Multi-Model Assessment of Regional Surface Temperature Trends: CMIP3 vs CMIP5 Historical (20C3M) Runs

Regional surface temperature trends from the CMIP3 and CMIP5 20th century runs are compared with observations, and assessed against a backdrop of internal climate variability as estimated from model control runs. The simulated internal climate variability is used to assess whether observed trends are “detectable” and whether the models' historical run trends are consistent with observed trends. The trend tests focus on various periods (e.g., 1901-2010, 1951-2010, 1981-2010) and are applied at scales from global averages to individual grid points. For trends-to-2010 beginning in start years from 1901 to 1981, warming in the CMIP3 and CMIP5 simulations with volcanic forcing is consistent with observations over roughly 40-55% of the global area analyzed,. The consistent area in the CMIP5 ensemble is about 5% larger than in the CMIP3 ensemble, for trends-to-2010 that begin before 1960. The fraction of analyzed global area with no detectable trend in the observations is less than 10% for trends covering 1901-2010, but this fraction gradually grows to over 50%, and is generally slightly higher for CMIP5 than CMIP3, as the trend start date advances toward 1991. Especially for the trends beginning earlier in the record (e.g., 1901-2010) the ensemble historical run warming trend tends to be too large at lower latitudes and too small at higher latitudes. The analysis identifies regions where detection of warming trends is less robust (North Atlantic and North Pacific, the eastern tropical and subtropical Pacific), vs. areas with more robust warming signals (regions from about 40N-40S except for the eastern tropical Pacific).