The Arctic is currently experiencing significant warming and climate model projections indicate that this warming will continue and is likely to accelerate [Chapman and Walsh, 2006]. The predicted warming is particularly large over Arctic land areas where the NCAR Community Climate System Model (CCSM3) predicts between a 4 and 9°C increase above 1990's temperatures by the year 2100. A critical question is how much this warming will affect near-surface permafrost and how large-scale changes to permafrost will, in turn, feedback on the climate system.
 Burn and Nelson  (hereinafter referred to as BN) argue that the projected near-surface permafrost degradation in CCSM3 is exaggerated. We acknowledge that there are considerable uncertainties in both the timing and extent of near-surface permafrost (defined as perennially frozen soil present in upper 3.43 m of soil) degradation shown by Lawrence and Slater  (hereinafter referred to as LS) and that there are ways that the model can be improved to better represent permafrost dynamics. It is important to recognize, however, that deficiencies in the model's permafrost dynamics are balanced by the fact that coupled feedbacks, such as with the hydrologic cycle, are naturally incorporated and can therefore jointly influence future permafrost and climate.
 The projected change in near-surface permafrost does not seem wholly unreasonable when it is considered in the context of recent evidence that soil temperatures are warming, active layer thickness (ALT) is increasing, and permafrost is degrading rapidly [Osterkamp and Jorgenson, 2006; Osterkamp and Romanovsky, 1999; Smith et al., 2005]. Continuous permafrost in Alaska, which has been stable for hundreds or thousands of years has suffered an abrupt increase in degradation since 1982 that “appears beyond normal rates of change in landscape evolution” [Jorgenson et al., 2006]. Discontinuous permafrost in Canada has shown a 200–300% increase in the rate of thaw over the 1995–2002 period relative to 1941–1991 [Camill, 2005]. Payette et al.  find accelerated thawing of subarctic peatland permafrost over the last 50 years. Apparently, permafrost is changing at rates not previously seen.
 BN claim that near-surface permafrost degradation simulated in CCSM3 is inconsistent with previous studies. However, few studies have evaluated the fate of permafrost under a climate change of the magnitude generated in CCSM3 (∼+9°C over Arctic land areas under the A2 emission scenario). In those that have, roughly equivalent changes to ALT are seen. For example, using a sophisticated permafrost model that includes an insulating organic mat and both low (10%) and high (30%) excess ice contents, Anisimov and Poliakov [2003, Figure 2] show that current ALT's (0.4–3.0 m) increase to 2.5–14 m for a 8°C warming over 100 years. Using an advanced finite element model that accounts for thaw settlement, Buteau et al.  simulate downward thawing of the permafrost table at rates of up to 13 cm yr−1 in ice rich permafrost under a 5°C warming over 100 years.
 To demonstrate the importance of strong climatic forcing, we compare the CCSM3 projection to projections using the Surface Frost Index (SFI) method as applied by Anisimov and Nelson . Maximum and minimum CCSM3 annual temperature and winter snow depth are used to determine permafrost conditions using the SFI method (Table 1). Snow density of 250 kg m−3 is assumed, though densities of 200–400 kg m−3 produce similar results. The SFI method generates a similar result to that of the embedded CCSM3 soil model, i.e., under the A2 emission scenario, large-scale degradation of near-surface permafrost occurs in Russia and Alaska while the Canadian Arctic maintains continuous permafrost. The issue of climate forcing is also important when comparing future permafrost conditions in CCSM3 to permafrost conditions during the warmest period of the Holocene. Summer temperatures during the Holocene optimum, over a range of 16 terrestrial Arctic sites, are only 1.6°C ± 0.8°C warmer than 20th century temperatures [Kaufman et al., 2004], although Briner et al.  report up to +5°C change on Baffin Island. Conditions at the end of the 21st century are projected to be much warmer than at any time during the Holocene, thus it is not surprising that relic permafrost that survived the Holocene is under threat.
|Variable||Russian Arctic (66.5–90°N, 70–170°E)||Alaskan Arctic (66.5–72°N, 170–140°W)||Canadian Arctic (66.5–90°N, 120–60°W)|
|1980–1999||2080–2099 (A2)||1980–1999||2080–2099 (A2)||1980–1999||2080–2099 (A2)|
|Max. Monthly Mean Temp (°C)||8.0||14.0||4.0||8.0||2.0||7.0|
|Min. Monthly Mean Temp (°C)||−30.0||−18.0||−22.0||−9.0||−32.0||−19.5|
|Mean Winter Snow Depth (m)||0.30||0.30||0.36||0.19||0.23||0.30|
|Surface Frost Index||0.7317||0.5324||0.7860||0.5108||0.8945||0.6824|
 It is important also, in the context of temperature change forcing of permafrost, to consider where CCSM3 stands compared to other global climate model (GCM) temperature change projections. Globally, the climate sensitivity of CCSM3 is 2.7°C, which is in the middle of the range of comparable GCMs (2.1–4.4°C). Polar amplification of climate change, however, is quite strong in CCSM3 (global warming is average but pan-Arctic warming is among the highest of 14 GCMs). This strong warming may be related to a rapid decrease in summer sea-ice extent in the early part of the 21st century in CCSM3 [Holland et al., 2006].
