Arctic Minimum 2007 A Climate Model Perspective What makes these two special? Do models ever have 1 year decline as great as observed from September 2006 to 2007? Is there evidence for tipping points in models? What controls sea ice sensitivity in models? Only 2 of ~20 models have any ensemble members that can keep up with trend Faster than Forecast? Stroeve et al 2007

Low extent models tend to retreat faster (N) 5 models with ITD tend to retreat faster (Y), but with considerable spread

April (mid-21st minus late 20th c) (Y!)

TREND in atmospheric heat flux is negligible in the two models (N!) (Don’t know how much variability it explains) Cloud TRENDS are not unusual in the two models (N) Ice thickness in the late 20th c is high in the two models (N)

red = observations colors = 7 SRES A1B runs w/ CCSM3 black = ensemble mean Holland, Bitz, & Tremblay 2006 T85 (1.4 deg) in atmosphere and land deg in ocean and sea ice

1) Increase in absorbed shortwave 2) Increase in Ocean Heat Transport through Fram Strait Two strong positive feedbacks? Ocean Transport Absorbed Sunlight

CCSM3 Single Year Decline at Least as Big as Observed in 2007

CCSM A1B Scenario - 7 Runs Even with TREND skew is positive, though probably not significant 4 X in 350 years the drop is as big as observed in In any decade Observed Histogram of 1 yr September Sea Ice Change

red = observations colors = 7 SRES A1B runs w/ CCSM3 black = ensemble mean

Same runs only smoothed A1B Scenario

2000 Commitment 2020 Commitment 2030 Commitment

Histogram of 400 yr CCSM3 1990s Control No significant skew so positive and negative 1 year changes are equally likely Histogram of 1 yr September Sea Ice Change

Histogram of 320 yr CCSM3 Pre-Industrial Contol Variance is ~2/3 of modern (Because thickness is ~50% greater) Histogram of 1 yr September Sea Ice Change

Autocorrelation of 400 yr CCSM3 1990s Control September ice cover observed

R=0.5 (0.7 skipping 2 outliers) Correlation Coefficient of linear trend and mean from Model uncertainty grows

Climate Models from IPCC AR4 (CMIP3) Trends in sea ice thickness depend on the mean state R=-0.86 Model uncertainty shrinks

all Feedbacks - ∆H No Ice-Albedo Feedback - ∆H 0 “GAIN” G= ∆H / ∆H 0 = 1.25 (on average) ∆H Sea Ice Thickness Change from doubling CO2

Summary Two models that keep up with forecast have unusually high increase in ocean heat transport Sea ice anomalies like 2007 occur about 1% of the time in 21st century CCSM3 runs. Anomalies are not negatively skewed and there is little memory. Anomalies increase in size as ice thins. Sea Ice albedo feedback causes sea ice to thin 10-30% faster Although positive, the feedback is not large enough to cause much uncertainty in thickness prediction Instead model errors are probably more a function of error in the mean state. (Present day thickness spread in AR4 models is 1-3m)

For a blackbody Earth-like planet ∆T o ≈ 1.2 K = feedback factor = gain Now with additional physical processes ∆T = ∆T o + f ∆T When CO 2 has been doubled

September Ice Extent in one ensemble member Holland, Bitz, and Tremblay, 2006 A1B Scenario with CCSM km 2 September Concentration

hPa hPa CCSM3 JJA Composite 2007 JJA Reanalysis Sea Level Pressure Cloud anomalies too see Culather et al for more details