Notice: Wiley Online Library will be unavailable on Saturday 30th July 2016 from 08:00-11:00 BST / 03:00-06:00 EST / 15:00-18:00 SGT for essential maintenance. Apologies for the inconvenience.
 The projected effects of rising CO2 levels on Northern Hemisphere extratropical cyclone activity and cyclone-associated precipitation are examined for September–May, using output from version 3 of the Community Climate System Model (CCSM3). A cyclone identification algorithm was applied to a five member ensemble of CCSM3 20th and 21st century output, along with a method of isolating precipitation produced by each cyclone. Mean seasonal statistics describing cyclone activity and the character of associated precipitation were calculated over several study regions for 20 a periods. The dominant change in cyclone activity is a marked midlatitude decrease in frequency during autumn, winter, and spring. Few significant shifts in storm tracks or cyclone intensity were identified. Total daily precipitation from these events is found to increase into the 21st century, largely because of increases in available atmospheric moisture with rising temperatures. This thermodynamic increase in precipitation leads to large rises in total seasonal cyclone-associated precipitation over high latitudes, while over midlatitudes the thermodynamic increase is offset by the dynamic effect associated with decreased cyclone frequency.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Extratopical cyclones exert strong influences on the weather and climate of middle- and high-latitude regions. In addition to affecting regional cloudiness, temperatures, and winds, the passage of these synoptic-scale systems accounts for the majority of cold-season precipitation that falls outside of the tropics. Considering the magnitude of their influence, it is important to understand how cyclone activity and the precipitation produced by these systems may respond to increasing greenhouse gas (GHG) concentrations. Specific aspects of concern include potential changes in cyclone frequency, intensity, or distribution, and changes in the amount and character of precipitation produced by these systems.
 Cyclones develop as baroclinic instabilities that typically travel along well-defined storm tracks located near regions where the meridional temperature gradient is particularly strong. Throughout their life cycle cyclones draw energy from this gradient, while acting to reduce it. This is accomplished in two ways. First, cyclones promote meridional mixing, displacing warm air with cold and vice versa. Secondly, rising motion within cyclones promotes condensation of atmospheric moisture, inducing precipitation. As much of this moisture originates from evaporation over warm, low-latitude locations, this process represents a poleward transfer of latent heat [Trenberth, 1999]. Latent heat release also represents a secondary energy source for cyclones, acting to decrease static stability within the system and intensifying the resulting surface pressure depression.
 Global warming has the potential to affect cyclones through several competing mechanisms, many of which involve increases in atmospheric moisture. Rising temperatures imply an increase in the moisture-holding capacity of the atmosphere. Together with greater rates of evaporation associated with higher surface temperatures, this suggests that atmospheric moisture content should increase. This has already been observed in several regions [Hense et al., 1988; Ross and Elliott, 1996; Zhai and Eskridge, 1997]. With more moisture available the upper limit on possible latent heat release within cyclones increases, potentially influencing the intensity of these events [Trenberth, 1999; Held, 1993]. However, this also increases the efficiency of poleward heat transport by storms, which may modify the magnitude and/or frequency of atmospheric eddies. The issue is further complicated by adjustments in mean meridional temperature gradients with rising greenhouse gases. Modeling studies show that the largest increase in surface temperature will occur over the Arctic, especially in autumn and winter, in large part because of reductions in sea ice which promote large vertical heat fluxes to the atmosphere [Holland and Bitz, 2003]. This acts to reduce the meridional temperature gradient and baroclinicity in the lower troposphere, along with the energy available for cyclone development. However, temperature gradients and baroclinicity increase in the upper troposphere, in response to differential decreases in lapse rates over low and high latitudes. Near the equator, increases in temperature can substantially decrease the moist adiabatic lapse rate. However, low temperatures over the poles limit deviations from the dry adiabatic lapse rate. The result is greater upper atmosphere warming over the tropics than the poles as GHG concentrations rise [Yin, 2005].
 The cumulative effect of these influences remains uncertain, although a variety of modeling studies using general circulation models (GCMs) have been conducted to address the issue. A popular approach involves the use of an objective cyclone identification technique, sometimes coupled with a tracking algorithm, to evaluate simulated cyclone climatologies [Lambert, 1995; König et al., 1993; Murray and Simmonds, 1995; Sinclair and Watterson, 1999; Lambert and Fyfe, 2006]. Other studies have focused on examining storm tracks using variability of an atmospheric variable on a synoptic timescale (e.g., 2–8 d) [Yin, 2005], or the frequency of atmospheric eddies [Stephenson and Held, 1993]. More recently new techniques have been applied to the problem, including the use of self-organizing maps to create an objective Arctic synoptic climatology [Cassano et al., 2006].
