A transparency fusion-based methodology for meteorological drought prediction

Accurate prediction of drought is essential for assessing agricultural production, water resources management, and early risk warning. Various machine learning models have been developed to enhance the accuracy of drought prediction. However, most drought models do not account for data uncertainty. Novel approaches such as the stacking model consider the predictor uncertainty and include multi-source satellite-based products. Here, we develop and test a fusion-based ensemble stacking model that integrates extreme gradient boosting (XGBoost), random forecast (RF), and light gradient boosting machine (LightGBM) for drought modeling and prediction. Multi-source data, including meteorological, vegetation, anthropogenic, landcover, climate teleconnection patterns, and topological characteristics are incorporated in the proposed stacking model. The modeling framework forecast the one-month lead standardized precipitation evapotranspiration index (SPEI) on 12 month scale. In particular, data uncertainty is taken into account allowing for a rigorous model performance evaluation. The proposed method is applied and tested in the German federal states of Brandenburg, and Berlin. The results show that the ST model outperforms XGBboost, RF, and LightGBM, achieving an average coefficient of determination (R²) value of 0.845 in each month of the year 2018. The spatial-temporal Moran’s I method indicates that the ST model captures non-stationarity in modeling the relationship between predictors and the meteorological drought index and outperforms the other three models. Counterfactual sensitivity analysis indicated that extreme precipitation, soil moisture, runoff, and precedent SPEI explain more than 80 % of the total variance of the prediction. Based on the accuracy and flexibility of the method, it seems to be a promising approach for predicting other environmental phenomena.