FASAL: Ensemble Crop Recommendation System using Google Earth Engine and XAI Interpretability

Agriculture is transforming at a very fast pace, previously being one of the biggest sources of livelihood for individuals. But with technology coming into play, becoming part of one’s daily life and with its implementation of cutting-edge technologies like machine learning, satellite remote sensing, and explainable artificial intelligence (XAI). In this paper, we outline a complete intelligent agricultural solution to help farmers both choose crops and to calculate the NDVI index. Our method utilises an ensemble-based classification model, Random Forest, Support Vector Classifier (SVC), and Gradient Boosting Classifier to suggest the best crops in line with environmental and soil factors such as NPK values, temperature, humidity, rainfall, and pH values. The model shows superior classification performance, with macro and weighted averages of around 0.98 and an accuracy of 98% overall. To promote transparency and trust, we integrate Local Interpretable Model-Agnostic Explanations (LIME) for interpreting model predictions and additionally use a Large Language Model (LLM) to explain these interpretations in farmer-friendly natural language. We utilise Google Earth Engine (GEE) to, compute NDVI over time. The convergence of explainable AI, natural language interfaces, and remote sensing provides a comprehensive, data-driven decision support system for new-age agriculture in our endeavour to assist the backbone of our country - the hardworking farmers.