Mapping seasonal flood-recession cropland extent in the Senegal River Valley

Mapping seasonal flood-recession cropland extent in the Senegal River Valley Laurent Bruckmann a,b,* , Andrew Ogilvie c , Didier Martin c , Finda Bayo Diakhat´ e d , Amaury Tilmant a,b a Department of Civil and Water Engineering, Universit´ e Laval, Quebec, Canada b CentrEau, Water Research Center, Quebec, Canada c G-EAU, AgroParisTech, BRGM, Cirad, INRAE, Institut Agro, IRD, Univ Montpellier, Montpellier, France d ISRA, BAME, Dakar, Senegal ABSTRACT Flood-recession agriculture (FRA) represents a crucial source of livelihood for numerous communities across Africa who reside near expansive floodplains and wetlands. However, it is currently insufficiently monitored. In this study, we present a methodology for mapping FRA harvested areas in the Senegal River Valley that is both reproducible and scalable. Our methodology entails the integration of optical and radar data from Sentinel platforms, conducted through a multitemporal analysis with a seasonal focus, and the application of the Random Forest algorithm. The results, supported by a kappa coefficient of 91.9%, demonstrate the first comprehensive mapping of FRA in the Senegal River valley, conducted between 2019 and 2023. This mapping facilitates the identification of the hydrological factors that influence FRA harvesting. The results of the analyses have demonstrated the importance of interannual variability in the cultivated areas of FRA, which range from 14,000 to 75,000 ha depending on the intensity of the annual flood. The duration and flooded extension are the primary factors that regulate the cropping pattern of FRA over the floodplain. The flood duration must be around 35 days to permit the cultivation, with growth generally starting between 10 and 30 November. In consideration of these findings, we recommend that future water management strategies and rural development initiatives give due consideration to FRA, to enhance the visibility of farmers. 1. Introduction A global analysis reveals that 24% of croplands are situated in floodplain areas (Dryden et al., 2021). Floodplains of large tropical rivers are essential areas for the provision of ecosystem services (De Groot et al., 2018) and for livelihoods (Asare et al., 2022). In Africa, flood-based farming systems cover approximately 25–30 million ha (Kool et al., 2018), and 3 million people depend directly on floodplains for agriculture (Richter et al., 2010). One of the most prevalent systems is floodplain agriculture (FRA), also known as recession agriculture. This system utilizes the residual moisture from extended annual floods to cultivate flood plains following the flood’s recession (Mollard and Walter, 2008). In Africa, this practice is observed along major rivers, including the Senegal River. FRA supports the production of rice in Niger (Barbier et al., 2011), maize and sorghum in Senegal (Saarnak, 2003; Bruckmann, 2018), and various other crops, including fruits and vegetables (Adamczewski et al., 2011). For example, flood-recession sorghum covers approximately 2 million ha in Africa (Kebe et al., 2001). FRA is part of complex livelihood systems that also include fishing and grazing, enhancing food production and diversity (Motsumi et al., 2012). As FRA is primarily harvested during the dry season in semi-arid environments, it significantly contributes to food security, providing either household consumption or cash income (Sidib´ e et al., 2016; Balana et al., 2019). In the mid-Zambezi floodplain, FRA constitutes 40% of the income for agricultural households * Corresponding author. Department of Civil and Water Engineering, Universit´ e Laval, Quebec, Canada E-mail address: [email protected] (L. Bruckmann). Contents lists available at ScienceDirect Remote Sensing Applications: Society and Environment journal homepage: www.elsevier.com/locate/rsase https://doi.org/10.1016/j.rsase.2025.101473 Received 9 October 2024; Received in revised form 10 January 2025; Accepted 23 January 2025 Remote Sensing Applications: Society and Environment 37 (2025) 101473 Available online 30 January 2025 2352-9385/© 2025 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license ( http://creativecommons.org/licenses/by-nc-nd/4.0/ ). (Chimweta et al., 2022). However, flood-based farming systems in Africa have been largely overlooked by development policies, with greater focus placed on full control irrigation systems (Kool et al., 2018; Steenbergen et al., 2011; Sidib´ e et al., 2016), even though considering them could offer opportunities on the long-term sustainability of floodplains (Zenebe et al., 2022). In the Senegal River Valley (SRV), flood-recession agriculture was the dominant agricultural system before irrigation was introduced in the 1960s, with cultivated areas ranging from 110,000 to 180,000 ha annually (Crousse et al., 1991). The establishment of irrigation systems in the 1970s, supported by the Manantali and Diama dams in the 1980s, was aimed at mitigating the effects of prolonged droughts and regulating river flows. However, these dams have altered the hydrological regime, reduced flood volumes, and increased irregularity (Sall et al., 2020; Bruckmann, 2021). Despite this, FRA remains crucial for local food security and complements irrigated crops (Bruckmann, 2018; Poussin et al., 2020). The recent Master Plan (OMVS, 2022) has identified at least four proposed dam developments that give rise to concerns about the potential for flooding to be lost as a result of upstream discharge control. Monitoring annually cultivated FRA surfaces is essential to assess the impacts of future dams and optimize water allocation. While irrigation monitoring is managed by Senegalese and Mauritanian national companies (SAED and SONADER), flood-recession monitoring is underdeveloped, partly due to challenges and lack of stakeholder interest. Although there have been remote sensing experiments in the early 2000s (Man´ e and Fraval, 2001), and short-term research programs like AGRICORA have explored hydro-agrological functioning on specific basins (Poussin et al., 2020; Sall et al., 2020), a comprehensive, large-scale, interannual FRA monitoring and mapping program has yet to be implemented. Mapping and monitoring FRA can help balance agricultural and water needs, and guide water management decisions on how to meet the contradictory water demands of hydroelectric production and peak flow support (Fraval et al., 2002; Bader et al., 2003; Tilmant et al., 2020). Accurate data can aid in modeling optimal trade-offs, benefiting local livelihoods and ecosystems, and potentially reducing hydroelectric production impacts from upstream dams (Raso et al., 2020). The research on nature-based solutions demonstrates the significant value