 Commonly these studies suggest that increasing GHG concentrations lead to decreases in Northern Hemisphere cyclone frequency [Stephenson and Held, 1993; Lambert, 1995; Sinclair and Watterson, 1999; Lambert and Fyfe, 2006]. In several studies this frequency decrease is accompanied by an increase in the number of intense cyclones [Lambert, 1995; Lambert and Fyfe, 2006; Cassano et al., 2007], while in others the frequency of intense cyclones declines [Sinclair and Watterson, 1999]. Use of different intensity metrics may partially explain these disagreements. Some GCM studies also show shifts in storm track locations. Yin  found that a suite of GCMs showed a consistent poleward shift in modeled storm tracks, using synoptic timescale filtering of eddy kinetic energy. In contrast, Lambert and Fyfe  found no evidence of such a shift in cyclones as measured at the surface using the same model runs, although the authors note that the coarse resolution used in their study may mask subtle changes in storm track location. Together these GCM experiments illustrate a range of possible responses to climate change, and have inspired studies looking for similar changes in observations and reanalysis products. Using the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP-NCAR) reanalysis, both McCabe et al.  and Zhang et al.  found that high-latitude cyclone activity had increased over the past few decades at the expense of midlatitude activity. McCabe et al. additionally found that cyclone intensities had increased across the northern hemisphere.
 Uncertainty in the response of cyclone activity to climate change leads to uncertainty in projected changes in precipitation. Simulations suggest that increases in extreme precipitation events may be more dramatic than increases in annual total precipitation, raising concerns of more frequent and severe flooding [Cusbach et al., 2001; Meehl et al., 2005]. The expected increases in both total precipitation and extreme precipitation events involve a convolution of influences from changing atmospheric circulation (“dynamic” adjustments) and increased moisture holding capacities with higher air temperatures (“thermodynamic” adjustments). Several studies have attempted to identify the relative magnitudes of these contributions. On an annual basis it appears that thermodynamic adjustments dominate [Emori and Brown, 2005; Cassano et al., 2007]. However, cold-season dynamic effects in these annual analyses are likely masked, as much of the precipitation in middle and even high latitudes falls during strong summer convective storms [Serreze et al., 2002].
 Potential changes in the Arctic are of particular interest. Discharge from the Eurasian portion of the vast Arctic terrestrial drainage (which extends well into middle latitudes) has increased over the past century [Peterson et al., 2002], likely driven by increased cold-season precipitation [Holland et al., 2006a]. These studies suggest that the Arctic Ocean may be freshening. Numerous investigations indicate that the oceanic meridional overturning circulation (MOC) is sensitive to freshwater exports into the North Atlantic, which can act to stratify the upper ocean in convective regions of the Greenland-Iceland-Norwegian (GIN) and Labrador Seas [Broecker, 1997; Manabe and Stouffer, 1999; Holland et al., 2000]. Modeling studies have shown that anomalous freshwater exports from the Arctic Ocean are capable of influencing the MOC [e.g., Delworth et al., 1997; Mauritzen and Hakkinen, 1997; Hakkinen, 1993; Holland et al., 2001; Mysak et al., 2005], and it is possible that changes in precipitation can contribute to such anomalies, particularly during the autumn, winter and spring. In present conditions, summer precipitation exerts a limited influence on Arctic river runoff, as high evaporation rates return much of the water to the atmosphere. Most high-latitude runoff is due to cold-season precipitation, which is stored as snowpack and then melts in the spring [Serreze et al., 2002]. Changes in precipitation directly over the Arctic Ocean itself may also alter the freshwater budget, and affect interocean freshwater exchanges.
 Using output from a state-of-the-art general circulation model (GCM), this study directly connects cold-season cyclone activity with precipitation events in a simple, intuitive manner to produce an expanded cyclone climatology. By comparing seasonally averaged metrics describing cyclone activity and associated precipitation from 20th and 21st century model simulations, we identify the effects of rising greenhouse gases on this expanded climatology, and relate this to changing Arctic hydrology. In addition to considering statistics averaged over the entire Northern Hemisphere, midlatitudes (30°–60°N) and high latitudes (60°–90°N), a number of smaller regions important to cyclone activity and Arctic hydrology are examined (Figure 1). These include the major storm track locations (the North Atlantic and Pacific), the GIN Sea, and both middle- and high-latitude regions of North America and Eurasia. Finally, we quantify the relative magnitudes of thermodynamic and dynamic influences on cold-season precipitation associated with cyclones.
