Improving Weather Predictions Accuracy With a Hybrid LSTM-Transformer Model: A Study on Indian Climate

Weather prediction is a challenging task, particularly in dealing with the diverse geography and weather patterns of various Indian cities throughout the year. This research presents an optimized hybrid model (OHM) that combines long-short-term memory (LSTM) and transformer models to increase the prediction performance for weather variables. The primary objective was to provide accurate weather forecasts for various Indian cities, which contributes to better disaster management, better agricultural sector, and daily lives of people. Weather forecasting was conducted for multiple Indian cities using advanced hybrid models and traditional machine learning algorithms. Data were collected, pre-processed, benchmarked against existing algorithms, and the results were evaluated. OHM was trained and tested using open source Kaggle weather data covering several Indian cities. The proposed hybrid model took advantage of the capabilities of LSTM and transformer to learn temporal dependencies, enhance feature extraction, and manage long-range dependencies more effectively. The OHM was compared with standard machine learning models, comprising random forest, XGBoost, support vector regression, linear regression, and an isolated LSTM model. The results showed an increase of $0.2~^{\circ }$ C in the average annual temperature over the next 10 years and a variation in rainfall of 8% during the monsoon season. An increase in cyclone occurrence 15% was also expected. The OHM outperformed traditional learning models by achieving lower root mean square error (RMSE) values when tested in various Indian cities and provided better forecasts for different weather parameters, such as rainfall, wind speed and temperature. This research proposes a novel hybrid deep learning model combining LSTM and transformer architectures to forecast key weather parameters such as temperature, humidity, precipitation, wind speed, and atmospheric pressure, for various Indian cities.