of floodplains for ecosystems and food security, emphasizing the importance of floodplain agriculture for livelihoods and sustainable development goals (Singh et al., 2021). Conversely, large-scale irrigation programs have seen limited success in the SRV (Poussin, 1998; Garcia-Bola˜ nos et al., 2011), prompting a reevaluation of rural development strategies in the region. Recent efforts have aimed at mapping potential FRA areas, such as identifying flood-prone zones in Ethiopia using multisource remote sensing (Gumma et al., 2022) or applying a Bayesian approach to estimate FRA probability in Kenya and Ethiopia (Liman Harou et al., 2020). While these studies provide valuable insights, they don’t capture actual FRA cultivation. Other research has focused on mapping flooded areas in West African floodplains, including Senegal, as the foundational basis for supporting FRA using datasets like MODIS, Landsat, and Sentinel (Ogilvie et al., 2015; Ogilvie et al., 2020; Bruckmann et al., 2022;Ogilvie et al., 2025 ). However, mapping a single type of agriculture in a heterogeneous agricultural landscape poses significant challenges. In the Senegal River basin, FRA occurs during the season following both the rainy and flood seasons. The landscape features a mix of irrigation, rainfed crops, forest, and grass, which can overlap with FRA between late October and early December. Furthermore, FRA crops are sown at different times along the floodplain as the agricultural calendar is defined by the local floodwater dynamics. Effective FRA mapping must prioritize a methodology that combines flood and agricultural mapping, grounded in a deep understanding of the hydro-agroecological processes that drive it. The intricate water-vegetation dynamics, also seen in areas like irrigated perimeters, require sophisticated approaches. Various methods have been applied to map seasonal vegetation or agricultural land use in such complex environments. For instance, rice has been a focal point due to its overlap with other crops during the growing stage. Mapping efforts should consider using specific spectral bands like the red edge (Jiang et al., 2021) or a multitemporal approach (Gumma et al., 2014; Dong et al., 2015). Integrating data from multiple sources can enhance the accuracy of rice field detection (Zhao et al., 2021). Temporal analysis is also valuable for examining flood pulse dynamics and their effects on landscape and vegetation (Hu et al., 2015; Kool et al., 2022). Machine learning is increasingly utilized to identify and map specific land uses across the globe, particularly in diverse agricultural landscapes. The choice of classification methodology is critical, and numerous studies have sought to determine the most effective approaches in various contexts. Gxokwe et al. (2020) reviewed methods for semi-arid wetlands mapping using Sentinel-2 and found that Random Forest (RF) and artificial neural networks, combined with classification and regression tree (CART), delivered the best results. This aligns with the findings by Thanh Noi and Kappas (2017), who compared non-parametric classifiers like RF, k-Nearest Neighbor (kNN), and Support Vector Machine (SVM) for land cover mapping in the Red River Delta. Another approach to map seasonal vegetation involves using multitemporal image series to aid classification. In the Cambodian Mekong floodplains, Orieschnig et al. (2021) identified RF and Gradient Boosting Trees (GTB) as the most accurate classifiers. Incorporating intra-seasonal data, such as phenology, can further improve wetland classification precision (Tian et al., 2016). One limitation of using machine learning and multitemporal data over large floodplains is the computational demand of analyzing numerous images. Cloud computing platforms, like Google Earth Engine (GEE), can overcome these challenges by offering vast storage and computational power. GEE is widely used today, with many case studies highlighting its effectiveness in applications, such as large-scale wetland mapping (Wang et al., 2020; Gxokwe et al., 2022). This study aims to develop an approach capable of mapping seasonal flood-recession agriculture in the Senegal River Valley to monitor the temporal trends of this vital agricultural livelihood. The main contributions of this work are first the development of an adaptable workflow for FRA mapping that combines the benefits of using well-known machine learning algorithms and a multitemporal approach inside GEE cloud-computing platform. The workflow is designed to be easily replicated and adapted in other floodrecession agricultural regions. Furthermore, this approach utilizes open-access data from Sentinel-1 and Sentinel-2 satellites, thereby ensuring that the methodology is not only accessible but also cost-effective for researchers and practitioners worldwide. Secondly, we utilize spatial analysis and extensive locally collected ground truth data to provide a detailed description of the spatial distribution and extent of FRA areas in the Senegal floodplain during recent years, from 2019 to 2023. Furthermore, we analyze the spatial drivers conditioning the implementation of FRA in SRV with optical data, including the characteristics of flooding in space (extent) and time L. Bruckmann et al. Remote Sensing Applications: Society and Environment 37 (2025) 101473 2 Fig. 1. – Map of the Senegal River Basin and Valley. White boxes are the four focus areas presented below; Black circles are the main cities of the valley; Dark blue is permanent waters defined using Sentinel-2 and MNDWI. Background from Google Satellite. L. Bruckmann et al. Remote Sensing Applications: Society and Environment 37 (2025) 101473 3 (start, end, duration), to assess the threshold that initiates the cultivation of FRA areas. The study aims to analyze spatial patterns and temporal changes in FRA, providing scientific evidence to support the sustainable development of floodplains and address the waterenergy-food-ecosystem nexus conflicts in the region. 