Predicting long term regional drought pattern in Northeast India using advanced statistical technique and wavelet-machine learning approach

Abstract

Understanding drought and its multifaceted challenges is crucial for safeguarding food security, promoting environmental sustainability, and fostering socio-economic well-being across the globe. As a consequence of climate change and anthropogenic factors, the occurrence and severity of drought has risen globally. In India, droughts are regular phenomenon affecting about 16% area of country each year which leads to a loss of about 0.5–1% of country’s annual GDP. Hence, the study aims to analyse and predict the meteorological drought in northeast India during 1901 to 2015 using standardised precipitation index (SPI) and analytical techniques such as Mann–Kendall test (MK), innovative trend analysis (ITA), and wavelet approach. In addition, the periodicity of the drought was estimated using Morlet wavelet technique, while discrete wavelet transform (DWT) was applied for decomposing the time series SPI-6 & SPI-12. Study shows that the northeast India experienced moderate drought conditions (SPI-6) in short term and two significant severe droughts (SPI-12) in long term between 1901 and 2015. The trend analysis shows a significant increase in SPI-6 & SPI-12 (p-value 0.01). Further, the combination of parameters i.e. approximation and levels result in the best drought prediction model with higher correlation coefficient and lower error. By using PSO-REPTtree, this study pioneers the use of decomposed parameters to detect trends and develop a drought prediction model. The study is the first step towards establishing drought early warning system that will help decision-makers and farmers to mitigate the impact of drought at the regional level.