Ecophysiological variables retrieval and early stress detection: insights from a synthetic spatial scaling exercise

International Journal of Remote Sensing ISSN: (Print) (Online) Journal homepage: www.tandfonline.com/journals/tres20 Ecophysiological variables retrieval and early stress detection: insights from a synthetic spatial scaling exercise Javier Pacheco-Labrador, M.Pilar Cendrero-Mateo, Shari Van Wittenberghe, Itza Hernandez-Sequeira, Gerbrand Koren, Egor Prikaziuk, Szilvia Fóti, Enrico Tomelleri, Kadmiel Maseyk, Nataša Čereković, Rosario Gonzalez-Cascon, Zbyněk Malenovský, Mar Albert-Saiz, Michal Antala, János Balogh, Henning Buddenbaum, Mohammad Hossain Dehghan-Shoar, Joseph T. Fennell, JeanBaptiste Féret, Hamadou Balde, Miriam Machwitz, Ádám Mészáros, Guofang Miao, Miguel Morata, Paul Naethe, Zoltán Nagy, Krisztina Pintér, R. Reddy Pullanagari, Anshu Rastogi, Bastian Siegmann, Sheng Wang, Chenhui Zhang & Daniel Kopkáně To cite this article: Javier Pacheco-Labrador, M.Pilar Cendrero-Mateo, Shari Van Wittenberghe, Itza Hernandez-Sequeira, Gerbrand Koren, Egor Prikaziuk, Szilvia Fóti, Enrico Tomelleri, Kadmiel Maseyk, Nataša Čereković, Rosario Gonzalez-Cascon, Zbyněk Malenovský, Mar AlbertSaiz, Michal Antala, János Balogh, Henning Buddenbaum, Mohammad Hossain DehghanShoar, Joseph T. Fennell, Jean-Baptiste Féret, Hamadou Balde, Miriam Machwitz, Ádám Mészáros, Guofang Miao, Miguel Morata, Paul Naethe, Zoltán Nagy, Krisztina Pintér, R. Reddy Pullanagari, Anshu Rastogi, Bastian Siegmann, Sheng Wang, Chenhui Zhang & Daniel Kopkáně (28 Oct 2024): Ecophysiological variables retrieval and early stress detection: insights from a synthetic spatial scaling exercise, International Journal of Remote Sensing, DOI: 10.1080/01431161.2024.2414435 To link to this article: https://doi.org/10.1080/01431161.2024.2414435 © 2024 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group. View supplementary material Published online: 28 Oct 2024. Submit your article to this journal Article views: 760 View related articles View Crossmark data Full Terms & Conditions of access and use can be found at https://www.tandfonline.com/action/journalInformation?journalCode=tres20 Ecophysiological variables retrieval and early stress detection: insights from a synthetic spatial scaling exercise Javier Pacheco-Labrador a,b , M.Pilar Cendrero-Mateo c , Shari Van Wittenberghe c , Itza Hernandez-Sequeira d , Gerbrand Koren e , Egor Prikaziuk f , Szilvia Fóti g,h , Enrico Tomelleri i , Kadmiel Maseyk j , Nataša Čereković k , Rosario GonzalezCascon l , Zbyněk Malenovský m , Mar Albert-Saiz n , Michal Antala n , János Balogh g , Henning Buddenbaum o , Mohammad Hossain Dehghan-Shoar p , Joseph T. Fennell j , Jean-Baptiste Féret q , Hamadou Balde r , Miriam Machwitz s , Ádám Mészáros g , Guofang Miao t , Miguel Morata c , Paul Naethe u , Zoltán Nagy g,h , Krisztina Pintér g,h , R. Reddy Pullanagari v , Anshu Rastogi n , Bastian Siegmann w , Sheng Wang x,y , Chenhui Zhang z and Daniel Kopkáně aa a Environmental Remote Sensing and Spectroscopy Laboratory (SpecLab), Spanish National Research Council, Madrid, Spain; b Max Planck Institute for Biogeochemistry, Jena, Germany; c Laboratory of Earth Observation, Image Processing Laboratory, University of Valencia, Paterna (Valencia), Spain; d Institute of New Imaging Technologies, University Jaume I, Castellón de la Plana (Castellón), Spain; e Copernicus Institute of Sustainable Development, Utrecht University, Utrecht, The Netherlands; f Faculty of Geo-Information Science and Earth Observation (ITC), University of Twente, Enschede, The Netherlands; g Department of Plant Physiology and Plant Ecology, Institute of Agronomy, Hungarian University of Agriculture and Life Sciences, Gödöllő, Hungary; h HUN-REN-MATE Agroecology Research Group, Gödöllő, Hungary; i Faculty of Agricultural, Environmental and Food Sciences, Free University of Bozen/Bolzano, Bozen-Bolzan, Italy; j School of Environment, Earth and Ecosystem Sciences, The Open University, Milton Keynes, UK; k University of Banja Luka, Centre for Development and Research Support, Institute of Genetic Resources, Banja Luka, Bosnia and Herzegovina; l INIA-CSIC, Department of Environment and Agronomy, Madrid, Spain; m Remote Sensing Research Group, Department of Geography, University of Bonn, Bonn, Germany; n Laboratory of Bioclimatology, Department of Ecology and Environmental Protection, Faculty of Environmental and Mechanical Engineering, Poznań University of Life Sciences, Poznań, Poland; o Environmental Remote Sensing and Geoinformatics, Trier University, Trier, Germany; p School of Agriculture and Environment, Massey University, Palmerston North, New Zealand; q TETIS, INRAE, AgroParisTech, CIRAD, CNRS, Université Montpellier, Montpellier, France; r Laboratoire de Météorologie Dynamique, Sorbonne Université, IPSL, CNRS/L’École polytechnique, Palaiseau Cedex, France; s Remote sensing and natural resources modelling, Department for Environmental Research and Innovation (ERIN), Luxembourg Institute of Science and Technology (LIST), Belvaux, Luxembourg; t School of Geographical Sciences, Fujian Normal University, Fuzhou City, China; u JB Hyperspectral Devices, Düsseldorf, Germany; v Australian Plant Phenomics Facility (APPF), The Plant Accelerator, School of Agriculture, Food and Wine, University of Adelaide, Urrbrae, Australia; w Institute of Bioand Geosciences, IBG-2: Plant Sciences, Jülich, Germany; x Department of Agroecology, Aarhus University, Department of Agroecology, Aarhus, Denmark; y Agroecosystem Sustainability Center, University of Illinois Urbana-Champaign, Agroecosystem Sustainability Center, Institute for Sustainability, Energy, and Environment, University of Illinois Urbana-Champaign, Urbana, IL, USA; z Institute for Data, Systems, and Society, Massachusetts Institute of Technology, Cambridge, MA, USA; aa Global Change Research Institute of the Czech Academy of Sciences (CzechGlobe), Bělidla, Czech Republic CONTACT Javier Pacheco-Labrador [email protected] Environmental Remote Sensing and Spectroscopy Laboratory (SpecLab), Spanish National Research Council, Albasanz 26-28, Madrid 28037, Spain This article has been corrected with minor changes. These changes do not impact the academic content of the article. Supplemental data for this article can be accessed online at https://doi.org/10.1080/01431161.2024.2414435 INTERNATIONAL JOURNAL OF REMOTE SENSING https://doi.org/10.1080/01431161.2024.2414435 © 2024 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/ licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The terms on which this article has been published allow the posting of the Accepted Manuscript in a repository by the author(s) or with their consent. ABSTRACT The ability to access physiologically driven signals, such as surface temperature, photochemical reflectance index (PRI), and suninduced chlorophyll fluorescence (SIF), through remote sensing (RS) are exciting developments for vegetation studies. Accessing