1. Introduction
Simulations of the present-day, and projections of future, climate require the use of atmospheric general circulation models (GCMs), frequently coupled to ocean and sea ice models, on a domain that covers the globe. These coupled atmosphere–ocean general circulation models are computationally expensive and must be used to simulate long time periods: both factors that increase the computational cost and therefore limit the horizontal resolution of the model. There is also a need to quantitatively describe climate change on a sufficiently fine spatial scale for use in evaluating social and economic impacts—a requirement that has lead to the development of different methods of deriving finer spatial-scale information from the relatively coarse resolution fields calculated by GCMs.
Statistical downscaling is one method that has been widely used to project GCM fields to finer scales (Karl et al. 1990; Cubasch et al. 1996). The use of limited-area regional climate models (RCMs), using boundary conditions provided by a separate GCM simulation, provides an alternative method of producing finer spatial-scale fields. Such an approach is referred to as dynamical downscaling. As RCMs contain a numerical description of the physical processes within the climate system, the RCM is capable of producing an internally, physically consistent set of variables. Assuring an internally, physically consistent set of variables through statistical downscaling techniques is not as straightforward. Dynamical downscaling is, however, considerably more computationally expensive than statistical downscaling.
RCMs, operating on a limited domain but at a much higher spatial resolution than many present-day GCMs, offer the advantage of more accurately simulating mesoscale features that are either absent or crudely represented in the GCM due to the necessarily coarse horizontal resolution of the global model. Improvements in the RCM representation of climate are particularly evident for features strongly forced by topography or land–sea contrasts (e.g., Giorgi et al. 1998; Laprise et al. 2003). While RCMs offer the ability to resolve small-scale forcing more accurately than the driving GCM, the climate produced by a RCM is still strongly influenced by the driving GCM (e.g., Räisänen et al. 2004).
Nested RCMs are increasingly used to produce climate change information on a regional scale. Simulations of regional climate and climate change with RCMs have been reported for various regions of Europe (Machenhauer et al. 1998; Christensen et al. 2001; Räisänen et al. 2004; Giorgi et al. 2004) and North America (Giorgi et al. 1998; Pan et al. 2001; Laprise et al. 2003; Liang et al. 2004), as well as for Australia (Whetton et al. 2001), Africa (Joubert et al. 1999; Arnell et al. 2003), and Asia (Kato et al. 1999).
Presented here is an analysis of transient time slice experiments for greenhouse gas and aerosol forcing representative of present (1971–90) and future (2041–60) climate on a model domain covering most of North America. These simulations were made using an updated version of the Canadian Regional Climate Model version 3 (CRCM3) (Laprise et al. 2003) driven with atmospheric and oceanic data from the coupled Canadian General Circulation Model version 2 (CGCM2) (Flato and Boer 2001). A suite of experiments is presented to investigate the effects of internal variability of the combined CGCM–CRCM models and changes to the domain on which the CRCM is run, the impacts of boundary conditions, and the effects of changes to the physical parameterizations used in the CRCM on the model climatology. The effects of these changes on the model representation of the current climate are assessed by comparing the different simulations with an observed climatology for seasonally averaged precipitation and surface air temperature and the interannual variability of these two quantities. The effects of these changes on the model-projected climate change are also presented.
2. Model description
The regional model used for this study is an updated version of the Canadian Regional Climate Model (CRCM-I: Caya and Laprise 1999; Laprise et al. 1998; CRCM-II: Laprise et al. 2003). This limited-area model, developed at the Université du Québec à Montréal, uses a dynamical kernel that is based on the fully elastic nonhydrostatic equations solved by a noncentered semi-implicit semi-Lagrangian three-time-level marching scheme with a weak running time filter (Laprise et al. 1997). The horizontal grid is uniform in a polar stereographic projection, and the vertical resolution is variable, using a Gal-Chen scaled-height terrain-following coordinate (Gal-Chen and Sommerville 1975). An Arakawa-C-type grid is used for the location of the atmospheric variables, with staggering in the vertical as well as in the horizontal.
Lateral boundary conditions are provided through the one-way nesting method of Davies (1976), applied over a nine-gridpoint-wide buffer zone. Within the interior of the model domain, large-scale features in the temperature and wind fields are weakly nudged toward the driving model fields, similar to the approach of von Storch et al. (2000). As detailed more fully in Riette and Caya (2002), at every model time step the large-scale features in the driving and model fields are extracted with a spectral method that retains only features with a wavelength greater than ∼1400 km. An additional tendency, proportional to the difference between the large-scale features in the driving and model fields, is then applied to the model fields. The strength of the nudging tendency increases linearly from 500 hPa for the wind fields and 50 hPa for temperature, to a maximum at the model lid. At the model lid, the magnitude of the nudging tendency is sufficient to relax the model fields to the driving fields with a characteristic time constant of approximately 10 h.
The standard version of the CRCM3, referred to here as version 3.6, utilizes most of the subgrid-scale physical parameterization package of the second-generation Canadian Atmospheric General Circulation Model (GCM2; McFarlane et al. 1992; Boer et al. 1992). The coupled version of the same model (CGCM2; Flato and Boer 2001) combines a dynamical ocean and sea ice model (Flato et al. 2000) with the atmospheric and land surface components of the earlier uncoupled version of the model. The reader is referred to McFarlane et al. (1992) for an extensive description of the physical parameterizations, to Caya and Laprise (1999) for a summary of its implementation in CRCM I, and to Laprise et al. (2003) for modifications appearing in CRCM-II.
Notable changes to version 3.6 of the CRCM used here, as compared with that used in Laprise et al. (2003), include the addition of an interactive mixed-layer lake model developed by Goyette et al. (2000), which simulates the time-dependent evolution of surface water temperature and ice coverage for the Great Lakes of eastern North America. The treatment of deep convection has been modified as well, with the original Kain and Fritsch scheme (Kain and Fritsch 1990) replaced with that of Bechtold et al. (2001).
A set of current and future climate simulations is also presented from a version of CRCM3 using a modified set of physical parameterizations. The updated version of the model is referred to here as version 3.7 of the CRCM. The modified physics package includes the use of a constant soil water capacity of 10 cm for the “bucket” representation of the soil, which is a decrease of a factor between 2 and 8 from the original, spatially varying soil water capacity. The soil water capacity was decreased to improve the representation of summertime surface fluxes (evaporation and precipitation) over land (Verseghy 1996) and to shorten the time to freeze the ground in the autumn, a problem that had been observed to unrealistically delay the onset of snow cover (Frigon et al. 2002). We note that a value of 10 cm for soil water capacity is somewhat low compared with typically employed values. For example, Delworth and Manabe (1989) quote a constant value of 15 cm as the soil water capacity used for a bucket representation in an earlier version of the Geophysical Fluid Dynamics Laboratory GCM, and a constant value of 20 cm was used in the Canadian GCM (Boer et al. 1984).
