Development of gridded monthly reference evapotranspiration dataset for Germany for long term trend analysis

Estimation of reference evapotranspiration (ET_o) is necessary for hydrological modeling, water stress adaptation and agricultural water management. While numerous studies have addressed changes in temperature and precipitation patterns at different spatiotemporal scales for assessing hydroclimatic variability, similar analyses on regional evapotranspiration trends are limited. This can be attributed to lack of observed records of ET_o at regional scales, and highlights the need for developing better models for estimating this variable. To address this research gap, we developed monthly 0.5° gridded ET_o dataset for entire Germany using machine learning techniques and then investigated the temporal trends in ET_o over the region. We utilized flux tower data (sensible heat flux, net radiation, soil heat flux, and latent heat flux) from multiple locations in the region to compute observed ET_o using the surface energy balance method. Then, fed-forward back-propagation method (BPNN) is used for predicting monthly ET_o with easily available input predictors such maximum temperature, minimum temperature, precipitation, soil moisture, short wave radiation, and wind speed. The BPNN is trained with various input combinations in order to estimate ET_o with minimal input predictors, and their performance is assessed using metrics: coefficient of determination, mean absolute error, and root mean square error. The results showed that with all the input parameters, the coefficient of determination  for training and testing are 0.89 and 0.93 respectively, while the best parsimonious model (precipitation and downward shortwave radiation as predictors) gives 0.88 and 0.93 respectively. Gridded ET_o estimated using the best parsimonious model is then used for assessing spatially varying trends in the variable at monthly and annual time scales over Germany using Mann-Kendall test and Sen’s slope. The long term analysis helps us to identify critical regions in the study area that needs attention for water resources management, drought mitigation and improved adaptation to changing climate.

Keywords: Reference Evapotranspiration, Flux Tower, Surface Energy Balance, Feed Forward Back Propagation (BPNN), Trend Analysis, Mann-Kendall Test