Quantification of precipitation and temperature uncertainties simulated by CMIP3 and CMIP5 models

Abstract Assessment of climate change impacts on water resources is extremely challenging, due to the inherent uncertainties in climate projections using global climate models (GCMs). Three main sources of uncertainties can be identified in GCMs, i.e., model structure, emission scenario, and natural variability. The recently released fifth phase of the Coupled Model Intercomparison Project (CMIP5) includes a number of advances relative to its predecessor (CMIP3), in terms of the spatial resolution of models, list of variables, and concept of specifying future radiative forcing, among others. The question, however, is do these modifications indeed reduce the uncertainty in the projected climate at global and/or regional scales? We address this question by quantifying and comparing uncertainty in precipitation and temperature from 6 CMIP3 and 13 CMIP5 models. Uncertainty is quantified using the square root of error variance, which specifies uncertainty as a function of time and space, and decomposes the total uncertainty into its three constituents. The results indicate a visible reduction in the uncertainty of CMIP5 precipitation relative to CMIP3 but no significant change for temperature. For precipitation, the GCM uncertainty is found to be larger in regions of the world that receive heavy rainfall, as well as mountainous and coastal areas. For temperature, however, uncertainty is larger in extratropical cold regions and lower elevation areas.