The “snow-masking depth”—the specified depth of snow where surface elements are assumed to be covered by snow and the surface albedo changes from the albedo of land to that of snow—has been changed from the spatially varying field used in version 3.6 to a constant value of 3.0 m over land surfaces other than tundra and swamp. Solar radiative heating is calculated using an improved method with four bands in the visible and near-infrared, replacing an earlier two-band parameterization, and a better representation of the water vapor continuum (Puckrin et al. 2004). The updated parameterization of radiation results in increased atmospheric absorption as compared with the earlier version of the solar radiation. The treatment of cloud cover has been modified in the version-3.7 physics, resulting in an increased reflection of incoming solar radiation. Vertical mixing in the boundary layer has also been modified to include nonlocal mixing of heat and moisture under conditions where the buoyancy flux at the surface is upward (Jiao and Caya 2006).
a. CGCM2
The GCM-driven CRCM simulations use data from the CGCM2 (Flato et al. 2000), a model that uses spectral dynamics and was run, for the simulations presented here, at a resolution of T32 with 10 vertical levels between the surface and 10 hPa. The ocean model is a three-dimensional gridpoint model based on the GFDL Modular Ocean Model version 1.1 (MOM1.1) code (Pacanowski et al. 1993) and uses a horizontal resolution of 1.875° × 1.875° and 29 vertical levels. The isopycnal eddy-stirring parameterization of Gent and McWilliams (1990) has been included, and sea ice is simulated using the dynamical cavitating-fluid scheme of Flato and Hibler (1992).
b. Description of simulations
The CRCM simulations presented here were driven by data from a CGCM2 transient simulation covering the period 1850–2100 (Flato and Boer 2001). The specified greenhouse gas and aerosol evolution follows that of Mitchell et al. (1995), which is a modified version of the Intergovernmental Panel on Climate Change IS92a scenario (Houghton et al. 1992). The version of the physics used in the CGCM2 and the CRCM use an “effective” CO2 concentration as the basis for calculating radiative forcing. For the simulations presented here, the effective CO2 concentration increases from 408 to 476 ppmv between 1971 and 1990, and was specified to increase from 789 to 977 ppmv over the period 2041–60. The direct effect of sulphate aerosols is included in both CGCM2 and CRCM3 by increasing the surface albedo in a geographically varying fashion, as done in Reader and Boer (1998), where the annual average aerosol loading patterns are based on the slow-oxidation version of the chemistry model of Langner and Rodhe (1991), as described by Boer et al. (2000). The treatment of greenhouse gas and aerosol forcing is the same as that used in Laprise et al. (2003).
Two of the CRCM future-climate simulations presented below were driven using data from a CGCM run performed with radiative forcing specified by the Special Report on Emission Scenarios A2 scenario (Houghton et al. 2001). The A2 simulation of the CGCM used to drive the CRCM started in year 1990, using initial conditions from the same IS92a scenario run discussed above. For the SRES-A2 scenario the CO2 concentration was set to 476 ppmv in 1990 (the same value as used in the IS92a scenario for 1990) and was specified to increase from 714 to 860 ppmv between 2041 and 2060. Over the 20-yr period analyzed here, the radiative forcing specified for the CGCM using the IS92a and A2 scenarios is not significantly different, and we consider these GCM simulations to represent two realizations of the same climate.
Atmospheric fields calculated by the CGCM were archived at a frequency of 6 h and linearly interpolated in time to provide boundary conditions for the CRCM at intermediate times. Sea surface temperatures and sea ice cover as simulated by the CGCM2 were interpolated in space and time to serve as time-dependent lower boundary conditions for the CRCM over oceans. As the North American Great Lakes are not resolved by the CGCM2 at its resolution, lake surface temperatures and ice cover were calculated from the aforementioned lake model. Geophysical fields over land points, such as liquid and frozen soil water content, snow amount, and ground temperature, were initialized with monthly mean values from a CGCM2 climatology.
The CRCM simulations were performed at a horizontal resolution of 45 km (true at 60°N). Two domains were used: the Pan-Canadian (PC) domain is a 193 × 145 gridpoint domain covering Canada, the continental United States, and the northern part of Mexico; and a larger domain comprising 201 × 193 grid points, which extends the PC grid to the south and east and is referred to as the Amérique du Nord, or simply AN, domain. These domains and the topography as resolved by the CRCM are shown in Fig. 1. We note that both of these domains are large in the context of regional climate simulations, which have typically used domains on the order of 100 × 100 grid points.
In the vertical, 29 unequally spaced Gal-Chen scaled-height levels were used; the lowest thermodynamic level is about 25 m above the surface, and the computational rigid lid was located near 29 km. A time step of 15 min was used with a decentering coefficient of 0.01 in the time averaging of the semi-implicit algorithm and a running time filter of 0.05.
For present-climate conditions, a set of five CRCM simulations has been compared: three driven by data from the CGCM2 and two driven by reanalysis. A summary of the full suite of runs is provided in Table 1. A present-climate CGCM2-driven simulation on the PC domain and a similar run on the AN domain (PCpr1 and ANpr1) have been completed. Note that, owing to a change in supercomputer facilities, the GCM simulations used to drive these two simulations are not identical, though the two simulations do produce similar climates for North America. In addition, a simulation was performed on the AN domain using CRCM version 3.7 with modified physics (ANpr2) and driven by the same set of boundary conditions as the other AN domain simulation. To explore the impacts of boundary conditions on the model climate, additional simulations have been performed on the AN domain driven by data from the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis (Kalnay et al. 1996) and using Atmospheric Model Intercomparison Project (AMIP; Gates et al. 1999) ocean temperatures. The NCEP–NCAR reanalysis–driven simulations have been performed with both the standard set of physical parameterizations (ANra1) and the updated version-3.7 physics (ANra2). The aforementioned simulations were begun at different dates, though a common 20-yr period from 1971 to 1990 has been selected for analysis that allows for at least a 2-yr spinup for all simulations.
A set of four simulations for future-climate conditions has also been completed to allow for an analysis of the model-projected climate change. Two simulations have been performed to pair with the present-climate simulations on the PC and AN domains (PCfr1 and ANfr1a, respectively). Both of these simulations used a radiative forcing specified by the IS92a scenario, described above. An additional simulation on the AN domain using the SRES A2 radiative forcing has also been performed (ANfr1b). The final simulation was performed using the set of modified physical parameterizations, CRCM version 3.7, and the SRES A2 forcing on the AN domain (ANfr2). Data from 2041–60 were analyzed from these simulations, which again allowed for at least a 2-yr spinup period for all simulations.
3. Results
We first present a general overview of the current climate as simulated by the CRCM driven by data from the CGCM2. The CRCM representation of large-scale atmospheric features is assessed by comparing the model December–February (DJF) and June–August (JJA) seasonal average 500-hPa geopotential height with the same quantity taken from the NCEP–NCAR reanalysis. Figure 2 presents the comparison of these fields for the period 1971–90 using results from the ANpr1 simulation on the AN domain with a 10 gridpoint ribbon around the perimeter of the domain, where nesting was applied, removed. For the DJF season, the large-scale features in the 500-hPa height field are generally well placed though the model has a tendency to underestimate the amplitude of the circulation. Ridging along the west coast of North America is underestimated, and the trough extending southward over Hudson Bay is not deep enough, or as sharply defined, as compared with the NCEP climatology. These errors result in the model 500-hPa height field being too low over the northwestern portion of the model domain and too high, by as much as 10 dm, over eastern Canada. Heights over the southern regions of the model domain are underestimated, with the magnitude of the underestimate increasing to 6 dm over the extreme southern portions of the model domain.
For the JJA season, the model underestimates the large-scale gradient in the 500-hPa height field, with heights being too high over northern regions and the magnitude of the subtropical high slightly underestimated over southern portions of the domain.
A comparison of the winter and summer average surface air temperature from the ANpr1 simulation of the present climate with the Climatic Research Unit (CRU) gridded observations (CRU TS 2.0; Mitchell and Jones 2005) for the period 1971–90 is presented in Fig. 3. For the DJF season the CRCM shows a warm bias over a large part of the continent. Over more southerly, snow-free regions, this bias is generally less than 2°C, though over the northern United States and across central and eastern regions of Canada the warm bias is larger than 6°C. The winter season warm bias results, to some extent, from the model underestimating the depth of the longwave trough over North America, as was seen above in the comparison of 500-hPa heights. North of 60°N the warm bias rapidly decreases, and across the Canadian Archipelago the model is too cold by 2°–3°C. An important caveat to the comparison over northern regions is the extremely limited number of observing stations that exist across northern portions of Canada for use in the CRU climatology (New et al. 1999). The model demonstrates a cold bias of between 3° and 4°C over the arid regions of the U.S. Southwest and northern Mexico.
