Characterizing CMIP5 model spread in simulated rainfall in the Pacific Intertropical Convergence and South Pacific Convergence Zones

Abstract

Current-generation climate models exhibit various errors or biases in both the spatial distribution and intensity of precipitation relative to observations. In this study, empirical orthogonal function analysis is applied to the space-model index domain of precipitation over the Pacific from Coupled Model Intercomparison Project Phase 5 (CMIP5) simulations to explore systematic spread of simulated precipitation characteristics across the ensemble. Two significant modes of spread, generically termed principal uncertainty patterns (PUPs), are identified in the December-January-February precipitation climatology: the leading PUP is associated with the meridional width of deep convection, while the second is associated with tradeoffs in precipitation intensity along the South Pacific Convergence Zone, the Intertropical Convergence Zone (ITCZ), and the spurious Southern Hemisphere ITCZ. An important factor distinguishing PUPs from the analogy to time series analysis is that the modes can reflect either true systematic intermodel variance patterns or internal variability. In order to establish that the PUPS reflect the former, three complementary tests are performed by using preindustrial control simulations: a bootstrap significance test for reproducibility of the intermodel spatial patterns, a check for robustness over very long climatological averages, and a test on the loadings of these patterns relative to interdecadal sampling. Composite analysis based on these PUPs demonstrates physically plausible relationships to CMIP5 ensemble spread in simulated sea surface temperatures (SSTs), circulation, and moisture. Further analysis of atmosphere-only, prescribed SST simulations demonstrates decreased spread in the spatial distribution of precipitation, while substantial spread in intensity remains.