2. Description of Model and Simulations
2.1. Model Description
 Version three of the Community Climate System Model (CCSM3) is a state-of-the-art coupled general circulation model, with individual atmosphere, ocean, land, and sea ice components that communicate via a fifth coupling component; the model uses no flux correction. A brief description of the model is provided here. Detailed descriptions of the model and its simulated climate are given by Collins et al. [2006a] and a recent special issue of the Journal of Climate, 19(11).
 The atmospheric component model used by CCSM3 is version three of the Community Atmosphere Model (CAM3), which features a T85 (approximately 1.4°) horizontal resolution and 26 vertical levels [Collins et al., 2006b]. The ocean [Danabasoglu et al., 2006] is treated via a level coordinate model based on the Parallel Ocean Program model [Smith and Gent, 1994]. It employs a roughly 1° horizontal resolution, rotated orthogonal grid in which the north pole is displaced over Greenland, and features 40 vertical levels. Isopycnal transport is parameterized following Gent and McWilliams , and vertical mixing follows Large et al. . The Community Sea Ice Model (CSIM) [Holland et al., 2006b] acts as the sea ice component. CSIM uses an elastic-viscous-plastic sea ice rheology [Hunke and Dukowicz, 1997], energy-conserving thermodynamics [Bitz and Lipscomb, 1999] and subgridscale ice thickness distribution [Thorndike et al., 1975; Bitz et al., 2001; Lipscomb, 2001]. It has the same horizontal resolution as the ocean model. The land component used is the Community Land Model (CLM3) [Bonan et al., 2002; Dickinson et al., 2006]. With the exception of the river routing scheme, which uses a resolution of 0.5°, CLM3 uses the same horizontal resolution of CAM3. CLM3 features 10 soil levels that explicitly treat liquid water and ice, a multilayer snowpack, and a subgridscale mosaic representation of land cover types.
 Several authors have compared cyclone climatologies from CCSM3 output to observations, most notably the 40+ years (a) reanalysis (ERA-40) of the European Centre for Medium Range Weather Forecasts. These studies have used several measures of storm activity, including variability of the upper atmosphere on synoptic (2–8 d) timescales [Hurrell et al., 2006; Alexander et al., 2006], synoptic climatologies of sea level pressure [Cassano et al., 2006], and mean tracks identified using cyclone identification and tracking algorithms [Alexander et al., 2006]. In general CCSM3 compares well with ERA-40 [Cassano et al., 2007], showing similar storm track locations, though there are indications that these tracks are too vigorous [Hurrell et al., 2006; Cassano et al., 2006]. Modeled cyclone frequencies also compare well with ERA-40, although CCSM3 tends to underestimate cyclone activity at the entrance and exit regions of the major storm tracks, in the lee of the Rocky Mountains, and over the Mediterranean Sea. Finally, in comparison with ERA-40 these climate features exhibit realistic variability, such as that associated with the North Atlantic Oscillation and the El Nino Southern Oscillation [Alexander et al., 2006].
Hack et al.  examined CCSM3's representation of the global hydrologic cycle, and found that it also generally compares well with observations over middle- and high-latitude regions. The primary biases of concern to the current study are a tendency toward excessive precipitation in the vicinity of storm tracks, and excessive winter precipitation near 60°N. This latter bias is centered primarily over land surrounding the North Atlantic, and leads to a high bias in terrestrial runoff into this basin.
2.2. Run Descriptions
 We use two sets of CCSM3 simulations. The first is a five member ensemble of simulations of the twentieth century (termed here 20C), part of a larger ensemble prepared for the IPCC 4th Assessment Report. Each ensemble member was started with slightly different initial conditions obtained from different years of a multicentury control integration representative of the late 19th century, and run with identical natural and anthropogenic forcings matching available observations. These forcings include solar variability, volcanic activity, and evolving greenhouse gas and sulfate aerosol concentrations. Full descriptions of the general climate and forcings used for these simulations are given by Meehl et al. .