In general, model performance for surface air temperature is much better for the summer season. Over the eastern two-thirds of the continent, east of the cordillera along the western edge of North America, the model displays a warm bias in average surface air temperature of generally less than 2°C. The warm bias increases in magnitude, reaching from 3° to 4°C over more northern and western regions of Canada. Over mountainous regions of the western United States the model shows a cold bias of 3°–4°C, becoming larger in the arid regions of the U.S. Southwest and northern Mexico, reaching 5°–6°C over this region.
We note that the elevation dependence of surface air temperature introduces an additional source of uncertainty when comparing model output with the gridded CRU observations over mountainous regions. The CRU data have been corrected to a gridded elevation dataset, though no attempt was made to correct the CRCM surface air temperature to account for differences between the CRU and CRCM topography.
The DJF and JJA seasonal average precipitation is, likewise, compared with the gridded climatology of precipitation from CRU for the 1971–90 period and presented in Fig. 4. Although somewhat obscured by the presentation of fractional differences given in Fig. 4, the model captures the large-scale features in the observed distribution of DJF precipitation. The maximum along the west coast of North America is well represented, though arguably placed slightly too far north with large values of precipitation extending farther north along the coast of Alaska than shown in the observations and not extending far enough south into California. The secondary maximum inland along the Rocky Mountains on the east side of the Okanagan Valley in British Columbia is also well represented in the model, though too much precipitation is simulated inland of the Coastal Mountain range, inland in Washington state, and northward into the Okanagan Valley itself. The large region in the interior of the continent with DJF precipitation averaging less than 1 mm day−1, stretching from northern Mexico northward to the Arctic Ocean and eastward to Hudson Bay, is generally well reproduced by the model with errors of less than ±40% over much of this area. Northern Mexico and the eastern slopes of the Rocky Mountains, as well as far northern regions, are exceptions with the model predicting too much precipitation over these regions. Finally, the increasing amounts of precipitation eastward toward the Atlantic Ocean are also well captured by the CRCM. The observed maximum in precipitation inland along the north coast of the Gulf of Mexico is notably absent in the model, a problem that has been found with other RCMs as well (Giorgi et al. 1994; Pan et al. 2001).
Summer precipitation is overestimated across a large portion of the continent by between 1 and 3 mm day−1 during the summer season. This overestimate is shown as a widespread positive bias of more than 40% of the observed amounts over eastern regions and overpredictions of greater than 80%, to more than 200%, of the observed amount over western regions where the observed JJA precipitation is smaller. The problem appears to result from an unrealistically large number of days with between 0.5 and 3 mm of precipitation. As discussed further below, the problems with summer precipitation result from biases in the shortwave radiation scheme used in this version of the CRCM and the simple bucket representation of soil moisture. Results from a version of the model with an updated set of physical parameterizations, including shortwave radiation, are presented below. It is noted that a more complex surface scheme (Verseghy et al. 1993) has recently been implemented in the model and is currently being assessed.
One exception to the overprediction of JJA seasonal precipitation is a region stretching north from the coast of the Gulf of Mexico into the southern plains of the United States, where the modeled precipitation is in good agreement with the observations or shows a slight underestimation.
a. Regional analysis
From the general features of the CRCM representation of the present climate, we move to a more detailed investigation of the seasonal averages, interannual variability, and the influence of driving data and modifications to the physical parameterizations on the model representation of the present climate. For this analysis a set of 11 physiographic-based regions were defined, shown in Fig. 5, taking into account dominant topographic features and climatic variations. Out of necessity, to ensure that the regions did not become too small or numerous, important differences in climate can occur within individual regions, though the set of regions used should capture the large-scale variations in climate. Furthermore, to allow for a comparison of results from the sets of simulations performed on the PC and AN domains, all of the defined regions fall well within the PC domain, the smaller of the two overlapping domains.
Shown in Fig. 6 are the regionally averaged seasonal surface air temperatures from the five different current-climate simulations of the CRCM, together with the CRU observations. We note the very large geographic and seasonal variation in temperature over the model domain. As compared to the CRU observations, the CRCM captures these large variations in temperature across the domain and throughout the year. To more clearly illustrate the differences between the simulations the seasonal temperature bias, defined against the regional averages calculated from the CRU observations, is given in Fig. 7.
For the DJF season, both GCM-driven simulations using version 3.6 of the physics (PCpr1 and ANpr1) display a warm bias of between 3° and 4°C over many regions to as much as 6°–8°C over the northeast (NE) forest and northwest (NW) forest regions of central Canada. The GCM-driven simulations also show a warm bias of between 2° and 4°C in the September–November (SON) seasonal average over northern regions of the domain. We note that these biases are quite similar to those found in the original CGCM surface air temperature fields for both SON and DJF seasons (not shown). The NCEP–NCAR reanalysis-driven simulation using version 3.6 of the CRCM physics, denoted ANra1, does not display these warm biases. It is interesting to note that the GCM-driven simulation performed using version 3.7 of the physics (ANpr2) also does not display the large DJF season warm bias seen in other GCM-driven simulations. As noted above, version 3.7 of the physics uses a constant soil water capacity of 10 cm, which is between two and eight times smaller than that used in version 3.6. The larger soil water capacity of version 3.6 impedes the freezing of the soil in autumn and winter since the soil parameterization requires all water in the single soil layer to freeze before the surface temperature is allowed to fall below 0°C and snow is allowed to exist on the surface (Frigon et al. 2002). The reduced soil water capacity of version 3.7 gives a more rapid decrease in surface air temperatures through the autumn and early winter and results in better agreement with observations for surface air temperature in the CGCM-driven simulations. However, the tendency of the updated physics to produce colder surface air temperatures in DJF results in a cold bias of between 3° and 4°C when used for the reanalysis-driven simulations.
The combination of the colder winter surface air temperatures generated by the version-3.7 physics and an apparent warm bias from the GCM boundary conditions results, serendipitously, in winter surface air temperatures in good agreement with observations for the GCM-driven simulations. However, it has been standard practice within the CRCM modeling group to assess changes to the model using reanalysis-driven simulations, and thus the model with version-3.7 physics is judged to display a cold bias for winter surface air temperatures. As will be seen below, though, the updated physics also result in an improved representation of summer season precipitation.
The reduced soil water capacity used in version 3.7 may also be contributing to a larger warm bias than that found for simulations with version-3.6 physics for JJA seasonal temperature over the Great Lakes and Northern Plains regions. Many other regions show little difference in JJA temperature between simulations with versions 3.6 and 3.7 of the physics, and over far-northern regions the version-3.7 physics actually produce colder (by 1°–2°C) JJA seasonal average temperatures.
One other difference among the simulations worth noting is found over the Northern Plains region in the March–May (MAM) season. Here, the two simulations using updated physics (ANra2 and ANpr2) give MAM temperatures that are 3°C warmer than the other CRCM simulations. As noted above, in version 3.7 a constant snow-masking depth of 3.0 m was used over much of the model domain. The original version (3.6) of the physics used a value of 0.1 m over the agricultural regions of the Central Plains, and this value was found to result in an unrealistic persistence of snow cover in the spring. The larger snow-masking depth used in version 3.7 resulted in a more realistic timing of the spring snowmelt in the Northern Plains of the United States and southern portions of the Canadian Prairies. The change in snow-masking depth also resulted in warmer surface air temperatures for this region in MAM for both GCM- and reanalysis-driven simulations, bringing them into better agreement with observations.
