Forecasting temperature and rainfall using deep learning for the challenging climates of Northern India

Area of Science:

• Meteorology and Climatology
• Artificial Intelligence
• Data Science

Background:

• Accurate temperature and rainfall (T&R) forecasting is critical for climate-sensitive regions like Northern India.
• Traditional forecasting methods struggle with computational demands and localized, real-time predictions.
• A research gap exists in efficient, localized T&R forecasting for regions with volatile weather.

Purpose of the Study:

• To develop and evaluate a deep learning framework for precise T&R forecasting in Jammu, Kashmir, and Ladakh.
• To capture complex temporal dependencies in localized time-series weather data.
• To compare the performance of Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) architectures.

Main Methods:

• Utilized Indian Meteorological Department (IMD) data from 2000-2023 for Jammu, Srinagar, and Ladakh.
• Employed Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) for time-series forecasting.
• Evaluated models in Single Input Single Output (SISO) and Multiple Input Multiple Output (MIMO) configurations.

Main Results:

• Both RNN and LSTM showed robust performance in SISO setups.
• The RNN model consistently outperformed LSTM in capturing temporal relationships.
• RNN in MIMO configuration achieved significantly lower MAE, RMSE, and MSE across all stations.

Conclusions:

• The RNN model demonstrates superior precision for localized T&R forecasting.
• This deep learning framework offers a practical tool for real-time weather prediction.
• Findings contribute to improved climate adaptation, disaster preparedness, and sustainable development in the region.