2. Study area: the Senegal river valley The Senegal River basin is a large transboundary watershed (337,000 km 2 ) shared by four countries: Guinea, Mali, Mauritania, and Senegal. In the downstream part, the Senegal River Valley is a large floodplain of 10–20 km wide and 790 km long. Topography is flat, with an average decline of 1–3 cm per km (Bader et al., 2014). The valley water resource is allogenic due to its location downstream of the three primary tributaries: the Fal´ em´ e, Bafing, and Bakoye. The Senegal floodplain is delimited between the two towns of Bakel upstream and Dagana downstream (Fig. 1). The annual flood is fed by rainfall in the upper basin. In the Valley, rainfall is modest and varies from 250 mm in the north to 600 mm in the south. The flood occurs between the months of August and November, and the timing of the peak depends on the location along the river as well as the amplitude of the flood, with larger floods propagating less fast downstream. FRA is practiced in the floodplain after the flood, with farmers sowing seeds progressively as the flood recedes. Crops are sown between September and November depending on the date the water retreats from land in the floodplain. Farmers grow mostly sorghum, maize, cowpeas, and sweet potatoes. The harvest happens between February and March, depending on the year and the distance downstream. In the SRV, three types of crop farming are distinguished according to the origin and control of the water resource: rain-fed crops, irrigated crops under total control, and flood-recession crops. Irrigation is practiced throughout the Senegal Fig. 2. – Points for calibration and validation over the 4 focus areas. L. Bruckmann et al. Remote Sensing Applications: Society and Environment 37 (2025) 101473 4 River Valley, but mainly in the downstream part. Rainfed farming is found mostly in the southern part of the valley due to the minimal annual rainfall in the northern, Sahelian part of the watershed. The floodplain has a surface area of approximately 400,000 ha (4000 km 2 ) of which 200,000 ha are intended for irrigation (OMVS, 2018). To enhance understanding of FRA, the paper focuses on four distinct zones within the floodplain (Fig. 1). Zone 1 is situated in the vicinity of Podor, the northern city in Senegal (Fig. 2a). It encompasses a FRA area situated to the west of the urban area. This is the area examined by Ogilvie et al. (2020) and Poussin et al. (2020). The second zone is situated north of Ndioum city (Fig. 2b), and zones 3 and 4 are located in the central valley respectively around the villages of Mboumba (Fig. 2c) and Diaba Dekele (Fig. 2d). 3. Data and methods The FRA mapping methodology comprises four steps. The initial step is to collect ground truth points for different land uses. Spectral band ratio indices from optical and radar imagery are employed in step two to comprehend the behavior of flood recession crops on an intra-seasonal scale and to define sub-seasons specific to flood recession agriculture. In step three, a synthetic composite image of the indices is created in order to train a machine-learning algorithm. Finally, the accuracy of the results is evaluated prior to extrapolation to subsequent years. 3.1. Data 3.1.1. Calibration and validation data One crucial step is the establishment of a database of regions of interest, which is utilized by the algorithm to train and validate the classification. Data were collected from high-resolution Google Earth images in 2019, mainly in December. We gathered 2951 points across 7 land uses (Table 1 & Fig. 2), split 70/30 for calibration and validation. Additionally, 100 ground-truth points from floodrecession plots were included from a December 2022 field trip along the SRV floodplain between Bakel and Podor. Points were acquired using a handheld GPS device within plots where recession crops were grown in 2022. 3.1.2. Satellite images and cloud-computing Data were analyzed using GEE due to the extensive area under study and the numerous images available. We utilized Sentinel-2 MSI images accessible via GEE, employing the level 2 surface reflectance product, which became available for our area on December 16, 2018. Sentinel-2 images are processed with cloud masking by using the CLOUDY_PIXEL_PERCENTAGE property to retain images with less than 40% cloudy pixels and the cloud mask probability product. The COPERNICUS/S2_CLOUD_PROBABILITY Image Collection is derived from Sentinel-2 data and provides information about the likelihood of cloud presence for each pixel in an image. Images with more than 50% of cloud probability were removed from our image collection. Additionally, we employed the Sentinel-1 Ground Range Detected images. The integration of radar data and optical imagery enables the differentiation of vegetation, particularly forest and tree cover, during flood periods, thereby enhancing the accuracy of classification (Steinhausen et al., 2018; Slagter et al., 2020). All Sentinel-1 GRD in GEE have undergone preprocessing, which involves the steps implemented by the Sentinel-1 Toolbox to derive the backscatter coefficient in each pixel. We created a homogeneous subset by filtering using some metadata properties: all images are IW (Interferometric Wide Swath) and from an ascending orbit. While longer datasets like Landsat ETM+and OLI or MODIS provide broader coverage, Sentinel images are preferable due to their finer spatial resolution. This resolution is crucial for accurately mapping small flood-recession cultivated areas influenced by hydrological, pedological, and micro-topographic conditions. Additionally, Sentinel’s frequent revisits allow for intra-seasonal analysis, which helps in distinguishing and classifying land cover based on its spectral signature throughout the season. Both Sentinel-2 and Sentinel-1 have high revisit frequencies (up to 5 days for S2 and 6 days for S1) and high resolution from 10 to 20 m. The image collection used is composed of images from 7 tiles for Sentinel-2 which are half-shared between S2A and S2B. For Sentinel-1 all the images are from S1A. For the calibration year 2019, we used 568 Sentinel-2 and 120 Sentinel-1 images between July and February. Details about images used for each yearly analysis can be found in Appendix 1. Table 1 – Dataset used for calibration and validation. Land use Number of points Flood-recession agriculture 674 Flood-recession basin uncultivated 400 Irrigated perimeters 537 Bare soil 462 Forest 531 Other natural vegetation 160 Water 187 Total 2951 Calibration 2066 Validation 885 L. Bruckmann et al. Remote Sensing Applications: Society and Environment 37 (2025) 101473 5 3.1.3. Optical and radar indexes Several optical indices were tested for classification based on their temporal and spectral signatures. These indices were selected for their ability to detect water, moisture, vegetation, and bare soil (Table 2). MNDWI, a refinement of the Normalized Water Difference Index (NDWI), enhances water detection while minimizing the influence of built-up areas and soil, which is crucial in Sahelian zones where water is shallow (Campos et al., 2012). Studies have effectively used MNDWI to delineate flooded areas in the SRV (Ogilvie et al., 2020; Ogilvie et al., 2025). The Normalized Difference Moisture Index (NDMI) monitors soil and vegetation moisture, with higher values indicating healthy vegetation with greater water content. Three complementary vegetation indices are also used: the widely used Normalized Difference Vegetation Index (NDVI) and the Modified Soil Adjusted Vegetation Index (MSAVI). MSAVI, which incorporates a soil adjustment factor, is advantageous in regions such as the SRV, where it provides a more accurate assessment of sparsely vegetated pixels. The Normalized Difference Red Edge Index (NDRE) is sensitive to chlorophyll and accounts for different vegetation growth stages, reducing confusion in vegetation mapping (Jiang et al., 2021). Finally, the Bare Soil Index (BSI) is used to detect variations in land cover. Table 2 – Indexes used in the study. Indice Formula with Sentinel bands Pixel resolution Fullname Source MNDWI (B3-B11)/(B3+B11) 20 Modified Normalized Difference Water Index Xu, n.d. NDMI (B8-B11)/(B8+B11) 20 Normalized Difference Moisture Index Gao (1996) BSI ((B11+B4)-(B8+B2))/((B11+B4)+(B8+B2)) 20 Bare Soil Index Rikimaru et al. (2002) NDVI (B8-B4)/(B8-B4) 10 Normalized Difference Vegetation Index Rouse et al. (1973) MSAVI (2 * NIR +1 - sqrt(pow((2 * NIR +1), 2) - 8 * (NIR - RED)))/2 10 Modified Soil Adjusted Vegetation Index Qi et al. (1994) NDRE (B8-B5)/(B8+B5) 10 Normalized Difference Red Edge Barnes et al. (2000) VV/VH VV/VH 10 Polarization bands ratio Slagter et al., 2020 VV +VH VV +VH 10 Polarization bands sum  VV-VH VV-VH 10 Polarization bands difference  Fig. 3. – Temporal behavior of five different land uses on different optical and radar indices used in the study. (Number of sampling points used per land use: FRA =675, PI =542, Forest =619, Bare soil =675, Uncultivated flooded areas =407). L. Bruckmann et al. Remote Sensing Applications: Society and Environment 37 (2025) 101473 6 For radar, we use indices derived from arithmetic operations on the vertical transmit, horizontal receive (VH), and vertical transmit, vertical receive (VV) channels. Coand cross-polarized backscatter intensities (VV and VH) are useful for identifying crop types, especially rice (Nguyen et al., 2016; Mandal et al., 2018). SAR data are also commonly used for forest detection due to their stability within a single season. The indices used include the VV/VH ratio, VV-VH difference, and VV +VH addition, which enhance vegetation detection and forest area discrimination by accounting for soil structure and texture throughout the season. Previous studies have shown that combining SAR and optical data improves the accuracy of LULC mapping in wetlands (Slagter et al., 2020), as these data sources complement each other (Veloso et al., 2017; Orieschnig et al., 2021). Inclusion of the VV/VH difference or ratio can also improve accuracy (Abdikan et al., 2016). 3.2. Method 3.2.1. Definition of sub-seasons by using temporal series of indexes To define and map seasonal crops whose extent is partly determined by the annual flood, we focused on the intra-seasonal variations in the spectral signatures of various elements (water, humidity, vegetation, bare soil). By analyzing the sub-seasonal temporal (Fig. 3) and spatial (Fig. 4) variations of these spectral signals, we aimed to enhance the detection of FRA. The multi-temporal MNDWI curve for flood-recession agriculture (FRA) areas in 2019 shows distinct patterns: MNDWI values are positive between September and late October during flooding but remain around −0.4 otherwise. As expected, MNDWI remains negative in bare soils, while in flooded forests and irrigated zones MNDWI values can temporarily exceed zero but remain lower than in flooded areas. NDMI follows a similar trend, increasing during flooding and decreasing afterwards. In irrigated areas, the NDMI remains high for much longer than in FRA areas. The Bare Soil Index (BSI) is useful as flood recession areas are bare outside the cropping Fig. 4. – Average seasonal value for the various indexes used in Zone 2 – Ndioum from July 2019 to February 2020. L. Bruckmann et al. Remote Sensing Applications: Society and Environment 37 (2025) 101473 7 periods, with BSI around 0.2 when bare and dropping negative when cropped. The NDVI profile is low before the flood, negative during the flood, and increases during crop growth to 0.2–0.4 in December and January. In contrast, other vegetated areas maintain higher NDVI values throughout the season. MSAVI and NDRE mirror NDVI patterns, with irrigation resulting in higher values for the vegetation indices. Radar indices are effective in detecting flooded areas, as water causes significant signal changes, in contrast to the stable signal from forests. For instance, the VV/VH ratio increases in flooded areas due to strong VV backscatter and weak VH backscatter. Based on visual analysis, the 2019 flooded agricultural season can be divided into four sub-periods (Fig. 5): (1) Pre-flood (July to mid-August), with bare soil and no water or vegetation; (2) Floodi (mid-August to late October), featuring extensive water coverage; (3) Post-flood (November), with receding water and bare soil; (4) post-flood vegetative period (December to February), when crops and natural vegetation reach peak growth. This division helps in identifying land-use patterns and distinguishing FRAs from other areas. Indeed, the sharp and shorter peak observed in the water indices distinguishes flood recession areas from irrigated areas, forests, and bare soils, while the peak in the vegetation indices in January–February helps to distinguish FRA from uncultivated flooded areas. In other years, these sub-periods are adjusted to align with the flood period, using a consistent visual approach to define the temporal boundaries. The map of average index values for the different sub-seasons is presented in Fig. 6, which focuses on a specific area within the SRV. It can be observed that some indexes, such as MNDWI, NDMI, or vegetation indexes, exhibit significant changes between subseasons. In contrast, the pre-flood period demonstrates minimal spatial variations compared to the other seasons. 