Agricultural drought is an extreme event that threatens crop production, particularly in the Mediterranean region. To manage its cascading impacts sustainably, this study aims to evaluate the potential of a 1-km surface soil moisture (SSM) product and to intercompare it with SSM products at coarser spatial resolutions in order to identify periods of drought. The study covered two Mediterranean sites: the Occitanie region in France and several bioclimatic regions in Tunisia. The SSM dataset used in this paper is produced from Sentinel-1 and Sentinel-2 data based on a 100 m resolution land cover map (CGLS-LC100), then aggregated to 1 km (hereafter referred to as HRSM). First, the performance of the HRSM product was compared to that of two other SSM products: the Soil Moisture Active Passive (SMAP) and the European Space Agency Climate Change Initiative (ESA CCI). The results show an overall good coherence between the different time series of the HRSM and the other two SSM products over France and Tunisia. However, HRSM and SMAP were more correlated compared to HRSM and CCI SSM. The intercomparison of products based on different bioclimatic regions in Tunisia shows that over the humid and subhumid regions, the best agreement was observed between the CCI SSM and HRSM time series. Second, to identify drought events, the ESA CCI root zone soil moisture (RZSM) at a depth of 1 m was also used as confirmation of agricultural drought. The monthly averages of SSM and RZSM were used to compute a drought index. The results show that the drought events identified by the SSM drought index are also observed with the RZSM drought index. Among the SSM products, HRSM has the advantage of identifying drought events only in grasslands and agricultural fields. This is ensured by its native high spatial resolution, which allows non-agricultural and non-grassland fields to be masked before being aggregated to 1 km. However, thanks to their high revisit frequency, the CCI SSM and SMAP are better able than HRSM to capture the dynamics of SSM in response to rainfall. They therefore make it more efficient to identify all drought events. Consequently, the proposed identification method in this study is sensitive to the availability of SSM products and the revisit time required to sufficiently cover rainfall events and different hydrological processes.

Improving the spatial resolution of thermal imagery is essential for agricultural field management, especially in developing countries where fields are fragmented and heterogeneous. The existing downscale land surface temperature (LST) products are a promising solution while awaiting for further advances in satellite-based thermal sensors. In this study, a linear regression-based approach was proposed to downscale LST from 1 km to 10 m spatial resolution. The proposed method is hereafter referred to as Stratified and Adaptive Regression for Land Surface Temperature Downscaling (STAR-LST). Taking advantage of the linearity between LST and NDVI under homogeneous conditions, multiple linear regression models were first derived for different regions in the (LST, NDVI) feature space at coarse resolution. These regions were defined by partitioning the triangular form of the (LST, NDVI) feature space into sub-triangles, sharing a common top vertex. Each model is then applied to the corresponding NDVI range, allowing thus to derive LST at finer resolution. LST estimates were validated against in-situ measurements collected over olive trees, Kernza, wheat, and barley. Good results were obtained with RMSE values ranging from 2.79 to 4.49 degrees C for STAR-LST. This presents an improvement of an average of 33 % compared to the two classical methods DisTrad and TsHARP (RMSE in the range 4.18-6.83 degrees C), which showed similar accuracy. In addition, STAR-LST enabled daily LST estimation whenever coarse-resolution data were available, using interpolated NDVI to generate a high spatio-temporal LST product suitable for managing agricultural fields, particularly the small size fields.

This work presents a hybrid physics-guided machine learning framework designed to improve the simulation of Sentinel-1 backscatter over winter wheat fields in the semiarid region of Chichaoua, Morocco. The approach integrates physical constraints derived from the Water Cloud Model (WCM) within a neural-network architecture, enabling the estimation of soil and vegetation scattering components while maintaining physical consistency. By embedding WCM-based relationships into both the model structure and the loss function, the proposed method enhances interpretability and robustness under strong environmental variability. Applied to multi-season datasets combining Sentinel-1 observations and in-situ measurements, the framework demonstrates superior accuracy compared to purely data-driven models, supporting reliable monitoring of soil-vegetation water dynamics in water-limited agricultural systems.