 Output from the 20C simulations are compared to five associated 21st century simulations (termed here 21C), each of which was continued from one of the 20C ensemble members. These simulations follow IPCC's SRESA1B climate change scenario, in which greenhouse gas concentrations undergo moderate increases, with carbon dioxide reaching about 720 ppm by the end of the century [Intergovernmental Panel on Climate Change, 2001]. Model output used includes daily snapshots of sea level pressure (SLP) and daily mean precipitation.
 Twenty annums of data have been used from each ensemble member, including 1980–1999 for the 20C runs, and 2080–2099 for the associated 21C runs. All analysis was conducted on the model's native T85 resolution.
 We make use of identified cyclone centers, with no attempt to track individual cyclones from one day to another. Cyclone tracking has proven difficult using the daily output available to us, as the distance covered by a model cyclone in a day can be on the same order as the separation between individual cyclones. Hence, rather than examining cyclones through their lifecycle, this study considers each cyclone and each day separately, an approach that has been used in previous GCM studies [e.g., Lambert and Fyfe, 2006]. Cyclone frequency refers to the number of times cyclone centers are identified in the once daily model output.
 Cyclones are identified using the daily sea level pressure (SLP) fields, recorded at the end of each day. This is done following techniques outlined by Serreze et al. . The method identifies local minima in the SLP field, and then uses an iterative search to compare SLP values at surrounding grid cells in a series of outwardly expanding shells. A minimum search radius, maximum search radius, and the number of shells between the two are specified in calls to the program. If the difference between the greatest SLP value in an inner shell and the minimum SLP value in a neighboring outer shell is less than a selected threshold (ΔPthresh), the SLP minimum is declared a cyclone center. If this condition is not met but all inner shell values are less than all those in the outer shell, the algorithm compares the next two outlying shells. The SLP minimum is discounted as a cyclone center when either none of the shell sets pass the minimum difference test, or when an outer shell is found that contains one or more SLP values lower than the minimum in the neighboring inner shell before the minimum difference test is passed.
 For consistency, the choice of algorithm parameters used are identical to those employed by Serreze et al. , and include a minimum search radius of 400 km, a maximum radius of 1200 km, and three evenly spaced shells between the two. The results are not particularly sensitive to these parameter values. The ΔPthresh value selected for the minimum difference test was 2.0 mb. With shell spacings of ∼260 km, this implies that cyclone candidates with geostrophic velocities of less than about 7.5 m s−1 will be rejected. The algorithm was subjected to extensive testing with these settings before application. It was found to capture the majority of manually identified cyclones, regardless of cyclone shape or orientation. The majority of weak SLP depressions and spurious lows produced near regions of prominent topography were rejected.
 The intensity of each identified cyclone was calculated as the Laplacian of SLP at the cyclone center. The Laplacian is proportional to relative vorticity. As outlined by Serreze et al.  this measure of intensity is better than SLP at the cyclone center, since the SLP can be strongly influenced by variations in the background pressure field.
 Precipitating regions were identified as patches of connected grid cells in which total daily large-scale (i.e., nonconvective) precipitation (PRECL) exceeds 1.5 mm. This threshold allows for adequate separation of modeled precipitation systems without excluding regions of significant precipitation. Precipitating regions intersecting an area of radius 250 km (approximately 1/4 Rossby radius) about a cyclone center occurring at the end of the 24 h period over which precipitation was calculated were then associated with that cyclone. Any convective precipitation (PRECC) in the identified region was then added back to each cyclone-related precipitation system, and the horizontal extent, total daily output, mean daily intensity, and maximum daily intensity of each precipitation feature were recorded. This represents an intuitive, objective means of directly connecting each cyclone to significant, simultaneous precipitation events. As with cyclone identification parameters, the results proved relatively insensitive the choice of search radius and precipitation threshold.
4.1. Cyclone Frequency and Intensity
 The majority of CCSM3 cold-season extratropical cyclones are found in narrow, zonally oriented storm tracks centered over the North Pacific and North Atlantic. Following Blackmon , these tracks have been approximated as regions of high synoptic timescale variability in the mean sea level pressure field. Mean winter standard deviations of synoptic timescale (2–7 d) filtered sea level pressure from all runs are illustrated in Figure 2a; high-variability storm tracks are shaded. Figure 2b shows differences in this field between the 21C and 20C periods. The dominant changes are a decrease in variability across the majority of the storm track, with increases over the North Atlantic and Northern Europe.
