Comparison of Sentinel-2 and MODIS for estimating GPP along an ecosystem gradient in eastern Germany

Accurate gross primary production (GPP) estimation is crucial for understanding ecosystem carbon dynamics and their responses to climatic variability, particularly in climate-sensitive regions, such as eastern Germany. This study evaluated the performance of three approaches, the vegetation photosynthesis model (VPM), vegetation indices (VI) and random forest (RF) models using Sentinel-2 (S2) and MODIS data to identify a suitable framework for estimating GPP across agricultural, forest and grassland ecosystems. Among all approaches, the EVI-based RF model using S2 data demonstrated superior performance across varying climatic conditions, seasonal patterns, and water stress periods across all ecosystems, achieving the highest accuracy and stability. In contrast, the MODIS-based RF model performed better under wet conditions but showed limited responsiveness during peak growth phases and under water stress because of its coarse spatial resolution. The EVI and LSWI were key predictors under dry conditions, while EVI dominated under drought conditions, and LST and PAR under wet conditions. These patterns reflect the realistic shift in limiting factors between energy- and water-limited states. Upscaling to the ecosystem level using this S2-based model estimated 19%–28% higher GPP values than that of the MODIS model and captured the spatial heterogeneity of GPP along vegetation cover gradients. Overall, these findings emphasize that integrating S2 data with optimized RF modelling provides a robust framework for precise GPP estimation and improved understanding of ecosystem responses under changing climatic conditions.