Sources of uncertainty in remote sensing based water productivity from evapotranspiration and net primary production inputs

Remote sensing based water productivity indicators are increasingly used in agricultural and ecosystem monitoring, yet their accuracy is constrained by uncertainties in evapotranspiration (ET) and net primary production (NPP) data. Here we evaluate four global ET and NPP products against eddy covariance (EC) flux tower data from AmeriFlux, ICOS, and OzFlux for 2018-2022. Tower latent heat flux is used to derive ET, while tower gross primary production (GPP) is converted to NPP. All satellite products are harmonized temporally and evaluated at dekadal scale using correlation, bias, and RMSE, with stratification by land cover classes. In cropland and forest land covers, WaPOR shows the highest overall performance against EC data for both ET and NPP, with strong correlations and low systematic bias. NPP products show much stronger site-to-site variability than ET, and no product is consistently superior at all locations. Overall, the comparison suggests a clear imbalance: ET estimates are relatively consistent across products, while NPP remains the main source of disagreement between datasets at many sites. This matters directly for any water productivity calculation based on ET and NPP, because water productivity can shift simply with the choice of NPP dataset. The next step of this work will use these results to quantify how much ET versus NPP drives uncertainty in remotely sensed water productivity over cropland and forest land covers.