this analysis, the R software environment (Version 4.2.2; R Core Team 2020) was used, incorporating the "dplyr," "tidyr," "rstatix," "purrr," and "broom" packages. The assumptions of normality and homogeneity of variance were tested using the Shapiro-Wilk test (α = 0.05; Royston, 1995) and Levene's test (α = 0.05; Fox, 2015), respectively. While most parameters exhibited homogeneity of variance, not all did. Therefore, the Welch test (α = 0.05; Welch, 1947) was selected to compare the means of the two CGZs, as this test does not need homogeneity of variance and only two groups were compared. For the case of three CGZs, the Kruskal-Wallis (α = 0.05) test was chosen with Dunn-BonferroniTests (α = 0.05) as post hoc test. Both tests lack the assumption of normality and homogeneity of variance (Hollander, 1999). Scientific publications within the context of this work 54 Results NDRE maps and management zones The mean NDRE of PowerWeg was 0.9 on June 12, 2020 and 0.8 on August 27, 2020, with a coefficient of variation (CV) of 6.9 and 14.0 respectively. The eastern rim shows lower values at both dates (Fig 15). The first observation of Kiesschacht showed a mean NDRE of 0.49 (CV: 10.6) on May 14, 2021, which decreased to 0.28 (CV: 18.8) at the second observation on August 26, 2021. On May 14, 2021, the western part shows lower values, while on August 26, 2021, this pattern reverses. The mean NDRE values of BremerStraße are lower at 0.08 (CV: 60.6) on the first date (June 16, 2022) than on the second date (July 28, 2022), which shows an NDRE of 0.26 (CV: 26.1). On both dates, lower NDRE values were observed in the northeast. Fig 15. NDRE-Maps (normalized difference red edge) of clover-grass for the different fields and selected dates. The clustering algorithm revealed two CGZs per field. At the PowerWeg field, 80.3 % and 42 of 48 plots belong to CGZ 1. CGZ 2 can be found at the eastern rim and in some smaller spots, as shown in Fig 16. In the Kiesschacht field, 63.7 % and 35 plots are part of CGZ 1, mostly located in the eastern part of the field. CGZ 2 can be found along the slope and in the lower areas in the western part of the field. The BremerStraße field shows a more fragmented pattern in the southwest and southeast; most of CGZ 2 can be found in the centre and northwest, while CGZ 1 is located in the northeast. In this field, 45.8 % of the observed area and 30 plots belong to CGZ 2. The delineation into three CGZs did not show additional differences in the soil and agronomic parameters, the second CGZ takes an intermediate position between CGZ 1 and 3, with no significant differences to one or both CGZs. Therefore the results are not shown further. Scientific publications within the context of this work 55 Fig 16. Maps for the delineated of clover-grass zones (CGZ) for each field. Light blue is CGZ 1, dark blue is CGZ 2. Rectangles show the plots, filling indicates the zone. Clover-grass growth The clover-grass yield of the PowerWeg field varied between 1285.9 and 1562.7 g m-2, (Fig 17) but did not differ between the CGZs on any date. The third harvest showed a higher clover rate in CGZ 2 of 82.5 % compared to 68.1 % in CGZ 1 (Fig 18). Fig 17. Yield of clover, grass, and weeds are compared between the two clover-grass zones (CGZ) for different fields and timestamps. Dots represent the individual values of the clover-grass yield. The "Harvest" column describes the sampling before harvesting. The "Sample" column describes the sampling between harvests. The "Year" column is the annual sum of yield. Red signs indicate Welch-Test results: "ns" (p > 0.05), "*" (p < 0.05), "*" (p < 0.005), "**" (p < 0.0005), "****" (p < 0.00005). The annual yield of Kiesschacht field was 4386.8 g m-2. On the first harvest date (May 14, 2021), CGZ 2 of the Kiesschacht field achieved 320 g m-2 higher total biomass (p < 0.002, Welch test) in comparison to 1061.2 g m-2 in CGZ 1. The annual grass yield was also higher in CGZ 2, but not the clover biomass. The annual yield of BremerStraße was lower than the other fields with 1320.5 g m-2. CGZ 2 shows higher total biomass on the first (CGZ 2: 84.1, CGZ 1: 63.7 g m-2, p = 0.029, Welch test) and second sample date (CGZ 2: 69.1, CGZ 1: 51.1 g m-2, p = 0.046, Welch test) after the first harvest, due to higher clover yield (p = 0.017 Scientific publications within the context of this work 62 et al., 2025). Another possibility is site-specific seeding (Munnaf et al., 2022). Additionally, crop variety or species could be selected based on the MZs. For example, summer spelt is known for its lower soil quality requirements compared to winter wheat. A mixture of species, also known as intercropping, can balance out lower soil fertility, with supporting crops like peas. Legumes can transfer biologically-fixed N to the partner crop (Zhao et al. 2022) or compensate for yield gaps if the main crop performs poorly (Munz et al., 2023). An additional approach could be to manage the low yield areas less intensive, with lower fertilisation, weeding and tillage to promote biodiversity. There is an ongoing debate on whether lower-performing areas should receive more support through seeding or fertilization (the "Robin Hood" approach) or if high-potential zones should receive more resources to maximize their potential (the "King" approach). For sitespecific seeding of maize, the "King" approach is more suitable (Munnaf et al., 2022). Meanwhile, Corti et al. (2023) used site-specific manure application to even out SOM. In contrast, Morari et al. (2018) achieved higher yields with the "Robin Hood" approach in durum wheat, though this carries the risk of higher losses due to leaching. The delineation of fields based on the pre-crop has been done multiple times (Breunig et al., 2020; Girz and Mattila, 2024; Oldoni et al., 2025; Ouazaa et al., 2022), but to our knowledge, it has not been tested in practice. Future research should test site-specific applications (e.g. site-specific seeding or fertilization) in on-farm experiments. Additionally, the "King" and "Robin Hood" approaches should be evaluated for fertilization and variable seeding. Conclusions This study demonstrates the use of UAV-based NDRE imagery for the delineation of clover-grass fields and the subsequent cereal crops. The delineated CGZs were different both in the clover-grass period, as well as in subsequent cereal period. For all three fields, high productivity zones and low productivity zones were found. The differences are mostly due to spatial heterogenous soil conditions. Despite these results, the fields are very different to each other regarding fertility, soil texture and relief, further influencing the quality of clover-grass and cereal. Image acquisition with UAVs offers a flexible and accurate base for delineation of sub fields, the NDRE was suited for this task. UAVs additionally offers the opportunity to use the data for other applications, like site-specific weeding or plant monitoring. Contrary to the hypothesis, higher clover yield did not result in higher SNC, differences in the subsequent crop are therefore due to soil differences, relief and other effects like improved SOM. Adapted management decisions based on the MZs should be evaluated in on-farm trials in future research. Scientific publications within the context of this work 63 Supplementary Material Table S 1: Sampling dates field data and UAC campaigns. “X” marks if the parameters was collected on this date. Field Date Time Crop Biomass Plant height Plant Number Yield Proportion VS WC SP AD Soil Sampling UAV Images Power Weg 202006-12 1.1 sample Clovergrass X X X Power Weg 202007-10 2. harvest Clovergrass X X X Power Weg 202007-21 2.1 sample Clovergrass X X X Power Weg 202007-31 2.2 sample Clovergrass X X X Power Weg 202008-11 2.3 sample Clovergrass X X Power Weg 202008-27 2.4 sample Clovergrass X X X Power Weg 202009-10 3. harvest Clovergrass X X X Power Weg 202104-09 Soil sampling Summer spelt X Power Weg 202104-29 1. sample Summer spelt X X Weed cover X Power Weg 202105-20 2. sample Summer spelt X X Weed cover X Power Weg 202106-03 3. sample Summer spelt X X Weed cover X X Power Weg 202106-09 Soil sampling Summer spelt X Power Weg 202106-24 4. sample Summer spelt X + Ear length X Weed cover X X Power Weg 202108-11 Harvest Summer spelt X X + Ear length X Power Weg 202108-31 Soil sampling Summer spelt X Kiesschacht 202105-14 1. harvest Clovergrass X X X X X Kiesschacht 202106-10 Soil sampling Clovergrass X Kiesschacht 202106-14 1.1 sample Clovergrass X X X X Kiesschacht 202107-08 2. harvest Clovergrass X X X X Kiesschacht 202107-29 2.1 sample Clovergrass X X X X X Kiesschacht 202108-11 2.2 sample Clovergrass X X X X X Kiesschacht 202108-26 2.3 sample Clovergrass X X X X X Kiesschacht 202109-02 Soil sampling Clovergrass X Kiesschacht 202109-07 3. harvest Clovergrass X X X X X X Kiesschacht 202204-25 1. sample Summer spelt X X Weed cover X X Kiesschacht 202205-12 2. sample Summer spelt X X Weed cover X X Kiesschacht 202206-10 3. sample Summer spelt X X Weed cover X X Kiesschacht 202206-21 Soil sampling Summer spelt X Kiesschacht 202207-14 4. sample Summer spelt X X Weed cover X X Kiesschacht 202207-25 Harvest Summer spelt X X + Ear length X Weed cover Scientific publications within the context of this work 64 Field Date Time Crop Biomass Plant height Plant Number Yield Proportion VS WC SP AD Soil Sampling UAV Images Kiesschacht 202209-01 Soil sampling Summer spelt X Bremer Straße 202205-03 0.1 sample Clovergrass X X X X X Bremer Straße 202205-18 1. harvest Clovergrass X X X X Bremer Straße 202206-16 1.1 sample Clovergrass X X X X X Bremer Straße 202206-29 1.2 sample Clovergrass X X X X X Bremer Straße 202207-12 1.3 sample Clovergrass X X X X X Bremer Straße 202207-28 2. harvest Clovergrass X X X X X Bremer Straße 202208-17 2.1 sample Clovergrass X X X X X Bremer Straße 202208-31 2.2 sample Clovergrass X Bremer Straße 202209-15 2.3 sample Clovergrass X X X X X Bremer Straße 202209-29 3. harvest Clovergrass X X X X X X Bremer Straße 202211-08 1. sample Winter wheat X X X Bremer Straße 202303-30 2. sample Winter wheat X X Weed cover X X Bremer Straße 202304-03 Soil sampling Winter wheat X Bremer Straße 202305-17 3. sample Winter wheat X X Weed cover X X Bremer Straße 202305-22 4. sample Winter wheat X X Weed cover X X Bremer Straße 202306-18 Harvest Winter wheat X X + Ear length X Weed cover X X Scientific publications within the context of this work 65 Fig S 1. Indexes for finding best number of clusters (k) for fuzzy C-means. A higher value of Calinski-Harabasz index and the silhouette index indicates a matching with the own cluster. A high partition entropy indicates a fuzzy result. Fig S 2. Maps for the delineated of clover-grass zones (CGZ) for the single observed dates and for each field. Light blue is CGZ 1, dark blue is CGZ 2. Scientific publications within the context of this work 66 2.3 Effects of mixed intercropping on the agronomic parameters of two organically grown malting barley cultivars (Hordeum vulgare) in Northwest Germany Tobias Reuter1, Therese Brinkmeyer1, Johann Schreiber1, Valentin Freese1, Dieter Trautz1, Insa Kühling2 1Osnabrück University of Applied Sciences, Am Krümpel 31, 49090 Osnabrück, Germany 2Kiel University, Hermann-Rodewald-Str. 9, 24118 Kiel, Germany Abstract The market for organic malting barley (Hordeum vulgare L.) is growing. One goal of producing malting barley is to attain a defined protein content. This is challenging in organic farming because nutrient uptake is unpredictable. Since malting barley is quality sensitive to high amounts of available nitrogen (N) during the grain filling phase, artificial competition for N from a second crop within a mixed intercropping system could help to limit N uptake during these later growing stages. The objective of this study was to evaluate the effects of intercropping on the brewing quality parameters of malting barley within the framework of organic farming. In a field trial, with two spring barley cultivars (Odilia, Marthe) as sole crops and in intercropping