Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Spatio-Temporal Deep Learning Approach For High-Resolution Gridded Monsoon Prediction

Parashjyoti Borah, Sanghamitra Sarkar, Ranjan Phukan

TL;DR

This work reframes long-range, high-resolution monsoon prediction as a spatio-temporal video-to-image task by treating pre-monsoon multi-variable fields as a multi-channel video to forecast gridded rainfall for June–September and JJAS. It introduces a 3D-CNN framework with a spatiotemporal encoder, temporal collapse, and a regression head, trained on 85 years of ERA5 predictors and IMD rainfall targets, with extensive data preprocessing and augmentation to compensate for limited samples. Quantitative results show robust generalization across architectures, with simple models achieving competitive MSE and MAE, while regional extremes (notably in complex terrains) remain challenging. The study offers a scalable, data-driven baseline for intra-seasonal to seasonal monsoon forecasts and points to future work in probabilistic forecasts, Vision Transformers, data augmentation, and explainable AI for deeper monsoon insights.

Abstract

The Indian Summer Monsoon (ISM) is a critical climate phenomenon, fundamentally impacting the agriculture, economy, and water security of over a billion people. Traditional long-range forecasting, whether statistical or dynamical, has predominantly focused on predicting a single, spatially-averaged seasonal value, lacking the spatial detail essential for regional-level resource management. To address this gap, we introduce a novel deep learning framework that reframes gridded monsoon prediction as a spatio-temporal computer vision task. We treat multi-variable, pre-monsoon atmospheric and oceanic fields as a sequence of multi-channel images, effectively creating a video-like input tensor. Using 85 years of ERA5 reanalysis data for predictors and IMD rainfall data for targets, we employ a Convolutional Neural Network (CNN)-based architecture to learn the complex mapping from the five-month pre-monsoon period (January-May) to a high-resolution gridded rainfall pattern for the subsequent monsoon season. Our framework successfully produces distinct forecasts for each of the four monsoon months (June-September) as well as the total seasonal average, demonstrating its utility for both intra-seasonal and seasonal outlooks.

A Spatio-Temporal Deep Learning Approach For High-Resolution Gridded Monsoon Prediction

TL;DR

This work reframes long-range, high-resolution monsoon prediction as a spatio-temporal video-to-image task by treating pre-monsoon multi-variable fields as a multi-channel video to forecast gridded rainfall for June–September and JJAS. It introduces a 3D-CNN framework with a spatiotemporal encoder, temporal collapse, and a regression head, trained on 85 years of ERA5 predictors and IMD rainfall targets, with extensive data preprocessing and augmentation to compensate for limited samples. Quantitative results show robust generalization across architectures, with simple models achieving competitive MSE and MAE, while regional extremes (notably in complex terrains) remain challenging. The study offers a scalable, data-driven baseline for intra-seasonal to seasonal monsoon forecasts and points to future work in probabilistic forecasts, Vision Transformers, data augmentation, and explainable AI for deeper monsoon insights.

Abstract

The Indian Summer Monsoon (ISM) is a critical climate phenomenon, fundamentally impacting the agriculture, economy, and water security of over a billion people. Traditional long-range forecasting, whether statistical or dynamical, has predominantly focused on predicting a single, spatially-averaged seasonal value, lacking the spatial detail essential for regional-level resource management. To address this gap, we introduce a novel deep learning framework that reframes gridded monsoon prediction as a spatio-temporal computer vision task. We treat multi-variable, pre-monsoon atmospheric and oceanic fields as a sequence of multi-channel images, effectively creating a video-like input tensor. Using 85 years of ERA5 reanalysis data for predictors and IMD rainfall data for targets, we employ a Convolutional Neural Network (CNN)-based architecture to learn the complex mapping from the five-month pre-monsoon period (January-May) to a high-resolution gridded rainfall pattern for the subsequent monsoon season. Our framework successfully produces distinct forecasts for each of the four monsoon months (June-September) as well as the total seasonal average, demonstrating its utility for both intra-seasonal and seasonal outlooks.
Paper Structure (26 sections, 4 equations, 3 figures, 2 tables)

This paper contains 26 sections, 4 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Statistical and Dynamical Forecasting
  4. Statistical Forecasting
  5. Dynamical Forecasting
  6. Machine Learning in Climate Prediction
  7. Index Prediction
  8. Gridded Spatio-Temporal Forecasting
  9. A Deep Learning Framework for Gridded Monsoon Forecasting
  10. Dataset Construction and Preprocessing
  11. Predictor Dataset: Pre-Monsoon Spatio-Temporal Fields
  12. Target Variable: Gridded Monsoon Rainfall
  13. Evaluation Metrics
  14. Mean Squared Error (MSE)
  15. Mean Absolute Error (MAE)
  16. ...and 11 more sections

Figures (3)

  • Figure 1: Schematic diagram of the proposed 3D CNN-based model architecture.
  • Figure 2: Example of a strong prediction for the JJAS seasonal average (Year 1988, Bottleneck size- 512, 4 Blocks). The model (right) closely predicts the spatial patterns of the ground truth (left).
  • Figure 3: Example of a weaker prediction for the JJAS seasonal average (Year 2004, Bottleneck size- 512, 4 Blocks). The model (right) fails to capture the high-intensity rainfall in the Northeast India seen in the ground truth (left).