Cloud cover creates frequent data gaps in high-resolution satellite imagery, particularly from Sentinel-2. These disrupt its continuity and reliability for time-sensitive applications such as water resource management, irrigation scheduling and crop health or yield prediction. Several approaches have been proposed in the literature, but there is a lack of performance comparisons. To address this challenge, this study aims to establish an evaluation framework to investigate the effectiveness of spatial, temporal, spatio-temporal, and spatio-spectral gap-filling approaches for restoring cloud-induced gaps in Sentinel-2 imagery using only satellite derived data. The evaluation was conducted using simulated cloud scenarios for both spatial within a single image and temporal for a time series cube of images. The methods were tested on key Sentinel-2 bands visible (B02, B03, B04), near-infrared (B08), and shortwave infrared (B11 and B12). The performance of each method was assessed using the coefficient of determination (R²), rRMSE and bias. Among the evaluated methods, spatio-temporal category, especially clustering and linear regression (CLR), showed the highest accuracy and robustness across all gap scenarios and types. Spatio-temporal Deep learning (DL) also performed well but required more training effort and failed in generalizing over all scenarios. Also, spatio-spectral approaches like SSRF showed strong results in visible and NIR bands. In contrast, spatial methods such as kriging struggled with larger or irregular gaps. These findings offer a comparative evaluation of gap-filling methods, highlighting the strengths and limitations of each approach to guide the selection of appropriate methods and explore new prospects for improving gap-filling techniques. To ensure reproducibility all codes used in this study are publicly available on https://github.com/said-grich/gap-filling.

In semi-arid regions, irrigation represents over 85% of water use, making its efficient management crucial. Accurate estimation of evapotranspiration (ETR) is essential to determine crop water requirements. The FAO-56 method, based on crop coefficient (Kc), is widely used for ETR estimation. While NDVI from optical data has been applied to estimate Kc, it shows limited variation for tree crops despite phenological changes. Radar data, however, are sensitive to structural vegetation changes at the wavelength scale. This study evaluates the capability of C-band radar data to estimate Kc for olive trees. Empirical relationships were assessed between Kc and temporal coherence (ρ) in VV and VH polarizations, as well as backscattering coefficients (σ0) from Sentinel-1 over the 2021 seasonal cycle. A high agreement is observed between σ0VV and Kc, with a correlation coefficient of 0.76 and an RMSE of 0.11, highlighting the capability of radar data for improving evapotranspiration estimation in tree crops.

High-resolution evapotranspiration mapping plays a key role in water management across heterogeneous agricultural areas where surface conditions change rapidly. Contextual trapezoidal models offer a practical solution, and the optical trapezoidal formulation in particular provides high-resolution retrievals because it relies solely on optical data, requires minimal inputs, and is straightforward to apply. This approach operates within the STR-NDVI space, where STR is the shortwave-infrared transformed reflectance, and its accuracy depends on how the dry and wet edges are positioned within this space. Fixed-edge methods, including the Spatiotemporal Aggregation and Regression (STAR) algorithm, do not explicitly capture scene-specific temporal variability driven by soil and vegetation dynamics. We addressed this limitation by proposing a scene-adaptive alternative, the SpatioDynamic Aggregation and Regression (SDAR) algorithm, which retrieves new edges for every Sentinel-2 acquisition. Over an olive orchard in Morocco, SDAR increased the explained variance from 0.59 to 0.69 (24.4%) and reduced the RMSE and bias by 0.31 mm/day (44%) and 0.51 mm/day ($\approx 94 \%$), respectively. Spatial patterns showed the same improvement, with the dynamic approach maintaining realistic surface contrasts while the static option consistently overestimated the retrieved values. These findings indicate that dynamic edges strengthen OPTRAM-ET and enable automatic edge parametrization without in-situ calibration, supporting transferable 10 m ET mapping for operational monitoring in complex agricultural landscapes.