For more southern regions in the SON season, as well as for most regions for the MAM and JJA seasons, the regional averages produced by the CRCM are within ±2°C of the CRU observations. We find differences of typically less than 1°C in seasonal surface air temperatures between the simulations performed on the PC and AN domains.
Note that these runs have been performed using driving data from the second-generation Canadian GCM, which has a recognized wintertime warm bias over northern regions. The third-generation CGCM displays a much-reduced warm bias for these regions (Boer et al. 2000).
A similar comparison of the five CRCM simulations and the CRU observations has been made for seasonal precipitation and is shown in Fig. 8. Note that no corrections have been made to the CRU data to account for known biases in precipitation measurements, most notably wind-induced gauge undercatch of solid precipitation (e.g., Legates and Willmott 1990). Adam and Lettenmaier (2003) have shown that this correction can approach a factor of 2 during the winter over northern regions of Canada.
As discussed above, the three CRCM simulations performed using version 3.6 of the physics strongly overpredict precipitation during the summer (JJA) season for all regions. Excepting the summer season, the model seasonal average precipitation is generally within ±40% of the observations. The Yukon region and the mountainous region over western North America, referred to here as the Western Cordillera region, are exceptions. The CRCM has difficulty reproducing the very small precipitation amounts observed for the Yukon in seasons outside of JJA, though as noted above there are open questions concerning the accuracy of the gridded observations for all three of the regions across the north of Canada. The overprediction of precipitation for the Western Cordillera region reflects, to some extent, the fact that the model produces too much precipitation in the interior valleys and plateaus of southern British Columbia and the northwestern United States.
The CRCM simulations performed using version 3.7 of physics, the ANra2 and ANpr2 simulations, show much better agreement with observations for JJA season precipitation. Several important parameterizations have been changed between versions 3.6 and 3.7, though for JJA seasonal precipitation tests have found that the largest effects on precipitation result from the changed shortwave radiation scheme. The updated four-band visible and near-infrared radiation results in more absorption of incoming solar radiation by the atmosphere. Associated changes to the way clouds are specified in the radiation also result in a higher planetary albedo. These two changes result in a decrease in the solar radiation absorbed at the ground by approximately 60 W m−2 over the central regions of the model domain, averaged over the month of June, and bring the model radiation budget at the ground into better agreement with Surface Radiation Budget version 2 (SRB2; data were obtained from the NASA Langley Research Center Atmospheric Sciences Data Center) estimates. The decrease in surface radiative heat flux results in decreased latent and sensible heat fluxes at the surface, which act to reduce the almost daily occurrence of precipitation found in the version-3.6 simulations. As the problems with JJA precipitation result from locally generated processes, the effects of the modified physics are quite similar for both the GCM- and reanalysis-driven simulations.
Regarding differences between the two simulations performed on different model domains, we find generally very similar precipitation amounts predicted for both simulations. The most significant exception being the Mid-Atlantic region along the east coast of North America, where the PC domain simulation has precipitation from 20% to 40% higher than the AN domain simulation for the DJF and MAM seasons. The AN domain extends the PC domain farther to the south, covering much of the Gulf of Mexico, and to the east over the Atlantic Ocean. These changes may have allowed for a fundamentally different evolution of synoptic systems along the east coast of North America to occur in the larger AN domain. A similar pattern of differences in precipitation was found between simulations on the PC and AN domains for simulations driven by NCEP–NCAR reanalysis (results not shown). Here, as the two simulations were driven by two different GCM realizations of the climate, internal variability of the GCM cannot be ruled out.
An interesting aspect of the comparison of precipitation is the finding that the seasonal precipitation climatologies produced using nesting data from the CGCM and the NCEP–NCAR reanalysis generate almost identical precipitation patterns on the regional scale analyzed here. For the DJF and MAM seasons, the average precipitation from all five CRCM simulations differ from one another by less than 0.4 mm day−1 for most regions. The Maritimes and Mid-Atlantic regions, along the east coast of North America, show larger differences of up to 1 mm day−1 for these seasons. For the JJA season, and to a lesser extent the SON season as well, the simulated precipitation amounts fall into two groups according to which version of the physical parameterizations were used.
Ignoring for the moment the differences related to changes in the physics, the similarity of simulated precipitation for the JJA season is not surprising. As many researchers have noted, simulations for the summer are more locally controlled than for other seasons (e.g., Giorgi et al. 1994; Caya and Biner 2004); additionally, the version-3.6 model physics generates a significant positive bias that is common to all simulations irrespective of the nesting data used. However, for other seasons when larger-scale forcing can expected to be more dominant, the model simulations driven by NCEP–NCAR reanalysis and CGCM data also show very similar amounts of precipitation. In the interior of the continent the CRCM produces precipitation amounts that agree well with observations whether CGCM or NCEP–NCAR reanalysis is used. Perhaps more surprisingly, for regions where the CRCM does not correctly predict precipitation when driven by the CGCM, the use of NCEP–NCAR reanalysis produces little improvement in the model predictions. For example, the Canadian Maritimes along the east coast of North America is a region where the model underpredicts precipitation by between 20% and 40% for the SON and DJF seasons, with simulations made using nesting data from either the CGCM or NCEP–NCAR reanalysis producing similar amounts. Similarly, for regions in western North America, the West Coast and Western Cordillera, the comparison with observations shows a tendency to overpredict precipitation, and the use of NCEP–NCAR or CGCM data for nesting produces little change in these predictions.
One possible explanation is that, due to the spatial scale of the regions and the use of seasonal averages, the variability in precipitation between regions shown in the analysis is controlled by large-scale features, such as topography and proximity to oceans, that are captured by the model similarly whether NCEP–NCAR reanalysis or CGCM data are used. For regions where the model poorly captures the observed seasonal precipitation, it is possible that common biases introduced by the model treatment of dynamics and physical processes produce similar errors in the precipitation fields, again somewhat irrespective of which driving data are used.
It is noted in passing that, though the regional precipitation averages are quite similar between the different CRCM simulations, the CGCM precipitation (not shown) is typically 0.5–1.0 mm day−1 larger than the CRCM-predicted amounts. The JJA season is an exception, with CGCM precipitation amounts being comparable to those predicted by version 3.6 of the CRCM.
b. Interannual variability
In addition to the seasonal averages, interannual variability of the temperature and precipitation has also been investigated. The interannual standard deviation of the seasonal-mean surface air temperatures is shown in Fig. 9. From the observations we note that the interannual variability of temperatures is greatest for the SON season and is largest for the interior and northern regions of North America. Overall the CRCM reproduces the spatial and seasonal patterns of the variability. However, the CGCM-driven simulations underestimate the interannual variability in the SON and DJF seasons. Given the large-scale forcing responsible for interannual variability in these seasons, changes to the model physics have little effect on the model variability though the use of NCEP–NCAR reanalysis boundary conditions does produce a markedly better representation of the interannual variability. Other seasons show a more modest improvement for the representation of interannual variability when NCEP–NCAR reanalysis is used. For certain regions in certain seasons the NCEP–NCAR-driven simulation provides improved interannual variability, though for many other cases the CRCM simulation driven by NCEP–NCAR reanalysis is similar to the simulations driven by CGCM data.
The most significant changes in interannual variability associated with the modified physics are found for the JJA season. The reduced soil water capacity results in an unrealistically increased sensitivity of the model surface air temperature to changes in precipitation, particularly for regions in the interior of North America.