 In order to better identify and characterize the changes in cyclone activity contributing to the shifts illustrated in Figure 2b, mean cyclone frequencies and intensities have been compared over the study regions shown in Figure 1. Changes in cyclone characteristics are evaluated from the difference in mean seasonal statistics between the 21C and 20C study periods, normalized by the pooled variance of the two populations:
Here μ21C and μ20C are the seasonal means from 21C and 20C study periods, respectively, σ21C and σ20C are the standard deviations, and n is the sample size; in this case, 20 a. Absolute values greater than 1.96 represent a statistically significant difference between the means at the 95% significance level; positive values indicate the 21C mean is higher than that of the 20C. Results for each of the five ensemble member pairs and regions considered are shown in Figure 3 (frequency) and Figure 4 (intensity). Robust indications of change are found wherever a majority of ensemble pairs agree on the direction and significance of a shift. Mean seasonal statistics have been compared for autumn (SON), winter (DJF), and spring (MAM).
 For all seasons, and for most ensemble pairs, there is a decrease in cyclone frequency in the 21C period for the Northern Hemisphere as a whole. This shift is most pronounced in autumn and becomes steadily weaker through winter and into spring. It is also evident from Figure 3 that the Northern Hemisphere signal is primarily due to decreases over middle latitudes. Over high latitudes few changes are significant, and there is little agreement between the five ensemble members.
 Statistics for the various subregions (as defined in Figure 1) indicate that autumn decreases in frequency are primarily centered over the North Pacific and midlatitude Eurasia. Additional weaker decreases are seen over the North Atlantic and midlatitude North America. No regions show compelling evidence of increased cyclone frequency. These patterns weaken somewhat into winter, although there remain moderate indications of decreases over both continental and oceanic midlatitudes. Into the spring the Pacific and continental signals weaken, while the North Atlantic shows much stronger declines in frequency.
 Several past studies have suggested that storm tracks may respond to climate change by shifting poleward [Yin, 2005; McCabe et al., 2001]. To explore this possibility, cyclone frequencies have also been examined over 10° latitude bands, both covering the entire longitudinal extent (Figure 5) as well as over separate Eurasian, North American, Atlantic and Pacific sectors (not shown). Results show that latitude bands typically associated with storm tracks experience robust decreases in frequency. The shift is most pronounced between 30°–50°N in winter and autumn, becoming weaker with increasing latitude. No hemispheric latitude bands show robust increases in frequency. Only over the North American sector is there any indication of increased frequency; this is found between 70°–80°N, and though all ensemble members show an increase, few have statistical significance. Also shown in Figure 5 are total seasonal cyclone counts for latitude bands defined by the model's meridional grid spacing (∼1.4° latitude bands) between 30° and 90°N for the 20C and 21C ensembles. The latitudes at which cyclone counts peak show little shift, although a widespread decrease in center counts is apparent throughout the midlatitudes.
 In comparison to frequency, mean cyclone intensities show few compelling changes. The only strong signal is seen in the hemisphere-wide statistics during autumn; for all but one ensemble member, mean Northern Hemisphere cyclone intensity shows a significant increase in the 21C runs. Again, this signal is dominated by changes in midlatitudes, particularly the North Pacific. However, while all ensemble members show rising intensities for these regions, few shifts are significant. With the exception of the admittedly weak North Pacific signal, indications of rising intensity weaken into the winter. By spring no regions show any notable difference between the 21C and 20C study periods.
4.2. Precipitation Systems
 As cyclone frequency decreases, it is reasonable to expect precipitation frequency to decrease as well. However, if rising air temperatures and evaporation provide additional moisture to these systems, it may lead to an increase in their precipitation output. Here we evaluate several statistics describing precipitating systems associated with cyclones.
Figure 6 shows changes in total seasonal precipitation for autumn, winter, spring, and summer across the Northern Hemisphere. For all four seasons precipitation shows a widespread increase, which is particularly pronounced over high latitudes. However, some midlatitude regions show a decrease in precipitation, with the greatest decreases occurring over parts of the North Atlantic and the North Pacific associated with the primary storm tracks. Similarly, the number of wet days (days with 1 mm or more of precipitation) decreases across much of the middle latitudes (Figure 7), with storm track regions showing the largest decrease. This is consistent with a decrease in cyclone activity. Over high latitudes the number of wet days increases sharply.
