Joint wavelet decomposition of predictors and target variables for drought forecasting

Area of Science:

• Environmental Science
• Climate Science
• Data Science

Background:

• Drought prediction is crucial for water resource management and risk assessment.
• Existing time-series methods may suffer from leakage of future information.
• Understanding hydroclimatic interactions across multiple scales is essential.

Purpose of the Study:

• To develop a leakage-free wavelet-based methodology for drought prediction.
• To improve the interpretability of predictor-target relationships in drought forecasting.
• To merge local and large-scale climate predictors across multiple temporal scales.

Main Methods:

• Application of wavelet transform within moving windows to prevent future information use.
• Merging of multi-scale local and large-scale climate predictors.
• Evaluation using 76 years of drought data from Illinois, USA, with metrics like RMSE, SMAPE, and directional accuracy.

Main Results:

• The proposed model frequently outperforms naïve and univariate wavelet baselines.
• Achieved improvements up to 26.5% in RMSE and 4.95% in SMAPE.
• Demonstrated higher directional accuracy and provided insights into scale-dependent drought dynamics.

Conclusions:

• The leakage-free wavelet methodology offers competitive drought forecasting performance.
• The approach enhances the interpretability of hydroclimatic interactions and drought processes.
• This method provides a robust tool for understanding and predicting drought evolution across different scales.