this ecophysiological information requires considering processes operating at scales from the top-of-the-canopy to the photosystems, adding complexity compared to reflectance index-based approaches. To investigate the maturity and knowledge of the growing RS community in this area, COST Action CA17134 SENSECO organized a Spatial Scaling Challenge (SSC). Challenge participants were asked to retrieve four key ecophysiological variables for a field each of maize and wheat from a simulated field campaign: leaf area index (LAI), leaf chlorophyll content (C ab ), maximum carboxylation rate (V cmax,25 ), and non-photochemical quenching (NPQ). The simulated campaign data included hyperspectral optical, thermal and SIF imagery, together with ground sampling of the four variables. Non-parametric methods that combined multiple spectral domains and field measurements were used most often, thereby indirectly performing the top-of-the-canopy to photosystem scaling. LAI and C ab were reliably retrieved in most cases, whereas V cmax,25 and NPQ were less accurately estimated and demanded information ancillary to RS imagery. The factors considered least by participants were the biophysical and physiological canopy vertical profiles, the spatial mismatch between RS sensors, the temporal mismatch between field sampling and RS acquisition, and measurement uncertainty. Furthermore, few participants developed NPQ maps into stress maps or provided a deeper analysis of their parameter retrievals. The SSC shows that, despite advances in statistical and physically based models, the vegetation RS community should improve how field and RS data are integrated and scaled in space and time. We expect this work will guide newcomers and support robust advances in this research field. ARTICLE HISTORY Received 20 June 2024 Accepted 29 September 2024 KEYWORDS Remote sensing; plant physiology; spatial scaling; temporal mismatch; fluorescence; thermal; hyperspectral; top of the canopy; photosystem; downscaling 1. Introduction Remote sensing (RS) is widely used to monitor a range of vegetation attributes across scales. Mapping plant ecophysiological traits from spectral imagery is necessary for enhancing agricultural production (Weiss, Jacob, and Duveiller 2020), forest management (Lechner, Foody, and Boyd 2020), and sustaining ecosystem services (Del Río-Mena et al. 2020). The measurement and regular monitoring of both natural and managed ecosystems across time and space play a key role in our efforts to understand drivers and variations in productivity and ecosystem–climate interactions (Rocchini and Lenoir 2021; J. Yang et al. 2013). As more frequent and intense stress events are being recorded due to climate change and environmental variations (IPCC 2023), the need for early stress detection and recommendations for resource use efficiency in agriculture and forestry (Herrmann and Berger 2021) also increases. Light and carbon dioxide are two resources that drive vegetation (primary) productivity, which depends on the absorption and use efficiency of light energy (Haxeltine and Prentice 1996) and carbon dioxide assimilation through stomata (Gates 1968). 2J. PACHECO-LABRADOR ET AL. Characterizing the ecophysiological parameters related to photosynthesis is fundamental to understanding vegetation functions and responses to the environment. Parameters that are typically quantified from field and laboratory measurements include the fraction of photosynthetically active radiation absorbed by leaves, leaf chlorophyll a and b content (C ab ), the maximal rate of CO 2 carboxylation at 25°C (V cmax,25 ), and photosystem II quantum yields. Light energy that is not used for CO 2 assimilation and is dissipated by xanthophyll cycle pigments is characterized by the level of non-photochemical quenching (NPQ) or the thermal dissipation of absorbed photosynthetically active radiation (PAR) (Demmig-Adams et al. 2012). As NPQ increases under energy-limiting conditions, it can be used as a stress or potential stress indicator. In addition, photosynthetic efficiency depends on temperature, and optimal functioning involves stomatal control to balance CO 2 assimilation against water loss in the process of transpirational cooling (Farquhar and Sharkey 1982; Gates 1968). Usually, determining these parameters involves intensive field measurements at the leaf scale. While these measurements offer fundamental insights into the leaf-level processes, time and other resource constraints limit the extent and frequency of acquisition and, therefore, the capability of becoming monitoring tools. The fundamental link to the absorption and use of visible light enables some of these parameters to be inferred from top-of-canopy (TOC) RS observations and, thus, the opportunity to measure and monitor vegetation status and productivity across spatial scales from field to global and at high temporal frequency. Traditionally, these observations have relied on reflected visible and infra-red radiance and derived vegetation indices (VI) related to vegetation structure, biochemistry, and photosynthetic capacity that vary over medium time scales (Nemani et al. 2003; Running et al. 2004). Recent advances in satellite technology are providing the opportunity to collect signals more closely related to dynamic vegetation properties and ecophysiological traits (Schimel et al. 2019). These include hyperspectral Visible-Near InfraRed (VNIR) reflected radiance that enables the quantification of solar-induced chlorophyll fluorescence (SIF) and the Photochemical Reflectance Index (PRI), as well as thermal infrared data for land surface temperature (LST) mapping. These signals are influenced by biophysical properties and ecophysiological processes at both canopy and leaf levels, exhibiting variations from diurnal to seasonal scales (Farella et al. 2022). Sound physiological understanding of these signals requires solving the spatial and temporal-scale mismatches between RS and field data (Ma et al. 2020). Facing such problems requires down-scaling information from canopy to leaf and even photosystem scales (Van Wittenberghe et al. 2021) for reliable interpretation of these signals and reconciling the large temporal variation of many processes with the periodic image acquisition of RS data. Incoming radiation reaching the leaf is absorbed, reflected, and transmitted through the leaf, and part of the absorbed fraction is emitted as fluorescence or thermal radiation, with all these processes carrying ecophysiological information. However, light is scattered and reabsorbed within the plant canopy’s vertical profile before escaping back to space, making it difficult to relate leaf-level physiological processes to spectral signals measured from the TOC. The RS community has developed various methodologies for estimating structural and foliar vegetation properties (Verrelst et al. 2015). Irrespective of the approach taken, the combination of RS and field information remains challenging