Soil salinization will affect 50% of global cropland areas by 2050 and represents a major threat to agricultural production and food sovereignty. As soil salinity monitoring is costly and time consuming, many regions of the world undertake very limited soil salinity observation (in space and time), preventing the accurate assessment of soil salinity hazards. In this context, this study assesses the relative performance of Sentinel-1 radar and Sentinel-2 optical images, and the combination of the two, for monitoring changes in soil salinity at high spatial and temporal resolution, which is essential to evaluate the mitigation measures required for the sustainable adaptation of agriculture practices. For this purpose, an improved learning database made of 863 soil electrical conductivity (i.e., soil salinity) observations is considered for the training/validation step of a Random Forest (RF) model. The RF model is successively trained with (1) only Sentinel-1, (2) only Sentinel-2 and (3) both Sentinel-1 and -2 features using the Genetic Algorithm (GA) to reduce multi-collinearity in the independent variables. Using k-fold cross validation (3-fold), overall accuracy (OA) values of 0.83, 0.88 and 0.95 are obtained when considering only Sentinel-2, only Sentinel-1 and both Sentinel-1 and -2 features as independent variables. Therefore, these results highlight the clear complementarity of radar (i.e., Sentinel-1) and optical (i.e., Sentinel-2) images to improve soil salinity mapping, with OA increases of approximately 10% and 7% when compared to Sentinel-2 and Sentinel-1 alone. Finally, pre-sowing soil salinity maps over a five-year period (2019–2023) are presented to highlight the benefit of the proposed procedure to support the sustainable management of agricultural lands in the context of soil salinization on a regional scale.

Soil moisture content (SMC) is a critical parameter for agricultural productivity, particularly in semi-arid regions, where irrigation practices are extensively used to offset water deficits and ensure decent yields. Yet, the socio-economic and remote context of these regions prevents sufficiently dense SMC monitoring in space and time to support farmers in their work to avoid unsustainable irrigation practices and preserve water resource availability. In this context, our study addresses the challenge of high spatial resolution (i.e., 20 m) SMC estimation by integrating remote sensing datasets in machine learning models. For this purpose, a dataset made of 166 soil samples’ SMC along with corresponding SMC, precipitation, and radar signal derived from Soil Moisture Active Passive (SMAP), Integrated Multi-satellitE Retrievals for GPM (IMERG), and Sentinel-1 (S1), respectively, was used to assess four machine learning models’ (Decision Tree—DT, Random Forest—RF, Gradient Boosting—GB, Extreme Gradient Boosting—XGB) reliability for SMC mapping. First, each model was trained/validated using only the coarse spatial resolution (i.e., 10 km) SMAP SMC and IMERG precipitation estimates as independent features, and, second, S1 information (i.e., 20 m) derived from single scenes and/or composite images was added as independent features to highlight the benefit of information (i.e., S1 information) for SMC mapping at high spatial resolution (i.e., 20 m). Results show that integrating S1 information from both single scenes and composite images to SMAP SMC and IMERG precipitation data significantly improves model reliability, as R² increased by 12% to 16%, while RMSE decreased by 10% to 18%, depending on the considered model (i.e., RF, XGB, DT, GB). Overall, all models provided reliable SMC estimates at 20 m spatial resolution, with the GB model performing the best (R² = 0.86, RMSE = 2.55%).

