ML-DPIE: comparative evaluation of machine learning methods for drought parameter index estimation: a case study of Türkiye

Finding solutions for long-term drought parameter index estimation (DPIE) is very crucial since there is a rising trend of drought which has a huge impact on water supplies, various ecosystems, public health, agriculture, and the tourism industry. Therefore, researchers developed a variety of indices to describe the frequency, intensity, duration, and geographic distribution of droughts. In addition, a variety of physical/conceptual models are proposed and used for DPIE. However, the scientific community has recently focused on machine learning (ML) for a variety of problems including DPIE. This is because data-driven ML models learn through experience and are reliable for hydrological and meteorological forecasting. In this study, we therefore performed a comparative evaluation of regression versus deep learning methods for the task of DPIE. We performed experiments on three stations located in Türkiye and considered three different indices. Our results show that traditional regressors often provide better results than deep learners. In addition, the effect of indices on the results is limited especially for the regression algorithms. Deep learning models on the other hand outperform regression algorithms in some cases and they have a disadvantage in finding the optimum structure. To the best of our knowledge, this study is the first of its kind concerning the number of employed algorithms and indices, the extent of experiments, and considered stations. Besides, the findings of this study can be used to deploy an ML model to monitor drought in a highly accurate way for related stations in Türkiye.