The interannual standard deviation of the seasonal-mean precipitation, presented here normalized by the corresponding seasonal average and referred to as the coefficient of variation, is shown for the different regions in Fig. 10. The coefficient of variation does not show a clear seasonal cycle for most regions, with values typically falling between 0.1 and 0.25 throughout the year. For the DJF season the different model simulations generally reproduce the observed interannual variability in precipitation. The Mid-Atlantic region is notable as one instance where the GCM-driven simulations greatly underestimate the variability, while the reanalysis-driven simulations perform much better in comparison to the observed variability. In the MAM season the interannual variability is generally underestimated for all simulations, though the reanalysis-driven simulations produce a better representation of the variability for eastern regions. The comparison for the JJA season is affected by the biases in precipitation for this season, with the version-3.6 simulations significantly underestimating the coefficient of variation and simulations with version 3.7 producing a coefficient of variation closer to that found from the observations. For a few regions, however, the version-3.7 simulations greatly overestimate the coefficient of variation for the JJA season. The comparison for the SON season is more mixed, with no consistent differences between the various simulations.
c. Propagation of bias
From the set of simulations that have been completed, there exist two reanalysis-driven simulations performed with different versions of the physical parameterizations and a complementary pair of GCM-driven simulations with the same two sets of physics. Figure 11 presents a comparison of the change in the regionally averaged seasonal surface air temperature and precipitation caused by the change in the set of physical parameterizations. The change resulting from the modified physics found for the NCEP–NCAR reanalysis–driven simulations is compared with the corresponding change found for the pair of GCM-driven simulations.
As shown in Fig. 11, the changes introduced in the physics result in nearly identical changes in temperature and precipitation whether the CRCM is driven by NCEP–NCAR reanalysis or by fields from the GCM. That the model reacts to changes in the physics similarly for both sets of driving data suggests that the biases introduced in the RCM simulation by the physical parameterizations are, to a large degree, independent of the driving data. Assuming there are only two sources of bias that vary in these simulations, as the model dynamics and domain are not changed, a corollary to the independence of biases generated by physical parameterizations would be the independence of biases passed to the RCM from the driving data. Therefore, we suggest that if a RCM is developed to reproduce the present-day climate using driving data from reanalysis, any biases in a set of GCM fields (as compared to the reanalysis data) subsequently used to drive the RCM will result in RCM biases of a comparable magnitude, particularly for RCM fields that are strongly influenced by boundary conditions.
We note, however, that these results apply strictly only for the set of changes to the physical parameterizations introduced here: that is, we have only considered changes to precipitation and surface air temperature and that the response of the physics was only tested for changes between runs driven with NCEP–NCAR reanalysis and the Canadian GCM for presentclimate conditions.
d. Projected climate change
The general ability of the CRCM to simulate present-day seasonal mean temperature and precipitation has been investigated above. We now briefly describe the CRCM projections for the change in the seasonal averages for these two quantities over the period 1971–90 to 2041–60.
Figure 12 presents the projected changes in seasonal-average surface air temperature calculated from four pairs of present and future climate CRCM simulations. Two of these pairs have been completed using the IS92a forcing scenario—one on the PC domain (PC92a-1) and one on the AN domain (AN92a-1). Recall that these PC and AN simulations were driven by different GCM realizations of the present and future climate. An additional pair of simulations has been defined using the current climate of the IS92a simulation on the AN domain and a future simulation following the SRES-A2 scenario on the AN domain (ANa2–1). A final climate change experiment has been defined using current and future climate simulations performed with the modified set of physics parameterizations, following the SRES A2 forcing on the AN domain (ANa2–2). The definition of these present and future climate pairs is summarized in Table 2.
Also shown in Fig. 12 is the projected climate change derived from the original CGCM simulations used to provide the driving data for these CRCM simulations. The climate change signal was calculated from the CGCM by interpolating the present- and future-climate fields onto the CRCM grid and calculating regional averages in an identical manner as was performed for the CRCM output.
The change in seasonal-average surface air temperature is seen to be the largest over high northern latitude regions in winter and to a lesser degree in spring. A warming of between 3° and 4°C is found for regions in the center of the continent, such as the Northern Plains, in winter and spring. Changes in the JJA and SON seasons are more homogeneous across the different regions and generally fall between +1.5° and +3.0°C. These general patterns of warming are found for all CRCM and GCM simulations.
For each region and season the warming projected by the different CRCM simulations is largely similar. Note, however, that the PC92a-1 simulation shows a larger warming for many regions than does the AN92a-1 simulation. These differences are most pronounced over the northern regions of the Yukon, Mackenzie, and east Arctic in the MAM season, where the PC92a-1 simulations shows a warming of 1°C more than that given by the AN92a-1 simulation. It is interesting to note that the differences in the CRCM-projected warming over these regions is paralleled by similar differences found in the CGCM fields used to drive these two simulations. As shown in Fig. 12, the CGCMa-92a simulation, used to drive the PC92a-1 simulation of the CRCM, shows an increased warming as compared to the CGCMb-92a simulation, which was used to nest AN92a-1. Additional instances, where the CGCMa simulation shows a greater warming than the CGCMb simulation, with similar differences found in the CRCM projections, can be found for other seasons and regions as well. Thus, differences in projected warming caused by the internal variability of the GCM result in similar differences in the CRCM-projected warming.
Projected changes found from the CRCM with version-3.7 physics are largely similar to the projected changes from the other CRCM simulations, though the JJA season is an exception. Owing to the decreased soil water capacity, the ANa2–2-projected warming for JJA is typically 0.5°–1.0°C larger than that projected by the other CRCM simulations for all regions except those over the far north. The same physical processes that gave rise to the increased interannual variability of JJA temperature found for the current climate of CRCM version 3.7 have also resulted in a larger climate change warming signal for the JJA seasonal temperature.
One other significant difference is found for the ANa2–2 simulation with the updated physics over the Northern Plains region in MAM. The other CRCM simulations and the CGCM simulations all project a warming of approximately 4.0°C for this region, while the ANa2–2 simulation has a smaller warming of 2.0°C. As noted above, the modified physics included changes to the snow-masking depth that resulted in an earlier melting of snow over the Northern Plains. The original version of the physics used a smaller snow-masking depth and maintained snow cover much later into the spring. By maintaining snow cover later into the spring under current-climate conditions, the original version of the physics produces a larger snow albedo feedback for a warmed climate. Given the errors in the present-climate simulation of spring snow cover, it seems reasonable to assume that the larger MAM warming of 4.0°C over the Northern Plains is erroneous, even though the larger warming is present in all the other climate change experiments.
The projected changes in seasonal average precipitation between 1971–90 and 2041–60 are shown in Fig. 13 as a fractional change with respect to the present-climate seasonal-mean precipitation. In general the results show changes in precipitation of less than ±10% with a tendency to project a small increase in precipitation over most regions and seasons. Larger increases are projected by several of the CRCM simulations over the NW Forest and Northern Plains regions, in the interior of the continent, in the MAM season, though the pair of simulations performed using the updated physics, the ANa2–2 projection, shows little change in MAM precipitation for these regions.
A second large increase is found over the east Arctic region for the winter (DJF) season, where the different CRCM simulations are unanimous in projecting an increase in seasonal average precipitation of between 20% and 30%. The absolute increase in precipitation is not large, as the DJF seasonal average for the east Arctic is approximately 0.5 mm day−1, though the increase is interesting since the different GCM simulations project a near-zero change in precipitation. Sea ice predicted by the CGCM for 2041–60 is much reduced from the current-climate distribution, which would give rise to larger surface fluxes of water vapor to the atmosphere. We believe the increased source of water vapor in the far north is a likely explanation for the increased precipitation seen in the CRCM simulations. The CRCM uses the sea surface temperature and sea ice distribution simulated by the CGCM, though the response of local precipitation to these changes is markedly different in the CRCM and CGCM. The CGCM may be projecting little change in precipitation owing to the much different representation of land and sea in the Canadian Archipelago.