 To better examine the contribution of cyclone-associated precipitation to the changes described above, we examined precipitation near cyclone centers exclusively. As in the above discussion of cyclone statistics, high latitudes, middle latitudes, and the sub regions shown in Figure 1 were examined separately. Any precipitating system connected to a cyclone in a given region is associated with that region. As these precipitating systems can both span several study regions and contain several cyclones, certain systems will be considered more than once.
 Probability distributions of precipitation properties are typically non-Gaussian, complicating comparisons between population means. For this reason we compare precipitation properties between the 20C and 21C study periods using the nonparametric Wilcoxon Rank-Sum test [Wilcoxon, 1945], which operates well for large sample sizes and populations with very similar probability distributions, as is the case with the data being discussed here. In this method values from the two populations are combined, then each value is assigned a rank on the basis of its magnitude; this has been done for all daily cyclone-associated precipitation events from the 20C and 21C study periods for a given season. By then comparing the sum of the ranks of values from each population, the degree to which the populations are separated can be determined. As with cyclone frequency and intensity, rank sum z scores with values of 1.96 or greater indicate a statistically significant difference at the 95% confidence level. Positive (negative) z scores denote increases (decreases) in the 21C ensemble member compared to the associated 20C run.
Figure 8 shows z scores for 21C versus 20C comparisons of total daily precipitation from cyclone-associated precipitation systems. During autumn, daily precipitation increases over both middle and high latitudes, with the most robust changes occurring over the Eurasian sector. High-latitude North America and the North Pacific also show significant increases in all five ensemble pairs. Increases are also pronounced during spring, when all regions but midlatitude North America and the GIN Sea show significant change in at least three ensemble pairs. In winter only Eurasia and the North Pacific show evidence of increased cyclone-associated precipitation, and indications of high-latitude change weaken considerably.
 Much of the increase in daily precipitation is related to an increase in precipitation intensities; mean storm intensities generally increase for all study regions and seasons with the exception of high-latitude North America during winter (Figure 9). Maximum single grid cell intensity (not shown) also increases almost universally during autumn and spring, although the GIN sea and high latitudes outside of the Eurasian sector show little evidence of change during winter.
 In some study regions, the average horizontal extent of cyclone-associated precipitating systems increases and contributes to the increased daily precipitation (Figure 10). This change occurs exclusively over continental regions. Robust areal increases are seen primarily over high latitudes during autumn and spring, and midlatitudes during winter; throughout the year the Eurasian sector shows the most consistent change.
4.3. Precipitation Frequency Versus Precipitation Output
 The total change in cyclone-associated precipitation between the 21C and 20C periods can be separated into two components: (1) changes due to shifts in the frequency of cyclones, and (2) changes related to shifts in the amount of precipitation produced by these events. This decomposition can be calculated as follows:
Where p is seasonal precipitation, n is the average number of cyclones in a region, and q is the average precipitation output by an event. The subscripts 21C and 20C denote population.
Figure 11 shows the results of the frequency/output decomposition of precipitation changes. As would be expected from the earlier discussion of cyclone activity and precipitation characteristics, the frequency and output changes typically act in opposition in midlatitude regions. Here a decrease in frequency related precipitation acts to offset related output increases. Over midlatitude Eurasia and the North Pacific the magnitude of the output shift is greatest, resulting in net increases in precipitation. Over midlatitude North America and the North Atlantic frequency shifts dominate, leading to net decreases in precipitation. High-latitude regions show increases from both decomposition terms, although the contribution from frequency is usually negligible. Only during winter, when the high-latitude output increase drops significantly, do the two terms have similar magnitudes (not shown). These high-latitude results closely reflect the findings of Cassano et al. , discussed elsewhere in this issue.
5.1. Cyclone Activity
 Shifts in the synoptic timescale SLP variability from the 20C to 21C runs (Figure 2) show pronounced decreases over much of the Pacific and North America, suggesting a decline in the strength of the storm track in these regions. North Atlantic and Northern European portions of the storm track show increases in variability, which may be related to increased storm track activity, a shift in the location of the tracks, or an eastward track extension. However, these changes in variability represent a convolution of shifts in cyclone frequency and intensity, along with changes in anticyclone activity. Examining the results of regional changes in cyclone frequency (Figures 3 and 5) and intensity (Figure 4) helps to illuminate the contribution of changing cyclone activity to these shifts in synoptic variability. Figure 3 indicates that cyclone frequency declines significantly across midlatitudes in the 21C period, with particularly robust decreases over the North Atlantic, the North Pacific, and the Northern Hemisphere as a whole. Together these changes suggest a widespread weakening of storm tracks. The Pacific track is most affected in autumn and winter, while the Atlantic track is most affected in spring.
