Agricultural ecosystem is composed of various components and the system is affected by both climate condition and human activity. Therefore, decision-making support tool for agriculture should consider both the climate change impact and adaptation. However, it is difficult to integrate individual models using fully mechanistic approach because of limited data availability and identification of complex interactions. Meta-modeling, which is a simplification methodology for complex original models, could be an appropriate method for integrating individual models. The DSSAT (Decision Support System for Agrotechnology Transfer), the DNDC (DeNitrification-DeComposition), and the PPI (Pest Pressure Index) were used in this study as an original model. Each sub-model represents key agricultural components which are crop productivity, greenhouse gas emission, and pest pressure, respectively. In order to develop an integrated meta-model, we used the Random Forest machine learning algorithm. As a case study, we applied this integrated meta-model to the Korean rice production system.
Results/Conclusions
The “percent variance explained” of the meta-model was 83% for rice yield, 96%, 71%, 75% for CO2, CH4, and N2O emission, respectively. In case of PPI, it ranged from 63% to 91% depending on the species. Soil organic carbon was the most important variable of rice productivity, CO2, CH4 emission. The minimum temperature of the growing stage of rice was the most important variable for the PPI except one species. Irrigation option was the most important variable for N2O. The meta-model predicted that rice productivity was reduced under climate change scenarios when maintaining current management. Although there was regional variance, productivity optimization increased not only rice productivity but also greenhouse gas emissions and pest pressure. The results showed that maximizing rice productivity was a partially winning strategy. In this study, the meta-model was able to assess the impacts of climate change and consequences of specific adaptation strategies at the sectoral level and well reflected the key characteristics of the original model. Our results suggest that meta-modeling approach can be applied to the development of complex sectoral decision-making tool.