3.2.2. Algorithm processing and postprocessing Before classification, a single image is generated to capture the spectral characteristics of the four sub-seasons. Six reducers—minimum, maximum, mean, standard deviation, skew, and kurtosis—are used to create one image per each sub-season, reflecting pixel distributions and central values. This process results in 36 images per sub-season from nine indices (three radar, six optical), totaling a 216-band image for each year. Subsequently, four machine learning algorithms integrated into GEE were tested to identify the most robust one. The algorithms were Random Forest (RF), Gradient Tree Boost (GTB), Classification and Regression Tree (CART), and k-Nearest Neighbor (KNN). The RF algorithm produced the most accurate results, although the difference in accuracy between it and GTB was minimal (Table 3). These findings are consistent with those of previous studies conducted by Gxokwe et al. (2020), (Thanh Noi and Kappas, 2017), and Orieschnig et al. (2021) on floodplains in Africa and Asia. Additionally, these findings are consistent with those of other studies conducted in semi-arid environments (Zhao et al., 2024) and in crop classification (Gupta et al., 2024). The Random Forest algorithm was therefore selected based on its performance in terms of accuracy score and its frequency as an optimal classification model in the existing literature, thus ensuring the desired level of reproducibility. The Random Forest algorithm is then applied to the composite image. Attempting the same method for a single season from August to March resulted in lower Fig. 5. – Average temporal behavior of different optical indexes over flood-recession agriculture areas in 2019/2020 (n =675). L. Bruckmann et al. Remote Sensing Applications: Society and Environment 37 (2025) 101473 8 Fig. 6. Average value for each sub-season of the various indices used in Zone 2 - Ndioum from July 2019 to February 2020. L. Bruckmann et al. Remote Sensing Applications: Society and Environment 37 (2025) 101473 9 5. Discussion 5.1. Water governance implications The results on FRA cultivated areas presented here are the first observations since studies completed in the 1980s and early 2000s under the Programme d’Optimisation de Gestion des R´ eservoirs (POGR). This program estimated, based on partial national statistics, that the average annual area of cultivated FRA between 1946 and 1999 was 52,000 ha (OMVS & IRD, 2002). This area varied according to the hydrological periods, reaching 110,000 ha (1100 km 2 ) over 1946–1970 during high-flood years and reducing to 46,000 ha (460 km 2 ) over 1970–1999 during the subsequent drought period when flood amplitudes reduced substantially (Bruckmann et al., 2022). Based on this data, Bader et al. (2003, 2015) proposed a model to estimate FRA areas based on discharge data during September at Bakel gauging station. This model has significantly influenced water allocation policies in the basin. Ogilvie et al. (2025) recently updated this relationship based on hydrological regressions and earth observations of the flooded areas in the SRV (Fig. 14). Our comparison of EO-based FRA estimates with this modeling approach reveals that our estimates are generally close to those derived from discharge data. The mapping proposed in this work has the potential to enhance comprehension of the annual and seasonal agricultural dynamics of the Senegal River valley. Moreover, it can optimize the water resource management at the basin level, as demonstrated by illustrating the importance of flood characteristics, such as duration and extent. Indeed, at the time of the Manantali dam construction, operational rules had set a target to implement an artificial flood to maintain 55,000 ha (550 km2) of FRA areas each year to balance flood support with energy production. This artificial flood has rarely been implemented, and since 2003 the Manantali dam has ceased to support flooding, with OMVS relying on unregulated tributaries to meet this objective. Our results suggest that this approach may be insufficient, as FRA areas were less than 22,000 ha (220 km 2 ) in 2021 and 2023. The 55,000 ha target was established in the 1980s to address drought-induced reductions in FRA potential while avoiding negative impacts on energy production. The target is therefore inappropriate for the current hydrological period and is now pursued with great urgency in all official documents, including the Integrated Water Management Project (2016) and the latest Master Plan (OMVS, 2022). Nevertheless, our recent results show that this target has been exceeded in three of the last five years, with a notable 36% increase in 2022, and 12 times since the early 2000s (Fig. 14), highlighting the potential and local interest in increasing FRA. This underscores the need for improved consideration of FRA areas in future water management, especially with the anticipated increase in dam construction, which may reduce flood volumes. Current water management and agricultural policies in the Senegal Fig. 14. – Annual FRA areas in SRV from 1950 to 2022 based on previous modelling by IRD compared to our study and average discharge at Bakel from August to October. Data: Ogilvie et al. (2025). L. Bruckmann et al. Remote Sensing Applications: Society and Environment 37 (2025) 101473 16 River Basin have primarily addressed drought, even as precipitation levels have risen since the late 1990s (Bodian et al., 2020). Updating these policies can better accommodate sectors like flood-based agriculture and fishing is essential. Our results emphasize the significant impact of water availability on FRA patterns. The future hydrological regime in the Senegal River Valley will be critical for FRA dynamics (Ogilvie et al., 2025; Bruckmann, 2018). However, there is a conflict between the water needs for FRA and the increasing water demands due to energy production and dry season irrigation. When flood conditions are favorable, FRA can cover over 75,000 ha (750 km2), which is considerable compared to the 130,000 ha (1330 km2) used annually for irrigation (PARACI data, OMVS, 2018). Finally, our study focuses on cultivated areas, but previous research highlighted the multifunctionality of flooded areas, including their crucial role in fishing and grazing (Horowitz et al., 1990; Bruckmann, 2018). Addressing trade-offs within the WEFE nexus must account for this complexity. 5.2. Assessing socio-economic and food security impacts in agriculture: the need for field-based data EO can significantly facilitate/enhance the assessment of large-scale cultivated areas, supporting informed decision-making. However, a detailed understanding of crop practices (type, distribution, and yield) is urgently needed. In the Senegal River Valley, primary crops include sorghum, cowpeas, and maize, but their distribution varies across the valley. In the upper regions, maize is the predominant crop, while downstream areas primarily cultivate sorghum, often alongside cowpeas. Yield data for these crops are sparsely documented. They can vary widely based on several factors, including the duration and timing of flooding, soil water reserves, pest presence, sowing period, weed prevalence, and available labor. Previous studies have reported notable discrepancies in yields. For instance, sorghum yields have been reported as low as 178 kg/ha (Saarnak, 2003) and as high as 280–330 kg/ha (Bruckmann, 2018; Poussin et al., 2020). These yields are relatively low, partly due to a simplified technical itinerary without fertilizers or plant protection products. The lack of detailed field-based data makes it challenging to ascertain the impact of annual FRA cropping on food security or household incomes. FRA plays a significant role in supplying households, with a large