and requires accounting for the existing spatial and temporal mismatches and the differences in environmental and physiological conditions of the different measurements. INTERNATIONAL JOURNAL OF REMOTE SENSING 3 Furthermore, these approaches might not have matured to estimate the most sensitive physiological variables (e.g. V cmax,25 , NPQ), whose weak effects on the spectral signals present an additional challenge. The interpretation of RS information without prior knowledge of the vegetation state is rare (e.g. factorial experiments), and the capability of the RS community in this regard is largely untested (Ma et al. 2020). Several studies have evaluated the strengths and weaknesses of the retrieval methods to estimate biophysical variables featuring low diurnal variability and strong effects on spectral information, such as Cab and leaf area index (LAI) (Verrelst et al. 2019). However, there is limited information on how these methods perform when down-scaling physiological variables with subtle contributions to spectral signals and are driven by diurnal meteorology, such as V cmax,25 and NPQ (Damm et al. 2022). It is equally important to understand the level of uncertainty associated with these estimates (Malenovský et al. 2019). Furthermore, it is worth considering how the RS scientific community integrates a series of sequential field measurements of these dynamic ecophysiological variables with usually instantaneous RS imagery. An additional problem is the interpretation of these sources of information, as most studies are carried out in controlled experiments, where the sources and levels of stress are known beforehand (e.g. Damm et al. (2022)). However, RS interpretations are expected to provide reliable conclusions from decontextualized situations, which have rarely been systematically assessed and reported in publications. Within the context of the COST Action CA17134 SENSECO ‘Optical synergies for spatiotemporal SENsing of Scalable ECOphysiological traits’ (www.senseco.eu), we set out to challenge the scientific community to retrieve relevant biophysical and physiological variables from a simulated dataset of remotely sensed reflectance factors, chlorophyll sun-induced fluorescence, and thermal imagery, as well as field measurements. Specifically, the objectives for every scientist joining this study were to: (1) down-scale the biophysical and plant physiological variables C ab , LAI, V cmax,25 , and NPQ from simulated hyperspectral and thermal imagery and (2) evaluate the capability of RS data to diagnose and discriminate between different plant ecophysiological states and translate the retrieved variables into low or high vegetation efficiency or stress. To assess the quality of the results, we asked the participants of this study to provide the uncertainty for each of the required parameters and stress maps. The overarching goal of the study was to gather the knowledge and experience of the RS community regarding the scaling problems in the context of ecophysiological monitoring, identify gaps, strengths, and weaknesses, and guide RS scientists who want to expand their research from vegetation biophysical properties to physiology. 2. Material and methods 2.1. The spatial scaling challenge The Spatial Scaling Challenge (SSC) was organized by the coordinators of the SENSECO Working Group 1, ‘Closing the scaling gap: from leaf measurements to 4J. PACHECO-LABRADOR ET AL. satellite images’, who anonymized the identity of all participants in all cases and hid the problem ‘solution’ until participation was closed (Figure 1). The SSC was announced in April 2022 among various RS networks and research communities worldwide working on the retrieval of vegetation ecophysiological parameters. The datasets and documentation were published open-access through ZENODO. The main dataset (Pacheco-Labrador et al. 2022b) included airborne scene imagery (NetCDF files) and field data sets. The field ancillary data was provided in two CSV files: one containing the spatial data acquired in the field sampling plots and a second including the time series of NPQ and meteorological variables measured in a nearby Eddy Covariance tower. Matlab (.m), R (.R), and Python (.py) scripts were provided to import the SSC data, as well as to export the outcome results and products in a standardized format. Moreover, the SSC included template reports describing the retrieval and scaling methods applied to estimate each requested variable: C ab , LAI, V max,25 , and NPQ. The SSC also provided a bonus dataset (PachecoLabrador et al. 2022a) that included a half-hourly time series of down-welling spectral irradiance (W m −2 µm −1 ) in the visible and near-infrared domains as measured by a field spectroradiometer operating in a nearby ecosystem station. The information provided in the reports was complemented later with a questionnaire (supplementary material S1). The SSC contributors and SENSECO members collaborated to evaluate the results and obtained the conclusions presented in this manuscript (Figure 1). Figure 1. Spatial scaling challenge (SSC) workflow. The SCC coordinators designed and simulated the different SSC datasets privately, and published the ‘problem’ dataset in a zenodo repository. The contributors ‘solved’ the SSC and sent their contributions to the SSC coordination. These evaluated their results, anonymized them, and shared them with the participants and other COST action CA17134 SENSECO working group 1 collaborators for the elaboration of the manuscript, some of which contributed to evaluating some of the anonymized results. INTERNATIONAL JOURNAL OF REMOTE SENSING 5 2.2. Scene simulation Except for the meteorological variables provided by an automated weather station in Spain, all the variables were synthetic, i.e. produced artificially with models and empirical equations and sampled from a normal distribution with assigned means and standard deviations. Despite their synthetic nature, those values throughout the manuscript are referred to as ground truth, reference data, baseline dataset, and measured data. This section provides a comprehensive overview of what parameters were modelled and the principles behind them. For a more thorough exploration, please consult the supplementary materials S2 (Supplementary S2.1 and Table S2.1). The SSC simulated two crop fields, maize (Zea mays L.) and wheat (Triticum aestivum L.), in different growth stages. The fields were exposed to varying fertilization and soil moisture gradients. The scene was 100 m × 100 m, with a bare soil corridor separating the fields. Meteorological data from Majadas de Tiétar was used for the simulation, which took place during the summer under high radiation and vapour pressure deficit conditions. The SSC campaign simulated ground sampling and airborne imagery datasets with the Soil Canopy Observation, Photochemistry and Energy model (SCOPE, v1.73) (van der Tol et al. 2009). 