DEEP LEARNING-BASED SEASONAL NDVI FORECASTING FOR ENHANCED AGRICULTURAL SUSTAINABILITY: A CASE STUDY IN THE KUTTANAD REGION, KERALA, INDIA

India's agricultural sector, despite transitioning across economic domains, remains heavily reliant on its agrarian foundation, which faces substantial challenges from natural and human-induced disruptions. Leveraging technological advancements such as Artificial Intelligence (AI), Machine Learning (ML) and Deep Learning holds immense promise in bolstering agricultural productivity by providing actionable insights. This study focuses on harnessing Deep Learning techniques to forecast seasonal Normalized Difference Vegetation Index (NDVI) in paddy fields using satellite imagery and an in-depth analysis of paddy field health based on the predicted NDVI values with a specific emphasis on the Kuttanad region in Kerala, India. Integration of Deep Learning models, particularly Long Short-Term Memory (LSTM) networks, the research aims to deliver precise and timely predictions of NDVI, facilitating a holistic evaluation of crop health dynamics. Time series satellite imagery with optimal temporal resolution and suitable spectral bands for NDVI computation is utilized for analysis. The results demonstrate the effectiveness of LSTM models in accurately predicting seasonal NDVI. The predicted NDVI values are correlated with ground truth data, allowing for a comprehensive assessment of paddy field health. The developed model demonstrated superior performance, achieving notable accuracy of an R2 value of 0.978. Comparing anticipated NDVI values to benchmarked periods of optimal vegetation health, one can assess the current state of crops and predict future trends in vegetation conditions.