 Over high latitudes cyclone frequencies show a weak tendency to increase in the 21C runs, although few changes are statistically significant. It is possible that this is an indication of a poleward migration of the simulated storm track, similar to those found in past studies [Yin, 2005; McCabe et al., 2001]. However, it remains unclear whether this occurs in the present study. Analysis of cyclone counts over 10° latitude bands show only robust midlatitude frequency decreases, with no compensating high-latitude frequency increase. Further, the latitudes of peak cyclone frequency remain unchanged in the 21C runs (Figure 5). However, these results do not necessarily discount storm track migration, as relatively subtle shifts may be missed in the large-scale analyses used here. Weakening storm track activity may also mask shifts in location, emphasizing frequency decreases at low latitudes while deemphasizing higher-latitude increases. Contrasting results with Yin  may also be related to differences in both the fields studied and the definition of storm track used. While this study examines near-surface cyclone center counts to identify shifts in the primary latitudes of cyclone activity, Yin considers synoptic timescale variability of eddy kinetic energy, with a focus on the upper troposphere. It is also possible that storm track behavior undergoes very different changes in the upper and lower troposphere, perhaps in response to differing temperature adjustments to rising GHG concentrations.
 Mean cyclone intensity shows no robust change, aside from autumn and winter increases over the North Pacific. This limited increase is less pronounced and widespread than intensity increases reported in past GCM studies [e.g., Lambert and Fyfe, 2006], largely because different measures of cyclone intensity have been used. In this study cyclone intensity was measured as the Laplacian of SLP at the cyclone's center, as opposed to the often used cyclone central sea level pressure. This value is proportional to relative vorticity, and therefore the winds about a cyclone, giving an absolute measure of intensity. In contrast, the intensity of a cyclone with a given central pressure can vary significantly depending on the background pressure field. When central pressures are used to measure intensity shifts from 20C to 21C, results suggest more robust and widespread autumn increases in intensity (not shown). However, these shifts are likely amplified by changes in the mean SLP field, which are not exclusively the result of changing cyclone activity. For example, an often cited response of the SLP field to climate change is an increased tendency toward a positive phase of the North Atlantic Oscillation (NAO) [Fyfe et al., 1999], which is not entirely explained in observations by an increase in cyclone counts [Serreze et al., 1997]; rather, many recent studies point to important stratospheric influences on the NAO [e.g., Ambaum and Hoskins, 2002].
5.2. Precipitation Changes
 Changes in precipitation from 20C to 21C agree with a decline in storm track activity and midlatitude cyclone frequency, as seasonal total precipitation declines over regions of the North Atlantic and Pacific associated with storm tracks (Figure 6), and the frequency of precipitation events declines across much of the midlatitudes (Figure 7).
 Although the frequency of cyclones decreases, the amount of precipitation delivered by these systems increases in the 21C runs across the majority of the Northern Hemisphere. Output from cyclone-associated precipitation systems increases significantly in many of the study regions (Figure 8) because of increases in both mean precipitation intensity (Figure 9) and areal extent (Figure 10).
 The relative effects of event frequency and precipitation output per event are quantified in Figure 11. Through choice of methodology and limited robust shifts in cyclone intensity, this decomposition also represents an approximate separation of the 20C to 21C precipitation change into dynamic and thermodynamic components. As only cyclone-associated precipitation is being considered, the frequency term is entirely the result of changing atmospheric circulation; fewer cyclones imply less cyclone-associated precipitation. The second term of the decomposition combines effects of both dynamics and thermodynamics, convoluting precipitation changes related to shifts in cyclone intensity and to increases in atmospheric moisture. With few significant shifts in cyclone intensity, this term can be attributed primarily to thermodynamic influences, i.e., increased atmospheric moisture at higher air temperatures. For all months between September and May, CCSM3 predicts substantial increases in atmospheric moisture across the Northern Hemisphere, with the greatest increases occurring over the pole. This change is illustrated in Figure 12.