2.2.1. Water and nutrient availability treatments The simulations comprise two crop management treatments. The first was a biochar amendment with a nitrogen enrichment along the fields’ North–South (NS) axis (Fig. S2.1), which reduced albedo (Fig. S2.2) and increased nitrogen availability. Over the long term, it influenced the vegetation properties, such as leaf area index, maximum carboxylation rate, and pigment content. The second was a drip irrigation system that created varying soil moisture zones within each field shortly before the campaign, inducing physiological fast stress responses along the East–West (EW) direction (Fig. S2.3). Smooth soil moisture transitions between water regimes were simulated using log functions. Biochar also positively impacted soil water retention. 2.2.2. Soil properties and soil moisture stress simulation The soil input variables of the Brightness – Shape – Moisture (BSM) model of SCOPE (Verhoef, van der Tol, and Middleton 2018) were spatially defined, with Gaussian noise added to simulate natural variability. Some soil properties, such as soil brightness, field capacity (FC), and moisture content (SMC), were affected by fertilization and water treatments. A decreasing relationship between the relative soil water content (SMC/FC) and the soil resistance for evaporation from the pore space (r ss ) was defined to reduce soil evaporation under low moisture. In contrast, soil thermal emissivity was modulated based on SMC to modify energy balance and heat stress (Table S2.1, Table S2.1.3.1). 2.2.3. Vegetation input variables and stress-based radiative transfer modelling The simulations modelled vegetation parameters based on crop-specific literature. Taxonomy, soil moisture, and nitrogen availability were the key drivers of the vegetation spatial variability of properties and physiological state (Table S2.1, Table S2.1.4.1). Most vegetation properties such as foliar nitrogen content, leaf dry matter content, pigment contents, the leaf structural parameter, V cmax,25 , LAI, or canopy height were simulated as a function of nitrogen fertilization or nitrogen-dependent variables (Fig S2.4-7). Other 6J. PACHECO-LABRADOR ET AL. variables, such as leaf angle distributions, stomatal sensitivity, fluorescence quantum efficiency, and photosystem II-to-I ratio, were estimated or adopted from the literature. The rate coefficient of non-photochemical quenching was modelled as a function of relative light saturation and soil moisture content (Fig. S2.1.4.1, S2.8). Gaussian noise was added to simulate natural variability in all variables. 2.2.4. Airborne campaign The simulations modelled airborne hyperspectral optical and thermal imagery acquisition over a simulated scene. The virtual flight took place at midday (11:45 UTC, 13:45 local time) under specific conditions of solar radiation, temperature, and vapour pressure deficit. Three hyperspectral sensors were used, similar to the HyPlant+TASI system (Hanuš, Fabiánek, and Fajmon 2016; Siegmann et al. 2019). Simulated hemisphericaldirectional reflectance factors (HDRF, Fig. S2.9–11) and outgoing thermal radiance in the observation direction of each pixel were convolved to the spectral characteristics reported for the visible to near-infrared and shortwave infrared (VNIR + SWIR module) as well as the thermal imaging spectrometers. Fluorescence radiances (Fig. S2.12–13) were provided in the O 2 -B and O 2 -A bands (F 687 and F 760 , respectively). LST was retrieved using the TES algorithm (Hanuš, Fabiánek, and Fajmon 2016) from the bottom of the atmosphere radiances simulated by SCOPE (Fig. S2.14). Uncertainty levels were considered realistic based on previous studies (Table S2.1.5.1). All signals were simulated at 1 m spatial resolution, then degraded for fluorescence (4 m) and thermal (2 m) radiances. LST was retrieved from the 2 m resolution imagery. 2.2.5. Ancillary data from a field campaign During the simulated flight campaign, we modelled vegetation sampling at various times. C ab , LAI, V cmax,25 , and NPQ were measured at nine points in each field. A randomized sampling scheme was employed to minimize bias (Figure 2a). Measurements were conducted at different times to account for temporal discrepancies and limitations in Figure 2. (a) Location of the field sampling plots and permanent moni-pam. (b) Temporal distribution of the different measurements (time ranges) and incoming shortwave incoming radiation (SW in ). The time on the x-axis is in UTC, local time was UTC + 2, taking the flight one hour before maximum solar height. INTERNATIONAL JOURNAL OF REMOTE SENSING 7 variability were generally attributed to slower growth associated with different soil types having lower nitrogen and water content towards the south (e.g. #5, #7, #8, #9, #12). Heterogeneous soil conditions could also explain such gradients (#12). In addition to the LAI and C ab gradients, the NS stress trend was also reported based on V cmax,25 , and NPQ, as shown in Fig. S4.5. & S4.7. (e.g. #1, #5, #6, #8, #9). Higher LAI, C ab , and V cmax,25 values were considered to be closely related and linked to the plant nutrient status (#13, #14), and the NPQ trend was described as ‘inverse’ to the others (e.g. #1, #6) if detected. In general, it was reported that the maize field NPQ did not show any spatial gradient, unlike wheat (#12). In addition, #10 detected an NS gradient in the Water Deficit Index (WDI) (Moran et al. 1994), an index developed to detect crop water deficit using surface–air temperature difference and a spectral vegetation index (NDVI), also revealed an NS gradient (#10). Figure 5. Taylor plots comparing simulated and estimated LAI (a), C ab (b), V cmax,25 (c), and NPQ (d). Each code corresponds to the SSC participants’ contributions ranging from 1 to 15. The symbols identify the retrieval approach (‘o’ parametric regression, ‘x’ non-parametric regression, ’*’ physical; y-based, and ‘◇’ hybrid regression). 14 J. PACHECO-LABRADOR ET AL. 3.3.2. Short-term acute physiological stress responses (EW gradient) The decreasing EW soil moisture content gradient (Fig. S2.3) proved more difficult to detect and was only reported by a few participants through the patterns observed in the retrieved NPQ (#13, #14) or the F imagery (#4) (Fig. S4.7). Specifically, the use of F relied on F 687 normalized by sunlit leaf area (F 687,sl , obtained by retrieving leaf angle distribution factors from the SCOPE model first, then calculating the sunlit leaf area by using the Figure 6. Root mean squared error (RMSE) between simulation and estimation for (a) LAI; (b) C ab ; (c) V cmax,25 ; and (d) NPQ. Colours are based on the method type (similar to fig. 4), and the texture of the bars refers to the vegetation type. Note that in panel (a), the y-scale was truncated to improve the visibility; the RMSE for maize of model #8 is 6.82 m 2 m −2 . Figure 7. Spearman’s rank correlation (ρ) between simulation and estimation for (a) LAI; (b) C ab ; (c) V cmax,25 ; and (d) NPQ. Colours are based on the method type (similar to fig. 4), and the texture of the bars refers to the vegetation type. INTERNATIONAL JOURNAL OF REMOTE SENSING 15 retrieved LIDF values), assuming aPAR is the main factor dominating the F signal. A few participants also used a combination of parameters/methods (#3, #4, #10) to produce stress maps (Fig. S4.9). The fluorescence yield (F 687,sl ) obtained after post-processing the F 687 values showed an EW gradient, as did the combined (V cmax,25 , NPQ, and nadirnormalized F 687-yield and F 760-yield ) stress map (#4). It was also pointed out that the fluorescence yield maps showed the EW but not NS stress gradients. The EW stress gradient was attributed to acute or management-driven stress (#4). Although a few methods were adequate to assess EW trends or soil moisture zones, several reports differentiated between long-term (nutrient status) and acute (heat, drought) stress effects, and the output maps were considered as a result of the combinations of both (#5, #6, #7, #8, #9, #10 #11, #12). 