YieldSense-X: Temporal Deep Learning Framework for Crop Yield Prediction Using Climate and Soil Dynamics

Abstract

Precise yield prediction of crops plays a vital role in food security, proper management of resources and sustainable agriculture. This paper proposes the YieldSense-X, a time-dependent deep learning model that optimally estimates crop yield based on the dynamics of climate and soil. The suggested model combines Long Short-Term Memory (LSTM) and Temporal Attention to obtain success in obtaining sequential dependencies and focus on key moments that contribute to crop development. Using multivariate data sets which cover weather factors like temperature, rainfall and humidity and soil characteristics such as moisture, pH and nutrient composition, the framework offers a broad picture of the effects of the environment on yield. The data is then pre sacrificed, normalized and converted into time series sequences so as to make good use of time learning. Through experiments, it can be proved that the proposed model is superior to the traditional machine learning and standalone deep learning methods with respect to accuracy, RMSE and MAE. Moreover, the model facilitates forecasting of yields at an early stage, thus proactive agricultural decision making. The interpretable, scale-oriented, and climate-conscious design guarantees the robustness, interpretability, and scalability of the vision to the real-world accuracy applications of precision agriculture.