Accurate estimation of Gross Primary Productivity (GPP) in semi-arid regions is challenging because vegetation is primarily limited by water availability rather than light. Traditional light-use-efficiency (LUE) models, such as MOD17, rely on meteorological stress scalars based on vapor pressure deficit (VPD) and temperature, which represent atmospheric demand but not the actual water stress experienced by the canopy. This results in GPP underestimation during drought. To address this limitation, a stress constraint based on the Crop Water Stress Index (CWSI) was incorporated into the MOD17 LUE framework. CWSI was computed using the Two-Source Energy Balance (TSEB) model. The modified model was evaluated across five eddy-covariance systems in the Tensift Basin (Morocco) representing old olive orchards, medium-age olive trees, young olive trees, and bare soil. Results show that the CWSI based LUE model consistently outperformed the VPD-temperature stress constraint, reduced the overall RMSE from 1.70 to $1.22 \text{gC} \mathrm{m}^{-2} \text{day}^{-1}$ and corrected the systematic underestimation of the VPD-temperature formulation from −1.21 to $0.12 \text{gC} \mathrm{m}^{-2}$ day $^{-1}$. Under drought conditions, CWSI stress constraint maintained strong performance, reducing RMSE from 2.32 to 1.01 gC $\mathrm{m}^{-2}$ day ${}^{-1}$ and bias shifted from −2.12 to $0.44 \text{gC} \mathrm{m}^{-2}$ day ${}^{-1}$. The findings demonstrate that the CWSI constraint effectively reduces model error and resolves the underestimation problem by capturing canopy-level water stress more accurately than atmospheric stress scalars.

Accurate evapotranspiration (ET) estimation at high spatial resolution is essential for resolving fine-scale water and energy exchanges across ecosystems, where soil moisture (SM) exerts strong, spatially variable control on soil evaporation. The Priestley–Taylor Jet Propulsion Laboratory (PT-JPL) model is widely used for ET monitoring, yet its reliance on relative humidity and vapor pressure deficit to constrain soil evaporation can misrepresent fine-scale moisture limitations where SM and atmospheric humidity decouple. To address this, we introduced an SM constraint from the Optical Trapezoid Model (OPTRAM). Because broader OPTRAM application is challenged by uncertainties in edge geometry, parameterization strategy, and sensitivity to surface heterogeneity, we developed the SpatioDynamic Aggregation and Regression (SDAR) algorithm, which adapts trapezoid edges for each Sentinel-2 scene using only its spectral feature space, enabling an automated SM constraint for PT-JPL. Across 78 sites spanning diverse climates and land covers, SDAR provided a highly stable and transferable SM constraint compared to the static STAR scheme: it increased R² by 7%, reduced RMSE by 6.27%, and lowered bias by 71%. SDAR proved robust to edge shape by requiring only a linear edge, and it optimized accuracy at a Region of Interest (ROI) of approximately 5 km, beyond which sensitivities to both edge shape and ROI size vanished. SDAR's superiority was most pronounced over sparsely vegetated surfaces, indicating that transferability is governed primarily by surface characteristics rather than climate. Benchmarking PT-OPTRAM-SDAR against PT-JPL and five SWIR-based constraints shows ET bias reductions of 77.7% and 66.4% relative to PT-JPL and the best SWIR-based variant, and lowered RMSE by 9% against both, while improving R² by 10.3% over PT-JPL and matching the best SWIR-based variant. These findings demonstrate that contextual SM information strengthens PT-JPL and position PT-OPTRAM-SDAR as a physically consistent, scale-adaptive framework for reliable 10 m ET mapping with limited inputs

This study aims to investigate the potential of sub-diurnal C-band radar data for maize crop monitoring. To this end, a tower-based experiment equipped with a C-band radar antenna (15-min acquisition interval) was conducted in a maize field in 2021. In situ measurements of soil and vegetation were carried out between September and December 2021. To analyze the radar response to plant water stress, a period of induced stress was implemented around the development peak. Both the temporal coherence (rho) and radar backscattering coefficient (sigma (0)) are analyzed. The analysis shows that the temporal coherence exhibits a marked daily cycle. More specifically, a morning drop (Delta|rho|(morning)) was observed prior to the onset of wind. Previous studies on wheat crops and olive orchards have attributed this drop to the onset of the plant's physiological activity at dawn. The daily mean sigma(0) values across all three polarizations were strongly correlated with both evapotranspiration (ETR) and evaporative fraction during both the growth and maturity stages (r > 0.60). This observed correlation is due to structural changes in the canopy as vegetation grows, associated with increased transpiration. Despite significant differences in acquisition configurations between tower-based radar measurements and Sentinel-1, a remarkable overall agreement is observed between both datasets. In contrast to rho, Sentinel-1 sigma (0) detects the water stress period, with a decrease of about 2 dB coinciding with intensifying water stress. These findings emphasize the importance of disentangling physiological drivers from structural indicators when interpreting sigma(0) dynamics in relation to plant-water interactions.