portion of the produce consumed locally. In Mauritania, it is estimated that in a good year, 26% of national cereal production comes from FRA sorghum production (Kebe, 2002). Furthermore, harvesting in February or March helps to ensure a steady distribution of cereals throughout the year. However, reports from the CILSS indicate that the region frequently faces food insecurity. Low flood years have a notable negative impact on food security across the valley. For instance, following years with particularly low floods, such as 2017/2018 or 2021/2022, regions in Senegal and Mauritania along the valley are classified by the CILSS as experiencing a nutritional crisis. Conversely, years with normal floods are typically considered under pressure. While other factors also influence food security, and a direct correlation between nutritional security and FRA is complex, the broader pattern is clear. These insights underscore the necessity for regular monitoring of flooded and flood-recession cultivated areas, as well as the collection of data on their production. Integrating and fusing localized insights of ground data with the extensive spatial coverage provided by EO data is crucial for the assessment of food security, the economic functionality of these areas, and the enhancement of overall water and food security policies. The provision of up-to-date, high-resolution data through remote sensing mapping can facilitate the sustainable and efficient management of water resources, which is of vital importance for farming communities that depend on natural flood irrigation. 6. Conclusion Flood Recession Agriculture (FRA) is a vital livelihood in African floodplains. Dependent on the annual flood, it remains highly vulnerable to hydrological variations over time and upstream changes including dam construction. The objective of this study was to provide an accurate mapping of the cultivated areas in the Senegal River valley. To achieve this, existing and readily available methodologies based on machine learning algorithms were mobilized for this novel application, which holds significant relevance for water and rural management in West Africa. The results of the study reveal that Random Forest (RF) algorithms combined with optical and radar indices allows for robust mapping and monitoring of annual FRA areas at a precise scale, facilitated by the pixel resolution of Sentinel images. While the classification accuracy is strong, additional ground truth data for other years and environments will allow refinement for a range of hydrological and land-use conditions. This consideration is important in the Sahelian context where spectral signature variability across years can be pronounced. Increasing the number of regions of interest is key to enhancing accuracy and facilitating more precise and context-specific monitoring. Over the past five years, FRA cultivation areas in the Senegal River valley have shown significant variability, ranging from 13,000 to 75,000 ha. The considerable degree of variability observed highlights the necessity for regular monitoring of FRA to manage water resources and mitigate the potential for food crises. The research also highlighted the interrelationship between inundation characteristics (timing and duration) and the dynamics of the FRA in space and time, indicating that FRA cultivation requires a flood duration L. Bruckmann et al. Remote Sensing Applications: Society and Environment 37 (2025) 101473 17 of between 20 and 40 days. Understanding these patterns is crucial, as it provides critical insights into how changes in flood regimes affect agricultural productivity and food security in flood-prone regions. The lack of comprehensive FRA monitoring has left farmers invisible and often neglected in water and rural policies in the Senegal basin. In this context, Earth Observation (EO) can play a pivotal role in bridging the data gap and fostering cooperation over water resources in large transboundary basins. The present study uses open-source and replicable tools such as GEE and the Random Forest algorithm to provide an objective, scalable, robust, and cost-effective application of EO for monitoring annual agricultural dynamics in large floodplains. EO data are critical for informing water and rural policies that support these communities. As evidenced by studies in the Senegal and Nile basins, a careful delineation of trade-offs between the demands for hydropower and agricultural demands is imperative. EO data for FRA monitoring provides a valuable tool to build trust among stakeholders and address the multifaceted demands of the Water-Energy-Food-Environment nexus. CRediT authorship contribution statement Laurent Bruckmann: Writing – review & editing, Writing – original draft, Visualization, Validation, Software, Resources, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Andrew Ogilvie: Writing – review & editing, Validation, Resources, Methodology, Investigation, Funding acquisition. Didier Martin: Validation, Resources, Investigation. Finda Bayo Diakhat´ e: Resources, Investigation. Amaury Tilmant: Supervision, Project administration, Funding acquisition. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgements This work was supported by the GoNEXUS project, which is funded by the European Union Horizon Programme call H2020-LCCLA-2018-2019-2020 - Grant Agreement Number 101003722, and the AFD Cycle de l’Eau et Changement Climatique (CECC) project. The participation of Canadian researchers in GoNEXUS was made possible by the New Frontiers Research Fund program (Canada) (grant NFRFG-2020-00430) and by the Fonds Nouvelles Frontieres en Recherche (Quebec) (grant 2022-FNFR310131). We gratefully acknowledge Google Earth Engine for the cloud computing and we thank the editor and the two anonymous reviewers for their comments that helped strengthen the manuscript. APPENDIX 1– Number of images used per year of analysis Appendix 1a Number of Sentinel-2 images per month used for analysis between 2019/2020 to 2023/2024 Month 2019/2020 2020/2021 2021/2022 2022/2023 2023/2024 7 81 50 83 55 49 8 67 70 51 35 40 9 59 52 65 46 69 10 79 89 77 84 78 11 81 81 72 60 81 12 83 71 55 97 80 1 60 80 58 66 69 2 58 64 67 58 71 Total 568 557 528 501 537 Appendix 1b Number of Sentinel-1 images per month used for analysis between 2019/2020 to 2023/2024 Month 2019/2020 2020/2021 2021/2022 2022/2023 2023/2024 7 13 14 13 15 14 8 17 15 17 13 17 9 15 15 13 17 13 10 16 15 17 13 16 11 14 16 14 17 13 12 16 17 14 15 17 1 15 17 13 17 16 2 14 12 14 12 12 Total 120 121 115 119 118 L. Bruckmann et al. Remote Sensing Applications: Society and Environment 37 (2025) 101473 18 APPENDIX 2– Accuracy values and confusion matrices Appendix 1 c Number of images used for analysis between 2019/2020 to 2023/2024 Sentinel-2 