treatments of 4 intermediate mixing ratios with camelina (Camelina sativa), linseed (Linum usitatissimum) and pea (Pisum sativum), the effects on yield and malting quality were observed. The results from three growing seasons on a study site in north-western Germany showed an opposite protein response for mixtures with linseed and pea. However, sole stands of barley mostly performed as well as the mixtures. For mixtures with camelina, neither yield nor quality aspects were affected. Comparing the two cultivars, the well-established Marthe, from traditional conventional selection, showed the better overall performance with a 13 % higher grain yield, a 3 % higher hectolitre weight and 5 % higher proportion of size fraction >2.5 mm than Odilia, which was released later from an organic breeding programme. However, the yield component analysis showed a consistent yield determination for Odilia regardless of intercropping partner or mixing ratios which might indicate its suitability in polycultures. The observed average land equivalent ratios (LER) of mixtures with linseed (1.04) and pea (1.13) showed the potential to increase land-use efficiency but were lower compared to the mean LER found in a recent meta-analysis. PICTURE CREDIT: Therese Brinkmeyer (2017) Citation Reuter, T., Brinkmeyer, T., Schreiber, J., Freese, V., Trautz, D., & Kühling, I. (2022). Effects of mixed intercropping on the agronomic parameters of two organically grown malting barley cultivars (Hordeum vulgare) in Northwest Germany. European Journal of Agronomy, 134 (January), 126470. https://doi.org/10.1016/j.eja.2022.126470 Keyword mixed cropping, crop mixtures, yield, components, diversity, oilseeds, organic, agriculture, sustainability, Europe Author contributions T.R.: investigation, formal analysis, writing – original draft/review & editing. T.B.: investigation, methodology. J.S: investigation. V.F.: investigation. D.T.: conceptualisation, resources, supervision, writing – review & editing. I.K.: conceptualisation, methodology, validation, formal analysis, visualisation, writing – original draft/review & editing, supervision. Acknowledgements This study was supported by Osnabrück University of Applied Sciences. Received 31 August 2021; Received in revised form 18 January 2022; Accepted 21 January 2022 1161-0301/© 2022 Elsevier B.V. All rights reserved. Received 31 August 2021; Received in revised form 18 January 2022; Accepted 21 January 2022 1161-0301/© 2022 Elsevier B.V. All rights reserved. General discussion 67 Chapter 3 General discussion Picture 3: Summer spelt (Triticum aestivum subsp. Spelta) trial, 19th July 2022. General discussion 68 3.1 Integration of Precision Farming and organic agriculture: potential and challenges The global agricultural sector is confronted with the dual challenge of sustaining food production for an increasing world population (Willett et al., 2019) while coping with a decline in per capita agricultural land availability (FAO, 2024). This challenge is further compounded by significant losses in plant and animal biodiversity (Geiger et al., 2010; Willett et al., 2019). Both Precision Farming (PF) and organic agriculture offer possible solutions, as organic farming enhances soil fertility and promotes biodiversity, by prohibiting synthetic fertilizers and pesticides (Sanders et al., 2025). PF optimizes resource application based on field-specific conditions, thereby increasing resource use efficiency (Ammar et al., 2024). The integration of PF into organic agriculture can help to close the yield gap between organic and conventional farming and further increase the sustainability. Adoption of Precision Farming in organic systems Research that directly combines PF and organic farming methods is limited. For instance, Loewen (2023) has tested variable rate seeding in organic contexts, this approach reduced the weed pressure and increased net returns. Whereas site-specific mechanical weeding has been evaluated primarily under conventional farming conditions (Berge et al., 2024; Niemeyer et al., 2024). Additionally, Reumaux (2024) estimated within-field variation for the site-specific application of biogas digestate fertilization in organic wheat production in Sweden. This thesis is among the few studies that integrate both approaches of PF and organic farming, thereby substantially enhancing our understanding of sustainable agriculture This thesis research used commercial machinery, such as Unmanned aerial vehicles (UAV), hoes and sowing machines. The concepts presented can be applied in practical agriculture with commercial machinery, although modifications are needed to allow the hoe to automatically adjust based on the weed map. Currently, no data is available that specifically examines PF adoption on organic farms in Germany. However, surveys conducted in Germany indicate that approximately 70 % of all farms employ PF technologies. Among these, 95 % use automatic steering systems, 40 % rely on yield mapping, 45 % implement variable rate fertilization and 33 % practice variable rate seeding (Sonntag et al., 2022). In small-scale farming regions such as Bavaria, only 21 % of farms use digital field records, 17 % employ automatic steering systems and 14 % use satellite-derived maps (Gabriel and Gandorfer, 2023). Although these surveys include organic farms, they did not analyse the specific effects of organic management practices on PF adoption. Factors that positively influence adoption rates include higher education levels and larger farm sizes (Sonntag et al., 2022). Given that organic farms are typically smaller (about 51 ha) compared to conventional farms (approximately 66 ha; BMEL, 2023b), it can be expected that PF adoption in organic systems may be closer to the 21 % observed in Bavarian small-scale farms. The adoption rates of 21 % aligns with surveys on organic farms in the Czech Republic, Hungary, Poland and Slovakia, where 15 to 26 % of farms use Precision Farming technology. The highest usage was found in the Czech Republic (25 %) and Poland (26 %). The primary motivation for adopting Precision Farming technology was cost savings, while the main obstacle was the lack of financial support (Petrovic et al., 2025). For PF technology to be successfully adopted, it must deliver clear economic or ecological benefits, be reliable and user-friendly, while farmers must also possess the necessary knowledge for its application and evaluation (Mohr and Kühl, 2021; Sonntag et al., 2022). General discussion 69 The results presented in this thesis indicate that PF and Decision Support Systems (DSS) benefit organic farming by improving resource utilization—such as optimizing soil nutrient management and preserving wild plant communities. Table 9 provides an overview of the hypotheses tested across the three peer-reviewed papers. Of the eight hypotheses, six were accepted due to the field trial results. Table 9: Tested hypothesises in the three peer-reviewed paper. ✓ : accepted;  : rejected. SSMW: site-specific mechanical weed management. RWC: Relative Weed Cover. NDRE: Normalized Difference Red Edge. Hypothesis Result Site-specific mechanical weeding of maize (Zea mays) Hypothesis 1: SSMW does not influence maize yield compared to uniform weeding. ✓ Hypothesis 2: SSMW does not influence weed biomass compared to uniform weeding. ✓ Hypothesis 3: SSMW based on RWC results in less treated area than uniform weeding and SSMW based on weed cover. ✓ Delineation of management zones in clover-grass for site-specific management of subsequent crops Hypothesis 4: High productivity zones during the clover-grass period are also highly productive for the subsequent crop. ✓ Hypothesis 5: NDRE images are useful for delineating management zones in clovergrass fields. ✓ Hypothesis 6: Higher clover yields lead to higher fixed nitrogen (N) and, therefore, higher soil nitrogen content (SNC) during both the clover-grass period and the subsequent crop.  Effects of mixed intercropping on the agronomic parameters of two organically grown malting barley cultivars (Hordeum vulgare) Hypothesis 7: Intercropping barley with a non-legume partner can limit protein content and enhance malting quality. ✓ Hypothesis 8: Organic varieties intercropped with partners have better yield and quality performance.  The site-specific mechanical weeding of maize (Zea mays), examined in section 2.1, achieved a reduction in the treated area by 58 – 83 %, particularly in the Relative Weed Cover (RWC) treatment plots (supporting hypothesis 3). These savings were realized without any yield losses (supporting hypothesis 1) or increases in weed biomass (supporting hypothesis 2). This reduction in the treated area can decrease the risk of soil erosion and crop damage, while also sparing wild plants by applying weeding only when a harmful threshold is exceeded. In section 2.2, the delineation of management zones in clover–grass fields was successfully carried out using UAV-based NDRE images (supporting hypothesis 5). Across all fields, highly productive sub-fields were identified during both the clover–grass period and the subsequent cereal production period (supporting hypothesis 4). Contrary to hypothesis 6, these areas did not exhibit an increase in soil nitrogen content, even though higher clover yields were observed. Section 2.3 describes the intercropping of malting barley (Hordeum vulgare). Intercropping with a non-legume partner effectively limited the protein content (supporting hypothesis 7), whereas intercropping with pea (Pisum sativum) increased protein content. In contrast to the assumption of hypothesis 8, the conventional variety ‘Marthe’ achieved higher yields than the organic variety ‘Odelia’ under intercropping conditions. General discussion 70 In the following discussion, the results of the three papers are examined in a broader context with respect to their potential applications in sustainable agricultural practices. Site-specific mechanical weed management The application of site-specific mechanical weed management, as described in section 2.1, showed significant reduced managed area of around -58 % to -83 %, this results varied between the years. These reductions can enhance plant biodiversity (Sowiński, 2023), decrease soil erosion risk (Seitz et al., 2019) and reduce machinery wear by minimizing tillage time (Gry et al., 2019). Although initial costs in mapping weeds and acquiring specialized machinery may be higher, the method can save money in the long term through up to a 97 % reduction in fuel consumption and equipment wear (Gry et al., 2019). For arable crops such as maize and cereals—which are typically hoed two to three times per growing season—the benefits of site-specific weeding are significant. Vegetables like lettuce (Lactuca sativa), cabbage (Brassica oleracea) or carrots (Daucus carota), which require more intensive regulation due to lower competitive ability (Gerhards et al., 2025), may see multiplicative economic and agronomic benefits. Recent advances, including sensor-guided inter-row hoes, have successfully reduced weed cover within crop rows. However, crop row identification generally requires slower operating speeds (approximately 1 km h-1; Gerhards et al., 2025). This limitation could be overcome by partially raising the hoe in areas with low weed cover and where regulation is unnecessary. Another opportunity is the deployment of autonomous weeding robots—where labour costs are less critical compared to manned tractors (Griepentrog and Stein, 2024). Field robots also offer the potential to seed crops in uniform patterns, ensuring consistent spacing between plants. This uniformity optimizes nutrient and light accessibility while facilitating both interand intra-row weeding, as the robots can operate in any direction (Wegener et al., 2019). Despite potential future advancements in mechanical weeding techniques, this approach will always carry the risk of causing plant injuries (Machleb et al., 2020) and reducing the abundance of wild plants (Sowiński, 2023). The research presented here offers