Drought

Random Forest (RF) and recurrent neural network models such as LSTM predict field-scale ET more accurately than process-based models. Vapor pressure and crop coefficients are key predictors for irrigated crops. ET forecasting for non-irrigated crop requires enhanced vegetation index. Short-term (3-day) forecasts have lower uncertainty, higher accuracy using RF.

A new framework based on a machine learning data-driven network to estimate and forecast evapotranspiration in agricultural areas was proposed.

The SWAT-MODFLOW coupled model approach was applied at large spatio-temporal scale to study the cumulative effects of changing precipitation patterns, groundwater withdrawals, and forest evapotranspiration to improve projections of the future of lake levels and water availability in agricultural regions.