 The results shown in Figure 11 emphasize the importance of changing atmospheric circulation on cold-season midlatitude precipitation, an influence likely masked by heavy convective summer precipitation in studies separating thermodynamic and dynamic influences on total annual precipitation. As net precipitation (precipitation minus evapotranspiration) tends to be highest over land areas during the cold season, it is precipitation at this time of year that has the most influence on river runoff [Holland et al., 2006a], particularly the spring runoff peaks. As the land area draining into the Arctic Ocean extends as far south as 45°N, midlatitude dynamic decreases in precipitation act to offset increases in Arctic river discharge associated with thermodynamic increases in precipitation. Over high latitudes the dynamic effect is minimal, and thermodynamic precipitation effects contribute to the rising trend in river discharge identified by Holland et al. [2006a].
 It is somewhat counterintuitive that precipitation, and associated latent heat release, increases widely in the 21C runs, while cyclone intensities show little change. These results suggest that either latent heating exerts limited influence on extratropical cyclones in CCSM3, or that an opposing influence on cyclone intensity, such as reduced temperature advection, is offsetting the effects of increased precipitation. As the precipitation systems considered here extend across very large areas, it is likely that the rise in latent heating is spread too thinly, or released too far from the cyclone center, to have a significant impact on intensity.
 Changes in both cyclone activity and precipitation show some seasonality over the September–May period studied here. Most notably, differences between the 20C and 21C runs tend to be most pronounced and widespread during the autumn. The strength of precipitation changes also declines over high latitudes during the winter, before rebounding in spring. This seasonality may be related to changes in sea ice cover. In the 21C runs sea ice cover disappears almost completely by the end of summer [Holland et al., 2006a], and thick multiyear ice is almost nonexistent. The loss of this insulating cover allows for greater heat and moisture exchange between the ocean and atmosphere, providing additional fuel to any cyclones passing through the region. During autumn and winter, a thin sea ice cover is reestablished, limiting the Arctic Ocean as a source of moisture and heat. Increased spring precipitation may be related to an earlier loss of the thin ice cover.
 Fall reductions in cyclone activity may also be related to sea ice reductions, as the loss of autumn ice cover greatly reduces meridional temperature gradients in the lower troposphere. Further work is underway to address this possibility.
 Using a cyclone identification algorithm along with a simple method of connecting cyclones to precipitation systems, the effects of anthropogenic climate change on extratropical cyclone activity and cyclone-associated precipitation between September and May have been assessed. The results provide possible explanations for recent changes in river runoff into the Arctic, and indications of how the Arctic freshwater budget may change in the future.
 The most pronounced change in cyclone activity in climate projections for the 21st century is a marked decrease in midlatitude cyclone frequency, indicative of weakened primary storm tracks. Although high-latitude cyclone frequency shows some tendency to increase, the statistical significance of this increase is relatively weak. These changes are apparently not a result of a poleward migration of storm tracks as measured near the surface, although the analysis used here may simply be missing subtle shifts in track location. Cyclone intensities show very few significant changes, although some weak indications of intensity increases were found.
 Decreases in cyclone frequency are accompanied by a decline in the number of days with measurable precipitation in midlatitudes. This is particularly pronounced near regions associated with the North Pacific and Atlantic storm tracks. However, this dynamic decrease in precipitation is compensated in most regions by an increase in the precipitation output by storms. Given minimal shifts in cyclone intensity, this increase can be attributed primarily to thermodynamic increases in available atmospheric moisture. High latitudes experienced a similar rise in storm output during spring and autumn, although cyclone-associated precipitation changed little during winter. It is hypothesized that this seasonality is related to the loss of sea ice in the 21st century simulations.
 These results emphasize the importance of cold-season cyclone activity to changes in the Arctic hydrologic cycle. Increases in Arctic river discharge identified in CCSM3 in response to rising GHG concentrations [Holland et al., 2006a] appear to be the result of thermodynamic precipitation increases, which are partially offset by dynamic precipitation decreases over southern regions of the Arctic drainage basin. The study also highlights the importance of dynamic influences on precipitation from autumn through spring, an effect that is likely masked by high thermodynamic increases in summertime convective precipitation in annual analyses.
 This study was supported by NSF grants OPP-0242125, OPP-0240948, and OPP-0229649, as well as NASA contract NNG04GH04G. The cyclone identification algorithm used was adapted from the cyclone_loc.pro IDL program written by J. Lin (email@example.com). The authors also wish to thank Jeffrey Yin and an anonymous reviewer for their helpful comments during the review process.