valuable insights into DSS for site-specific mechanical weeding, thereby mitigating some of these drawbacks. In particular, the Relative Weed Cover (RWC) approach is recommended because it resulted in the smallest weeded area. The RWC metric takes into account both weed and crop cover, enabling an effective estimation of crop competitiveness. Management zones and intercropping in organic fields Three organic fields were successfully delineated into management zones, based on NDRE-Maps of clover-grass (Section 2.2). Both the clover-grass and the subsequent cereal crops exhibited significant differences in yield and quality parameters across management zones. This zonation permits more targeted management approaches. Fertilizer can be applied based on the yield potential, as they are limited in organic farming and are a reason for the yield gap to organic agriculture (Döring and Neuhoff, 2021). As Reumaux (2024) showed, site-specific biogas digestate can improve the yield and protein content for organic wheat. Furthermore, management zones can serve as the basis for selecting crop varieties and intercropping strategies. Intercropping with legumes, for example, can enhance protein content (Section 2.3) and mitigate the effects of nutrient limitations in low-yield areas. Intercropping has been reported to offer greater yield stability than sole cropping, mitigating both spatial and temporal variability (Munz et al., 2023; Raseduzzaman and Jensen, 2017). This approach is General discussion 71 particularly beneficial in unstable field zones—often found in depressions with a higher risk of waterlogging—where crops exhibit lower yields during high-precipitation years and improved yields during dry periods. Inverse, areas with low water-holding capacity benefit with higher precipitation and experience low yields under dry conditions (Maestrini and Basso, 2018a; McEntee et al., 2020). Additionally, winter crops, which generally possess better-developed root systems, are less sensitive to dry summer conditions compared to summer crops (Hernández-ochoa et al., 2025). Research in the midwestern USA indicates that approximately 18 % of fields are unstable, with unpredictable annual variability. In such cases, intercropping can stabilize yields as different crop species complement one another by compensating for losses due to their varying requirements (Raseduzzaman and Jensen, 2017). This dynamic adjustment of crop composition has also been termed “ecological precision farming” (Jensen et al., 2015). Munz et al. (2023) recommends adjusting the legume sowing ratio based on soil conditions to improve their competitiveness against cereals. Increased legume proportions are beneficial in low nitrogen areas. Another approach involves identifying areas with high weed pressure. The weed recognition could be conducted with UAV-images described in section 2.1. In these zones, crop mixtures and species that exhibit rapid early growth and provide dense ground cover can indirectly suppress weed populations. In contrast, regions with lower weed pressure might benefit from sowing species and mixtures optimized for higher yields. 3.2 Variation of agricultural parameters between years and sites In all trials, significant annual differences were observed. In paper 1 (Section 2.1) on sitespecific mechanical weeding, a maize yield of 1879 g m-2 was recorded in 2021 compared to 533 g m-2 in 2022. In the control treatment, weed biomass was with 13.6 g m-2 in 2021 lower as in 2022 with 58.8 g m-2. The reduction in treated area for the site-specific treatments varied widely from 0 % to 100 %, resulting in average savings of 58 % in 2021 and 83 % in 2022. For the delineation of management zones reported in paper 2 (Section 2.2), trials were conducted across different sites, where seasonal differences were observed in addition to inter-field variations. The clover–grass yield varied considerably: 1320.5 g m-2 at PowerWeg (2020), 4386.6 g m-2 at Kiesschacht (2021) and 983.7 g m-2 at BremerStraße (2022). Similarly, cereal yields differed among sites, with yields of 158.8 g m-2 at PowerWeg (2021), 141.2 g m-2 at Kiesschacht (2022) and 287.6 g m-2 at BremerStraße (2023). In paper 3 (Section 2.3), an intercropping trial revealed that barley yields in sole stands varied over the years: 142.8 g m-2; 113.5 g m-2; and 50.0 g m-2 in 2017, 2018 and 2019, respectively. The protein content exhibited an annual pattern, with values of 9.6 % in 2017, 10.4 % in 2018 and 8.5 % in 2019. Annual differences have been documented in other studies on site-specific applications as well. Corti et al. (2023) reported variations in yield and nitrogen use efficiency in trials of sitespecific manure application. The simulations of site-specific herbicide application similarly showed varying efficiencies between years (Maillot et al., 2023). Temporal and spatial variability was anticipated; therefore, all trials were repeated over two to three years. Nevertheless, this inherent variability makes it challenging to accurately analyse the effects of PF technology. Temporal variation Temporal variation in these trials is primarily attributed to weather differences. Especially drought stress can vary dynamically from year to year (Johnen et al., 2014). During the growing Summary 78 Summary The global agriculture is facing the challenge of providing enough food for a growing population while addressing declining biodiversity. Organic farming benefits the environment but typically yields 20 % less than conventional methods. Precision Farming (PF) can help to close this gap by tailoring inputs like fertilizers to variations within fields. This dissertation presents a series of trials—reported in three peer-reviewed papers— aimed at increasing the productivity and sustainability of organic farming. Conducted near Osnabrück in northwestern Germany, these field studies employed either randomized block designs or on-farm trials over two to three years. In these trials, PF technologies such as Unmanned aerial vehicles (UAVs) and Decision Support Systems (DSS) were integrated with conventional organic management practices including mechanical weeding, crop rotation and intercropping. In paper 1 (Section 2.1), the application of site-specific mechanical weeding based on UAV imagery reduced the managed area by 58 % to 83 % without compromising yield or influencing weed biomass (Supporting hypothesis 1 & 2). Two decision-support strategies were compared against a uniform weeding approach. Among these, the Relative Weed Cover emerged as a promising metric (Supporting hypothesis 3), since it considers both maize and weed cover, the crop competitiveness can be estimated and the hoed area was minimized. This approach not only reduces input use but can also enhance plant biodiversity, decrease soil erosion and reduce machinery wear, offering long-term cost savings despite higher initial mapping and equipment costs. Paper 2 (Section 2.2) focuses on delineating management zones within clover–grass fields using Vegetation Indices. Three organic fields were successfully partitioned into two distinct management zones each. The used NDRE (Normalized Difference Red Edge)-Maps and fuzzy C-means clustering algorithms were appropriated for this task (Supporting hypothesis 5). Significant differences in clover–grass parameters and in the yield of subsequent cereal crops were observed between zones (Supporting hypothesis 4). These differences are largely attributable to variations in soil properties and topography. This zonation supports targeted management strategies, such as site-specific fertilization, which is particularly important in organic systems where fertilizer availability is limited. In contrast to hypothesis 6 no significant difference in soil nitrogen content between the zones were observed, despite differences in clover proportion and biomass. The final paper (Section 2.3) compares the performance of malting barley (Hordeum vulgare) intercropped with various partners. Trials revealed that intercropping barley with peas (Pisum sativum) improved protein content and land-use efficiency, whereas intercropping with linseed (Linum usitatissimum) reduced protein levels (Supporting hypothesis 7); nevertheless, sole stands of barley generally performed similarly to mixed systems. These results inform the selection of optimal crop mixtures according to different management zones and underscore the importance of intercropping in stabilizing yields in areas with variable fertility. The organic barley variety did not show benefits in regard of yield and quality compared to the conventional variety, therefore the hypothesis 8 can be rejected. Summary 79 Substantial temporal variability was observed across all trials—for example, maize yields ranged from 533 to 1879 g m-2, clover–grass yields varied between 984 g m-2 and 4387 g m-2 and cereal yields ranged from 50 to 143 g m-2. Such variability, driven largely by weather conditions, along with differences between sites, highlights the importance of multi-year and multi-site trials for robust evaluation of PF technologies. Finally, the integration of on-farm trials with a co-design approach involving stakeholders is crucial for enhancing the acceptance and practical implementation of innovative management practices. Recent advances in data analytics and field robotics offer further potential for transforming agricultural practices toward biodiversity-based systems. Future approaches may involve subdividing fields into smaller management units, which are integrated with landscape elements to promote biodiversity. This thesis is among the few studies that integrate PF-technologies with organic farming practices, thereby significantly enhancing our understanding of sustainable agriculture. The results indicate that the combined use of PF and DSS benefits organic farming by improving resource utilization. This is achieved by applying targeted, site-specific weeding in areas where it is necessary and by delineating fields into management zones to apply fertilization and seeding based on sub-field conditions. Zusammenfassung (German summary) 80 Zusammenfassung (German summary) Die Landwirtschaft steht vor der Herausforderung gleichzeitig für Ernährungssicherheit und ein intaktes Ökosystem zu sorgen. Der ökologische Landbau gilt als umweltverträglicher, erntet jedoch im Schnitt 20 % weniger als der konventionelle Landbau. Precision Farming (PF) bietet die Möglichkeit die Ertragslücke durch effizientere Ressourcenverteilung basieren auf Ungleichheiten im Feld zu verringern. In dieser Dissertation werden drei peer-reviewed Artikel vorgestellt, die das Ziel verfolgen, die Produktivität und Nachhaltigkeit des ökologischen Landbaus zu verbessern. PFAnwendungen wie der Einsatz von Unmanned aerial vehicles (UAVs) und Entscheidungsunterstützungssystemen (englisch Decision Support System; DSS) wurden mit klassischen Methoden des ökologischen Landbaus, wie mechanische Beikrautregulierung, Leguminosenanbau und Mischkulturanbau, kombiniert. Dazu wurden im Raum Osnabrück in Nordwestdeutschland verschiedene Versuche durchgeführt, die entweder als randomisierte Blockversuche oder als On-farm-Versuche realisiert wurden. Jeder Versuch wurde in zwei oder drei Versuchsjahren wiederholt. Artikel 1 (Abschnitt 2.1) befasst sich mit der teilflächenspezifischen mechanischen Beikrautregulierung. Die Erkennung der Beikräuter erfolgte anhand von UAV-Aufnahmen. Mit diesem Verfahren konnte die bearbeitete Fläche um 58 % bis 83 % reduziert werden, ohne dass Ertragsverluste auftraten oder die Beikrautbiomasse beeinflusst wurde, was Hypothese 1 & 2 belegt. Von den untersuchten Entscheidungsstrategien führte der relative Unkrautbedeckungsgrad (englisch: Relative Weed Cover), der sowohl den Beikrautals auch den Maisbedeckungsgrad berücksichtigt, zu der geringsten bearbeiteten Fläche (Bestätigung Hypothese 3). Die Reduzierung der bearbeiteten Fläche durch diese Methode bietet das Potenzial, Wildpflanzen zu schützen, Erosion zu verhindern und den Verschleiß von