Leaf chlorophyll content (Cab) is an important indicator of crop physiological status and nitrogen condition. This study investigates the retrieval of Cab from Sentinel 2 data over winter wheat using two approaches: an empirical red-edge spectral index method and a physically based PROSAIL inversion. The empirical method is locally calibrated using in situ measurements and evaluated on an independent field. The PROSAIL approach estimates Cab without site calibration. Results indicate that the empirical method performs well on the site where it was calibrated but shows limited transferability when applied to another site. In contrast, the PROSAIL inversion provides more robust and consistent Cab estimates across fields and better captures seasonal chlorophyll dynamics. These results demonstrate the potential of physically based approaches for reliable and operational monitoring of leaf chlorophyll content using Sentinel-2 images.

Surface soil moisture (SSM) products at high spatial resolution are increasingly available, either from the disaggregation of coarse-resolution products such as SMAP and SMOS, or from high-resolution radar data such as Sentinel-1. In contrast to coarse resolution products, there is a lack of intercomparison studies of high spatial resolution products, which are more relevant for applications requiring the plot scale. In this context, the objective of this work is the evaluation and intercomparison of three high spatial resolution SSM products on a large database of in situ SSM measurements collected on two different sites in the Urgell region (Catalonia, Spain) in 2021. The satellite SSM products are: i) SSMTheia product at the plot scale derived from a synergy of Sentinel-1 and Sentinel-2 using a machine learning algorithm; ii) SSMρ product at 14 m resolution derived from the Sentinel-1 backscattering coefficient and interferometric coherence using a brute-force algorithm; and iii) SSMSMAP20m product at 20 m resolution obtained from the disaggregation of SMAP using Sentinel-3 and Sentinel-2 data. Evaluation of the three products over the entire database showed that SSMTheia and SSMρ yielded a better estimate than SSMSMAP20m, and SSMρ is slightly better than SSMTheia. In particular, the correlation coefficient is higher than 0.4 for 72%, 40% and 27% of the fields using SSMρ, SSMTheia and SSMSMAP20m, respectively. The lower performance of SSMTheia compared to SSMρ is due to the saturation of SSMTheia at 0.3 m3/m3. The time series analysis shows that SSMSMAP20m is able to detect rainfall events occurring at large scale while irrigation at the plot scale are not caught. This is explained by the use of Sentinel-2 reflectances, which are not linked to surface water status, for the disaggregation of Sentinel-3 land surface temperature. The approach can therefore be improved by using high spatial and temporal resolution thermal data in the perspective of new missions such as TRISHNA and LSTM. Finally, the results show that although reasonable estimates are obtained for annual crops using SSMTheia and SSMρ, poor performance is observed for trees, suggesting the need for better representation of canopy components for tree crops in SSM inversion approaches.

Root zone soil moisture (RZSM) is a key variable controlling the soil-vegetation-atmosphere exchanges. Its estimation is vital for monitoring hydrological, meteorological and agricultural processes. A number of large-scale products exist but with a coarse resolution (>1 km), which is not suitable for plot-scale studies. The aim of this work is to map RZSM, for the first time, at very high spatial resolution using a very high spatial resolution surface soil moisture (SSM) product and a recursive exponential filter. SSM is estimated from Sentinel-1 data using the water cloud model at a resolution of approximately 50 m. The approach was evaluated on a database consisting of 12 fields, including 7 winter wheat and 5 summer maize fields, irrigated using different techniques. The results show that the approach performs reasonably well using Sentinel-1 SSM product with correlation coefficient (R) between 0.3 and 0.82, root-mean-square error (RMSE) between 0.05 and 0.12 m(3)/m(3) and a bias in the range -0.1-0.07 m(3)/m(3), at 15-20 cm depth. This is equivalent to R = 0.6, RMSE = 0.12 m(3)/m(3) and bias = 0.07 m(3)/m(3) using the entire database, which is quite low compared to the use of in situ SSM measurements (R = 0.81, RMSE = 0.07 m(3)/m(3) and bias = 0.03 m(3)/m(3)). This is related to inaccuracies in the SSM product, where fields with good SSM estimation also resulted in good RZSM estimation and conversely. In addition to SSM, the approach is also sensitive to its time constant T. Analysis of RZSM sensitivity to T shows that the optimum T value depends on soil texture, climate and measurement depth. In particular, low optimum T values (1 day) are obtained for loamy and sandy loam soils, while higher values (5-10 days) are optimal for soils with a high clay fraction, at 15-20 cm depth. These values increase with soil depth and are influenced by seasonal atmospheric demand. Combined to reasonable statistical metrics, the spatial variability depicted by the RZSM maps opens up prospects for high-resolution RZSM mapping from Sentinel-1 SSM data using a simple approach over annual crops. This is of prime relevance for agricultural applications requiring very high-resolution estimation at plot scale, such as crop yield, irrigation and fertilizer management, as well as for the assessment of inter-plot variability.

