Monthly Rainfall Prediction of Kerala Using Lag-Llama And Lstm

Abstract

Accurate monthly rainfall prediction is important for climate resilience and water resource management in Kerala, India. This study presents a novel rainfall forecasting methodology using the Lagged Language Meta AI (Lag-LLama) model, an open-source foundation model designed specifically for time series forecasting. The proposed approach applies Lag-LLama for monthly rainfall prediction and evaluates its performance against established machine learning and deep learning models, namely Random Forest (RF) and Long Short-Term Memory (LSTM) networks.

Results show that while RF and LSTM outperform traditional statistical methods, the fine-tuned Lag-LLama model achieves a substantial improvement of approximately 41–42% in prediction accuracy over its zero-shot configuration and consistently surpasses both RF and LSTM models. These findings demonstrate the transformative potential of foundation models for complex climate time series forecasting and highlight the importance of domain-specific fine-tuning. The proposed methodology establishes a new benchmark for rainfall prediction in Kerala and offers a transferable framework for similar climatic regions under increasing climate variability.