Comparison of the carbon cycle models in DICE/RICE

Abstract

Modeling of the carbon cycle is one of the most important issues in research of the Integrated Assessment Model (IAM). The carbon-cycle module can not only implement the carbon balance among different carbon reservoirs, but can also provide an interface for climate adaptation and mitigation through management of carbon sinks and land use change. As one of the most popular IAMs in the world, the dynamic integrated model of climate and the economy / regional integrated model of climate and the economy (DICE/RICE) model has two versions of carbon-cycle models. These are the soloreservoir (N1-N) and three-reservoir (N3-N) models. However, there is an obvious drawback of the two models. This is that terrestrial carbon storage is not considered. Therefore, this work examines the effectiveness of the carbon-cycle models within DICE/RICE, and compares the two models with another carbon-cycle model presented by Svirezhev (S-N model). By inputting global historical emission data into the three models (N1-N, N3-N and S-N), we obtain simulations of historical temperature and CO2 concentration during 2001 to 2008. The results are calibrated with observed historical CO2 concentrations and temperature changes, by developing a correlation test. The results show that correlations of CO2 concentration based on the N1-N, N3-N and S-N models are 0.9967, 0.9971 and 0.9970, respectively, and corresponding correlations of temperature are 0.452, 0.447 and 0.451. It was found that there was a significant correlation between simulated and observed CO2 concentration data, but simulated and observed temperature data were uncorrelated. This result is verified by an analysis of variance for the simulated and observed data. Although the correlations between the N1-N, N3-N and S-N models are very similar, the standard errors of CO2 concentration data are 2.53, 2.76 and 0.89, respectively. This shows that the simulation based on the S-N model is much more accurate in relation to the observed data. The N1-N, N3-N and S-N models were used to project climate change by the year 2100, for which the temperatures are 2.98, 3.54°C and 2.91°C, respectively, and the CO2 concentrations are 608.04,733.04, 594.70 ppm, respectively. Projections with smallest values were produced by the terrestrial ecosystem and ocean carbon reservoir represented in the S-N model. This indicates that when carbon is absorbed by the terrestrial ecosystem and ocean in that model, atmospheric carbon is less than those in the other models. Although there are also three reservoirs in the DICE/RICE N3-N model, its results deviate substantially from actual observations. The climate response model used by DICE/RICE was also found to be sensitive to the initial value of land surface temperature, when applied in a short-term projection. However, the sensitivity becomes weaker when applied in a long-term projection. Therefore, the S-N model turns out to be superior to the other models in terms of a much more detailed model mechanism and more accurate modeling performance. In spite of the simplification of the N1-N model, its simulation results are still better than those of the N3-N, in which carbon in the biosphere and ocean is significantly different from observations.