Improvement of SVR-Based Drought Forecasting Models using Wavelet Pre-Processing Technique

Abstract

Drought is a damaging natural hazard due to the lack of precipitation from the expected amount for a period of time. Mitigations are required to reduced its impact. Due to the difficulty in determining the onset and offset of droughts, accurate drought forecasting approaches are required for drought risk management. Given the growing use of machine learning in the field, Wavelet-Boosting Support Vector Regression (W-BS-SVR) was proposed for drought forecasting at Langat River Basin, Malaysia. Monthly rainfall, mean temperature and evapotranspiration for years 1976 -2015 were used to compute Standardized Precipitation Evapotranspiration Index (SPEI) in this study, producing SPEI-1, SPEI-3 and SPEI-6. The 1-month lead time SPEIs forecasting capability of W-BS-SVR model was compared with the Support Vector Regression (SVR) and Boosting-Support Vector Regression (BS-SVR) models using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), coefficient of determination (R 2 ) and Adjusted R 2 . The results demonstrated that W-BS-SVR provides higher accuracy for drought prediction in Langat River Basin.

Predicting Standardized Streamflow index for hydrological drought using machine learning models

“…The superior agreement between observed and predicted SPEI indicated the potential of the developed models for contributing more in understanding the potential future of drought-risks in eastern Australia. Fung, Huang, and Koo (2018) studied the improvement of SVR-based drought forecasting models using wavelet preprocessing techniques in the Langat River basin. Prediction of the SPEIs of hybrid wavelet models compared with SVR and the Boosted-support Vector Regression (BS-SVR) models showed that hybrid wavelet BS-SVR model provided more accuracy for the prediction of drought in the Langat River basin.…”