3.3.3. Stress detection across species Differences were found between maize and wheat stands under stress conditions, as observed in both the F imagery and derived variables such as F 687,sl (#4), LAI, C ab , V cmax,25 , and NPQ maps (#4, #5, #6). Additionally, distinctions were evident in the LST imagery (#5, #6, #7, #12) and a water deficit index (#10). The stand differences were also evident in the derived relationships and normalized signals that the participants were able to interpret. Maize was found to be generally less stressed than wheat, based on relationships between NPQ and nadir-normalized F 687,nadir values (Hao et al. 2022) and from the lower nadir normalized F 760,nadir values seen in wheat compared to maize. It was also reported that wheat plants experienced more stress than maize due to their C3 photosynthetic pathway. Maize, being a C4 plant, exhibited a higher light compensation point (#13), temperature tolerance (#5, #6, #7, #12, #13), and water use efficiency (#12). The increased stress level in the case of the wheat stand was found by the participants to be accompanied by increased variability or lower estimation accuracy of V cmax,25 (#5, #7, #8, #9), while on the contrary, maize showed a less variable and higher photosynthetic performance (#5, #6, #12) irrespective of the daytime high radiation and temperature. 4. Discussion 4.1. Addressing the scaling challenges The SSC asked the RS community to estimate four variables – LAI, C ab , V cmax,25 , and NPQ – featuring different effects on vegetation optical and thermal spectral properties and different spatial and temporal scales of variability. Beyond the performance evaluation, the SSC sought to assess to what extent and how the participants addressed the different scaling problems. Since these scaling issues already limit the exhaustive measurement of these variables in the field, the SSC relied on detailed simulations of RS and field observations to generate a dataset that enabled an adequate assessment of the results. While we acknowledge that the simulation framework produced simplified representations that can be biased or less challenging than the real world, we consider the SSC dataset sufficient to assess the degree of understanding and consideration of the scaling issues performed by participants. In particular, we simulated different stress types linked to vegetation properties and functioning using relationships previously reported in the 16 J. PACHECO-LABRADOR ET AL. literature. In addition, we included randomness originating from the natural variability of individuals in all vegetation variables and uncertainties related to all simulated measurements. The specificities of these relationships and the model SCOPE v1.73 selected for conducting the simulation could bias estimates relying on PB models (i.e. SCOPE2.0 (P. Yang et al. 2021) and different versions of PROSPECT (J. B. Féret et al. 2017; J.-B. Féret et al. 2021) + SAIL (Verhoef 1984)), featuring different assumptions or representations of vegetation and related signals. 4.1.1. Methods NPR models were most used and achieved lower relative RMSE values than the rest of the approaches (~15% (LAI), ~62% (C ab ), ~64% (V cmax,25 ), and ~98% (NPQ)). However, these flexible but data-hungry algorithms were trained with relatively small datasets (9–18 field observations), likely at the expense of a lack of generalization (Verrelst et al. 2019; Wang, Guan, Wang, Ainsworth, Zheng, Townsend, Liu, et al. 2021). PR was the second most used approach, and unlike the others, it used only a small fraction of the information available. Still, these achieved acceptable performances in most of the cases. We hypothesize that the smoothness of the vegetation gradients favoured the performance of NPR and PR approaches, particularly when models were trained per species. In any case, the use of these models is limited to the availability of field data, which does not necessarily prevent extrapolation (e.g. Moreno-Martínez et al. (2018)). Alternatively, HR approaches only require knowledge of the forward use of a physical model since inputs and outputs are integrated into an NPR model. HR was used more than PB and often performed better, but not as well as NPR. One of the sources of uncertainty for PB and HR is the discrepancy between the model and reality (or, in this case, the model used for the simulations). For example, contributor #2 used PROSPECT-PRO (J.-B. Féret et al. 2021) RTM to estimate V cmax,25 exploiting the absorption features of proteins in the leaf. However, the SCOPE simulations did not present such features, which could explain the low correlation obtained by #2 (Figure 7c). Nonetheless, this approach could perform better when applied to real vegetation measurements (e.g. Wang, Guan, Wang, Ainsworth, Zheng, Townsend, Li, et al. (2021), Camino et al. (2022)). While the number of contributions using PB and HR is too low to make clear conclusions, HR training error might relax the underlying assumptions of the physical model, improving, in some cases, the retrieval (Verrelst et al. 2019; Wang, Guan, Wang, Ainsworth, Zheng, Townsend, Liu, et al. 2021). Overall, better results were obtained when retrieving biophysical parameters with a strong effect on the optical signals (i.e. LAI and C ab ) than the two physiological variables indirectly related to F, LST, and PRI (i.e. V cmax,25 , and NPQ) (Figures 5–7). Our results demonstrate that modelling and data assimilation improvements are needed to adequately characterize plant physiological variables from RS, particularly when no field data is available to calibrate empirical models. Finally, none of the contributors used the observational uncertainties of the field measurements nor asked about the uncertainty estimates of RS imagery. Uncertainties could have been employed to assign weights to each data point’s contribution during model training and to subsequently propagate these uncertainties to the predictions. Uncertainty maps were