Maschinen zu verringern. Artikel 2 (Abschnitt 2.2) beschreibt, wie mithilfe des drohnenbasierten Vegetationsindexes (Normalized Difference Red Edge), Kleegrasflächen in zwei Managementzonen unterteilt werden konnten. Diese Einteilung wurde erfolgreich an drei Flächen durchgeführt, was Hypothese 5 bestätigt. Diese Zonen unterschieden sich signifikant im Kleegrasertrag und - zusammensetzung. In den anschließenden Getreidekulturen konnten zudem Differenzen in Ertrag und Qualität zwischen den Zonen festgestellt werden (Bestätigung Hypothese 4). Diese Unterschiede sind hauptsächlich auf Variationen in Bodenbeschaffenheit und Topografie zurückzuführen. Die so ermittelten Managementzonen können als Grundlage für teilflächenspezifische Managementmaßnahmen dienen, etwa einer gezielten Düngung, welche für den ökologischen Landbau besonders relevant ist, da Düngemittel nur begrenzt vorhanden sind. Entgegen der Hypothese 6, gab es jedoch keine Unterschiede zwischen den Zonen in Bezug auf mineralischen Stickstoffgehalt im Boden. Im letzten Artikel (Abschnitt 2.3) wurden Braugerste (Hordeum vulgare) im Gemengeanbau mit verschiedenen Mischungspartnern (Erbse, Pisum sativum; Leindotter, Linum usitatissimum; und Öllein, Camelina sativa) untersucht. Der Gemengeanbau mit Erbse (Pisum sativum) erhöhte den Proteingehalt und die Landnutzungseffizienz, während die Mischung mit Leindotter (Linum usitatissimum) den Proteingehalt verringerte (Bestätigung der Hypothese 7). Die Reinsaat von Braugerste (Hordeum vulgare) erzielte hinsichtlich des Ertrages und der Qualität ähnliche Ergebnisse wie die Mischkulturen. Diese Erkenntnisse können Grundlage für die Kulturauswahl in den unterschiedlichen Managementzonen sein. Insbesondere in ertragsschwachen Bereichen können Gemenge mit Erbse (Pisum sativum) Zusammenfassung (German summary) 81 zu einer Verbesserung der Ertragsstabilität beitragen, da in Gemengeanbau, die Partnerkulturen sich ergänzen und Ertragsausfälle kompensieren können. Dies prädestiniert den Gemengeanbau für Bereichen mit schwankender Ertragspotential. Die ökologische Sorte zeigten keine Vorteil gegenüber der konventionellen Sorte, weder in Bezug auf Ertrag noch Qualität, was zur Ablehnung von Hypothese 8 führte. In allen Versuchen wurden erhebliche Schwankungen zwischen den Jahren festgestellt: So variierte der Maisertrag zwischen 533 und 1879 g m-2, der Kleegrasertrag zwischen 984 und 4387 g m-2 und der Getreideertrag zwischen 50 und 143 g m-2. Diese Variationen sind hauptsächlich auf Wetterunterschiede zurückzuführen – ein Befund, der auch in anderen Untersuchungen bestätigt wurde. Darüber hinaus wurden große Unterschiede zwischen den Flächen festgestellt, was ebenfalls in weiteren Feldversuchen festgestellt wurde. Dies unterstreicht die Bedeutung von mehrjährigen Versuchen an verschiedenen Standorten für eine robuste Evaluation von PF-Technologien. Für die Akzeptanz und Implementierung von PF und innovativen Anbausystemen ist es wichtig Stakeholder partizipative in die Planung und Durchführung von On-farm-Versuchen miteinzubeziehen. Die aktuellen Entwicklungen in der Datenanalyse und Feldrobotik ebnen den Weg zu einer biodiversitätsbasierten Landwirtschaft. Zukünftige Agrarsysteme könnten so gestaltet werden, dass große Flächen in kleinere Bewirtschaftungseinheiten mit ähnlichem Ertragspotenzial unterteilt und entsprechend ihren spezifischen Bedingungen bewirtschaftet werden. Dabei wird nicht nur das einzelne Feld betrachtet, sondern die gesamte Agrarlandschaft, welche durch Landschaftselemente wie Hecken und Bäume miteinander verbunden werden. Ein solches System weist komplexe Wechselwirkungen auf und stellt hohe Anforderungen an das Management. DSS können dabei helfen, diese Zusammenhänge zu verstehen und fundierte Entscheidungen zu treffen. Die hier vorgestellte Dissertation gehört zu den wenigen Arbeiten, die PF-Technologien im ökologischen Landbau anwenden. Die durchgeführten Versuche zeigen, dass diese Ansätze dazu beitragen, die Landwirtschaft nachhaltiger zu gestalten und Ressourcen effizienter zu nutzen. So konnte teilflächenspezifische Beikrautregulierung die behandelte Fläche signifikant reduzieren, während die Einteilung der Flächen in Managementzonen eine fundierte Grundlage für standortspezifische Düngung und Aussaat bietet. References 82 References Adeux, G., Vieren, E., Carlesi, S., Bàrberi, P., Munier-Jolain, N., Cordeau, S., 2019. Mitigating crop yield losses through weed diversity. Nat. Sustain. 2, 1018–1026. https://doi.org/https://doi.org/10.1038/s41893-019-0415-y Aksakal, E.L., Barik, K., Angin, I., Sari, S., Islam, K.R., 2019. Spatio-temporal variability in physical properties of different textured soils under similar management and semi-arid climatic conditions. Catena 172, 528–546. https://doi.org/10.1016/j.catena.2018.09.017 Ali, A., Streibig, J.C., Andreasen, C., 2013. Yield loss prediction models based on early estimation of weed pressure. Crop Prot. 53, 125–131. https://doi.org/10.1016/j.cropro.2013.06.010 Ali, A., Streibig, J.C., Christensen, S., Andreasen, C., 2015. Image-based thresholds for weeds in maize fields. Weed Res. 55, 26–33. https://doi.org/10.1111/wre.12109 Allmendinger, A., Spaeth, M., Saile, M., Peteinatos, G.G., Gerhards, R., 2024. Agronomic and Technical Evaluation of Herbicide Spot Spraying in Maize Based on High-Resolution Aerial Weed Maps—An On-Farm Trial. Plants 13. https://doi.org/10.3390/plants13152164 Ammar, E.E., Aziz, S.A., Zou, X., Elmasry, S.A., Ghosh, S., Khalaf, B.M., EL-Shershaby, N.A., Tourky, G.F., AL-Farga, A., Khan, A.N., Abdelhafeez, M.M., Younis, F.E., 2024. An indepth review on the concept of digital farming. Environ. Dev. Sustain. https://doi.org/https://doi.org/10.1007/s10668-024-05161-9 Arbeitsgruppe Boden (Soil working Group), 2024. Bodenkundliche Kartieranleitung KA6, 6. Edition. ed. Schweizerbart Science Publishers, Stuttgart, Germany. Azimi, S., Kaur, T., Gandhi, T.K., 2021. A deep learning approach to measure stress level in plants due to Nitrogen deficiency. Measurement 173, 108650. https://doi.org/10.1016/j.measurement.2020.108650 Bai, Z., Caspari, T., Ruiperez, M., Batjes, N.H., Mäder, P., Bünemann, E.K., Goede, R. De, Brussaard, L., Xu, M., Ferreira, C.S.S., Reintam, E., Fan, H., Miheličh, R., Glavan, M., Tóth, Z., 2018. Effects of agricultural management practices on soil quality: A review of long-term experiments for Europe and China. Agric. Ecosyst. Environ. 265, 1–7. https://doi.org/10.1016/j.agee.2018.05.028 Balasundram, S.K., Shamshiri, R.R., Sridhara, S., Rizan, N., 2023. The Role of Digital Agriculture in Mitigating Climate Change and Ensuring Food Security: An Overview. Sustainability 15, 5325. https://doi.org/10.3390/SU15065325 Ball, B.C., Crawford, C.E., 2009. Mechanical weeding effects on soil structure under field carrots (Daucus carota L.) and beans (Vicia faba L.). Soil Use Manag. 25, 303–310. https://doi.org/10.1111/j.1475-2743.2009.00226.x Banniza, S., Vandenberg, A., 2003. The influence of plant injury on development of Mycosphaerella pinodes in field pea. Can. J. Plant Pathol. 25, 304–311. https://doi.org/10.1080/07060660309507083 Bàrberi, P., Burgio, G., Dinelli, G., Moonen, A.C., Otto, S., Vazzana, C., Zanin, G., 2010. Functional biodiversity in the agricultural landscape: Relationships between weeds and arthropod fauna. Weed Res. 50, 388–401. https://doi.org/10.1111/j.13653180.2010.00798.x Bareth, G., Lussem, U., Menne, J., Hollberg, J., Schellberg, J., 2019. Potential of noncalibrated UAV-based RGB imagery for forage monitoring: Case study at the rengen longterm grassland experiment (rge), Germany. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. - ISPRS Arch. 42, 203–206. https://doi.org/10.5194/isprs-archives-XLII-2-W13203-2019 References 83 Barrasso, C., Krüger, R., Eltner, A., Cord, A.F., 2024. Mapping indicator species of segetal flora for result-based payments in arable land using UAV imagery and deep learning. Ecol. Indic. 169, 112780. https://doi.org/10.1016/j.ecolind.2024.112780 Barreto, A., Ispizua Yamati, F.R., Varrelmann, M., Paulus, S., Mahlein, A.K., 2023. Disease Incidence and Severity of Cercospora Leaf Spot in Sugar Beet Assessed by Multispectral Unmanned Aerial Images and Machine Learning. Plant Dis. 107, 188–200. https://doi.org/10.1094/PDIS-12-21-2734-RE Barzin, R., Pathak, R., Lotfi, H., Varco, J., Bora, G.C., 2020. Use of UAS multispectral imagery at different physiological stages for yield prediction and input resource optimization in corn. Remote Sens. 12. https://doi.org/10.3390/RS12152392 Basso, B., Antle, J., 2020. Digital agriculture to design sustainable agricultural systems. Nat. Sustain. 3, 254–256. https://doi.org/10.1038/s41893-020-0510-0 Bedoussac, L., Journet, É.-P., Hauggaard-Nielsen, H., Naudin, C., Corre-Hellou, G., Prieur, L., Jensen, E.S., Justes, E., 2014. Eco-functional Intensification by Cereal-Grain Legume Intercropping in Organic Farming Systems for Increased Yields, Reduced Weeds and Improved Grain Protein Concentration, in: Organic Farming, Prototype for Sustainable Agricultures. Springer Netherlands, Dordrecht, pp. 47–63. https://doi.org/10.1007/978-94007-7927-3_3 Bedoussac, L., Journet, E.P., Hauggaard-Nielsen, H., Naudin, C., Corre-Hellou, G., Jensen, E.S., Prieur, L., Justes, E., 2015. Ecological principles underlying the increase of productivity achieved by cereal-grain legume intercrops in organic farming. A review. Agron. Sustain. Dev. https://doi.org/10.1007/s13593-014-0277-7 Behera, S.K., Mathur, R.K., Shukla, A.K., Suresh, K., Prakash, C., 2018. Spatial variability of soil properties and delineation of soil management zones of oil palm plantations grown in a hot and humid tropical region of southern India. Catena 165, 251–259. https://doi.org/10.1016/j.catena.2018.02.008 Beiküfner, M., Kühling, I., Vergara-Hernandez, M.E., Broll, G., Trautz, D., 2024. Impact of mechanical weed control on soil N dynamics, soil moisture, and crop yield in an organic cropping sequence. Nutr. Cycl. Agroecosystems. https://doi.org/10.1007/s10705-02410370-9 Beillouin, D., Ben-Ari, T., Malézieux, E., Seufert, V., Makowski, D., 2021. Positive but variable effects of crop diversification on biodiversity and ecosystem services. Glob. Chang. Biol. 27, 4697–4710. https://doi.org/10.1111/gcb.15747 Berdeni, D., Turner, A., Grayson, R.P., Llanos, J., Holden, J., Firbank, L.G., Lappage, M.G., Hunt, S.P.F., Chapman, P.J., Hodson, M.E., Helgason, T., Watt, P.J., Leake, J.R., 2021. Soil quality regeneration by grass-clover leys in arable rotations compared to permanent grassland: Effects on wheat yield and resilience to drought and flooding. Soil Tillage Res. 212. https://doi.org/10.1016/j.still.2021.105037 Berge, T.W., Urdal, F., Torp, T., Andreasen, C., 2024. A Sensor-Based Decision Model for Precision Weed Harrowing. Agronomy 14, 1–14. https://doi.org/10.3390/agronomy14010088 Berry, P., Sylvester-Bradley, R., Philipps, L., Hatch, D., Cuttle, S., Rayns, F., Gosling, P., 2002. Is the productivity of organic farms restricted by the supply of available nitrogen? Soil Use Manag. 18, 248–255. https://doi.org/10.1079/SUM2002129 Blank, L., Rozenberg, G., Gafni, R., 2023. Spatial and temporal aspects of weeds distribution within agricultural fields – A review. Crop Prot. 172. https://doi.org/10.1016/j.cropro.2023.106300 BMEL, 2023a. Bio-Strategie 2030 Nationale Strategie für 30 Prozent ökologische Landund Lebensmittelwirtschaft bis 2030. Bundesministerium für Ernährung und Landwirtschaft. References 84 BMEL, 2023b. Visualisierung der Strukturdaten zum ökologischen Landbau in Deutschland. Bundesministerium für Ernährung und Landwirtschaft [WWW Document]. URL https://bmel-statistik.de/landwirtschaft/oekologischer-landbau BMEL, 2010. Getreideeinheitenschlüssel Überarbeitung. Bundesministerium für Ernährung und Landwirtschaft. Boenecke, E., Ueck, E., Ruehlmann, J., Gruendling, R., Franko, U., 2018. Determining the within-field yield variability from seasonally changing soil conditions. Precis. Agric. 