Deep Learning for Short-Term Precipitation Prediction in Four Major Indian Cities: A ConvLSTM Approach with Explainable AI

TL;DR

This work tackles the challenge of forecasting short-term precipitation in four major Indian cities with a transparent, data-driven approach. It introduces a hybrid Time-Distributed CNN–ConvLSTM model trained on multi-decadal ERA5 reanalysis data, with city-specific architectures and a loss that emphasizes extreme rainfall events. An integrated explainable AI framework (Grad-CAM, permutation importance, temporal occlusion, counterfactual analysis) reveals how input variables, space, and time drive predictions across Bengaluru, Mumbai, Delhi, and Kolkata, linking model behavior to known meteorological processes. The results show competitive short-term predictive skill and city-tailored interpretability, providing actionable guidance for data collection and urban flood preparedness while highlighting limitations related to resolution and aggregation that motivate future work with higher-resolution data and hydrological coupling.

Abstract

Deep learning models for precipitation forecasting often function as black boxes, limiting their adoption in real-world weather prediction. To enhance transparency while maintaining accuracy, we developed an interpretable deep learning framework for short-term precipitation prediction in four major Indian cities: Bengaluru, Mumbai, Delhi, and Kolkata, spanning diverse climate zones. We implemented a hybrid Time-Distributed CNN-ConvLSTM (Convolutional Neural Network-Long Short-Term Memory) architecture, trained on multi-decadal ERA5 reanalysis data. The architecture was optimized for each city with a different number of convolutional filters: Bengaluru (32), Mumbai and Delhi (64), and Kolkata (128). The models achieved root mean square error (RMSE) values of 0.21 mm/day (Bengaluru), 0.52 mm/day (Mumbai), 0.48 mm/day (Delhi), and 1.80 mm/day (Kolkata). Through interpretability analysis using permutation importance, Gradient-weighted Class Activation Mapping (Grad-CAM), temporal occlusion, and counterfactual perturbation, we identified distinct patterns in the model's behavior. The model relied on city-specific variables, with prediction horizons ranging from one day for Bengaluru to five days for Kolkata. This study demonstrates how explainable AI (xAI) can provide accurate forecasts and transparent insights into precipitation patterns in diverse urban environments.

Figures (7)

• Figure 1: Study area and data points
• Figure 2: Flowchart used in the present study
• Figure 3: Schematic illustration of the model used in the study
• Figure 4: Permutation-based feature importance for the ConvLSTM models in four cities: (a) Bengaluru, (b) Mumbai, (c) Delhi, and (d) Kolkata. The length and sign of each bar represent the change in RMSE when that input feature is randomly shuffled; positive values indicate a deterioration in performance (higher error), while values near zero or slightly negative indicate that the model can largely substitute that predictor with others.
• Figure 5: Counterfactual perturbation analysis for the ConvLSTM models in four cities: (a) Bengaluru, (b) Mumbai, (c) Delhi, and (d) Kolkata. Each panel shows how predicted rainfall changes when individual input features are systematically reduced, indicating model sensitivity to each meteorological variable.