Soil moisture is a critical variable in many fields of study, including meteorology, hydrology, and agricultural sciences. In this latter, surface soil moisture (SSM) is crucial for plant growth and development and consequently for yield estimation. Synthetic Aperture Radar (SAR) can be a reliable and trustworthy data source for SSM inversion through the use of empirical, semi-empirical and physically based models. Each of these methods has its own strengths and limitations, depending on the specific application and the environmental conditions. Likewise, there has been a recent surge in attention towards the use of machine learning regression algorithms in the SSM inversion process from SAR data. This work aims to assess the effectiveness of the two algorithms neural network (i.e. single-layer artificial neural network (ANN) and deep neural network (DNN)) for retrieving SSM by utilizing data gathered from diverse rainfall and irrigated (sprinkler) wheat fields located in Tunisia and Morocco. The comparison between predicted and measured SSM showed that the best retrieval results were obtained using sentinel-1 data at VV polarization with R of 0.75 and 0.76 for ANN and DNN, respectively. The RMSE was about 0.05 m^3/m^3 for both algorithms. Overall, the performance of single-layer ANN mimics the highly complex multi-layers DNN in terms of statistical results at VV polarization.

Recent studies have shown that radar temporal coherence over tropical and boreal forests undergoes a diurnal cycle as a result of a combined effect of the wind-induced motion of scatterers and of the change and displacement of water within the plant in response to the transpiration process. Within this context, the objective of this paper is to investigate, for the first time, the diurnal cycle of temporal coherence over wheat crops in relation to its development and physiological functioning throughout the agricultural season. A ground-based experiment was installed in Morocco, targeting a wheat field during the 2020 agricultural season. The radar system, essentially based on a Vector Network Analyzer (VNA) connected to 6C -band antennas installed at the top of a 20 m tower, has enabled quad-polarimetric acquisitions every 15 min. In parallel, evapotranspiration, soil moisture and meteorological variables are automatically measured in addition to above-ground biomass and vegetation water content collected during field campaigns. The results show that the temporal coherence with a 15 min baseline follows a marked diurnal cycle characterized by variable amplitude according to the phenological stage, with high values during the night, a significant morning drop to reach the lowest values in the late afternoon followed by an increase to recover the high nighttime values. The rate of the drop at dawn is shown to be related to the increase of evapotranspiration (r = 0.80 at VV polarization) when the wheat is covering the soil and the transpiration dominate the evapotranspiration process. This supports the assumption of a physiological effect related to water movement entailing a decorrelation. By contrast, the daily minimum of temporal coherence occurring in the late afternoon correlates well to the daily maximum of wind (r = 0.7). Interestingly enough, the amplitude of the diurnal cycle exhibit a marked seasonal evolution characterized by an increase of 85% from tillering to maturity in relation to the wheat development. At the early start of the season when the soil is almost bare, irrigation events impact slightly the diurnal cycle of temporal coherence. Likewise, it is shown that the presence of dew in the early morning has led to a decrease of the decorrelation rate. Temporal coherence dynamic has also been investigated for longer baselines up to 22 days. Results indicate a stronger decorrelation than what has been observed on tropical and boreal forests by previous studies with values below 0.4 for baselines above 2 days. Taken together, the results of this work demonstrate the unique potential of sub-daily Cband data for monitoring crop water status by future geostationary radar missions such as Hydroterra.