We note that for JJA seasonal precipitation, changes projected by CRCM simulations using the standard set of physical parameterizations and the simulations using the updated physics are quite similar, despite the fact that current-climate precipitation is considerably different when the updated physics is used.
To summarize the projected climate change from these four pairs of simulations, the average changes in screen temperature and precipitation are presented in Figs. 14 and 15, respectively, for the DJF and JJA seasons. Note that three of these climate change projections have been made with CRCM simulations on the larger AN grid, though these simulations have been analyzed on the smaller PC grid to allow for the inclusion of the fourth member. For the projected changes in seasonal temperature, the large-scale features are as described above; the largest warming is found over northern regions for the DJF season. Changes in the DJF season are also larger in the interior of the continent than nearer the coast, with changes over the eastern third of the United States showing a weaker warming of between 0.5° and 1.5°C. Temperature changes in the JJA season are generally smaller and more spatially uniform than those in DJF. Obscured by the regional analysis of changes presented above are some of the smaller-scale features in the climate change signal. Notable amongst these are larger, local, changes in screen temperature at higher elevations over the southwestern United States in DJF and over northwestern British Columbia in JJA. The projected changes over these regions also show considerably larger variation amongst the different model simulations, owing to the local positive feedbacks and differences in the treatment of snow cover. Variability among the different projections is also larger over the far north, owing partially to the internal variability of the different GCM simulations.
The average fractional change in seasonal-mean precipitation, defined as the average of the fractional changes projected by each of the four climate change simulations, is shown in Fig. 15 for the DJF and JJA season. Note that for certain regions where the precipitation rates are very small, rounding errors during data storage resulted in poorly defined fractional changes. These regions are left unshaded in Fig. 15 and are found over the far north in DJF and over portions of the U.S. Southwest in JJA.
Not clearly shown in the regional analysis is a broad region over the eastern United States where the CRCM simulations project a decrease in DJF season precipitation of between 5% and 15%. Localized decreases are also found over mountainous regions along the west coast of North America. Larger increases in DJF precipitation are also found over the U.S. Southwest and over the central Northern Plains. As was seen above, the largest relative changes in DJF precipitation are projected to occur over northern Canada, associated with decreases in the extent of sea ice. Variability in the projections of the change in DJF precipitation displays a lot of small-scale features associated with topography as well as differences in the climatological storm tracks, with the latter effect particularly evident over the Atlantic Ocean.
Changes for JJA show a widespread, though small, decrease in seasonal precipitation of between 0% and 10% over much of the United States. Small increases on the order of 10% are projected over much of Canada. Larger increases are projected over the far north and along topographic features over western North America and Greenland.
4. Conclusions
The current climate simulated by two variants of the third-generation Canadian Regional Climate model has been investigated. Simulations were performed on two different domains using different realizations of the current climate from the CGCM2. Additional simulations were performed on the larger Amerique du Nord (AN) domain: two simulations driven with NCEP–NCAR reanalysis and two driven with CGCM output, each pair containing one run with the standard set of physical parameterizations and one run with a modified set of physical parameterizations. The CGCM-driven simulations using the standard set of physical parameterizations displayed a significant warm bias in surface air temperature for the DJF season over much of the model domain and an overprediction of summertime (JJA season) precipitation. The use of boundary conditions from NCEP–NCAR reanalysis resulted in much better agreement for DJF surface air temperatures, though it had little effect on summertime precipitation. The use of a modified set of physical parameterizations in the CRCM resulted in better agreement for both DJF temperatures and JJA season precipitation for the CGCM-driven simulations. However, the use of the modified physics for reanalysis-driven simulations resulted in DJF surface air temperatures colder than observed, though still displaying the improved JJA precipitation. The smaller soil water capacity used in the modified physics acted to reduce the excessive thermal inertia of the soil and allowed snow cover to become established earlier in the season, in better agreement with observations. The use of a smaller soil water capacity has, however, also resulted in a model climate that appears too sensitive to changes in soil water content, as evidenced by the larger-than-observed interannual variability of surface air temperature during summer.
Differences in the current climate simulated on the two different model domains were generally small, though, since the driving data for these two simulations were not identical, it was difficult to separate changes due to differences in the domain from changes due to the internal variability of the GCM.
The model-projected climate change from a number of different CRCM simulations, defined as the change in the 20-yr averages from 1971–90 to 2041–60, has also been presented. The pattern of changes in temperature is similar to that found in other investigations; the largest warming is in winter at high northern latitudes with smaller warming found for regions farther south and for other seasons. In general, little variability in the projected climate change signal was found between the different CRCM simulations, though several important differences were noted. Internal variability of the CGCM used to drive the CRCM was found to have a significant impact on the CRCM projected climate-change signal for temperature. In particular, projected warming over northern regions in the MAM season in one of the CGCM realizations of climate change was considerably greater than that found for the other CGCM simulations. This difference in the climate-warming signal of the CGCM was mirrored by similar differences in the CRCM simulations. Modifications to the suite of physical parameterizations were also found to have important effects on the projected climate change for specific seasons and regions. Climate change for the spring season was found to be quite sensitive to the model treatment of snow cover. The standard version of the physics has a documented bias of maintaining snow cover for too long over the Northern Plains, which gives rise to a much stronger snow albedo feedback under climate warming. Modifications to the physics to reduce the bias in snow cover results in both better agreement for current-climate surface temperatures and a much smaller climate-warming signal. Changes to the soil water capacity result in a model climate that is more sensitive to soil moisture variations in the summer season, and also result in a model-projected summer warming that is larger than that found using the original soil water capacity (widespread warming of approximately 3°C for the modified, smaller values of soil water capacity versus projected warming of approximately 2°C for the original values).
Changes in seasonal precipitation between 1971–90 and 2041–60 projected by the different CRCM simulations were found to be generally small, less than ±10%, for most regions and seasons. In general, the different CRCM simulations gave reasonable agreement on the projected changes in seasonal precipitation, with the largest variation in projected changes found over the Northern Plains region for the MAM and SON seasons. The CRCM-projected changes in DJF season precipitation over the Canadian Archipelago were markedly different than the GCM-projected changes over this region. The CRCM uses the sea ice distribution calculated by the CGCM, though the representation of the distribution of land and sea through the archipelago is considerably different in the CRCM. The different representation of ocean and land in the CRCM likely explains the large increase in winter precipitation over this region projected by the CRCM, where the GCM is showing little change.
Acknowledgments
The authors wish to thank Claude Desrochers and Mourad Labassi for tirelessly maintaining the local computing environment at the Ouranos Consortium. The authors also wish to thank John Scinocca of the Canadian Centre for Climate Modelling and Analysis for constructive comments on the manuscript. This research was financially supported by the Ouranos Consortium.
REFERENCES
Adam, J. C., and D. P. Lettenmaier, 2003: Adjustment of global gridded precipitation for systematic bias. J. Geophys. Res., 108 .4257, doi:10.1029/2002JD002499.
Arnell, N. W., D. A. Hudson, and R. G. Jones, 2003: Climate change scenarios from a regional climate model: Estimating change in runoff in southern Africa. J. Geophys. Res., 108 .4519, doi:10.1029/2002JD002782.
Bechtold, P., E. Bazile, F. Guichard, P. Mascart, and E. Richard, 2001: A mass flux convection scheme for regional and global models. Quart. J. Roy. Meteor. Soc., 127 , 869–886.
Boer, G. J., N. A. McFarlane, R. Laprise, J. D. Henderson, and J-P. Blanchet, 1984: The Canadian Climate Centre Spectral atmospheric general circulation model. Atmos.–Ocean, 22 , 397–429.