19, 750–769. https://doi.org/10.1007/s11119-017-9556-z Bölenius, E., Stenberg, B., Arvidsson, J., 2017. Within field cereal yield variability as affected by soil physical properties and weather variations – A case study in east central Sweden. Geoderma Reg. 11, 96–103. https://doi.org/10.1016/j.geodrs.2017.11.001 BÖLW, 2024. Branchenreport 2024. Bund Ökologische Lebensmittelwirtschaft e.V. Booth, J.C., Mccall, D.S., Sullivan, D., Askew, S.A., Kochersberger, K., 2021. Investigating targeted spring dead spot management via aerial mapping and precision-guided fungicide applications. Crop Sci. 3134–3144. https://doi.org/10.1002/csc2.20623 Breiman, L., 2001. Random Forests. Mach. Learn. 45, 5–32. https://doi.org/https://doi.org/10.1023/A:1010933404324 Breunig, F.M., Galvão, L.S., Dalagnol, R., Dauve, C.E., Parraga, A., Santi, A.L., Della Flora, D.P., Chen, S., 2020. Delineation of management zones in agricultural fields using cover– crop biomass estimates from PlanetScope data. Int. J. Appl. Earth Obs. Geoinf. 85. https://doi.org/10.1016/j.jag.2019.102004 Brito Da Silva, L.E., Melton, N.M., Wunsch, D.C., 2020. Incremental Cluster Validity Indices for Online Learning of Hard Partitions: Extensions and Comparative Study. IEEE Access 8, 22025–22047. https://doi.org/10.1109/ACCESS.2020.2969849 Brooker, R.W., Bennett, A.E., Cong, W.-F.F., Daniell, T.J., George, T.S., Hallett, P.D., Hawes, C., Iannetta, P.P.M.M., Jones, H.G., Karley, A.J., Li, L., McKenzie, B.M., Pakeman, R.J., Paterson, E., Schöb, C., Shen, J., Squire, G., Watson, C.A., Zhang, C., Zhang, F., Zhang, J., White, P.J., 2015. Improving intercropping: A synthesis of research in agronomy, plant physiology and ecology. New Phytol. 206, 107–117. https://doi.org/10.1111/nph.13132 Brooker, R.W., Pakeman, R.J., Adam, E., Banfield-Zanin, J.A., Bertelsen, I., Bickler, C., FogPetersen, J., George, D., Newton, A.C., Rubiales, D., Tavoletti, S., Villegas-Fernández, Á.M., Karley, A.J., 2024. Positive effects of intercrop yields in farms from across Europe depend on rainfall, crop composition, and management. Agron. Sustain. Dev. 44. https://doi.org/10.1007/s13593-024-00968-2 BSA, 2021. Beschreibende Sortenliste Getreide, Mais Ölund Faserpflanzen Leguminosen Rüben Zwischenfrüchte. Federal Plant Variety Office (Bundessortenamt), Hannover. Buladaco II, M.S., Tandugon, H.M.F., Bunquin, M.A.B., Sanchez, P.B., Bugia, S.A.C., Yales, N.A.P., Casacop, S.M., 2024. Mapping and Assessment of Within-Field Spatial Variability of Soil pH, Electrical Conductivity, and Particle Size Distribution to Delineate Management Zones. Ecol. Eng. Environ. Technol. 25, 75–86. Buman, T., 2013. Opportunity now: Integrate conservation with precision agriculture. J. Soil Water Conserv. 68, 96–98. https://doi.org/10.2489/jswc.68.4.96A Carof, M., Godinot, O., Le Cadre, E., 2022. Biodiversity-based cropping systems: A long-term perspective is necessary. Sci. Total Environ. 838, 156022. https://doi.org/10.1016/j.scitotenv.2022.156022 Castaldi, F., Pelosi, F., Pascucci, S., Casa, R., 2017. Assessing the potential of images from unmanned aerial vehicles (UAV) to support herbicide patch spraying in maize. Precis. Agric. 18, 76–94. https://doi.org/10.1007/s11119-016-9468-3 References 85 Chambers, J.M., Freeny, A., Heiberger, R.M., 1992. Analysis of variance; Designed Experiments, in: Chambers, J.M., Hastie, T.J. (Eds.), Statistical Models in S. Wadsworth & Brooks, Cole. Chantre, G.R., 2020. Decision Support Systems for Weed Management. Springer-Verlag, Cham,. https://doi.org/10.1007/978-3-030-44402-0 Cheng, E., Zhang, B., Peng, D., Zhong, L., Yu, L., Liu, Y., Xiao, C., Li, C., Li, X., Chen, Y., Ye, H., Wang, H., Yu, R., Hu, J., Yang, S., 2022. Wheat yield estimation using remote sensing data based on machine learning approaches. Front. Plant Sci. 13, 1–16. https://doi.org/10.3389/fpls.2022.1090970 Cimpoiaşua, M.O., Kuras, O., Pridmore, T., Mooney, S.J., 2020. Potential of geoelectrical methods to monitor root zone processes and structure : A review. Geoderma 365. https://doi.org/https://doi.org/10.1016/j.geoderma.2020.114232 Corti, M., Cavalli, D., Pricca, N., Ferrè, C., Comolli, R., Marino, P., Davide, G., El, A., 2023. Site ‑ specific recommendations of cattle manure nitrogen and urea for silage maize. Nutr. Cycl. Agroecosystems 127, 155–169. https://doi.org/10.1007/s10705-023-10302-z Cowden, R.J., Shah, A.N., Lehmann, L.M., Kiær, L.P., Henriksen, C.B., Ghaley, B.B., 2020. Nitrogen fertilizer effects on pea–barley intercrop productivity compared to sole crops in Denmark. Sustain. 12, 1–17. https://doi.org/10.3390/su12229335 da Silva, E.R.O., Pereira, M.G., de Barros, M.M., dos Santos, L.M.M., Gomes, J.H.G., 2022. Soil Organic Matter Fractions and Multivariate Analysis in the Definition of Pasture Management Zones. Eng. Agric. 42. https://doi.org/10.1590/1809-4430ENG.AGRIC.V42N6E20220099/2022 Damian, J.M., De Castro Pias, O.H., Cherubin, M.R., Da Fonseca, A.Z., Fornari, E.Z., Santi, A.L., 2020. Applying the NDVI from satellite images in delimiting management zones for annual crops. Sci. Agric. 77, 1–11. https://doi.org/10.1590/1678-992x-2018-0055 Davis, A.S., Hill, J.D., Chase, C.A., Johanns, A.M., Liebman, M., 2012. Increasing Cropping System Diversity Balances Productivity , Profitability and Environmental Health. PLoS One 7, 1–8. https://doi.org/10.1371/journal.pone.0047149 de la Cruz, V.Y. V., Tantriani, Cheng, W., Tawaraya, K., 2023. Yield gap between organic and conventional farming systems across climate types and sub-types: A meta-analysis. Agric. Syst. 211. https://doi.org/10.1016/j.agsy.2023.103732 de Lara, A., Mieno, T., Luck, J.D., Puntel, L.A., 2023. Predicting site ‑ specific economic optimal nitrogen rate using machine learning methods and on ‑ farm precision experimentation. Precis. Agric. 24, 1792–1812. https://doi.org/10.1007/s11119-023-10018-8 De Ponti, T., Rijk, B., Van Ittersum, M.K., 2012. The crop yield gap between organic and conventional agriculture. Agric. Syst. 108, 1–9. https://doi.org/10.1016/j.agsy.2011.12.004 Dentika, P., Ozier-Lafontaine, H., Penet, L., 2021. Weeds as pathogen hosts and disease risk for crops in the wake of a reduced use of herbicides: Evidence from yam (Dioscorea alata) fields and colletotrichum pathogens in the tropics. J. Fungi 7. https://doi.org/10.3390/jof7040283 Dessureault-Rompré, J., Zebarth, B.J., Georgallas, A., Burton, D.L., Grant, C.A., Drury, C.F., 2010. Temperature dependence of soil nitrogen mineralization rate: Comparison of mathematical models, reference temperatures and origin of the soils. Geoderma 157, 97– 108. https://doi.org/10.1016/j.geoderma.2010.04.001 References 86 DESTATIS, 2024. Landwirtschaftliche Betriebe insgesamt und Betriebe mit ökologischem Landbau nach Bundesländern. Statistisches Bundesamt [WWW Document]. URL https://www.destatis.de/DE/Themen/Branchen-Unternehmen/LandwirtschaftForstwirtschaft-Fischerei/Landwirtschaftliche-Betriebe/Tabellen/oekologischer-landbaubundeslaender.html (accessed 3.14.25). Diacono, M., Castrignanò, A., Vitti, C., Stellacci, A.M., Marino, L., Cocozza, C., De Benedetto, D., Troccoli, A., Rubino, P., Ventrella, D., 2014. An approach for assessing the effects of site-specific fertilization on crop growth and yield of durum wheat in organic agriculture. Precis. Agric. 15, 479–498. https://doi.org/10.1007/s11119-014-9347-8 DMK, 2023. Flächenerträge von Körnermais und Silomais in Deutschland. Deutsches Maiskomitee e.V. [WWW Document]. URL https://www.maiskomitee.de/Fakten/Statistik/Deutschland/Flächenerträge (accessed 4.15.23). Döring, T.F., Neuhoff, D., 2021. Upper limits to sustainable organic wheat yields. Sci. Rep. 11, 1–11. https://doi.org/10.1038/s41598-021-91940-7 Duru, M., Therond, O., Martin, G., Martin-Clouaire, R., Magne, M.A., Justes, E., Journet, E.P., Aubertot, J.N., Savary, S., Bergez, J.E., Sarthou, J.P., 2015. How to implement biodiversity-based agriculture to enhance ecosystem services: a review. Agron. Sustain. Dev. 35, 1259–1281. https://doi.org/10.1007/s13593-015-0306-1 DWD, 2024. Index of /climate_environment/CDC/observations_germany/climate/daily/more_precip/historical. Deutscher Wetterdienst [WWW Document]. URL https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/daily /more_precip/historical/ DWD, 2020. Climate dataset - archive data station 342 Belm. Deutscher Wetterdienst [WWW Document]. Eli-Chukwu, N.C., 2019. Applications of Artificial Intelligence in Smart Agriculture: A Review. Eng. Technol. Appl. Sci. Res. 9, 4377–4383. https://doi.org/10.1007/978-981-16-82483_11 Elsalahy, H.H., Bellingrath-Kimura, S.D., Roß, C.L., Kautz, T., Döring, T.F., 2020. Crop Resilience to Drought With and Without Response Diversity. Front. Plant Sci. 11. https://doi.org/10.3389/fpls.2020.00721 Engels, A.M., Gaiser, T., Ewert, F., Grahmann, K., 2025. Simulating Soil Moisture Dynamics in a Diversified Cropping System Under Heterogeneous Soil Conditions. Agronomy 15. https://doi.org/10.3390/agronomy15020407 Esposito, M., Westbrook, A.S., Maggio, A., Cirillo, V., DiTommaso, A., 2023. Neutral weed communities: The intersection between crop productivity, biodiversity, and weed ecosystem services. Weed Sci. 1, 273–299. https://doi.org/10.1017/wsc.2023.27 European Union, 2020. Farm to Fork Strategy. https://doi.org/https://doi.org/10.2875/653604 Eurostat, 2025. Organic crop area by agricultural production methods and crops [WWW Document]. https://doi.org/https://doi.org/10.2908/ORG_CROPAR FAO, 2024. Land statistics 2001–2022 - Global, regional and country trends. Food and Agriculture Organization of the United Nations, Rome. https://doi.org/https://doi.org/10.4060/cd1484en FAO, 2022. In Brief to The State of Food and Agriculture 2022. Leveraging automation in agriculture for transforming agrifood systems. Rome,. Food and Agriculture Organization of the United Nations, Rome. https://doi.org/10.4060/cc2459en References 87 FAO, 2015. World reference base for soil resources 2014 international soil classification system for naming soils and creating legends for soil maps - Update 2015, World Soil Resources Reports No. 106. Food and Agriculture Organization of the United Nations, Rome. Fernández-Delgado, M., Cernadas, E., Barro, S., Amorim, D., 2014. Do we need hundreds of classifiers to solve real world classification problems? J. Mach. Learn. Res. 15, 3133– 3181. Fernández-Quintanilla, C., Peña, J.M., Andújar, D., Dorado, J., Ribeiro, A., López-Granados, F., 2018. Is the current state of the art of weed monitoring suitable for site-specific weed management in arable crops? Weed Res. 58, 259–272. https://doi.org/10.1111/wre.12307 Fortes, R., Millán, S., Prieto, M.H., Campillo, C., 2015. A methodology based on apparent electrical conductivity and guided soil samples to improve irrigation zoning. Precis. Agric. 16, 441–454. https://doi.org/10.1007/s11119-015-9388-7 Fox, J., 2015. Applied Regression Analysis and Generalized Linear Models, 2. ed. Sage publications. Francksen, R.M., Turnbull, S., Rhymer, C.M., Hiron, M., Bufe, C., Klaus, V.H., Newell-Price, P., Stewart, G., Whittingham, M.J., 2022. The Effects of Nitrogen Fertilisation on Plant Species Richness in European Permanent Grasslands: A Systematic Review and MetaAnalysis. Agronomy 12. https://doi.org/10.3390/agronomy12122928 Franzluebbers, A.J., Zentella, R., Kafle, A., 2025. Soil-profile fertility is altered by soil texture and land use across physiographic regions in the southeastern United States. Agron. J. 1–21. https://doi.org/10.1002/agj2.70041 Fussell, J., Rundquist, D., Harrington, J.A., 1986. On defining remote sensing. Photogramm. Eng. Remote Sens. 52, 1507–1511. Gabriel, A., Gandorfer, M., 2023. Adoption of digital technologies in agriculture—an inventory in a european small-scale farming region. Precis. Agric. 24, 68–91. https://doi.org/10.1007/s11119-022-09931-1 Gao, C., Ji, X., He, Q., Gong, Z., Sun, H., Wen, T., Guo, W., 2023. Monitoring of Wheat Fusarium Head Blight on Spectral and Textural Analysis of UAV Multispectral Imagery. Agric. 