Annual crop monitoring is a key parameter for managing agricultural strategies. Several studies have relied on remote sensing products such as the normalized difference vegetation index (NDVI) as a vegetation dynamic metric. However, the dependence of optical data on weather conditions limits its availability. In this study, we reconstruct the NDVI time series of wheat fields using the moving averages of the Sentinel-1 normalized VH/VV cross-polarization ratio (IN) and the interferometric coherence in VV polarization over wheat selected fields in a semiarid site in Tunisia during two seasons, from 2018 to 2020. The crop cycle is divided into two periods: before and after the heading phase, which occurs in approximately the middle of March. Due to the volume-scattering impact, the second phase is divided into the ripening and maturation phase (NDVI >= 0.4) and senescence phase (NDVI <0.4). To estimate the NDVI values, different methods are used: curve-fitting equations and machine learning regressors such as the random forest (RF) and the support vector regressor (SVR). Low root mean square error (RMSE) values characterize NDVI estimation during the first period. In the second period, the RMSE values reach 0.06 when the NDVI is lower than 0.4. When the NDVI values exceed 0.4 in the second period, lower accuracy marks the NDVI estimation using the curve-fitting equations as a function of IN or coherence. Relative low accuracy characterizes the regression algorithms' estimations when NDVI >= 0.4 compared to their performance during the aforementioned periods. The proposed approach was tested on different wheat fields. The NDVI estimations are characterized by RMSE values varying between 0.12 and 0.19. The use of RF and SVR outperformed the curve-fitting methods with an RMSE equal to 0.12. The present findings revealed the high accuracy of the proposed approach to estimate the missing values of wheat fields NDVI values during the vegetation development period until heading and the senescence phase. The presence of the mutual effect of the vegetation water content and its volume complicated the NDVI estimation using the C-band data.

This article aimed to monitor vegetation using C-band radar data at a subdaily time step. To this end, radar measurements using tower-mounted antennas with a 15-min time step, along with physiology-related information (sapflow and micrometric dendrometry), were acquired quasi-continuously from March 2020 to December 2021 in an olive orchard located near Marrakech, Morocco. The article focused on temporal coherence, whose clear diurnal cycle (highest at night and lowest at the end of the afternoon) had been highlighted over tropical and boreal forests in previous studies. The results showed that coherence was highly sensitive to: wind-induced movement of scatterers, since coherence was lowest when wind speed was highest in late afternoon, and vegetation activity, especially its water dynamics, since the morning coherence drop coincided with the onset of sapflow and the daily evapotranspiration cycle, as well as the good agreement between the temporal drop rate of coherence and the daily residual variation in trunk circumference (i.e., deviation from long-term trend). Finally, coherence remained high for temporal baselines of several days, showing that sentinel-1 data (when both satellites are operational) may be well suited for such studies, especially with acquisitions made during morning passes, when wind speed is low. These results open perspectives for monitoring tree crop physiology using high-revisit-time radar observations.

This work aims to assess the effectiveness of machine learning (ML) algorithms and semiempirical models for surface soil moisture (SSM) retrieval by exploring the Sentinel-1 backscatter and interferometric coherence data. First, three commonly used categories of ML algorithms are evaluated using data gathered from diverse rainfed and irrigated wheat fields located in Morocco and Tunisia. Specifically, these algorithms include artificial neural network (ANN), deep neural network, three support vector regression (SVR) models [radial basis function (SVR_rbf), linear (SVR_linear), and polynomial (SVR_quad) kernels], and two tree-based methods [random forest and eXtreme Gradient Boosting (XGBoost)]. The comparison between predicted and measured SSM showed that the best retrieval results were obtained using Sentinel-1 data at VV polarization with R ranging between $ 0.68$ and $ 0.76$ and root-mean-square error (RMSE) of $ \text{0.05}\,\text{m}^{3}/\text{m}^{3}$ and $ \text{0.06}\,\text{m}^{3}/\text{m}^{3}$. Second, to further assess their transferability, the ANN, SVR_rbf, and XGBoost, which demonstrated the most favorable results from each category, were evaluated and compared against the coupled Water Cloud and Oh models (WCM), using a second dataset collected over a drip-irrigated wheat field in Morocco. Overall, the best retrieval results were achieved by ANN and SVR_rbf with R and RMSE of $ 0.81$ and $ \text{0.034}\,\text{m}^{3}/\text{m}^{3}$, respectively. In addition, their performances were consistent with that of WCM, which yielded R and RMSE values of $ 0.81$ and $ \text{0.04}\,\text{m}^{3}/\text{m}^{3}$, respectively. Finally, due to its good compromise between retrieval accuracy of SSM, processing time, and simplicity, SVR_rbf was chosen to generate high-resolution SSM maps from Sentinel-1 data over irrigated wheat fields.

