Environmental Impact Analysis on Crop Yield using Feature-Aware Deep Learning and Remote Sensing

Traditional crop yield models have relied heavily on empirical and statistical approaches, such as linear regression and process-based models (e.g., DSSAT, APSIM). Although effective, these methods often struggle to capture the non-linear, near-close and dynamic relationships between environmental variables and crop growth. Additionally, its reliance on manual data collection limits scalability and real-time decision making in precision agriculture. Recent advances in remote sensing (e.g., SMAP, TRMM, MODIS, and Landsat) and machine learning provide new opportunities to overcome these limitations by enabling high-resolution, large-scale monitoring of agricultural variables. This study proposes a hybrid deep learning framework that combines Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks to model spatiotemporal crop performance dynamics. A Feature-Aware Remote Sensing (FARS) module was introduced to preprocess multisource satellite data, extracting key vegetation indices (NDVI, EVI), canopy features, and temporal phenological patterns. The CNN component processes the spatial features of the remote sensing image, whereas the LSTM network gains temporary dependence in the development of crops and environmental conditions. To increase interpretation, Explainable AI (XAI) technology (SHAP, Grad-CAM) was integrated, providing insight into the important variables affecting predictions. The experimental results demonstrate better performance than traditional methods, with an $\mathbf{R}^{\mathbf{2}}$ score of 0.93 and a reduction in the prediction error by 18%. The model effectively identifies major producers, such as soil moisture and NDVI, while the satellite exposes stress-affected areas in the imagery. This research contributes to accurate agriculture by detecting a scalable, data-driven approach, yield forecast, and stress for real-time crop monitoring, which facilitates informed decisions for permanent farming practices.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.