Boer, G. J., N. A. McFarlane, and M. Lazare, 1992: Greenhouse gas–induced climate change simulated with the CCC second-generation general circulation model. J. Climate, 5 , 1045–1077.
Boer, G. J., G. Flato, C. M. Reader, and D. Ramsden, 2000: A transient climate change simulation with greenhouse gas and aerosol forcing: Experimental design and comparison with the instrument record for the twentieth century. Climate Dyn., 16 , 405–425.
Caya, D., and R. Laprise, 1999: A semi-implicit semi-Lagrangian regional climate model: The Canadian RCM. Mon. Wea. Rev., 127 , 341–362.
Caya, D., and S. Biner, 2004: Internal variability of RCM simulations over an annual cycle. Climate Dyn., 22 , 33–46.
Christensen, J. H., J. Räisänen, T. Iverson, D. Bjorge, O. B. Christensen, and M. Rummukainen, 2001: A synthesis of regional climate change simulations—A Scandinavian perspective. Geophys. Res. Lett., 28 , 1003–1006.
Cubasch, U., H. von Storch, J. Waszkewitz, and E. Zorita, 1996: Estimates of climate change in southern Europe derived from dynamical climate model output. Climate Res., 7 , 129–149.
Davies, H. C., 1976: A lateral boundary formulation for multi-level prediction models. Quart. J. Roy. Meteor. Soc., 102 , 405–418.
Delworth, T., and S. Manabe, 1989: The influence of soil wetness on near-surface atmospheric variability. J. Climate, 2 , 1447–1462.
Flato, G. M., and W. D. Hibler III, 1992: Modeling pack ice as a cavitating fluid. J. Phys. Oceanogr., 22 , 626–651.
Flato, G. M., and G. J. Boer, 2001: Warming asymmetry in climate change simulations. Geophys. Res. Lett., 28 , 195–198.
Flato, G. M., G. J. Boer, W. G. Lee, N. A. McFarlane, D. Ramsden, M. C. Reader, and A. J. Weaver, 2000: The Canadian Centre for Climate Modelling and Analysis Global Coupled Model and its climate. Climate Dyn., 16 , 451–467.
Frigon, A., D. Caya, M. Slivitzky, and D. Tremblay, 2002: Investigation of the hydrologic cycle simulated by the Canadian Regional Climate Model over Québec/Labrador territory. Climatic Change: Implications for the Hydrological Cycle and for Water Management, M. Beniston, Ed., Advances in Global Change Research, Vol. 10, Kluwer, 31–55.
Gal-Chen, T., and R. C. Sommerville, 1975: On the use of a coordinate transformation for the solution of Navier-Stokes. J. Comput. Phys., 17 , 209–228.
Gates, W. L., and Coauthors, 1999: An overview of the results of the Atmospheric Model Intercomparison Project. Bull. Amer. Meteor. Soc., 80 , 29–55.
Gent, P. R., and J. C. McWilliams, 1990: Isopycnal mixing in ocean circulation models. J. Phys. Oceanogr., 20 , 150–155.
Giorgi, F., C. Shields, and G. T. Bates, 1994: Regional climate change scenario over the United States produced with a nested regional climate model. J. Climate, 7 , 375–399.
Giorgi, F., L. O. Mearns, C. Shields, and L. McDaniel, 1998: Regional nested model simulations of present-day and 2xCO2 climate over the central plains of the USA. Climate Change, 40 , 457–493.
Giorgi, F., X. Bi, and J. S. Pal, 2004: Mean, interannual variability and trends in a regional climate change experiment over Europe. 1. Present-day climate (1961–1990). Climate Dyn., 22 , 733–756.
Goyette, S., N. A. McFarlane, and G. M. Flato, 2000: Application of the Canadian Regional Climate Model to the Laurentian Great Lakes region: Implementation of a lake model. Atmos.–Ocean, 38 , 481–503.
Houghton, J. T., B. A. Callander, and S. K. Varney, 1992: Climate Change 1992: The Supplementary Report to the IPCC Scientific Assessment. Cambridge University Press, 205 pp.
Houghton, J. T., Y. Ding, D. J. Griggs, M. Noguer, P. J. van der Linden, X. Dai, K. Maskell, and C. A. Johnsons, 2001: Climate Change 2001: The Scientific Basis. Cambridge University Press, 881 pp.
Jiao, Y., and D. Caya, 2006: An investigation of summer precipitation simulated by the Canadian Regional Climate Model. Mon. Wea. Rev., 134 , 919–932.
Joubert, A. M., J. J. Katzfey, J. L. McGregor, and K. C. Nguyen, 1999: Simulating midsummer climate over southern Africa using a nested regional climate model. J. Geophys. Res., 104 , 19015–19025.
Kain, J. S., and J. M. Fritsch, 1990: A one-dimensional entraining/detraining plume model and application in convective parameterization. J. Atmos. Sci., 47 , 2784–2802.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437–471.
Karl, T. R., W-C. Wang, M. E. Schlesinger, R. W. Knight, and D. Portman, 1990: A method of relating general circulation simulated climate to the observed local climate. Part I: Seasonal statistics. J. Climate, 3 , 1053–1079.
Kato, H., H. Hirakuchi, K. Nishizawa, and F. Giorgi, 1999: Performance of the NCAR RegCM in the simulations of June and January climates over eastern Asia and the high-resolution effect of the model. J. Geophys. Res., 104 , 6455–6476.
Langner, J., and H. Rodhe, 1991: A global three-dimensional model of the tropospheric sulphur cycle. J. Atmos. Chem., 13 , 225–263.
Laprise, R., D. Caya, G. Bergeron, and M. Giguère, 1997: The formulation of André Robert MC2 (Mesoscale Compressible Community) model. Atmos.-Ocean, 35 .(Suppl.), 195–220.
Laprise, R., D. Caya, M. Giguère, G. Bergeron, H. Cote, J-P. Blanchet, G. J. Boer, and N. A. McFarlane, 1998: Climate and climate change in western Canada as simulated by the Canadian Regional Climate Model. Atmos.–Ocean, 36 , 119–167.
Laprise, R., D. Caya, A. Frigon, and D. Paquin, 2003: Current and perturbed climate as simulated by the second-generation Canadian Regional Climate Model (CRCM-II) over northwestern North America. Climate Dyn., 21 , 405–421.
Legates, D. R., and C. J. Willmott, 1990: Mean seasonal and spatial variability in gauge-corrected, global precipitation. Int. J. Climatol., 10 , 111–127.
Liang, X-Z., L. Li, K. Kunkel, M. Ting, and J. X. L. Wang, 2004: Regional climate simulations of U.S. precipitation during 1982–2002. Part I: Annual cycle. J. Climate, 17 , 3510–3529.
Machenhauer, B., M. Windelband, M. Botzet, J. Hesselbjerg, M. Déqué, G. R. Jones, P. M. Ruti, and G. Visconti, 1998: Validation and analysis of regional present-day climate and climate change simulations over Europe. MPI Rep. 275, Max Planck Institute for Meteorology, Hamburg, Germany, 87 pp.
McFarlane, N. A., G. J. Boer, J-P. Blanchet, and M. Lazare, 1992: The Canadian Climate Centre second-generation general circulation model and its equilibrium climate. J. Climate, 5 , 1013–1044.
Mitchell, J. F. B., T. C. Johns, J. M. Gregory, and S. F. B. Tell, 1995: Climate response to increasing levels of greenhouse gases and sulfate aerosols. Nature, 376 , 501–504.