To improve crop yields and incomes, farmers consistently adapt their practices to climate and market fluctuations, resulting in highly variable crop field distribution and coverage in space and time. As these dynamics illustrate farmers’ challenges, up-to-date crop-type mapping is essential for understanding farmers’ needs and supporting their adoption of sustainable practices. With global coverage and frequent temporal observations, remote sensing data are generally integrated into machine learning models to monitor crop dynamics. Unlike physical-based models that rely on straightforward use, implementing machine learning models requires extensive user interaction. In this context, this study assesses how sensitive the models’ outputs are to feature selection and hyperparameter tuning, as both processes rely on user judgment. To achieve this, Sentinel-1 (S1) and Sentinel-2 (S2) features are integrated into five distinct models (Random Forest (RF), Support Vector Machine (SVM), Light Gradient Boosting (LGB), Histogram-based Gradient Boosting (HGB), and Extreme Gradient Boosting (XGB)), considering several features selection (Variance Inflation Factor (VIF) and Sequential Feature Selector (SFS)) and hyperparameter tuning (Grid-Search) setup. Results show that the preprocess modeling feature selection (VIF) discards the features that the wrapped method (SFS) keeps, resulting in less reliable crop-type mapping. Additionally, hyperparameter tuning appears to be sensitive to the input features, and considering it after any feature selection improved the crop-type mapping. In this context a three-step nested modeling setup, including first hyperparameter tuning, followed by a wrapped feature selection (SFS) and additional hyperparameter tuning, leads to the most reliable model outputs. For the study region, LGB and XGB (SVM) are the most (least) suitable models for crop-type mapping, and model reliability improves when integrating S1 and S2 features rather than considering S1 or S2 alone. Finally, crop-type maps are derived across different regions and time periods to highlight the benefits of the proposed method for monitoring crop dynamics in space and time.

To improve crop yields and economic incomes, farmers consistently adapt their practices to climate and market fluctuations, resulting in highly variable crop field distribution and coverage in space and time. As these dynamics ilustrate, up-to-date crop-type mapping is essential to understand farmers’ needs and supporting them in adopting sustainable practices. With global coverage and frequent temporal observations, remote sensing data are generally integrated in machine learning models to monitor crop-type mapping dynamics. Unlike physical-based models that rely on straightforward use, the implementation of machine-learning approaches depends on deep interaction with users. In this context, the study assesses the output sensitivity of these models to features selection and hyper-parameter calibration, both of wich rely on user consideration. To do so, Sentinel-1 (S1) and Sentinel-2 (S2) features are integrated into five distinct models (RF, SVM, LGB, HGB, XGB), considering different features selection (VIF and SFS) and hyper-parameter calibration set-up. Results show that pre-process modeling VIF feature selection discards features that wrapped SFS feature selection keeps, resulting in less reliable crop-type mapping compared to using SFS. Additionally, hyper-parameter calibration appears to be sensitive to the input feature and its consideration after any the feature selection improved the crop-type mapping. In this context a three-step nested modelling set-up including a first hyper-parameters calibration followed by a wrapped feature selection (SFS) and another hyper-parameter calibration, lead to the most reliable model outputs. Across the considered region, LGB and XGB (SVM) are the most (less) suitable model for crop-type mapping and models reliability improved when integrated S1 and S2 features rather than the consideration of S1 or S2 alone. Finally, crop-type maps are derived across different regions and periods to highlight the benefits of the proposed method to monitor crops’ dynamics in space and time.

Due to its sensitivity to topographic and land use land cover features, air temperature (maximum, minimum, and mean—T_x, T_n, and T_mean) is extremely variable in space and time. The sparse and unevenly distributed meteorological stations observed across remote regions cannot monitor such variability. Freely available, gridded temperature datasets (T-datasets) are positioned as an opportunity to overcome this issue. Still, their coarse spatial resolution (i.e., ≥5 km) does not allow for the observation of air temperature variations on a fine spatial scale. In this context, a set of variables that have a close relationship with daily air temperature (MODIS maximum, minimum, and mean Land Surface Temperature—LST_x, LST_n, and LST_mean; MODIS NDVI; SRTM topographic features—elevation, slope, and aspect) are integrated in three regression machine-learning models (Random Forest—RF, eXtreme Gradient Boosting—XGB, Multiple Linear Regression—MLR) to propose a T-dataset estimates (T_x, T_n, and T_mean) spatial resolution downscaling framework. The approach consists of two main steps: firstly, the machine-learning models are trained at the native 5 km spatial resolution of the studied T-dataset (i.e., CHIRTS); secondly, the application of the trained machine-learning models at a 1 km spatial resolution to downscale CHIRTS from 5 km to 1 km. The results show that the method not only improves the spatial resolution of the CHIRTS dataset, but also its accuracy, with higher improvements for T_n than for T_x and T_mean. Among the considered models, RF performs the best, with an R², RMSE, and MAE improvement of 2.6% (0%), 47.1% (6.1%), and 55.3% (7%) for T_n (T_x). These results will support air temperature monitoring and related extreme events such as heat and cold waves, which are of prime importance in the actual climate change context.