Sahu et al. (2025) Deep Learning and Remote Sensing for Crop Yield Prediction and Decision Support

Identification

• Journal: Water Resources Management
• Year: 2025
• Date: 2025-12-26
• Authors: Shriya Sahu, Priyank Jain
• DOI: 10.1007/s11269-025-04355-8

Research Groups

• Department of Computer Science & Application, Atal Bihari Vajpayee Vishwavidyalaya, Bilaspur, Chhattisgarh, India
• Department of Computer Science Engineering, Indian Institute of Information Technology, Pune, Maharashtra, India

Short Summary

This study proposes a deep learning framework integrating multispectral satellite imagery and environmental variables for improved crop yield prediction and decision support. The framework, utilizing cGANs for data augmentation and DARTS for architecture optimization, significantly reduces prediction errors and demonstrates the potential to increase yields by 27% and reduce resource costs by 22% compared to conventional practices.

Objective

• To develop an advanced deep learning framework that integrates multispectral remote sensing data with environmental variables to enhance crop yield prediction and support data-driven agricultural decision-making.

Study Configuration

• Spatial Scale: Agricultural field plots across Punjab, India.
• Temporal Scale: 2018–2023 growing seasons.

Methodology and Data

• Models used:
  • Proposed: Deep learning framework combining Conditional Generative Adversarial Networks (cGANs) for synthetic data generation and Differentiable Architecture Search (DARTS) for neural network optimization.
  • Feature Selection: Recursive Feature Elimination (RFE) with RandomForestRegressor.
  • Baselines: Linear Regression (LR), Random Forest (RF), standard Convolutional Neural Networks (CNNs).
• Data sources:
  • Satellite imagery: Multispectral data from Sentinel-2 and Landsat-8 (Red, Green, Blue, Near-Infrared, Shortwave Infrared bands).
  • Ground truth: Crop yield data from agricultural surveys, government databases, and farm records.
  • Environmental variables: Temperature, precipitation, soil pH, soil moisture.
  • Derived features: Normalized Difference Vegetation Index (NDVI).

Main Results

• The proposed cGAN-DARTS model achieved superior performance compared to baselines, with a Mean Absolute Error (MAE) of 0.78 tonnes per hectare, a Root Mean Squared Error (RMSE) of 0.88 tonnes per hectare, and an R² of 0.79.
• It reduced MAE by 23%, RMSE by 19%, and increased R² by 17% relative to conventional baselines (Linear Regression, Random Forest, standard CNNs).
• Simulation-based assessments indicate that AI-informed strategies could increase crop yields by approximately 27% (± 3.4%) and reduce resource costs by 22% (± 2.7%) compared to conventional practices.
• The model demonstrated a Mean Absolute Percentage Error (MAPE) of 14.5% and a Nash–Sutcliffe Efficiency (NSE) of 0.75.
• Temperature (35%) and soil moisture (30%) were identified as the most influential factors for crop yield prediction.

Contributions

• Integration of multispectral remote sensing and environmental variables through advanced deep learning techniques for enhanced crop yield prediction.
• Application of Conditional GANs (cGANs) for synthetic data generation, improving model robustness against limited and variable datasets.
• Optimization of neural network architecture using Differentiable Architecture Search (DARTS) for automated design of efficient predictive models.
• Implementation of Recursive Feature Elimination (RFE) for identifying the most informative features, ensuring high-quality inputs.
• Deployment within a decision support system (DSS) that translates predictive analytics into actionable recommendations for irrigation, fertilization, and resource allocation.

Funding

• The authors received no funding from an external source.

Citation

@article{Sahu2025Deep,
  author = {Sahu, Shriya and Jain, Priyank},
  title = {Deep Learning and Remote Sensing for Crop Yield Prediction and Decision Support},
  journal = {Water Resources Management},
  year = {2025},
  doi = {10.1007/s11269-025-04355-8},
  url = {https://doi.org/10.1007/s11269-025-04355-8}
}