Improving subseasonal soil moisture and Evaporative Stress Index forecasts through machine learning: The role of initial land state versus dynamical model output

Authors: LORENZ, D - University Of Wisconsin; OTKIN, J - University Of Wisconsin; ZAITCHIK, B - Johns Hopkins University; HAIN, C - Nasa Marshall Space Flight Center; HOLMES, T - Goddard Space Flight Center; Anderson, Martha

Submitted to: Journal of Hydrometeorology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/28/2024
Publication Date: 8/1/2024
Citation: Lorenz, D., Otkin, J., Zaitchik, B.F., Hain, C., Holmes, T., Anderson, M.C. 2024. Improving subseasonal soil moisture and Evaporative Stress Index forecasts through machine learning: The role of initial land state versus dynamical model output. Journal of Hydrometeorology. 25(8):1147-1163. https://doi.org/10.1175/JHM-D-23-0074.1.
DOI: https://doi.org/10.1175/JHM-D-23-0074.1

Interpretive Summary: Flash drought events are characterized by a rapid decline in soil and crop water status due to a combination of lack of rainfall and hot, dry atmospheric conditions. Flash droughts are both difficult to predict and to respond to, given their rapid onset. Any improvement in flash drought prediction will have positive benefits to agriculture, giving advanced lead time to adjust water, herd, crop, and management decisionmaking. This paper explores a predictive approach combining short-term weather forecasting data with machine learning methods for incorporating knowledge about the current state of the land-surface, as derived from satellite remote sensing. The study looks at predictability of both soil moisture and evapotranspiration fraction, the latter serving as a proxy for crop health. We also explored new methods for defining samples used in the machine learning training process, which led to improvements in forecasts of both indicators. Using the new forecasting methodology, most improvement was actually realized in capturing soil moisture recovery post-drought. Findings from this study will help to advance subseasonal predictability of flash drought.

Technical Abstract: Accurate subseasonal-to-seasonal (S2S) forecast of flash drought has the potential to enhance drought preparation and mitigation. For this reason, numerous studies have applied dynamically based forecast systems and statistical prediction models to flash drought prediction. In this study, the two approaches are combined in the form of statistical models that draw predictors from both observed land surface conditions and S2S Prediction Project dynamically-based forecasts. Both standard regression models and nonlinear machine learning methods are considered. When the models are enhanced with machine learning and other improvements, the increases in skill are almost exclusively coming from predictors drawn from observations of current and past land surface states. This suggests that operational S2S flash drought forecasts should focus on optimizing use of information on current conditions rather than on integrating dynamically based forecasts, given the current state of knowledge. Nonlinear machine learning methods lead to improved skill over linear methods for soil moisture but not for evapotranspiration fraction. Improvements for both soil moisture and evapotranspiration fraction are realized by increasing the sample size by including surrounding grid points in training and increasing the number of predictors. In addition, all the improvements in the soil moisture forecasts predominantly impact soil moistening rather than soil drying—i.e., prediction of conditions moving away from drought rather than into drought—especially when the initial soil state is drier than normal.