Deep learning-based crop mapping from hyperspectral satellite data offers immense potential for capturing subtle phenological differences, yet leveraging sparse time series remains a major methodological challenge. This study evaluates the ability of the EnMAP sensor to identify nine major crop types in the intensive agricultural landscape of Southeastern Hungary. We utilized a limited time series (November, March, August) to benchmark two modeling strategies: a single-date dual-stream spatial–spectral 2D-CNN (DSS-2D) and a multi-temporal 3D-SE-ResNet. Model performance was assessed using parcel-level spatial cross-validation to ensure realistic accuracy estimates and reduce spatial autocorrelation bias. The results demonstrate that the DSS-2D model achieved superior single-date accuracy (OA > 97%), significantly outperforming pixel-based baselines. Furthermore, the multi-temporal 3D-SE-ResNet achieved a robust seasonal accuracy of 92.9%, effectively compensating for temporal sparsity by exploiting the deep spectral information of the SWIR domain. This study confirms that treating hyperspectral data as a 3D volume enables the extraction of phenological traits even from limited observations. These findings provide a strong proof-of-concept for the operational feasibility of future missions such as Copernicus CHIME for continental-scale food security monitoring.

Deep learning-based crop mapping from hyperspectral satellite data offers immense potential for capturing subtle phenological differences, yet leveraging sparse time-series remains a major methodological challenge. This study evaluates the EnMAP sensor for identifying nine major crop types in the intensive agricultural landscape of Southeastern Hungary. We utilized a limited time series (November, March, August) to benchmark two modeling strategies: a single-date Dual-Stream Spatial-Spectral 2D-CNN (DSS-2D) and a multi-temporal 3D-SE-ResNet. Model performance was assessed using parcel-level spatial cross-validation to ensure realistic accuracy estimates and reduce spatial autocorrelation bias. Results demonstrate that the DSS-2D model achieved superior single-date accuracy (OA > 97%), significantly outperforming pixel-based baselines. Furthermore, the multi-temporal 3D-SE-ResNet achieved a robust seasonal accuracy of 92.9%, effectively compensating for temporal sparsity by exploiting the deep spectral information of the SWIR domain. The study confirms that treating hyperspectral data as a 3D volume enables the extraction of phenological traits even from limited observations. These findings provide a strong proof-of-concept for the operational feasibility of future missions like Copernicus CHIME for continental-scale food security monitoring.

Groundwater recharge constrains sustainable aquifer use yet cannot be directly measured across entire basins. Recharge is often represented as a single recharge-potential surface, which can conflate favorable recharge conditions with the subsequent aquifer response. Here, we present a dual-indicator framework that explicitly separates climatic recharge opportunity from groundwater response. A climatic water-balance proxy (R_WB) was used to represent recharge opportunity, whereas median annual groundwater-level fluctuation (MedAF) provided an independent observation-based indicator of regional aquifer response. Machine-learning models were used to emulate the spatial organization of R_WB and to classify MedAF using climatic, soil-hydraulic, terrain, hydrogeomorphic, and land-cover predictors, with rigorous spatial validation. Applied to the Dél-Békés sector of the Maros alluvial fan (Hungary), both models achieved areas of applicability exceeding 95% of the basin, although observational support for MedAF varied with proximity to the 57 monitoring wells. Despite their physical connection, the two indicators showed limited spatial correspondence, agreeing across only 35.7% of the area. SHAP attribution for the R_WB emulator emphasized soil-hydraulic, hydrogeomorphic, and land-cover predictors, whereas the MedAF classifier was associated more strongly with climatic and topographic predictors. Applicability-constrained screening identified 100 candidate managed aquifer recharge (MAR) sites covering 18.03 km². By separating recharge opportunity from groundwater response, the framework provides a potentially transferable strategy for regional groundwater assessment and first-stage MAR screening, helping prioritize areas for subsequent site-specific investigation in heterogeneous alluvial aquifers.