exclusively generated based on the models’ fitting errors (Porwal and Raftery 2022). INTERNATIONAL JOURNAL OF REMOTE SENSING 17 4.1.2. Scaling challenges The SSC presented several spatial challenges. The first one was the heterogeneity of spatial resolutions in the RS imagery. The spatial resolution is relevant for the retrieval of vegetation ecophysiological properties since these relate non-linearly with the spectral signals and unresolved heterogeneity (spectral mixture) increase estimation uncertainty (Pacheco-Labrador et al. 2022; Vicente et al. 2021). Maybe partly due to the smoothness of the simulated crop field, none of the participants attempted to downscale the imagery to the finest resolution, which might be more relevant in heterogeneous environments. Still, larger errors were found in the borders than in the interior of the fields, where bare soil and vegetation signals were mixed (supplementary material S4). While the down-scaling of reflectance factors (e.g. fusion of multi and hyperspectral imagery (Vivone 2023)) relies on univocal relationships between the data at different scales, the down-scaling of physiologically regulated signals transfer (e.g. PRI, or thermal and SIF emissions) is more challenging. The latest would require ancillary information (e.g. canopy structure driving SIF escape and current incoming PAR) and accounting for the diurnal and seasonal dynamics of these signals and their relationship with plant physiology. For example, Guzinski et al. (2020) used physical models to downscale evapotranspiration retrievals, whereas Duveiller et al. (2020) used LST and semi-empirical stress models to down-scale coarse global SIF products. The second challenge was the species-dependent variability in the relationships between vegetation ecophysiological properties and the observed spectral signals for the vegetation types represented (i.e. wheat C3 and maize C4). This is particularly true for the C3 and C4 plants, which have developed different mechanisms to fix CO 2 from the atmosphere, resulting in differences in absorbing (LAI, C ab ), using (V cmax,25 ), and disposing (NPQ) radiation (Table S2.1.4.1). Half of the SSC participants modelled both vegetation types separately. While the number of contributions did not allow for a complete understanding of the impact of vegetation-type differentiation, previous studies show that it could improve the retrieval of ecophysiological properties (Moreno-Martínez et al. 2018). In the SSC, small sample sizes prevented most NPR approaches from characterizing each vegetation type separately. Such separation could be strongly recommendable for simple PR approaches that are unfit for generalization. In the case of PB and HR, physical models impose strong assumptions on the relationships between ecophysiological and spectral signals. However, radiative transfer theory offers different formulations (Widlowski et al. 2015), which could be selected as a function of the vegetation type (Wang, Guan, Wang, Ainsworth, Zheng, Townsend, Liu, et al. 2021). C3/C4 differentiation is inherent to physiological models often based on the Farquhar – von Caemmerer – Berry biochemical photosynthesis model. This must be accounted for by missions studying photosynthesis, such as FLEX, which aims to provide a light-response-centred approach for estimating photosynthetic efficiency (Liu et al. 2022) accounting for C3 and C4 vegetation types (ESA 2015). In all cases, vegetation type-oriented modelling relies on vegetation-type mapping, where the discrimination between C3 and C4 vegetation from RS remains particularly challenging (Poulter et al. 2015). A third challenge was the spatial scale of each retrieved variable (i.e. photosystem, leaf, or canopy). LAI is a canopy variable with strong control over all spectral signals, whereas C ab is a foliar variable with substantial radiation control in the visible spectral region. Both influence vegetation photosynthesis and transpiration via aPAR and gas exchange area. 18 J. PACHECO-LABRADOR ET AL. On the other hand, V cmax,25 is a foliar property with a weak manifestation in spectral signals (mainly in F). Finally, NPQ is not a vegetation property but a state that varies dynamically over time according to environmental conditions. SSC simulations only represented vertical profiles of V cmax,25 and NPQ, and the participants were informed these had been sampled from the upper layers of the canopy. While the participants did not consider the vertical gradients when training models or evaluating their estimates, no systematic biases were found in the retrievals (Figure 4), which suggests other factors dominated uncertainties. PB and HR approaches embed the down-scaling from TOC to leaf (i.e. C ab ) or photosystem (i.e. V cmax,25 , and NPQ) level in the RTM formulations. However, the complexity of coupled physiological-RTM models accounting for environmental factors and vertical gradients within the canopy made the use of fully coupled models rare. Despite their complexity, these approaches performed worse than the empirical ones. In all cases, the retrieval of biophysical (LAI and C ab ) and physiological variables (V cmax,25 , and NPQ) was separated into two steps where the first estimates informed the retrieval of the second, an approach that can reduce equifinality in the retrieval of least influencing variables (Pacheco-Labrador et al. 2019). PR and NPR do not provide an explicit-scale transfer. However, they can include biophysical variables relevant to the change of scale as predictors. Whereas LAI estimates did not support the retrieval of C ab , LAI or C ab , and even protein content estimates became predictors of V cmax,25 or NPQ. A second alternative could be to jointly model variables of different scales to exploit potential covariances, reducing equifinality and maybe providing more consistent estimates. The fourth challenge relates to the temporal mismatch between vegetation between variables and spectral observations. LAI and C ab were simulated as being static in time, as intra-daily variation is usually neglected. V cmax,25 measurements were simulated under conditions similar to those inside a leaf chamber. However, it must be considered that the actual carboxylation rate achieved by the plant is variable, which can bias the relationship between V cmax,25 and spectral signals influenced by physiology (i.e. F, PRI, and LST) over time, the links not necessarily being coherent. NPQ is highly dynamic; two datasets were provided: the first prior to the flight (NPQ 1 ) and the second centred around the overflight time (NPQ 2 ). Despite this temporal mismatch being most often disregarded, no systematic biases were found for the contributions using NPQ1, suggesting that the retrieval drives uncertainty. 