Mitchell, T. D., and P. D. Jones, 2005: An improved method of constructing a database of monthly climate observations and associated high-resolution grids. Int. J. Climatol., 25 , 693–712.
New, M. G., M. Hulme, and P. D. Jones, 1999: Representing twentieth-century space–time climate variability. Part I: Development of a 1961–1990 mean monthly terrestrial climatology. J. Climate, 12 , 829–856.
Pacanowski, R. C., K. Dixon, and A. Rosati, 1993: The GFDL modular ocean model user’s guide. GFDL Ocean Group Tech. Rep. 2, Geophysical Fluid Dynamics Laboratory, Princeton, NJ, 46 pp.
Pan, Z., J. H. Christensen, R. W. Arritt, W. J. Gutowski Jr., E. S. Takle, and F. Otieno, 2001: Evaluation of uncertainties in regional climate change simulations. J. Geophys. Res., 106 , 17735–17751.
Puckrin, E., W. F. J. Evans, J. Li, and H. Lavoie, 2004: Comparison of clear-sky surface radiative fluxes simulated with radiative transfer models. Can. J. Remote Sens., 30 , 903–912.
Räisänen, J., and Coauthors, 2004: European climate in the late twenty-first century: Regional simulations with two driving global models and two forcing scenarios. Climate Dyn., 22 , 13–31.
Reader, C. M., and G. J. Boer, 1998: The modification of greenhouse gas warming by the direct effect of sulphate aerosols. Climate Dyn., 14 , 593–608.
Riette, S., and D. Caya, 2002: Sensitivity of short simulations to the various parameters in the new CRCM spectral nudging. Research Activities in Atmospheric and Oceanic Modelling, H. Ritchie, Ed., WMO/TD-No. 1105, Rep. 32, 7.39–7.40.
Verseghy, D. L., 1996: Local climates simulated by two generations of Canadian GCM land surface schemes. Atmos.–Ocean, 34 , 435–456.
Verseghy, D. L., N. A. McFarlane, and M. Lazare, 1993: A Canadian Land Surface Scheme for GCMs: II. Vegetation model and coupled runs. Int. J. Climatol., 13 , 347–370.
von Storch, H., H. Langenberg, and F. Feser, 2000: A spectral nudging technique for dynamical downscaling purposes. Mon. Wea. Rev., 128 , 3664–3673.
Whetton, P. H., J. J. Katzfey, K. J. Hennessy, X. Wu, J. L. McGregor, and K. Nguyen, 2001: Developing scenarios of climate change for Southeastern Australia: An example using regional climate model output. Climate Res., 6 , 181–201.
The geographic extent of the Pan-Canadian (PC) and Amerique du Nord (AN) domains used for the CRCM experiments discussed here. Also shown is the topography used in the CRCM.
Citation: Journal of Climate 19, 13; 10.1175/JCLI3769.1
(a), (c) The DJF and JJA seasonal average geopotential height (dm) on the 500-hPa level from the NCEP–NCAR reanalysis and (b), (d) the difference between the CRCM fields and the NCEP–NCAR reanalysis. All panels represent a 20-yr climatology for the years 1971–90 with the CRCM results taken from the ANpr1 simulation.
Citation: Journal of Climate 19, 13; 10.1175/JCLI3769.1
(a), (c) The DJF and JJA seasonal average screen temperature from the CRU observations and (b), (d) the differences between the CRCM ANpr1 simulation and the CRU observations. All quantities are averaged over the period 1971–90. Note the change in contouring of the plots between the DFJ and JJA seasons.
Citation: Journal of Climate 19, 13; 10.1175/JCLI3769.1
(a), (c) The DJF and JJA seasonal average precipitation (mm day−1) from the CRU gridded observational dataset. (b), (d) Difference between the CRCM-simulated (ANpr1) seasonal precipitation and the CRU observations, expressed as a percentage of the CRU-observed amounts. All quantities are for the 1971–90 period.
Citation: Journal of Climate 19, 13; 10.1175/JCLI3769.1
The 11 climatic regions used in the analysis. The abbreviations are as used in the following figures and text and correspond to the following regions, WCoast: West Coast, WCrdra: Western Cordillera, NPlns: Northern Plains, GrtLks: Great Lakes, M-Atl: Mid-Atlantic, Mrtms: Canadian Maritimes, NEFor: North East Forest, NWFor: North West Forest, EArctc: east Arctic, Mack: Mackenzie Valley, and Yukon: Yukon Territory.
Citation: Journal of Climate 19, 13; 10.1175/JCLI3769.1
The seasonal average surface air temperature by region taken from the CRU climatology and five different CRCM simulations. All seasonal averages are computed over the period 1971–90. Details of the different CRCM simulations are given in Table 1 and the geographic extent of each region is displayed in Fig. 5.
Citation: Journal of Climate 19, 13; 10.1175/JCLI3769.1
A comparison of the bias in the regional seasonally averaged surface air temperature for the different CRCM simulations. Details of the individual model simulations and regions are as given in Fig. 6.
Citation: Journal of Climate 19, 13; 10.1175/JCLI3769.1
A comparison of the regional seasonally averaged precipitation from the CRU observations and five CRCM simulations; details as in Fig. 6.
Citation: Journal of Climate 19, 13; 10.1175/JCLI3769.1
The interannual standard deviation of the regional seasonally averaged surface air temperature from the CRU observations and four CRCM simulations; details as in Fig. 6.
Citation: Journal of Climate 19, 13; 10.1175/JCLI3769.1
The interannual coefficient of variation of the seasonal precipitation, defined as the interannual standard deviation of the seasonal average normalized by the average, data taken from the CRU observations and four CRCM simulations for present-climate conditions; details as in Fig. 6.
Citation: Journal of Climate 19, 13; 10.1175/JCLI3769.1
The change in regional seasonally averaged surface air temperature and precipitation between the two CRCM simulations with different sets of physical parameterizations. The changes found for the NCEP–NCAR reanalysis–driven simulations are plotted as a function of the corresponding changes found from the GCM-driven simulations.
Citation: Journal of Climate 19, 13; 10.1175/JCLI3769.1
The projected change in seasonal average temperature between the periods 2041–60 and 1971–90. The climate change signal has been defined using four different pairings of CRCM simulations, where the definition of these pairs are given in Table 2. Also shown are the changes from the two IS92A scenario CGCM simulations. Data from CGCMa-92a was used to drive the PC92a-1 simulation and data from CGCMb-92a to drive the AN92a-1 simulation; the ANa2–1 and ANa2–2 simulations were driven by the CGCM-a2 simulation.
Citation: Journal of Climate 19, 13; 10.1175/JCLI3769.1
The change in seasonal average precipitation, between the periods 2041–60 and 1971–90, projected by several different CRCM simulations. The change in precipitation is presented as a fractional change in the current-climate seasonal average. The set of runs is as described in Fig. 12.
Citation: Journal of Climate 19, 13; 10.1175/JCLI3769.1
The average change in seasonal screen temperature projected by the four pairs of CRCM simulations for (a) DJF and (c) JJA. Also shown is the standard deviation of the four climate change projections for the (b) DJF and (d) JJA screen temperature.
Citation: Journal of Climate 19, 13; 10.1175/JCLI3769.1
The average fractional change in seasonal precipitation projected by the four pairs of CRCM simulations for (a) DJF and (c) JJA. Also shown is the standard deviation of the fractional change for the four climate change projections for the (b) DJF and (d) JJA seasonal precipitation.
Citation: Journal of Climate 19, 13; 10.1175/JCLI3769.1
A summary of the CRCM simulations. The simulation names given to the present-climate simulations correspond with the names used in the subsequent figures.
The definition of the different climate change simulations used; names given to the climate change simulations correspond with the names used in the figures.