Analyzing urban footprints over four coastal cities of India and the association with rainfall and temperature using deep learning models

Abstract

Capabilities of deep-learning models CNN and ConvLSTM are explored for analyzing and projecting future urban growth in four Indian coastal cities, viz., Mumbai, Chennai, Kochi, and Vishakhapatnam. ConvLSTM performed better with higher overall accuracy, Cohen's kappa, macro F1-score (> 95 %), and R² and NSE values (> 0.87). The urbanization trends indicate a higher growth in Kochi (∼90.5 %), and greater projected rate for Mumbai (∼42.5 %). The monthly accumulated rainfall and mean temperature are analyzed and forecasted through the ConvLSTM model. The Mann-Kendall test-based analysis (with p-value <0.05), suggest no significant rainfall pattern for the cities except Mumbai, which exhibits an increasing trend. The projected rainfall trend is unaltered except for Vishakhapatnam, which is expected to increase in the coming years. The mean temperature over all the cities, shows an increasing trend (slope ∼ 0.02). However, in the coming years, the ‘increasing trend’ is expected to change into a ‘no significant trend’ for Mumbai and Vishakhapatnam. The forecasted results indicate a continuous urban expansion, which is expected to have a significant impact on rainfall and temperature trends. The outcome could assist urban planners in defining the timeline for improving drainage and include green spaces in the city development plans.