MS-YieldStackNet: multi-source data fusion for wheat yield estimation using a stacked ensemble neural network.

Area of Science:

• Agricultural Science
• Remote Sensing
• Machine Learning

Background:

• Accurate crop yield prediction is crucial for food security and agricultural policy.
• Manual estimation methods for wheat yield are labor-intensive and imprecise, especially in Pakistan.
• Integrating diverse data sources can improve yield forecasting accuracy.

Purpose of the Study:

• To develop and validate a novel algorithmic framework, MS-YieldStackNet, for high-resolution wheat yield prediction.
• To integrate multispectral satellite imagery, in-situ soil analytics, and meteorological variables for enhanced forecasting.
• To assess the model's performance using key statistical metrics.

Main Methods:

• Constructed a unified feature space using vegetation indices (NDVI, DVI), soil physicochemical attributes, and temporal climate data.
• Employed a stacked ensemble neural architecture (MS-YieldStackNet) combining three parallel feed-forward neural networks (FFNNs).
• Utilized a Random Forest meta-learner to integrate predictions from the FFNNs.

Main Results:

• Achieved a robust R-squared value of 0.81, indicating strong model performance.
• Reported Mean Squared Error (MSE) of 6,114.30 kg/ha and Root Mean Squared Error (RMSE) of 78.19 kg/ha.
• Demonstrated low prediction errors with Mean Absolute Error (MAE) of 59.07 kg/ha and Mean Absolute Percentage Error (MAPE) of 3.55%.

Conclusions:

• MS-YieldStackNet provides a precise and scalable solution for wheat yield forecasting.
• The integrated approach significantly improves prediction accuracy compared to traditional methods.
• The framework has strong potential for informing agricultural policy and ensuring food security.