Drought characteristics prediction using a hybrid machine learning model with correction

Drought prediction is a crucial aspect of drought risk assessment, playing a significant role in effectively minimizing losses associated with droughts. While many previous studies have focused on the variations of drought indices, understanding the impact of drought events requires a transformation into drought characteristics such as severity and duration. These drought features provide insights into the processes and hazards associated with drought events. The main objectives of this study are twofold: improving prediction accuracy and evaluating model performance based on drought characteristics. By decomposing the data into high and low-frequency sequences and effectively addressing these with optimally chosen models, a hybrid model (LSL) model is constructed. This model incorporates an error correction sequence to reduce errors during the prediction process and improve accuracy. Simultaneously, various aspects of drought characteristics, such as duration, start and end times, intensity, and peak value, were analyzed to enhance drought event prediction and impact assessment. The results show that, the LSL model is effective in predicting the drought class and all aspects of drought characteristics. Overall, the LSL model can be well applied to the prediction of drought events.