4.2. Stress assessment SSC considered three types of stress: (1) long-term nitrogen limitation (NS), resulting in slow plant responses; (2) short-term water limitation (EW), resulting in fast plant responses; and (3) inherent differences between wheat and maize due to different photosynthetic pathways. In SSC, changes in growth responses were more easily detected by the different retrieval methodologies of common biophysical variables than physiological stress. Plants balance the extent or size of roots and canopy and regulate the biosynthesis of photosynthetic pigments and the chlorophyll-to-carotenoid ratio in response to the longterm deficit of water or nutrient availability (Ru et al. 2023; Song et al. 2023). The longterm nitrogen-induced stress was evident in maps of retrieved variables with decreasing INTERNATIONAL JOURNAL OF REMOTE SENSING 19 values of LAI, C ab , V cmax,25 , and increasing values of NPQ from N to S, which were coherent with the simulated nutrient availability (Fig. S3.1). Most contributors identified the simulated long-term nutritional stress via the retrieved biophysical variables. Early stress induced by a sudden change in soil moisture content was evident in the provided airborne imagery: F 687 , F 760 , and more subtly PRI (Fig. S2.12, S2.13 and S2.11, respectively) presented clear EW stripe patterns (Fig. S3.3) despite being superimposed to the nutrient-stress induced patterns of LAI, C ab , V cmax,25 , and other variables. Nonetheless, the SSC simulated two relatively homogeneous crop fields, facilitating the interpretation, which could be more complex in natural ecosystems. Short-term stress was also evident in the NPQ (Fig. S2.8) was collected in the field at several locations. However, these patterns were unclear in the estimated NPQ maps (Fig. S4.7) and were less often identified. In a few cases, the spatial patterns of F or corrected versions were related to stress. Furthermore, NPQ retrieval errors did not systematically relate to the fact of not taking into account spatial and temporal mismatches. These results suggest that despite RS signals being able to identify differences in physiological stress, we still lack suitable methods for retrieving physiological variables that disentangle such information from the rest of the biophysical and structural factors affecting the RS signals. The standardization of the collected data and field sampling (e.g. F and NPQ) to a common measurement time and the F post-processing towards fluorescence quantum efficiency (f qe ) maps (Moncholi-Estornell et al. 2023; Wieneke et al. 2024) could contribute to overcoming the retrieval challenges. In the case of drought stress and different responses by C3 and C4 plants, plant hormone-driven (abscisic acid-mediated) stomatal closure is considered the first acclimation response with an immediate decrease in canopy evapotranspiration and an increase in surface temperature (Fahad et al. 2017; Zandalinas et al. 2018). However, the C3 and C4 plants differ in their drought tolerance due to different molecular, physiological, and developmental processes (Song et al. 2023), resulting in an altered sensitivity and timing of stomatal responses. This phenomenon was clearly detected in the wheat stand with a higher LST and a larger NPQ (in the field datasets NPQ 1 or NPQ 2 ) by the participants. The differences in F were also evident but were not used to discriminate drought tolerance between the two vegetation types. Despite the magnitudes being different, other biophysical parameters (e.g. LAI or C ab ) could challenge the interpretation of these differences, particularly when prior knowledge of the vegetation photosynthetic paths is unavailable. The results regarding the interpretation and mapping of short-term stress suggest that expertise on this topic is less generalized than the study of biophysical properties of vegetation. The advent of new RS missions (e.g. FLEX, LSTM, CHIME, etc.) will increase the combination of different signals (e.g. F, LST and hyperspectral reflectance factors) and, therefore, the possibilities of assessing vegetation physiology from space. More specific training, open-source tools, and group collaboration will be necessary to extend this specific knowledge to a broader RS community. 5. Conclusions In the context of the COST Action ‘Optical synergies for spatiotemporal SENsing of Scalable ECOphysiological traits’ (SENSECO) (CA17134), the SSC challenged the RS 20 J. PACHECO-LABRADOR ET AL. community to down-scale and retrieve two biophysical and two physiological plant variables from a simulated airborne and field data and diagnose the vegetation stress. Based on the SSC results and discussions held during the SENSECO COST Action activities, we have identified a number of areas where further research is necessary to foster operational down-scaling of biophysical/physiological variables as well as the early detection of vegetation stress. The following six prospective research areas of future work were summarized: (1) Synergetic multi-sensor detection of early stress and plant physiology characterisation. Vegetation expresses these responses to an ever-changing environment at different rates in different spectral regions. Therefore, no single spectral variable can be expected to characterize the physiological status and reveal early stress completely. Methods for combining and assimilating RS images of different resolutions and quality, acquired at different times of the day, that are suitable to characterize plant physiology still need to be developed. The best option would be to use high-quality aerial and field datasets while addressing the subsequent considerations. (2) Accounting for the temporal dynamics of the physiological parameters or resulting plant status. Light intensity and vapour pressure deficit regulate diurnal changes in leaf temperature, V cmax , NPQ, fluorescence, and photochemistry, resulting in diurnal changes in LTS, F, and PRI. These changes must be considered when combining spectral information and vegetation variables in physical or empirical models or when comparing observations at different times of the day. Also, long-term stress (e.g. sustained NPQ) should be included in the modelling. (3) Accounting for the variability of biophysical/physiological parameters through the plant canopy vertical profile. While optical and thermal RS imagery provides information about the upper canopy layers, the vertically varying environmental conditions within the canopy (light quality, wind speed, etc.) trigger different biophysical/physiological responses (e.g. C ab and V cmax profiles). This fact should be considered to determine vegetation status and/or early stress detection, particularly when top-of-canopy spectral information or field measurements of specific layers of the canopy are used to quantify or represent the whole vegetation’s volume physiology. (4) Accounting for the spatial heterogeneity and intrapixel variability. The inherent coarse resolution of most sensors able to capture physiologically driven signals makes the characterization of physiology particularly sensitive to the mixture of different surfaces, especially bare soil, inside