13, 1–16. https://doi.org/10.3390/agriculture13020293 Geiger, F., Bengtsson, J., Berendse, F., Weisser, W.W., Emmerson, M., Morales, M.B., Ceryngier, P., Liira, J., Tscharntke, T., Winqvist, C., Eggers, S., Bommarco, R., Pärt, T., Bretagnolle, V., Plantegenest, M., Clement, L.W., Dennis, C., Palmer, C., Oñate, J.J., Guerrero, I., Hawro, V., Aavik, T., Thies, C., Flohre, A., Hänke, S., Fischer, C., Goedhart, P.W., Inchausti, P., 2010. Persistent negative effects of pesticides on biodiversity and biological control potential on European farmland. Basic Appl. Ecol. 11, 97–105. https://doi.org/10.1016/j.baae.2009.12.001 Geologischer Dienst NRW, 2003. Geologie im Weserund Osnabrücker Bergland, 11th ed. Geologischer Dienst NRW, Krefeld. Gerhards, R., Andújar Sanchez, D., Hamouz, P., Peteinatos, G.G., Christensen, S., Fernandez-Quintanilla, C., 2022. Advances in site-specific weed management in agriculture—A review. Weed Res. 62, 123–133. https://doi.org/10.1111/wre.12526 Gerhards, R., Gutjahr, C., Weis, M., Keller, M., Sökefeld, M., Möhring, J., Piepho, H.P., 2012. Using precision farming technology to quantify yield effects attributed to weed competition and herbicide application. Weed Res. 52, 6–15. https://doi.org/10.1111/j.13653180.2011.00893.x References 94 Monti, M., Pellicanò, A., Santonoceto, C., Preiti, G., Pristeri, A., 2016. Yield components and nitrogen use in cereal-pea intercrops in Mediterranean environment. F. Crop. Res. 196, 379–388. https://doi.org/10.1016/j.fcr.2016.07.017 Moore, K.J., Moser, L.E., K. P. Vogel, S. S. Waller, B. E. Johnson, J. F. Pedersen, P., 1991. Describing and Quantifying Growth Stages of Perennial Forage Grasses. Agron. J. 1077, 1073–1077. Moran, M.S., Inoue, Y., Barnes, E.M., 1997. Opportunities and limitations for image-based remote sensing in precision crop management. Remote Sens. Environ. 61, 319–346. https://doi.org/10.1016/S0034-4257(97)00045-X Morari, F., Zanella, V., Sartori, L., Visioli, G., Berzaghi, P., Mosca, G., 2018. Optimising durum wheat cultivation in North Italy: understanding the effects of site-specific fertilization on yield and protein content. Precis. Agric. 19, 257–277. https://doi.org/10.1007/s11119-0179515-8 Munnaf, M.A., Haesaert, G., Mouazen, A.M., 2022. Site-specific seeding for maize production using management zone maps delineated with multi-sensors data fusion scheme. Soil Tillage Res. 220. https://doi.org/10.1016/j.still.2022.105377 Munnaf, M.A., Haesaert, G., Van Meirvenne, M., Mouazen, A.M., 2020a. Site-specific seeding using multi-sensor and data fusion techniques: A review, 1st ed, Advances in Agronomy. Elsevier Inc. https://doi.org/10.1016/bs.agron.2019.08.001 Munnaf, M.A., Haesaert, G., Van Meirvenne, M., Mouazen, A.M., 2020b. Map-based sitespecific seeding of consumption potato production using high-resolution soil and crop data fusion. Comput. Electron. Agric. 178. https://doi.org/10.1016/j.compag.2020.105752 Munz, S., Zachmann, J., Raj, I., Raj, N., Jens, D., Erik, H., Jensen, S., Carlsson, G., 2023. Yield stability and weed dry matter in response to field-scale soil variability in pea-oat intercropping. Plant Soil 506, 391–310. https://doi.org/10.1007/s11104-023-06316-9 Nahrstedt, K., Reuter, T., Trautz, D., Waske, B., 2024. Classifying Stand Compositions in Clover Grass Based on High-Resolution Multispectral UAV Images. Remote Sens. 16, 1– 18. https://doi.org/https://doi.org/10.3390/rs16142684 Neupane, J., Guo, W., 2019. Agronomic basis and strategies for precision water management: A review. Agronomy 9. https://doi.org/10.3390/agronomy9020087 Ngouajio, M., Lemieux, C., Leroux, G.D., 1999. Prediction of corn (Zea mays) yield loss from early observations of the relative leaf area and the relative leaf cover of weeds. Weed Sci. 47, 297–304. https://doi.org/10.1017/s0043174500091803 Niemeyer, M., Renz, M., Pukrop, M., Hagemann, D., Zurheide, T., Di Marco, D., Höferlin, M., Stark, P., Rahe, F., Igelbrink, M., Jenz, M., Jarmer, T., Trautz, D., Stiene, S., Hertzberg, J., 2024. Cognitive Weeding: An Approach to Single-Plant Specific Weed Regulation. KI - Künstliche Intelligenz. https://doi.org/10.1007/S13218-023-00825-6 Nikolić, N., Rizzo, D., Marraccini, E., Gotor, A.A., Mattivi, P., Saulet, P., Persichetti, A., Masin, R., 2021. Site-and time-specific early weed control is able to reduce herbicide use in maize-a case study. Ital. J. Agron. 16. https://doi.org/10.4081/ija.2021.1780 Oberson, A., Jarosch, K.A., Frossard, E., Hammelehle, A., Fliessbach, A., Mäder, P., Mayer, J., 2024. Higher than expected: Nitrogen flows, budgets, and use efficiencies over 35 years of organic and conventional cropping. Agric. Ecosyst. Environ. 362. https://doi.org/10.1016/j.agee.2023.108802 Odone, A., Popovic, O., Thorup-Kristensen, K., 2024. Deep roots: implications for nitrogen uptake and drought tolerance among winter wheat cultivars. Plant Soil 500, 13–32. https://doi.org/10.1007/s11104-023-06255-5 References 95 Oldoni, H., Magalhães, P.S.G., Oliveira, A.L.G., Lima, J.P., Figueiredo, G.K.D.A., Moro, E., Amaral, L.R., 2025. Management zones delineation: a proposal to overcome the croppasture rotation challenge. Precis. Agric. 26, 1–29. https://doi.org/10.1007/s11119-02410214-0 Oliveira, R.A., Näsi, R., Niemeläinen, O., Nyholm, L., Alhonoja, K., Kaivosoja, J., Jauhiainen, L., Viljanen, N., Nezami, S., Markelin, L., Hakala, T., Honkavaara, E., 2020. Machine learning estimators for the quantity and quality of grass swards used for silage production using drone-based imaging spectrometry and photogrammetry. Remote Sens. Environ. 246, 111830. https://doi.org/10.1016/j.rse.2020.111830 Oliver, M.A., Bishop, T.F.A., Marchant, B.P., 2013. Precision agriculture for sustainability and environmental protection, Precision Agriculture for Sustainability and Environmental Protection. https://doi.org/10.4324/9780203128329 Ouazaa, S., Jaramillo-Barrios, C.I., Chaali, N., Amaya, Y.M.Q., Carvajal, J.E.C., Ramos, O.M., 2022. Towards site specific management zones delineation in rotational cropping system: Application of multivariate spatial clustering model based on soil properties. Geoderma Reg. 30. https://doi.org/10.1016/j.geodrs.2022.e00564 Pahmeyer, C., Kuhn, T., Britz, W., 2021. ‘Fruchtfolge’: A crop rotation decision support system for optimizing cropping choices with big data and spatially explicit modeling. Comput. Electron. Agric. 181. https://doi.org/10.1016/j.compag.2020.105948 Palka, M., Manschadi, A.M., 2024. On-farm evaluation of a crop forecast-based approach for season-specific nitrogen application in winter wheat. Precis. Agric. 25, 2394–2420. https://doi.org/https://doi.org/10.1007/s11119-024-10175-4 On-farm Pang, X., Letey, J., 2000. Organic Farming: Challenge of Timing Nitrogen Availability to Crop Nitrogen Requirements. Soil Sci. Soc. Am. J. 64, 247–253. https://doi.org/10.2136/sssaj2000.641247x Pannacci, E., Tei, F., 2014. Effects of mechanical and chemical methods on weed control, weed seed rain and crop yield in maize, sunflower and soyabean. Crop Prot. 64, 51–59. https://doi.org/10.1016/j.cropro.2014.06.001 Papageorgiou, E.I., Markinos, A.T., Gemtos, T.A., 2011. Fuzzy cognitive map based approach for predicting yield in cotton crop production as a basis for decision support system in precision agriculture application. Appl. Soft Comput. J. 11, 3643–3657. https://doi.org/10.1016/j.asoc.2011.01.036 Pätzold, S., Hbirkou, C., Dicke, D., Gerhards, R., Welp, G., 2020. Linking weed patterns with soil properties: a long-term case study. Precis. Agric. 21, 569–588. https://doi.org/10.1007/s11119-019-09682-6 Paul, C., Bartkowski, B., Dönmez, C., Don, A., Mayer, S., Steffens, M., Weigl, S., Wiesmeier, M., Wolf, A., Helming, K., 2023. Carbon farming: Are soil carbon certificates a suitable tool for climate change mitigation? J. Environ. Manage. 330. https://doi.org/10.1016/j.jenvman.2022.117142 Paulsen, H.M., 2008. Mischfruchtanbausysteme mit Ölpflanzen im ökologischen Landbau 2. Ertragsstruktur des Mischfruchtanbaus von Lein (Linum ustitatissivum L.) mit Sommerweizen, Hafer oder Leindotter. Landbauforsch. - vTI Agric. For. Res. 4 4, 307– 314. Paulsen, H.M., 2007. Mischfruchtanbausysteme mit Ölpflanzen im ökologischen Landbau : 1. Ertragsstruktur des Mischfruchtanbaus von Leguminosen oder Sommerweizen mit Leindotter (Camelina sativa L. Crantz). Agric. For. Res. 57, 107–117. Paulsen, H.M., Schochow, M., 2007. Anbau von Mischkulturen mit Ölpflanzen zur Verbesserung der Flächenproduktivität im Ökologischen Landbau - Nährstoffaufnahme, Unkrautunterdrückung, Schaderregerbefall und Produktqualitäten. Bundesforschungsanstalt für Landwirtschaft (FAL), Westerau. References 96 Paulsen, H.M., Seling, S., 2007. Qualität von Getreide aus Mischfruchtanbausystemen mit Ölpflanzen im ökologischen Landbau. Landbauforsch. Völkenrode Sonderh. 309, 68–80. Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R., Dubourg, V., Vanderplas, J., Passos, A., Cournapeau, D., Brucher, M., Perrot, M., Duchesnay, É., 2011. Scikit-learn: Machine learning in Python. J. Mach. Learn. Res. 12, 2825–2830. Pei, H., Sun, Y., Huang, H., Zhang, W., Sheng, J., Zhang, Z., 2022. Weed Detection in Maize Fields by UAV Images Based on Crop. Agriculture 12, 975. https://doi.org/https://doi.org/10.3390/agriculture12070975 Peña, J.M., Torres-Sánchez, J., de Castro, A.I., Kelly, M., López-Granados, F., 2013. Weed Mapping in Early-Season Maize Fields Using Object-Based Analysis of Unmanned Aerial Vehicle (UAV) Images. PLoS One 8, 1–11. https://doi.org/10.1371/journal.pone.0077151 Peralta, N.R., Costa, J.L., Balzarini, M., Castro Franco, M., Córdoba, M., Bullock, D., 2015. Delineation of management zones to improve nitrogen management of wheat. Comput. Electron. Agric. 110, 103–113. https://doi.org/10.1016/j.compag.2014.10.017 Petrovic, B., Kononets, Y., Csambalik, L., 2025. Adoption of drone, sensor, and robotic technologies in organic farming systems of Visegrad countries. Heliyon 11. https://doi.org/10.1016/j.heliyon.2024.e41408 Pöttker, M., Kiehl, K., Jarmer, T., Trautz, D., 2023. Convolutional Neural Network Maps Plant Communities in Semi-Natural Grasslands Using Multispectral Unmanned Aerial Vehicle Imagery. Remote Sens. 15. https://doi.org/10.3390/rs15071945 R Core Team, 2020. R: A Language and Environment for Statistical Computing R. Foundation for Statistical Computing Version 4.0.2. Raatz, L., Bacchi, N., Pirhofer, K., Glemnitz, M., Müller, M.E.H., Joshi, J., Scherber, C., 2019. How much do we really lose ?— Yield losses in the proximity of natural landscape elements in agricultural landscapes. Ecol. Evol. 9, 7838–7848. https://doi.org/10.1002/ece3.5370 Rajcan, I., Swanton, C.J., 2001. Understanding maize-weed competition: Resource competition, light quality and the whole plant. F. Crop. Res. 71, 139–150. https://doi.org/10.1016/S0378-4290(01)00159-9 Rajmis, S., Karpinski, I., Pohl, J.P., Herrmann, M., Kehlenbeck, H., 2022. Economic potential of site-specific pesticide application scenarios with direct injection and automatic application assistant in northern Germany. Precis. Agric. 23, 2063–2088. https://doi.org/10.1007/s11119-022-09888-1 Raseduzzaman, M., Jensen, E.S., 2017. Does intercropping enhance yield stability in arable crop production? A meta-analysis. Eur. J. Agron. 91, 25–33. https://doi.org/10.1016/j.eja.2017.09.009 Rasmussen, J., Azim, S., Boldsen, S.K., Nitschke, T., Jensen, S.M., Nielsen, J., Christensen, S., 2021. The challenge of reproducing remote sensing data from satellites and unmanned aerial vehicles (UAVs) in the context of management zones and precision agriculture. Precis. Agric. 22, 834–851. https://doi.org/10.1007/s11119-020-09759-7 Rasmussen, Jesper, Griepentrog, H.W., Nielsen, J., Henriksen, C.B., 2012. Automated intelligent rotor tine cultivation and punch planting to improve the selectivity of mechanical intra-row weed control. Weed Res. 52, 327–337. https://doi.org/10.1111/j.13653180.2012.00922.x Rasmussen, Jim, Søegaard, K., Pirhofer-Walzl, K., Eriksen, J., 2012. N2-fixation and residual N effect of four legume species and four companion grass species. Eur. J. Agron. 