Animal-based CO2, CH4, and N2O emissions analysis: Machine learning predictions by agricultural regions and climate dynamics in varied scenarios

Abstract

Livestock is an essential source of livelihood and food. In the context of climate change, animal-based greenhouse gas (GHG) emissions are of great importance. This study predicted direct N₂O Emissions, indirect N₂O Emissions, CH₄ Emissions from manure management, CH₄ Emissions from enteric fermentation, and CO₂ Emissions as GHG emissions from animal sources for all provinces of Turkey using various machine learning algorithms. Animal populations, climate parameters, and agricultural area information are used to model GHG emissions. The proposed study includes two different analyses according to the number of features used. The CatBoost algorithm was primarily successful when using eight features according to Scenario-1 and twelve features according to Scenario-2. In Scenario-1, R² values for GHG emission predictions for 2021 are obtained as 0.996, 0.996, 0.992, 0.999, and 0.996, respectively, while in Scenario-2, they are obtained as 0.995, 0.996, 0.984, 0.996, and 0.996. In Scenario-1, R² values for GHG emission predictions for 2004–2009 are obtained as 0.976, 0.962, 0.982, 0.994, and 0.994, respectively, while in Scenario-2, they are obtained as 0.975, 0.957, 0.917, 0.993, and 0.993. According to the results, the model’s performance was not improved by increasing the number of features used. Using fewer features gave more successful results.