Extended lead time accurate forecasting of palmer drought severity index using hybrid wavelet-fuzzy and machine learning techniques

Abstract

Abstract Drought is a slowly developing phenomenon and possibly influences a wide domain. Drought index is one of the ways in monitoring and surveying drought, hence Palmer drought severity index (PDSI) has been used as a valid and operational model. In this study the Discrete Wavelet Transform (DWT) tool is incorporated with Fuzzy, k-Nearest Neighbour (kNN) and Support Vector Machine (SVM) modelling tools to improve forecasting accuracy and extend lead time. DWT is further used to decompose original PDSI data into wavelets (sub-series) which, in turn, are used as inputs into the Fuzzy, kNN, and SVM models for the development of a new model in forecasting PDSI for longer lead times from 1 to 12 months. DWT combined with Fuzzy, kNN and SVM models are termed as W-Fuzzy, W-kNN and W-SVM models. The predictive models are implemented in the Marmara region of Turkey. The accuracy of combined hybrid W-Fuzzy, W-kNN and W-SVM models are compared with stand-alone Fuzzy, kNN and SVM models by using Mean Square Error (MSE), Coefficient of Efficiency (CE) and Coefficient of Determination (R2) as performance indicators. The results of this study reveal that developed hybrid W-Fuzzy, W-kNN, and W-SVM models performed very well up to lead time of 6 months. Furthermore, combined W-Fuzzy, W-kNN and W-SVM models are performed better than stand-alone Fuzzy, kNN and SVM models. However, the prediction performance of W-Fuzzy model is slightly better than those of W-kNN and W-SVM models for all lead time predictions in terms of performance indicator criteria, MSE, CE and R2.