Climate Processes in CMIP5: The “warming hole” simulations in CMIP5 models–role of natural climate variability versus anthropogenic effects

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Tuesday, 4 February 2014: 9:15 AM
Room C101 (The Georgia World Congress Center )
Sanjiv Kumar, COLA, Fairfax, VA; and J. L. Kinter III, P. A. Dirmeyer, and D. M. Lawrence

A relative cooling in the east-central United States often referred as the “warming hole (WH)” is a prominent climate feature in the 20th century observations. Several studies have found limited WH skill in Coupled Model Intercomparison Project Phase 3 and Phase 5 (CMIP3 and CMIP5) climate simulations by employing conventional metrics, e.g. correlation between observations and simulated time series. Recent studies have emphasized the role of natural climate variability, particularly the North Atlantic Multi-Decadal Oscillation (AMO) in the WH phenomenon. The conventional metrics are not suitable to investigate the WH as a manifestation of natural variability, because a fixed time window is employed that is concurrent with the observations. Here we investigate the WH as low frequency climate variability in 20 CMIP5 climate models using a new metric. The new metric looks for the longest continuous negative temperature trends period in a 30-year running temperature trends time series calculated for each year using data from 1930 to 2004 and averaged over eastern half of the conterminous United States (i.e. for 1901 trend is calculated for 1901 to 1930, and so on), and compares them with the observations using the total area of the longest continuous negative temperature trends. We found that all models have some skill in simulating the WH, and 18 out of 20 models underestimate the WH. The average across 20 CMIP5 simulations underestimates the WH by 57±17%. Two models – GFDL-CM3 and HadGEM2-ES – overestimate the WH compared to observations by 8% and 61%, respectively.

We further investigate the role of low frequency climate variability in the North Atlantic and its relationship to the WH using pre-industrial control (piControl) and historical (20th century) climate simulations from the same 20 CMIP5 climate models. We have employed the Hurst Coefficient (HC) to quantify low frequency climate variability in the North Atlantic. The HC is a measure of long-term persistence, and it is calculated as a maximum likelihood estimator of a presumed fractional Gaussian noise process. Using equal length time series from the historical runs (1930 to 2004: total 75 years) and the last 75 years of the piControl runs, we found that most models show significant long-term persistence in North Atlantic sea surface temperature anomaly for both the time periods. Most climate models (18 out of 20) also show higher HC in historical simulations compared to piControl simulations. The average HC across 20 climate models for piControl is 0.71±0.04. The historical average HC is 0.83±0.02. The higher HC in historical simulations is in line with the observations (HC: 0.86). The CanESM2 and CCSM4 models are further investigated using 1000 years or longer piControl simulations. We selected 100 random 75-year blocks from their respective 1000 years piControl simulations and calculated HC for each random block. For both models, the historical HC falls outside the 95% HC range from 100 random piControl time series (> 97.5 percentile range). This result indicates a possible contribution of anthropogenic forcing in the observed North Atlantic low frequency climate variability during the 20th century. We also found significant correlation (correlation coefficient: -0.47) between North Atlantic HC and WH skill among 20 CMIP5 climate models in the 20th century simulations.