Enhancing rice yield prediction: a deep fusion model integrating ResNet50-LSTM with multi source data

Area of Science:

• Agricultural Science
• Computer Science
• Environmental Science

Background:

• Global food security relies heavily on rice production, a key crop in Pakistan facing climate change and pandemic impacts.
• Accurate rice yield prediction is crucial for economic stability and informed decision-making in Pakistan.
• Existing predictive models need enhancement to address the complexities of climate change and pandemics on crop yields.

Purpose of the Study:

• To develop and evaluate an innovative deep learning-based hybrid predictive model for forecasting rice yields in Pakistan.
• To utilize multi-modal data, including satellite imagery (EVI, LAI, FPAR) and meteorological/soil data, for enhanced prediction accuracy.
• To assess the performance of different ResNet50-LSTM configurations and feature combinations for optimal rice yield forecasting.

Main Methods:

• A hybrid deep learning model (ResNet50-LSTM) was developed, integrating ResNet50 for feature extraction from satellite data and LSTM for time-series prediction.
• Multi-modal data, including MODIS satellite indices (EVI, LAI, FPAR), meteorological, and soil data, were collected and preprocessed using Google Earth Engine.
• Three LSTM model configurations with varying layer architectures were tested, and feature importance was analyzed to identify optimal predictors.

Main Results:

• The ResNet50-LSTM model with two interconnected LSTM layers demonstrated superior prediction performance.
• The model achieved high accuracy with a selected feature set (EVI, FPAR, climate, and soil variables), yielding R² = 0.9903 and RMSE = 0.1854.
• The combination of EVI and FPAR was found to be particularly effective for rice yield prediction.

Conclusions:

• The developed ResNet50-LSTM framework offers a robust and accurate method for forecasting rice yields.
• The study highlights the potential of utilizing publicly available multi-source data for global crop yield estimation.
• This approach can support informed agricultural decisions, enhance productivity, and contribute to food security.