Identification of homogeneous rainfall zones through machine learning-based clustering algorithm for estimation of rainfall change over the Indo-Gangetic plains of India

The Indo-Gangetic Plain (IGP), India’s “breadbasket”; is essential for food security in India and South Asia. Rainfall patterns significantly influence the agriculture and water resources of this region. The current study aimed to identify homogeneous rainfall zones over IGP across different seasons through unsupervised machine learning methods like k-means clustering during 1961 – 2020 utilizing high-resolution gridded from (0.25° × 0.25°) rainfall data India Meteorological Department (IMD). Three homogeneous rainfall zones were identified during pre-monsoon and monsoon, and two for post-monsoon and winter. Seasonal rainfall trends for each homogeneous zone were analyzed using parametric (linear regression) and non-parametric (Mann-Kendall) methods. Both approaches revealed a significant increasing trend (2.07 – 2.13 mm/year) in pre-monsoon rainfall over zone 2 (Sub-Himalayan West Bengal, SHWB), a significant decreasing trend in monsoon rainfall (2.62 – 2.85 mm/year) over zone 1 (Trans-Gangetic and western Upper-Gangetic Plains) and post-monsoon rainfall (0.26 – 0.37 mm/year) over zone 1 (Trans-Gangetic, Upper-Gangetic, and western middle Gangetic Plains). No significant rainfall trend was observed during the winter season over the IGP. Changepoint detection was applied to identify shifts in the long-term inter-annual mean of seasonal rainfall. There was no significant shift in pre-monsoon rainfall time series over the IGP except for an increase in the SHWB after the changepoint year 1973. Monsoon rainfall over two other zones except SHWB decreased after the respective changepoint years (1996 and 2008). The identified trends over homogeneous rainfall zones may provide valuable inputs for multi-level stakeholders to implement different water management and adaptation strategies for agricultural development planning under changing scenarios.