36, 66– 74. https://doi.org/10.1016/j.eja.2011.09.003 References 97 Raymond Hunt, E., Daughtry, C.S.T., Eitel, J.U.H., Long, D.S., 2011. Remote sensing leaf chlorophyll content using a visible band index. Agron. J. 103, 1090–1099. https://doi.org/10.2134/agronj2010.0395 Redwitz, C. Von, Andert, S., Strehlow, B., Ulber, L., Bensch, J., Forster, R., Schatke, M., 2025. Enhancing Arable Weed Diversity by Reduced Herbicide Use ? J. Crop Heal. https://doi.org/10.1007/s10343-025-01127-7 Reiss, E.R., Drinkwater, L.E., 2018. Cultivar mixtures: a meta-analysis of the effect of intraspecific diversity on crop yield. Ecol. Appl. 28, 62–77. https://doi.org/10.1002/eap.1629 Reumaux, R., 2024. Constraints and opportunities for organic crop production in areas of high agricultural productivity. Acta Universitatis Agriculturae Sueciae. Swedish University of Agricultural Sciences. https://doi.org/10.54612/a.1gli1g6skv Reuter, T., Morales, J.C.S., Tieben, C., Nahrstedt, K., Kraatz, F., Meemken, H., Hünker, G., Lingemann, K., Broll, G., Jarmer, T., Hertzberg, J., Trautz, D., 2023. Evaluation of a decision support system for the recommendation of pasture harvest date and form, in: Informatik in Der Land-, ForstUnd Ernährungswirtschaft -- Referate Der 43. GILJahrestagung. Christa Hoffmann; Anthony Stein; Arno Ruckelshausen; Henning Müller; Thilo Steckel; Helga Floto, Bonn, pp. 489–494. Richardson, K., Steffen, W., Lucht, W., Bendtsen, J., Cornell, S.E., Donges, J.F., Drüke, M., Fetzer, I., Bala, G., von Bloh, W., Feulner, G., Fiedler, S., Gerten, D., Gleeson, T., Hofmann, M., Huiskamp, W., Kummu, M., Mohan, C., Nogués-Bravo, D., Petri, S., Porkka, M., Rahmstorf, S., Schaphoff, S., Thonicke, K., Tobian, A., Virkki, V., WangErlandsson, L., Weber, L., Rockström, J., 2023. Earth beyond six of nine planetary boundaries. Sci. Adv. 9, 1–16. https://doi.org/10.1126/sciadv.adh2458 Richner, N., Holderegger, R., Linder, H.P., Walter, T., 2015. Reviewing change in the arable flora of Europe: A meta-analysis. Weed Res. 55, 1–13. https://doi.org/10.1111/wre.12123 Riesinger, P., Herzon, I., 2010. Symbiotic nitrogen fixation in organically managed red clovergrass leys under farming conditions. Acta Agric. Scand. Sect. B Soil Plant Sci. 60, 517– 528. https://doi.org/10.1080/09064710903233870 Rodriguez Miranda, D.A., de Oliveira Alari, F., Oldoni, H., Bazzi, C.L., do Amaral, L.R., Graziano Magalhães, P.S., 2021. Delineation of management zones in integrated crop– livestock systems. Agron. J. 113, 5271–5286. https://doi.org/10.1002/agj2.20912 Roilo, S., Engler, J.O., Václavík, T., Cord, A.F., 2023. Landscape-level heterogeneity of agrienvironment measures improves habitat suitability for farmland birds. Ecol. Appl. 33, 1– 15. https://doi.org/10.1002/eap.2720 Roilo, S., Hofmeester, T.R., Frauendorf, M., Widén, A., Cord, A.F., 2024. The untapped potential of camera traps for farmland biodiversity monitoring: current practice and outstanding agroecological questions. Remote Sens. Ecol. Conserv. 1–12. https://doi.org/10.1002/rse2.426 Royston, P., 1995. Remark AS R94 : A Remark on Algorithm AS 181: The W-test for Normality. R. Stat. Soc. 44, 547–551. https://doi.org/https://doi.org/10.2307/2986146 Sadras, V.O., Calviño, P.A., 2001. Quantification of grain yield response to soil depth in soybean, maize, sunflower, and wheat. Agron. J. 93, 577–583. https://doi.org/10.2134/agronj2001.933577x Saikai, Y., Patel, V., Mitchell, P.D., 2020. Machine learning for optimizing complex site-specific management. Comput. Electron. Agric. 174, 105381. https://doi.org/10.1016/j.compag.2020.105381 References 98 Saile, M., Spaeth, M., Gerhards, R., 2022. Evaluating Sensor-Based Mechanical Weeding Combined with Pre-and Post-Emergence Herbicides for Integrated Weed Management in Cereals. Agronomy 12. https://doi.org/10.3390/agronomy12061465 Samuel, A.L., 2000. Some studies in machine learning using the game of checkers. IBM J. Res. Dev. 44, 206–226. https://doi.org/10.1147/rd.441.0206 Sanders, J., Brinkmann, J., Chmelikova, L., Ebertseder, F., Freibauer, A., Gottwald, F., Haub, A., Hauschild, M., Hoppe, J., Hülsbergen, K.J., Jung, R., Kusche, D., Levin, K., March, S., Schmidtke, K., Stein-Bachinger, K., Treu, H., Weckenbrock, P., Wiesinger, K., Gattinger, A., Heß, J., 2025. Benefits of organic agriculture for environment and animal welfare in temperate climates. Org. Agric. https://doi.org/10.1007/s13165-025-00493-w Sapkota, A., Verdi, A., Scudiero, E., Montazar, A., 2024. Assessing the effectiveness of satellite and UAV-based remote sensing for delineating alfalfa management zones under heterogeneous rootzone soil salinity. Smart Agric. Technol. 9. https://doi.org/10.1016/j.atech.2024.100583 Schaack, D., Rampold, C., 2021. AMI Markt Bilanz Öko-Landbau 2021 (in German). Bonn. Schaan, L.N., Finch, E.A., Wartenberg, A.C., Boettner, V.S., Bellingrath-Kimura, S.D., Bonn, A., Pe’er, G., 2025. Mapping and prioritising landscape feature restoration in agricultural landscapes: A case study in Brandenburg, Germany. Land use policy 154. https://doi.org/10.1016/j.landusepol.2025.107531 Schulp, C.J.E., Verburg, P.H., 2009. Effect of land use history and site factors on spatial variation of soil organic carbon across a physiographic region. Agric. Ecosyst. Environ. 133, 86–97. https://doi.org/10.1016/j.agee.2009.05.005 Schuster, J., Hagn, L., Mittermayer, M., Hülsbergen, K.J., 2024. After effects of historical grassland on soil organic carbon content and plant growth in croplands in southern Germany determined using satellite data. Sci. Total Environ. 947. https://doi.org/10.1016/j.scitotenv.2024.174507 Seitz, S., Goebes, P., Puerta, V.L., Pereira, E.I.P., Wittwer, R., Six, J., van der Heijden, M.G.A., Scholten, T., 2019. Conservation tillage and organic farming reduce soil erosion. Agron. Sustain. Dev. 39. https://doi.org/10.1007/s13593-018-0545-z Selfors, L., Werts, P., Green, T., 2018. Looking beyond the jug: Non-chemical weed seedbank management. Crop. Soils 51, 28–53. https://doi.org/10.2134/cs2018.51.0504 Shams, M.Y., Gamel, S.A., Talaat, F.M., 2024. Enhancing crop recommendation systems with explainable artificial intelligence: a study on agricultural decision-making. Neural Comput. Appl. 36, 5695–5714. https://doi.org/10.1007/s00521-023-09391-2 Sharma, L.K., Bali, S.K., 2018. A review of methods to improve nitrogen use efficiency in agriculture. Sustain. 10, 1–23. https://doi.org/10.3390/su10010051 Shukla, M.K., Sharma, P., 2023. Fuzzy K-Means and Principal Component Analysis for Classifying Soil Properties for Efficient Farm Management and Maintaining Soil Health. Sustain. 15. https://doi.org/10.3390/su151713144 Sirami, C., Gross, N., Baillod, A.B., Bertrand, C., Carrié, R., Hass, A., Henckel, L., Miguet, P., Vuillot, C., Alignier, A., Girard, J., Batáry, P., Clough, Y., Violle, C., Giralt, D., Bota, G., Badenhausser, I., Lefebvre, G., Gauffre, B., Vialatte, A., Calatayud, F., Gil-Tena, A., Tischendorf, L., Mitchell, S., Lindsay, K., Georges, R., Hilaire, S., Recasens, J., SoléSenan, X.O., Robleño, I., Bosch, J., Barrientos, J.A., Ricarte, A., Marcos-Garcia, M.Á., Miñano, J., Mathevet, R., Gibon, A., Baudry, J., Balent, G., Poulin, B., Burel, F., Tscharntke, T., Bretagnolle, V., Siriwardena, G., Ouin, A., Brotons, L., Martin, J.L., Fahrig, L., 2019. Increasing crop heterogeneity enhances multitrophic diversity across agricultural regions. Proc. Natl. Acad. Sci. U. S. A. 116, 16442–16447. https://doi.org/10.1073/pnas.1906419116 References 99 Skendži, S., Zovko, M., Leši, V., Živković, I.P., Lemic, D., 2023. Detection and Evaluation of Environmental Stress in Winter Wheat Using Remote and Proximal Sensing Methods and Vegetation Indices—A Review. Divers. 15, 481. https://doi.org/https://doi.org/10.3390/ d15040481 Skierucha, W., Wilczek, A., Szypłowska, A., Sławiński, C., Lamorski, K., 2012. A TDR-based soil moisture monitoring system with simultaneous measurement of soil temperature and electrical conductivity. Sensors (Switzerland) 12, 13545–13566. https://doi.org/10.3390/s121013545 Smith, O.M., Cohen, A.L., Rieser, C.J., Davis, A.G., Taylor, J.M., Adesanya, A.W., Jones, M.S., Meier, A.R., Reganold, J.P., Orpet, R.J., Northfield, T.D., Crowder, D.W., 2019. Organic Farming Provides Reliable Environmental Benefits but Increases Variability in Crop Yields: A Global. Front. Sustain. Food Syst. 3, 1–10. https://doi.org/10.3389/fsufs.2019.00082 Smolka, M., Puchberger-Enengl, D., Bipoun, M., Klasa, A., Kiczkajlo, M., Śmiechowski, W., Sowiński, P., Krutzler, C., Keplinger, F., Vellekoop, M.J., 2017. A mobile lab-on-a-chip device for on-site soil nutrient analysis. Precis. Agric. 18, 152–168. https://doi.org/10.1007/s11119-016-9452-y Soltani, N., Dille, A.J., Burke, I.C., Everman, W.J., VanGessel, M.J., Davis, V.M., Sikkema, P.H., 2016. Potential corn yield losses due to weeds in North America. Weed Technol. 30, 979–984. https://doi.org/https://doi.org/10.1614/WT-D-16-00046.1 Song, H., He, Y., 2005. Crop Nutrition Diagnosis Expert System Based on Artificial Neural Networks, in: Proceedings of the Third International Conference on Information Technology and Applications (ICITA’05). Sonntag, W.I., Wienrich, N., Severin, M., Schulze Schwering, D., 2022. Precision Farming – Nullnummer oder Nutzbringer? Berichte über Landwirtschaft. Sonobe, R., Yamaya, Y., Tani, H., Wang, X., Kobayashi, N., Mochizuki, K., 2018. Crop classification from Sentinel-2-derived vegetation indices using ensemble learning. J. Appl. Remote Sens. 12, 1. https://doi.org/10.1117/1.jrs.12.026019 Sowiński, J., 2023. Effect of mechanical tillage treatments intensity on weed infestation and yield of quinoa (Chenopodium quinoa Willd.). Crop Prot. 172. Spaeth, M., Sökefeld, M., Schwaderer, P., Gauer, M.E., Sturm, D.J., Delatrée, C.C., Gerhards, R., 2024. Smart sprayer a technology for site-specific herbicide application. Crop Prot. 177. https://doi.org/10.1016/j.cropro.2023.106564 Speranza, E.A., Mendonça, D., Manoel, C., Vaz, P., Ribeiro, L., Rabelo, L.M., Andr, L., Castro, D., Chagas, S., Schelp, M.X., Vecchi, L., 2023. Delineating Management Zones with Different Yield Potentials in Soybean – Corn and Soybean – Cotton Production Systems. AgriEngineering 5, 1481–1497. https://doi.org/https://doi.org/10.3390/agriengineering5030092 Stein-Bachinger, K., Preißel, S., Kühne, S., Reckling, M., 2022. More diverse but less intensive farming enhances biodiversity. Trends Ecol. Evol. 37, 395–396. https://doi.org/10.1016/j.tree.2022.01.008 Steinmann, H.H., 2002. Impact of harrowing on the nitrogen dynamics of plants and soil. Soil Tillage Res. 65, 53–59. https://doi.org/10.1016/S0167-1987(01)00278-1 Sutcliffe, L.M.E., Schellenberg, J., Meyer, S., Leuschner, C., 2024. Close to the edge: Spatial variation in plant diversity, biomass and floral resources in conventional and agrienvironment cereal fields. J. Appl. Ecol. 2075–2086. https://doi.org/10.1111/13652664.14737 References 100 Suter, M., Connolly, J., Finn, J.A., Loges, R., Kirwan, L., Sebastià, M.T., Lüscher, A., 2015. Nitrogen yield advantage from grass-legume mixtures is robust over a wide range of legume proportions and environmental conditions. Glob. Chang. Biol. 21, 2424–2438. https://doi.org/10.1111/gcb.12880 Swanton, C.J., Mahoney, K.J., Chandler, K., Gulden, R.H., 2008. Integrated Weed Management: Knowledge-Based Weed Management Systems. Weed Sci. 56, 168–172. https://doi.org/10.1614/ws-07-126.1 Thomsen, I.K., Schjønning, P., Olesen, J.E., Christensen, B.T., 2003. C and N turnover in structurally intact soils of different texture. Soil Biol. Biochem. 35, 765–774. https://doi.org/10.1016/S0038-0717(03)00093-2 Tilman, D., 2020. Benefits of intensive agricultural intercropping. Nat. Plants 6, 604–605. https://doi.org/10.1038/s41477-020-0677-4 Topographic-map.com, 2025. Topografische Karte Osnabrück [WWW Document]. URL https://de-de.topographic-map.com/maph66mt/Osnabrück/?center=52.21198%2C8.06741 Torres-Sánchez, J., López-Granados, F., De Castro, A.I., Peña-Barragán, J.M., 2013. Configuration and Specifications of an Unmanned Aerial Vehicle (UAV) for Early Site Specific Weed Management. PLoS One 8, 1–15. https://doi.org/10.1371/journal.pone.0058210 Tscharntke, T., Grass, I., Wanger, T.C., Westphal, C., Batáry, P., 2021. Beyond organic farming – harnessing biodiversity-friendly landscapes. Trends Ecol. Evol. 36, 919–930. https://doi.org/10.1016/j.tree.2021.06.010 Tuck, S.L., Winqvist, C., Mota, F., Ahnström, J., Turnbull, L.A., Bengtsson, J., 2014. Land-use intensity and the effects of organic farming on biodiversity: A hierarchical meta-analysis. J. Appl. Ecol. 51, 746–755. https://doi.org/10.1111/1365-2664.12219 Tucker, C.J., 1979. