Crop Yield Prediction using Hybrid Deep Learning and Geo-Spatial Data Integration

Agriculture is critical to food security, means of survival, and the economy of India. Hence there are challenges due to climatic variability, and at times worsening climatic systems which can lead to reduced stability in weather and available resources, we feel crop yield prediction is essential. Crop yield prediction does not only help farmers make improved decisions or more informed choices, it reasonable allows agri-tech stakeholders to optimise resource use and efforts for better productivity. The varying weather and insecure nature of weather have impacted agricultural productivity with respect to India, necessitating crop yield forecast with utmost accuracy. The proposed research put forth a new kind of hybrid deep learning forecast model that integrates Temporal Fusion Transformer (TFT) and the 1D Convolution Neural Network (CNN) models to increase forecasting accuracy through capturing both temporal and spatial dependencies in agricultural data. Differing from classic machine learning models considered earlier, the system incorporates geospatial inputs such as satellite-derived NDVI and soil moisture, along with historical weather variables including rainfall, temperature, and humidity. The Temporal Fusion Transformer takes advantage of the latest developments in attention mechanisms so that the model can pick up long-range temporal patterns and variable importance varying in time. District-level micro-climatic clustering has been incorporated to create localized prediction capability, and a crop recommendation module is provided considering both environmental and market factors. Explainable AI (XAI) techniques, especially SHAP, have been employed to interpret the feature contributions and thus render the model transparent and trustworthy. Comparative study against baseline models such as Random Forest and SVM has established that the presented framework is up to 12% more efficient in predictions, hinting at the real agricultural decision-making potential of the system. The study intends to furnish farmers with useful accurate data-based insights.