Tile 2019/2020 2020/2021 2021/2022/ 2022/2023 2023/2024 28PCC 30 33 28 25 27 28PDC 32 30 28 24 28 28PEC 34 32 34 27 32 28PFB 55 56 55 54 54 28PFC 63 61 61 54 58 28PGB 59 62 55 54 57 28PGC 67 56 59 54 58 28QCD 33 31 30 29 27 28QDD 65 66 60 64 67 28QED 33 35 27 28 33 28QFD 33 33 30 31 32 28QGD 64 62 61 57 64 Total 568 557 528 501 537 Appendix 2a Calibration Confusion Matrix for 2019/2020 using Random Forest algorithm Land cover FRA Irrigated perimeter Bare soil Flooded areas uncultivated Forest or aquatic vegetation Water FRA 463 0 0 0 1 0 Irrigated perimeter 0 365 0 0 1 0 Bare soil 0 0 318 0 0 0 Flooded areas uncultivated 0 1 1 285 0 0 Forest or aquatic vegetation 0 0 0 0 485 0 Water 0 0 0 0 0 128 Appendix 2b Validation Confusion Matrix for 2019/2020 using Random Forest algorithm Land cover FRA Irrigated perimeter Bare soil Flooded areas uncultivated Forest or aquatic vegetation Water FRA 203 0 0 5 2 0 Irrigated perimeter 1 155 0 0 15 0 Bare soil 1 1 135 5 2 0 Flooded areas uncultivated 16 0 7 83 2 5 Forest or aquatic vegetation 2 2 1 3 189 1 Water 0 0 0 2 0 57 L. Bruckmann et al. Remote Sensing Applications: Society and Environment 37 (2025) 101473 19 APPENDIX 3– Time series of water and vegetation indexes Appendix 3a. – Time serie of average MNDWI over FRA and flooded uncultivated areas between July 2019 to March 2024. Appendix 3b. – Time serie of average MSAVI over FRA and flooded uncultivated areas between July 2019 to March 2024. L. Bruckmann et al. Remote Sensing Applications: Society and Environment 37 (2025) 101473 20 Appendix 3c. – Time series of average MNDWI and NDVI values over different land cover during each year analyzed. L. Bruckmann et al. Remote Sensing Applications: Society and Environment 37 (2025) 101473 21 Appendix 3c. (continued). L. Bruckmann et al. Remote Sensing Applications: Society and Environment 37 (2025) 101473 22 Appendix 3c. (continued). L. Bruckmann et al. Remote Sensing Applications: Society and Environment 37 (2025) 101473 23 Data availability Data will be made available on request. References Abdikan, S., Sanli, F.B., Ustuner, M., Cal` o, F., 2016. Land cover mapping using sentinel-1 SAR data. Int. Arch. Photogram. Rem. Sens. Spatial Inf. Sci. 41, 757–761. Adamczewski, A., Hertzog, T., Dosso, M., Jouve, P., Jamin, J.Y., 2011. L’irrigation peut-elle se substituer aux cultures de d´ ecrue. Cah. Agric. 20, 97–104. https://doi. org/10.1684/agr.2011.0469. Asare, E., Mantyka-Pringle, C., Anderson, E., Belcher, K., Clark, R., 2022. Evaluating ecosystem services for agricultural wetlands: a systematic review and metaanalysis. Wetl. Ecol. Manag. 30, 1129–1149. https://doi.org/10.1007/s11273-022-09857-5. Bader, J.-C., Lamagat, J.-P., Guiguen, N., 2003. Gestion du barrage de Manantali sur le fleuve S´ en´ egal: analyse quantitative d’un conflit d’objectifs. Hydrol. Sci. J. 48, 525–538. Bader, J.-C., Cauchy, S., Duffar, L., Saura, P., 2014. Monographie hydrologique du fleuve S´ en´ egal. De l’origine des mesures jusqu’en 2011. In: Editio, I.R.D. (Ed.), IRD. Retrieved from. https://www.documentation.ird.fr/hor/fdi:010065190. Balana, B.B., Sanfo, S., Barbier, B., Williams, T., Kolavalli, S., 2019. Assessment of flood recession agriculture for food security in Northern Ghana: an optimization modelling approach. Agric. Syst. 173, 536–543. https://doi.org/10.1016/j.agsy.2019.03.021. Barbier, B., Ouedraogo, H., Demb´ el´ e, Y., Yacouba, H., Barry, B., Jamin, J.-Y., 2011. L’agriculture irrigu´ ee dans le Sahel ouest-africain. Cah. Agric. 20, 24–33. Appendix 3c. (continued). L. Bruckmann et al. Remote Sensing Applications: Society and Environment 37 (2025) 101473 24 Barnes, E., Clarke, T., Richards, S., Colaizzi, P., Haberland, J., Kostrzewski, M., Waller, P., Choi, C., Riley, E., Thompson, T., et al., 2000. Coincident detection of crop water stress, nitrogen status and canopy density using ground based multispectral data. Proceedings of the Fifth International Conference on Precision Agriculture 1619, 6. Bloomington, MN, USA, 16–19 July 2000. Bruckmann, L., 2018. Crue et d´ eveloppement rural dans la vall´ ee du S´ en´ egal : entre marginalisation et r´ esilience. Belgeo. Revue belge de g´ eographie. https://doi.org/ 10.4000/belgeo.23158. Bruckmann, L., 2021. La gestion partag´ ee de l’eau dans le bassin du fleuve S´ en´ egal : trajectoire, enjeux et perspectives. L’Ouest Saharien 13–14, 261–280. https://doi. org/10.3917/ousa.211.0261. Bruckmann, L., Delbart, N., Descroix, L., Bodian, A., 2022. Recent hydrological evolutions of the Senegal river flood (west Africa). Hydrol. Sci. J. 67, 385–400. https:// doi.org/10.1080/02626667.2021.1998511. Campos, J.C., Sillero, N., Brito, J.C., 2012. Normalized difference water indexes have dissimilar performances in detecting seasonal and permanent water in the Sahara–Sahel transition zone. J. Hydrol. 464–465, 438–446. https://doi.org/10.1016/j.jhydrol.2012.07.042. Chimweta, M., Nyakudya, I.W., Jimu, L., Musemwa, L., Mashingaidze, A.B., Musara, J.P., Kashe, K., 2022. Flood-recession cropping in the mid-Zambezi Valley: a neglected farming system with potential to improve household food security and income. Geogr. J. 188, 57–75. https://doi.org/10.1111/geoj.12418. Crousse, B., Mathieu, P., Seck, S.M., 1991. La Vall´ ee du fleuve S´ en´ egal: ´ evaluations et perspectives d’une d´ ecennie d’am´ enagements, 1980-1990. KARTHALA Editions. De Groot, D., Brander, L., Finlayson, C.M., 2018. Wetland ecosystem services. In: Finlayson, C.M., Everard, M., Irvine, K., McInnes, R.J., Middleton, B.A., Van Dam, A. A., Davidson, N.C. (Eds.), The Wetland Book. Springer, Netherlands, Dordrecht, pp. 323–333. https://doi.org/10.1007/978-90-481-9659-3_66. Dong, J., Xiao, X., Kou, W., Qin, Y., Zhang, G., Li, L., Jin, C., Zhou, Y., Wang, J., Biradar, C., Liu, J., Moore, B., 2015. Tracking the dynamics of paddy rice planting area in 1986–2010 through time series Landsat images and phenology-based algorithms. Remote Sens. Environ. 160, 99–113. https://doi.org/10.1016/j. rse.2015.01.004. Dryden, R., Anand, M., Lehner, B., Fluet-Chouinard, E., 2021. Do we prioritize floodplains for development and farming? Mapping global dependence and exposure to inundation. Global Environ. Change 71, 102370. https://doi.org/10.1016/j.gloenvcha.2021.102370. Fraval, P., Bader, J.-C., Man´ e, L.K., David-Benz, H., Lamagat, J.-P., Diop Diagne, O., 2002. The quest for integrated and sustainable water management in the Senegal River Valley. ENVIROWATER, 5` eme Conf´ erence Inter-R´ egionale sur l’Environnement et l’Eau, Ouagadougou. http://agritrop.cirad.fr/511230/. Gao, B., 1996. NDWI—a normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sens. Environ. 