a pixel (e.g. the SSC found large uncertainties in the crop boundaries). Such a mixture should be characterized and taken into account by physical or statistical models. (5) Down-scaling of spectral information from the top-of-canopy to the leaf and the photosystem levels. Characterising the physiological status requires understanding vegetation properties at different scales where spectral information is either generated or transformed. Therefore, the interpretation of the spectral signals should be considered explicitly (physical models with the right formulations) or implicitly (empirical approaches, which can be supported by the inclusion of estimates of foliar and structural biophysical variables). INTERNATIONAL JOURNAL OF REMOTE SENSING 21 (6) Accounting for uncertainties. Solving the former steps involves combining different datasets for model inversion or fitting. However, not all the observations are equally valuable. Further efforts are needed to quantify and account for the observational uncertainties, minimize equifinality or biased predictions, and quantify the uncertainty of the estimated ecophysiological variables. This is fundamental to determine the level of confidence and the significance of the difference between two estimates (and, hence, the capability to discriminate stress). Accounting for all these factors is demanding and requires deep knowledge. While very experienced research groups work and apply methods that comply with these needs, their use must be considered by less experienced researchers and popularized via openaccess code or protocols that facilitate applying specialized data processing for the study of vegetation physiology from RS. Disclosure statement No potential conflict of interest was reported by the author(s). Funding The work was supported by the Agence Nationale de la Recherche [ANR-17-CE32-0001]; Centre National d’Etudes Spatiales Deutsches Zentrum für Luftund Raumfahrt [50EE1912]; European Cooperation in Science and Technology [CA17134]; European Space Agency [C.N.4000140028/22/ I-DT-lr]; H2020 European Research Council [101041768,101086622]; Horizon 2020 Framework Programme [952396]; Ministerio de Ciencia e Innovación [IJC/2018/038039/I,PID2021-128794OBI00]; National Natural Science Foundation of China [42171353]; Universidad Jaume I [PREDOC/2020/ 50]; National Recovery and Resilience Plan [1.4 – D.D. 1032 17/06/2022, CN00000022]; German Federal Ministry of Education and Research [031B0918A]; National Science Centre of Poland [2020/ 37/B/ST10/01213]; Ministry of Agriculture Luxembourg Österreichische Forschungsförderungsgesellschaft [AustroSIF]. ORCID Javier Pacheco-Labrador http://orcid.org/0000-0003-3401-7081 M.Pilar Cendrero-Mateo http://orcid.org/0000-0001-5887-7890 Shari Van Wittenberghe http://orcid.org/0000-0002-5699-0352 Itza Hernandez-Sequeira http://orcid.org/0000-0002-1623-9337 Gerbrand Koren http://orcid.org/0000-0002-2275-0713 Egor Prikaziuk http://orcid.org/0000-0002-7331-7004 Szilvia Fóti http://orcid.org/0000-0003-3235-0948 Enrico Tomelleri http://orcid.org/0000-0001-6546-6459 Kadmiel Maseyk http://orcid.org/0000-0003-3299-4380 Nataša Čereković http://orcid.org/0000-0002-7195-5280 Rosario Gonzalez-Cascon http://orcid.org/0000-0003-3468-0967 Zbyněk Malenovský http://orcid.org/0000-0002-1271-8103 Mar Albert-Saiz http://orcid.org/0000-0001-5676-3750 Michal Antala http://orcid.org/0000-0003-1294-9507 János Balogh http://orcid.org/0000-0003-3211-5120 Henning Buddenbaum http://orcid.org/0000-0002-0956-5628 22 J. PACHECO-LABRADOR ET AL. Joseph T. Fennell http://orcid.org/0000-0001-6874-6667 Jean-Baptiste Féret http://orcid.org/0000-0002-0151-1334 Miriam Machwitz http://orcid.org/0000-0002-4999-673X Ádám Mészáros http://orcid.org/0009-0002-4971-3370 Guofang Miao http://orcid.org/0000-0001-5532-932X Miguel Morata http://orcid.org/0000-0002-0537-6803 Paul Naethe http://orcid.org/0000-0002-3649-2786 Zoltán Nagy http://orcid.org/0000-0003-2839-522X Krisztina Pintér http://orcid.org/0000-0001-8737-706X R. Reddy Pullanagari http://orcid.org/0000-0001-6560-986X Anshu Rastogi http://orcid.org/0000-0002-0953-7045 Bastian Siegmann http://orcid.org/0000-0002-1232-7102 Sheng Wang http://orcid.org/0000-0003-3385-3109 Chenhui Zhang http://orcid.org/0000-0003-3915-6099 Data availability statement The data simulated in this work have been published in Open Access repositories. References Camino, C., K. Araño, J. A. Berni, H. Dierkes, J. L. Trapero-Casas, G. León-Ropero, M. Montes-Borrego, et al. 2022. “Detecting Xylella Fastidiosa in a Machine Learning Framework Using Vcmax and Leaf Biochemistry Quantified with Airborne Hyperspectral Imagery.” Remote Sensing of Environment 282:113281. https://doi.org/10.1016/j.rse.2022.113281 . Damm, A., S. Cogliati, R. Colombo, L. Fritsche, A. Genangeli, L. Genesio, J. Hanus, et al. 2022. “Response Times of Remote Sensing Measured Sun-Induced Chlorophyll Fluorescence, Surface Temperature and Vegetation Indices to Evolving Soil Water Limitation in a Crop Canopy.” Remote Sensing of Environment 273:112957. https://doi.org/10.1016/j.rse.2022.112957 . Del Río-Mena, L. W. Trinidad, G. Tsegay Tesfamariam, O. Beukes, and A. Nelson. 2020. “Remote Sensing for Mapping Ecosystem Services to Support Evaluation of Ecological Restoration Interventions in an Arid Landscape.” Ecological Indicators 113:106182. https://doi.org/10.1016/j. ecolind.2020.106182 . Demmig-Adams, B., C. M. Cohu, O. Muller, and W. W. Adams. 2012. “Modulation of Photosynthetic Energy Conversion Efficiency in Nature: From Seconds to Seasons.” Photosynthesis Research 113 (1): 75–88. https://doi.org/10.1007/s11120-012-9761-6 . Duveiller, G., F. Filipponi, S. Walther, P. Köhler, C. Frankenberg, L. Guanter, and A. Cescatti. 2020. “A Spatially Downscaled Sun-Induced Fluorescence Global Product for Enhanced Monitoring of Vegetation Productivity.” Earth System Science Data 12 (2): 1101–1116. https://doi.org/10.5194/ essd-12-1101-2020 . ESA. 2015. Report for Mission Selection: FLEX. Noordwijk, The Netherlands: ESA. Fahad, S., A. A. Bajwa, U. Nazir, S. A. Anjum, A. Farooq, A. Zohaib, S. Sadia, et al. 2017. “Crop Production Under Drought and Heat Stress: Plant Responses and Management Options.” Frontiers in Plant Science 8. https://doi.org/10.3389/fpls.2017.01147 . Farella, M. M., J. B. Fisher, W. Jiao, K. B. Key, and M. L. Barnes. 2022. “Thermal Remote Sensing for Plant Ecology from Leaf to Globe.” Journal of Ecology 110 (9): 1996–2014. https://doi.org/10.1111/ 1365-2745.13957 . Farquhar, G. D., and T. D. Sharkey. 1982. “Stomatal Conductance and Photosynthesis.” Annual Review of Plant Physiology 33 (1): 317–345. https://doi.org/10.1146/annurev.pp.33.060182.001533 . Féret, J.-B., K. Berger, F. de Boissieu, and Z. Malenovský. 2021. “PROSPECT-PRO for Estimating Content of Nitrogen-Containing Leaf Proteins and Other Carbon-Based Constituents.” Remote Sensing of Environment 252:112173. https://doi.org/10.1016/j.rse.2020.112173 . INTERNATIONAL JOURNAL OF REMOTE SENSING 23