Improving prediction of N₂O emissions during composting using model-agnostic meta-learning

Area of Science:

• Environmental Science
• Agricultural Science
• Machine Learning

Background:

• Nitrous oxide (N₂O) is a potent greenhouse gas contributing to ozone depletion and global warming.
• Composting organic waste is beneficial but can release significant N₂O emissions.
• Accurate quantification of N₂O emissions from composting is crucial for environmental mitigation.

Purpose of the Study:

• To develop and validate an effective machine learning model for predicting N₂O emissions during manure composting.
• To assess the performance of the model-agnostic meta-learning (MAML) approach against other machine learning methods.
• To identify key factors influencing N₂O emissions in the composting process.

Main Methods:

• Employed the model-agnostic meta-learning (MAML) algorithm for N₂O emission prediction.
• Compared MAML performance with five other machine learning models: backpropagation neural network, extreme learning machine, ELM-random forest, gradient boosting decision tree, and extreme gradient boosting.
• Conducted feature analysis to determine the most influential factors affecting N₂O emissions.

Main Results:

• The MAML model achieved a high R² value of 0.939 and a low root mean squared error (RMSE) of 18.42 mg d^-1.
• MAML demonstrated superior performance compared to the other five machine learning methods evaluated.
• Moisture content of structural material and ammonium concentration were identified as the most significant predictors of N₂O emissions.

Conclusions:

• MAML is a highly effective method for predicting N₂O emissions in manure composting.
• Understanding the impact of material properties and process data is key to mitigating N₂O emissions.
• This study provides a robust framework for N₂O emission prediction and reduction strategies in composting.