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sens. Environ. 8, 127–150. https://doi.org/10.1016/00344257(79)90013-0 Tuomisto, H.L., Hodge, I.D., Riordan, P., Macdonald, D.W., 2012. Does organic farming reduce environmental impacts? - A meta-analysis of European research. J. Environ. Manage. 112, 309–320. https://doi.org/10.1016/j.jenvman.2012.08.018 Tursun, N., Datta, A., Sakinmaz, M.S., Kantarci, Z., Knezevic, S.Z., Chauhan, B.S., 2016. The critical period for weed control in three corn (Zea mays L.) types. Crop Prot. 90, 59–65. https://doi.org/10.1016/j.cropro.2016.08.019 Umweltstiftung Michael Otto, Deutscher Bauernverband e.V, 2023. F.R.A.N.Z. Zwischenbilanz 2023 – Aktuelle Erkenntnisse aus dem F.R.A.N.Z.- Projekt. 54. Usowicz, B., Lipiec, J., 2017. Spatial variability of soil properties and cereal yield in a cultivated field on sandy soil. Soil Tillage Res. 174, 241–250. https://doi.org/10.1016/j.still.2017.07.015 Vallentin, C., Dobers, E.S., Itzerott, S., Kleinschmit, B., Spengler, D., 2020. Delineation of management zones with spatial data fusion and belief theory. Precis. Agric. 21, 802–830. https://doi.org/10.1007/s11119-019-09696-0 Van Eerd, L.L., Chahal, I., Peng, Y., Awrey, J.C., 2023. Influence of cover crops at the four spheres: A review of ecosystem services, potential barriers, and future directions for North America. Sci. Total Environ. 858. https://doi.org/10.1016/j.scitotenv.2022.159990 Vandermeer, J., 1989. The ecology of intercropping. Cambridge University Press. References 101 Vannoppen, A., Gobin, A., 2022. Estimating Yield from NDVI, Weather Data, and Soil Water Depletion for Sugar Beet and Potato in Northern Belgium. Water (Switzerland) 14, 1–15. https://doi.org/10.3390/w14081188 Vasileiadis, V.P., Otto, S., van Dijk, W., Urek, G., Leskovšek, R., Verschwele, A., Furlan, L., Sattin, M., 2015. On-farm evaluation of integrated weed management tools for maize production in three different agro-environments in Europe: Agronomic efficacy, herbicide use reduction, and economic sustainability. Eur. J. Agron. 63, 71–78. https://doi.org/10.1016/j.eja.2014.12.001 VDLUFA, 2007. Handbuch der landwirtschaftlichen Versuchs-und Untersuchungsmethodik (VDLUFA-Methodenbuch). Verband Deutscher Landwirtschaftlicher Untersuchungsund Forschungsanstalten e. V.-Verlag, Germany. Vidican, R., Mălinaș, A., Ranta, O., Moldovan, C., Marian, O., Ghețe, A., Ghișe, C.R., Popovici, F., Cătunescu, G.M., 2023. Using Remote Sensing Vegetation Indices for the Discrimination and Monitoring of Agricultural Crops: A Critical Review. Agronomy 13, 1– 27. https://doi.org/10.3390/agronomy13123040 Vikuk, V., Spirkaneder, A., Noack, P., Duemig, A., 2024. Smart Agricultural Technology Validation of a sensor-system for real-time measurement of mineralized nitrogen in soils. Smart Agric. Technol. 7, 100390. https://doi.org/10.1016/j.atech.2023.100390 Viljanen, N., Honkavaara, E., Näsi, R., Hakala, T., Niemeläinen, O., Kaivosoja, J., 2018. A novel machine learning method for estimating biomass of grass swards using a photogrammetric canopy height model, images and vegetation indices captured by a drone. Agric. 8. https://doi.org/10.3390/agriculture8050070 Walker, E., Wooliver, R., Russo, L., Jagadamma, S., 2024. The context-dependent benefits of organic farming on pollinator biodiversity: A meta-analysis. J. Appl. Ecol. 62, 41–52. https://doi.org/10.1111/1365-2664.14826 Walsh, M.J., Squires, C.C., Coleman, G.R.Y., Widderick, M.J., McKiernan, A.B., Chauhan, B.S., Peressini, C., Guzzomi, A.L., 2020. Tillage based, site-specific weed control for conservation cropping systems. Weed Technol. 34, 704–710. https://doi.org/10.1017/wet.2020.34 Wegener, J.K., Urso, L.-M., von Hörsten, D., Hegewald, H., Minßen, T.-F., Schattenberg, J., Gaus, C.-C., de Witte, T., Nieberg, H., Isermeyer, F., Frerichs, L., Backhaus, G.F., 2019. Spot farming – an alternative for future plant production. J. für Kult. 71, 69–69. https://doi.org/10.5073/JfK.2019.04.01 Welch, B.L., 1947. The generalization of “students” problem when several different population variances are involved. Biometrika 34, 28–35. https://doi.org/10.1093/biomet/34.1-2.28 Wezel, A., Casagrande, M., Celette, F., Vian, J.-F., Ferrer, A., Peigné, J., 2014. Agroecological practices for sustainable agriculture. A review. Agron. Sustain. Dev. 34, 1–20. https://doi.org/10.1007/s13593-013-0180-7 Wierman, M.J., Dobransky, M.K., 1993. A review of: “NEURAL NETWORKS AND FUZZY SYSTEMS:” A Dynamic Systems Approach to Machine Intelligence. Int. J. Gen. Syst. 20, 37–41. https://doi.org/10.1080/03081079208945045 Wiesmeier, M., Urbanski, L., Hobley, E., Lang, B., von Lützow, M., Marin-Spiotta, E., van Wesemael, B., Rabot, E., Ließ, M., Garcia-Franco, N., Wollschläger, U., Vogel, H.J., Kögel-Knabner, I., 2019. Soil organic carbon storage as a key function of soils - A review of drivers and indicators at various scales. Geoderma 333, 149–162. https://doi.org/10.1016/j.geoderma.2018.07.026 Wiles, L.J., 2009. Beyond patch spraying: Site-specific weed management with several herbicides. Precis. Agric. 10, 277–290. https://doi.org/10.1007/s11119-008-9097-6 References 102 Willer, H., Trávníček, J., Meier, Cl., Schlatter, B. (Eds. ., 2021. The World of Organic Agriculture. Statistics and Emerging Trends 2021. Int. J. Sustain. High. Educ. https://doi.org/10.1108/ijshe.2009.24910aae.004 Willett, W., Rockström, J., Loken, B., Springmann, M., Lang, T., Vermeulen, S., Garnett, T., Tilman, D., DeClerck, F., Wood, A., Jonell, M., Clark, M., Gordon, L.J., Fanzo, J., Hawkes, C., Zurayk, R., Rivera, J.A., De Vries, W., Majele Sibanda, L., Afshin, A., Chaudhary, A., Herrero, M., Agustina, R., Branca, F., Lartey, A., Fan, S., Crona, B., Fox, E., Bignet, V., Troell, M., Lindahl, T., Singh, S., Cornell, S.E., Srinath Reddy, K., Narain, S., Nishtar, S., Murray, C.J.L., 2019. Food in the Anthropocene: the EAT–Lancet Commission on healthy diets from sustainable food systems. Lancet 393, 447–492. https://doi.org/10.1016/S0140-6736(18)31788-4 Williams, J.T., 1963. Chenopodium Album L. J. Ecol. 51, 711. https://doi.org/10.2307/2257758 Woźniak, A., 2020. Mechanical and chemical weeding effects on the weed structure in durum wheat. Ital. J. Agron. 15, 102–108. https://doi.org/10.4081/ija.2020.1559 Wuepper, D., Huber, R., 2022. Comparing effectiveness and return on investment of actionand results-based agri-environmental payments in Switzerland. Am. J. Agric. Econ. 104, 1585–1604. https://doi.org/10.1111/ajae.12284 Wulanningtyas, H.S., Gong, Y., Li, P., Sakagami, N., Nishiwaki, J., Komatsuzaki, M., 2021. A cover crop and no-tillage system for enhancing soil health by increasing soil organic matter in soybean cultivation. Soil Tillage Res. 205, 104749. https://doi.org/10.1016/j.still.2020.104749 Xu, K., Su, Y., Liu, J., Hu, T., Jin, S., Ma, Q., Zhai, Q., Wang, R., Zhang, J., Li, Y., Liu, H., Guo, Q., 2020. Estimation of degraded grassland aboveground biomass using machine learning methods from terrestrial laser scanning data. Ecol. Indic. 108, 105747. https://doi.org/10.1016/j.ecolind.2019.105747 Yu, Y., Stomph, T.-J., Makowski, D., van der Werf, W., 2015. Temporal niche differentiation increases the land equivalent ratio of annual intercrops: A meta-analysis. F. Crop. Res. 184, 133–144. https://doi.org/10.1016/j.fcr.2015.09.010 Yu, Y., Stomph, T.-J., Makowski, D., Zhang, L., van der Werf, W., 2016. A meta-analysis of relative crop yields in cereal/legume mixtures suggests options for management. F. Crop. Res. 198, 269–279. https://doi.org/10.1016/j.fcr.2016.08.001 Zeng, L., Chen, C., 2018. Using remote sensing to estimate forage biomass and nutrient contents at different growth stages. Biomass and Bioenergy 115, 74–81. https://doi.org/10.1016/j.biombioe.2018.04.016 Zhao, F., Jiao, L., Liu, H., 2013. Kernel generalized fuzzy c-means clustering with spatial information for image segmentation. Digit. Signal Process. 23, 184–199. https://doi.org/10.1016/j.dsp.2012.09.016 Zhao, Y., Tian, Y., Li, X., Song, M., Fang, X., Jiang, Y., Xu, X., 2022. Nitrogen fixation and transfer between legumes and cereals under various cropping regimes. Rhizosphere 22. https://doi.org/10.1016/j.rhisph.2022.100546 Zhu, H., Chen, X., Zhang, Y., 2013. Temporal and spatial variability of nitrogen in rice-wheat rotation in field scale. Environ. Earth Sci. 68, 585–590. https://doi.org/10.1007/s12665012-1762-4 Zhu, X.X., Tuia, D., Mou, L., Xia, G.S., Zhang, L., Xu, F., Fraundorfer, F., 2017. Deep Learning in Remote Sensing: A Comprehensive Review and List of Resources. IEEE Geosci. Remote Sens. Mag. 5, 8–36. https://doi.org/10.1109/MGRS.2017.2762307 Zingsheim, M.L., Döring, T.F., 2024. What weeding robots need to know about ecology. Agric. Ecosyst. Environ. 364, 108861. https://doi.org/10.1016/j.agee.2023.108861 Acknowledgements 103 Acknowledgements Throughout my research journey, including field trials, analysis, and writing, I received considerable assistance and support. Therefore, I extend my gratitude to everyone involved. Thank you for your feedback on my presentations and written work. I am grateful for the discussions regarding methodology and data interpretation, as well as the hours spent in the field, especially, sorting clover and grass leaves. You deserve life long cookies! Your open ears for my complaints and emotional support were invaluable. Furthermore, I appreciate the enjoyable moments shared during fieldwork, office hours, conferences, and at home. Thank you for your companionship! During this journey, I encountered numerous inspiring and nice individuals. I wish to express my sincere thanks to Hochschule Osnabrück University of Applied Sciences for providing the facilities, funding, and support. Specifically, I am grateful to the members of Trial Farm Waldhof and the Sustainable Agricultural Land Use working group: Maria, Mareike, David, Janis, Tim, Phyu Phyu, Tobias, Niko, Lisa, Hannes, Maike, and Maik. I greatly valued my time with this working group and their unwavering support. A significant portion of this thesis was contributed through the "Agro-Nordwest" project, funded by the Federal Ministry of Food and Agriculture (BMEL). The experiences within this project and the people I met therein were enriching, and I learned extensively. I extend special thanks to my co-authors, Insa Kühling, Thomas Jarmer, and Konstantin Nahrstedt. Insa Kühling has significantly aided in my academic development since my bachelor's studies. Thom and Konstantin provided new, “remote” perspectives, and our discussions on methodology, data interpretation, and Led Zeppelin were immensely constructive. From my perspective, Konstantin and I formed an excellent team, from data collection in the field to presenting the results. I also appreciate the help and proofreading efforts of my friends: Kathrin, Kaffee Kathi, Irmi, Levent, Juan-Carlos and Melanie. More importantly, I value their emotional support. Thank you! A special acknowledgment goes to Gabriele Broll for supervising my promotion. I have learned a great deal from you in writing scientific texts and appreciate your expertise and clear guidance. My profound thanks extend to Dieter Trautz for his unwavering support in resolving troubles and securing necessary funding. I particularly appreciate your dedication to fostering a welcoming working culture. I believe I had one of the best supervisors I could wish for. Lastly, but certainly not least, I am deeply grateful to my family for supporting me throughout my life and providing me the opportunity to attend university and complete my Ph.D. Dankeschön!