The surface soil moisture (SSM) is a key variable for monitoring hydrological, meteorological and agricultural processes. It can be estimated from active and passive microwave remote sensing data. While coarse-resolution SSM products (> 1 km) have already been evaluated for a large range of ecosystems, such assessments lack very high-spatial-resolution SSM products, although they are increasingly available thanks to very high-resolution radar data or disaggregation methods applied to coarse-scale products. Within this context, the aim of the current study is to carry out, for the first time, an intercomparison of high-spatial resolution SSM products using a large in situ SSM database collected from 33 fields located in the Ebro basin (Spain) that were cultivated with different crops and irrigated using different techniques. Three products are considered: (i) SSMTheia at the field scale derived from Sentinel-1 and Sentinel-2 data using a machine learning algorithm; ii) SSM rho at 50-m resolution derived from the Sentinel-1 data using both the backscattering coefficient and the interferometric coherence based on the inversion of a simple radiative transfer model; and iii) SSMSMAP20m at 20-m resolution obtained by disaggregating SMAP SSM using Sentinel-3 and Sentinel-2 data. The statistical metrics computed on the whole database show that the two Sentinel-1 products outperform the disaggregated approach and that the SSM rho product exhibits better statistical metrics than the SSMTheia product. This is mainly attributed to the inability of the SSMTheia approach to retrieve SSM >0.3 m(3)/m(3.) The correlation coefficients are >0.4 (up to 0.8) for 72%, 40% and 27% of the fields using SSM rho, SSMTheia and SSMSMAP20m, respectively. Similarly, 80% of the fields had RMSE values between 0.06 m(3)/m(3) and 0.1 m(3)/m(3) using SSM rho product against 36% using SSMTheia and 27% using SSMSMAP20m. In addition, the time series analysis showed that SSMSMAP20m was able to detect large-scale wetting events such as rainfall that impacted the whole SMAP pixel while irrigation at the field scale was not detected, mainly because the very high-resolution Sentinel-2 data used for the disaggregation of Sentinel-3 land surface temperature were not related to the hydric status of the surface. The results show that while both Sentinel-1 products perform reasonably well for cereals and, to a lesser extent, for annuals, a drastic drop of the metrics is observed for tree crops. Finally, the spatial SSM pattern over the study area is also better depicted by the Sentinel-1 products than by the SSMSMAP20m by comparison to the airborne GLORI GNSS-R (Global Navigation Satellite System Reflectometry) SSM maps. This study highlights the limitations of SSM products over tree crops and provides insights for improving irrigation scheduling at the field scale.

Irrigation is the most water consuming activity in the world. Knowing the timing and amount of irrigation that is actually applied is therefore fundamental for water managers. However, this information is rarely available at all scales and is subject to large uncertainties due to the wide variety of existing agricultural practices and associated irrigation regimes (full irrigation, deficit irrigation, or over-irrigation). To fill this gap, we propose a two-step approach based on 15 m resolution Sentinel-1 (S1) surface soil moisture (SSM) data to retrieve the actual irrigation at the weekly scale over an entire irrigation district. In a first step, the S1-derived SSM is assimilated into a FAO-56-based crop water balance model (SAMIR) to retrieve for each crop type both the irrigation amount (Idose) and the soil moisture threshold (SMthreshold) at which irrigation is triggered. To do this, a particle filter method is implemented, with particles reset each month to provide time-varying SMthreshold and Idose. In a second step, the retrieved SMthreshold and Idose values are used as input to SAMIR to estimate the weekly irrigation and its uncertainty. The assimilation approach (SSM-ASSIM) is tested over the 8000 hectare Algerri-Balaguer irrigation district located in northeastern Spain, where in situ irrigation data integrating the whole district are available at the weekly scale during 2019. For evaluation, the performance of SSM-ASSIM is compared with that of the default FAO-56 irrigation module (called FAO56-DEF), which sets the SMthreshold to the critical soil moisture value and systematically fills the soil reservoir for each irrigation event. In 2019, with an observed annual irrigation of 687 mm, SSM-ASSIM (FAO56-DEF) shows a root mean square deviation between retrieved and in situ irrigation of 6.7 (8.8) mm week-1, a bias of +0.3 (-1.4) mm week-1, and a Pearson correlation coefficient of 0.88 (0.78). The SSM-ASSIM approach shows great potential for retrieving the weekly water use over extended areas for any irrigation regime, including over-irrigation.