Recent above-normal temperatures, which exacerbated the impacts of precipitation deficits, are recognized as the primary driver of droughts in the Upper Colorado River Basin (UCRB), USA. This research aims to enhance drought prediction models by addressing structural changes in non-stationary temperature time series and minimizing drought misclassification through the ES-CBS-SVR model, which integrates ESSVR and CBS-SVR. The research investigates whether this coupling improves prediction accuracy. Furthermore, the model’s performance will be tested in a region distinct from those originally used to evaluate its generalizability and effectiveness in forecasting drought conditions. We used a change point detection technique to divide the non-stationary time series into stationary subsets. To minimize the chances of drought mis-categorization, category-based scoring was used in ES-CBS-SVR. In this study, we tested and compared the ES-CBS-SVR and SVR models in the Upper Colorado River Basin (UCRB) using data from the Global Land Data Assimilation System (GLDAS), where the periods 1950–2004 and 2005–2014 were used for training and testing, respectively. The results indicated that ES-CBS-SVR outperformed SVR consistently across of the drought indices used in this study in a higher portion of the UCRB. This is mainly attributed to variable hyperparameters (regularization constant and tube size) used in ES-CBS-SVR to deal with structural changes in the data. Overall, our analysis demonstrated that the ES-CBS-SVR can predict drought more accurately than traditional SVR in a warming climate.
Read full abstract