Predictive Neural Network Modeling for Almond Harvest Dust Control.

Area of Science:

• Agricultural Engineering
• Environmental Science
• Machine Learning

Background:

• Almond harvesting in California generates significant dust emissions, particularly fine particulate matter (PM2.5).
• Accurate prediction of these emissions is crucial for environmental compliance and mitigating health impacts.
• Existing methods for emission estimation may not fully capture the dynamic nature of harvesting operations.

Purpose of the Study:

• To develop and validate a predictive model for PM2.5 emissions during almond harvesting.
• To analyze the relationship between harvester operational parameters and PM2.5 generation.
• To provide a tool for proactive environmental management in the almond industry.

Main Methods:

• Utilized a feedforward neural network (FNN) for PM2.5 emission prediction.
• Preprocessed field data, including outlier removal and normalization.
• Trained the FNN with two hidden layers, tanh activation, MSE loss, and Adam optimization.

Main Results:

• The FNN model achieved high predictive accuracy for PM2.5 emissions.
• Demonstrated a Mean Squared Error (MSE) of 0.02 and Mean Absolute Error (MAE) of 0.01.
• Identified key operational parameters influencing dust generation.

Conclusions:

• The developed neural network model offers a precise method for forecasting PM2.5 emissions in almond harvesting.
• This approach integrates machine learning with agricultural practices for sustainable production.
• The study sets a precedent for using predictive analytics to reduce agricultural emissions.