Validation of CMIP5 multimodel ensembles through the smoothness of climate variables

Abstract

Smoothness is an important characteristic of a spatial process that measures local variability. If climate model outputs are realistic, then not only the values at each grid pixel but also the relative variation over nearby pixels should represent the true climate. We estimate the smoothness of long-term averages for land surface temperature anomalies in the Coupled Model Intercomparison Project Phase 5 (CMIP5), and compare them by climate regions and seasons. We also compare the estimated smoothness of the climate outputs in CMIP5 with those of reanalysis data. The estimation is done through the composite likelihood approach for locally self-similar processes. The composite likelihood that we consider is a product of conditional likelihoods of neighbouring observations. We find that the smoothness of the surface temperature anomalies in CMIP5 depends primarily on the modelling institution and on the climate region. The seasonal difference in the smoothness is generally small, except for some climate regions where the average temperature is extremely high or low.

Keywords

• composite likelihood
• Gaussian process
• NCEP/NCAR reanalysis
• restricted likelihood
• surface temperature anomaly
• uncertainty quantification
• variogram