58, 257–266. https://doi. org/10.1016/S0034-4257(96)00067-3. García-Bola˜ nos, M., Borgia, C., Poblador, N., Dia, M., Seyid, O.M.V., Mateos, L., 2011. Performance assessment of small irrigation schemes along the Mauritanian banks of the Senegal River. Agric. Water Manag. 98, 1141–1152. https://doi.org/10.1016/j.agwat.2011.02.008. Gibb, S.A., 1987. Partners, Electricit´ e de France & Euroconsult. Etude de la gestion des ouvrages communs de l’OMVS. OMVS, Dakar, S´ en´ egal. Gumma, M.K., Thenkabail, P.S., Maunahan, A., Islam, S., Nelson, A., 2014. Mapping seasonal rice cropland extent and area in the high cropping intensity environment of Bangladesh using MODIS 500m data for the year 2010. ISPRS J. Photogrammetry Remote Sens. 91, 98–113. https://doi.org/10.1016/j.isprsjprs.2014.02.007. Gumma, M.K., Amede, T., Getnet, M., Pinjarla, B., Panjala, P., Legesse, G., Tilahun, G., Van Den Akker, E., Berdel, W., Keller, C., Siambi, M., Whitbread, A.M., 2022. Assessing potential locations for flood-based farming using satellite imagery: a case study of Afar region, Ethiopia. Renew. Agric. Food Syst. 37, S28–S42. https:// doi.org/10.1017/S1742170519000516. Gupta, Priyanka, Gupta, Prateek, Singh, S.K., Thakur, B., Kumar, M., 2024. Ensemble methods-based comparative study of Landsat 8 operational land imager (OLI) and sentinel 2 multi-spectral images (MSI) for smart farming crop classification. Multimed. Tool. Appl. https://doi.org/10.1007/s11042-024-19737-0. Gxokwe, S., 2022. Leveraging Google Earth Engine platform to characterize and map small seasonal wetlands in the semi-arid environments of South Africa. Sci. Total Environ. Gxokwe, S., Dube, T., Mazvimavi, D., 2020. Multispectral remote sensing of wetlands in semi-arid and arid areas: a review on applications, challenges and possible future research directions. Rem. Sens. 12, 4190. https://doi.org/10.3390/rs12244190. Horowitz, M.M., Salem-Murdock, M., Magistro, J., Niasse, M., Nuttall, C., 1990. Senegal River Basin Monitoring Activity: Synthesis Report. Institute for Development Anthropology, Binghamton, New York. Hu, Y., Huang, J., Du, Y., Han, P., Huang, W., 2015. Monitoring spatial and temporal dynamics of flood regimes and their relation to wetland landscape patterns in dongting lake from MODIS time-series imagery. Rem. Sens. 7, 7494–7520. https://doi.org/10.3390/rs70607494. Jiang, X., Fang, S., Huang, X., Liu, Y., Guo, L., 2021. Rice mapping and growth monitoring based on time series GF-6 images and red-edge bands. Rem. Sens. 13, 579. https://doi.org/10.3390/rs13040579. Kebe, B., 2002. Le rˆ ole du sorgho de d´ ecrue dans la securit´ e alimentaire au Sahel. La culture du sorgho de d´ ecrue en Afrique de l’Ouest et du Centre 23. Agence Espagnole de Coop´ eration Internationale, ISBN 84-7232-898-8. Kool, M., van Steenbergen, F., Mehari Haile, A., Abbas, Y.M., Hagos, E., 2018. The promise of flood-based farming systems in arid and semi-arid areas. In: Leal Filho, W., de Trincheria Gomez, J. (Eds.), Rainwater-Smart Agriculture in Arid and Semi-arid Areas: Fostering the Use of Rainwater for Food Security, Poverty Alleviation, Landscape Restoration and Climate Resilience. Springer International Publishing, pp. 77–94. https://doi.org/10.1007/978-3-319-66239-8_5. Kool, J., Lhermitte, S., Hrachowitz, M., Bregoli, F., McClain, M.E., 2022. Seasonal inundation dynamics and water balance of the Mara Wetland, Tanzania based on multi-temporal Sentinel-2 image classification. Int. J. Appl. Earth Obs. Geoinf. 109, 102766. https://doi.org/10.1016/j.jag.2022.102766. Liman Harou, I., Whitney, C., Kung’u, J., Luedeling, E., 2020. Mapping flood-based farming systems with bayesian networks. Land 9, 369. https://doi.org/10.3390/ land9100369. Mandal, D., Kumar, V., Bhattacharya, A., Rao, Y.S., Siqueira, P., Bera, S., 2018. Sen4Rice: a processing chain for differentiating early and late transplanted rice using time-series sentinel-1 SAR data with Google earth engine. Geosci. Rem. Sens. Lett. IEEE 15, 1947–1951. https://doi.org/10.1109/LGRS.2018.2865816. Man´ e, L.K., Fraval, P., 2001. Suivi par t´ el´ ed´ etection des cultures de d´ ecrue dans la Vall´ ee du Fleuve S´ en´ egal en saison froide 2000/2001. SAED-IWMI-WARDA report. Motsumi, S., Magole, L., Kgathi, D., 2012. Indigenous knowledge and land use policy: implications for livelihoods of flood recession farming communities in the Okavango Delta, Botswana. Physics and Chemistry of the Earth, Parts A/B/C 50–52 185–195. https://doi.org/10.1016/j.pce.2012.09.013. Nguyen, D.B., Gruber, A., Wagner, W., 2016. Mapping rice extent and cropping scheme in the Mekong Delta using Sentinel-1A data. Remote Sensing Letters 7 (12), 1209–1218. Ogilvie, A., Belaud, G., Delenne, C., Bailly, J.-S., Bader, J.-C., Oleksiak, A., Ferry, L., Martin, D., 2015. Decadal monitoring of the Niger Inner Delta flood dynamics using MODIS optical data. J. Hydrol. 523, 368–383. https://doi.org/10.1016/j.jhydrol.2015.01.036. Ogilvie, A., Poussin, J.-C., Bader, J.-C., Bayo, F., Bodian, A., Dacosta, H., Dia, D., Diop, L., Martin, D., Sambou, S., 2020. Combining multi-sensor satellite imagery to improve long-term monitoring of temporary surface water bodies in the Senegal river floodplain. Rem. Sens. 12, 3157. https://doi.org/10.3390/rs12193157. Ogilvie, A., Fall, C.S., Bodian, A., Martin, D., Bruckmann, L., Dia, D., Leye, I., Ndiaye, P.M., Soro, D.D., Danumah, J.H., Bader, J.-C., Poussin, J.-C., 2025. Surface water and flood-based agricultural systems: mapping and modelling long-term variability in the Senegal river floodplain. Agric. Water Manag. 308, 109254. https://doi. org/10.1016/j.agwat.2024.109254. OMVS & IRD, 2002. Programme d’Optimisation et de Gestion des R´ eservoirs (POGR). Phase finale, rapport de synthese, Annexe 1 : Cultures de decrue. Dakar, Senegal, 2002. OMVS, 2018. Plan d’action r´ egional pour l’am´ elioration des cultures irrigu´ ees du bassin du fleuve S´ en´ egal. Dakar, S´ en´ egal., 2018 Orieschnig, C.A., Belaud, G., Venot, J.-P., Massuel, S., Ogilvie, A., 2021. Input imagery, classifiers, and cloud computing: insights from multi-temporal LULC mapping in the Cambodian Mekong Delta. European Journal of Remote Sensing 54, 398–416. https://doi.org/10.1080/22797254.2021.1948356. Poussin, J.-C., 1998. Diagnostic sur les syst` emes de riziculture irrigu´ ee dans la moyenne vall´ ee aval du fleuve S´ en´ egal. In: Biarnes, A. (Ed.), La conduite du champ cultiv´ e: points de vue d’agronomes. Orstom, Paris, pp. 133–161. Poussin, J.-C., Martin, D., Bader, J.-C., Dia, D., Seck, S.M., Ogilvie, A., 2020. Variabilit´ e agro-hydrologique des cultures de d´ ecrue. Une ´ etude de cas dans la moyenne vall´ ee du fleuve S´ en´ egal. Cah. Agric. For. 29, 23. https://doi.org/10.1051/cagri/2020022. Qi, J., et al., 1994. A modified soil adjusted vegetation index. Remote Sens. Environ. 48, 119–126. L. Bruckmann et al. Remote Sensing Applications: Society and Environment 37 (2025) 101473 25