Quantifying tree species effects on soil organic carbon using machine learning algorithm: A case study in tropical agroforestry system of gayo coffee, Indonesia

Gayo coffee based on tropical agroforestry systems in Aceh plays an ecological role in mitigating climate change and a socio-economic role as a social livelihood in Indonesia. Ecologically, coffee agroforestry systems can increase soil carbon stocks through complex vegetation that produces litter as a source of nutrients. However, studies on measuring the contribution of vegetation to soil carbon dynamics in agroforestry lands using advanced machine learning-based statistical models in Indonesia are still very rare. Therefore, this study involved 18 complex vegetation variables to prove their contribution to soil organic carbon (SOC) dynamics using machine learning-based predictions with the random forest (RF) algorithm and hyperparameter tuning settings. The SOC available in the study area reached 173.46 ± 60.34 Mg ha -1 , with 9.20% ± 3.87% organic C, in agroforestry systems characterized by vegetation density of 1752.94 ± 459.20 trees ha -1 (range: 850-2850 trees ha -1 ), and consisted of 11 overstory species. Based on two RF model tests (rf and ranger model), SOC dynamics were influenced by vegetation by 95% (R-squared) with an error rate of 0.05 (RMSE) and 0.04 (MAE). The contribution of vegetation focuses on the variable of agroforestry richness as the most important factor in predicting SOC, even though the species Leucaena leucocephala dominates around 88% of the species composition. These results recommend that increasing agroforestry species diversity is key to increasing SOC in coffee agroforestry. This information is expected to strengthen the implementation of SFA policies and enhance the sustainability of climate change mitigation-based social livelihoods in tropical Indonesia.