 BN question the “calibration” done for the permafrost simulation. In fact, the distribution of permafrost in CCSM3 has not been calibrated but is simply a diagnostic of the model's internally generated ground climate that itself reflects the simulated CCSM3 climate. All parameters are physically based and are assigned a priori. Soil temperatures evolve during a 550-yr spin-up forced with 1870 atmospheric CO2 concentrations. As BN argue, it may be more correct to consider the permafrost simulated in CCSM3 to represent the combination of continuous and discontinuous permafrost, although discontinuous permafrost cannot be explicitly simulated because grid box soils are spatially homogenous and each layer maintains only a single temperature. Based on this definition, the simulated areal extent of permafrost is low. However, our conclusion that the CCSM3 simulation of permafrost extent is reasonable is predominantly based on the fact that the spatial distribution of permafrost corresponds remarkably well to the International Permafrost Association (IPA) map. Some areas of discontinuous permafrost are captured by CCSM3 (e.g. central Alaska and northwestern Canada) while some areas of continuous permafrost are missed (southern boundary in eastern Canada, western boundary in Siberia) due in large part to biases in the simulated climate, but overall the spatial agreement with the IPA map is good. Note also that the total area is sensitive to the location of the southern permafrost boundary (a one grid box southward shift, ∼150 km, corresponds to an extra 1 million km2).
 BN also point out that CCSM3 does not simulate soil ice in excess of saturation, a condition which is often seen in permafrost soil, and that the presence of excess soil ice retards permafrost degradation. The question is how important this process is on a large scale. BN note that 50% excess ice would sharply reduce the rate of ALT increase. Excess ice values of 50%, however, appear to be relatively uncommon. Zhang et al.  calculate that only 10% of continuous and discontinuous permafrost contains high excess ice (>20%) whereas the majority of permafrost (57%) is characterized by low excess ice (0–10%). BN also argue that the absence of thaw consolidation further accelerates permafrost degradation in CCSM3. According to Buteau et al. , this isn't necessarily the case. Their study concludes that failing to consider thaw consolidation actually decreases thaw rate, not increases it, as thaw settlement brings frozen soil closer to the surface. Furthermore, while the strictly defined ALT may not deepen as rapidly when measured from the top of the consolidated soil surface, the important point is that previously frozen ground, ground that may contain previously locked soil carbon, thaws as consolidation progresses.
 BN argue that the negative trend over the 20th century in the area containing near-surface permafrost seen in CCSM3 is not supported by observations. First, we clarify a point about the diagnostic. The near-surface permafrost extent diagnostic is a concise way to illustrate how the area where permafrost is present within the upper 3.5 m of soil changes over time. It says nothing about what happens to deep permafrost, which is unlikely to thaw over the century timescale. Although the 20th century trend in near-surface permafrost extent may indeed indicate that CCSM3 near-surface permafrost is too sensitive to warming, there is no corresponding observationally-based dataset to which this diagnostic can be compared. The derivation of such a diagnostic, if possible, would provide a useful validation tool for climate model permafrost simulations. It is relevant to note that CCSM3 near-surface permafrost extent is stable and exhibits no drift over the last 250 years of a 550-year 1870 control simulation (11.7 ± 0.8 million km2), indicating that the trend is not due to a drift in soil temperatures. In fact, the trend between 1910 and 1950 reflects ALT deepening at the permafrost margins as air temperature recovers from a cold period simulated in the model (∼1°C degree cooler over Arctic land areas between 1890 and 1910 relative to 1870, leading to an ∼0.8 million km2 increase in near-surface permafrost extent).
 To summarize, observations indicate that Arctic climate is changing rapidly. The extent that permafrost degrades in response to this warming over the next 100 years remains highly uncertain. BN highlight a number of possible reasons why CCSM3 may overestimate the rate of degradation. There are other factors, though, that are not currently represented in the model that could actually accelerate the degradation rate. These feedbacks include carbon efflux from freshly thawed soil [Zimov et al., 2006], projected expansion of shrub cover [Sturm et al., 2001], and wildfire impacts [Yoshikawa et al., 2002]. Due to the coupled nature of the problem, climate models can and should be better exploited to study the influence of these feedbacks on permafrost and the climate system. Efforts are underway both to improve permafrost dynamics and to incorporate these feedbacks into the next versions of CCSM.