Monitoring changes in the landscape water balance: validation of satellite- and model-based evapotranspiration data...

Proper estimates of evapotranspiration rates and long-term totals of evapotranspiration (ET) are crucial for scientific research and practical issues such as sustainable water resources management and resilient land use planning. Traditionally, ground-based ET measurements and observations are used, which are precise and accurate but lack broad spatial coverage. Satellite-based ET products provide spatially comprehensive estimates, although the accuracy of products in certain regions and of certain land use types remains unclear. This study therefore presents a systematic multi-product validation of three publicly available satellite- and model-based ET products (CERv2, MODIS Terra Net Evapotranspiration (MOD16) and Landsat Provisional Actual Evapotranspiration Science Product), including two recently developed products that have not yet been comprehensively evaluated using long-term in situ measurements from lysimeters and eddy covariance sites in an anthropogenically shaped region such as the German lowland region. The lowest relative deviations from measured ET were found at grassland sites (Median relative deviation: 5-54%; Root Mean Square Difference (RMSD): 0.58-1.02 mm d^-1). Across all three datasets, MOD16 showed the lowest deviations at nearly all sites (Median relative deviation: 10-54%; RMSD: 0.58-0.90 mm d^-1). In addition to validating satellite- and model-based ET using in situ measurements, we evaluated whether the ET products can reliably represent the spatial and temporal dynamics of ET to assess their suitability for regional water management and hydrological applications. For this purpose, principal component analyses (PCA) of the time series of each satellite- and model-based dataset were performed. The results show comparable spatial and temporal patterns that can be attributed to land use, water availability and long-term land use changes. Yet, differences between products and land use types became evident in the absolute ET values, even though the PCA revealed consistent dominant patterns across datasets. This suggests that the characteristic spatial and temporal patterns are consistently and reliably represented by all three datasets, regardless of underlying modelling approaches and resolution. The findings highlight the practicability of satellite- and model-based ET estimates for analysing regional and mesoscale ET patterns while also revealing their limitations in estimating absolute values due to model assumptions and spatial aggregation effects.