An Investigation of Summer Precipitation Simulated by the Canadian Regional Climate Model

MARCH 2006 J I A O A N D C A Y A 919 An Investigation of Summer Precipitation Simulated by the Canadian Regional Climate Model YANJUN JIAO Department of Earth and Atmospheric Sciences, Université du Québec à Montréal, Montréal, Québec, Canada DANIEL CAYA Ouranos Consortium, Montréal, Québec, Canada (Manuscript received 2 March 2005, in final form 20 July 2005) ABSTRACT In the present paper, a 5-yr baseline integration for the period 1987 91 was carried out over a Pan- Canadian domain to validate the performance of the third-generation Canadian Regional Climate Model (CRCM). The CRCM simulated the large-scale circulation over North America well; it also correctly captured the seasonal variability of surface temperature and reproduced the winter precipitation over North America realistically. However, the CRCM systematically overestimated the summer precipitation over the continent when compared with the observed values. Extensive experiments have been conducted to trace down the sources of error of summer precipitation. Particular attention has been given to the water-vapor-related physical parameterization processes such as the mass flux convection scheme in the CRCM. Experiments involving spectral nudging of the specific humidity toward the values of large-scale driving data enabled the authors to link overestimation with abundant water vapor accumulated in the lower boundary layer resulting from an excessive amount of moisture stored in the soil. A strong boundary layer mixing process from the third generation of the Canadian Atmospheric General Circulation Model was then implemented into the CRCM along with an adjustment to the soil water holding capacity. A final analysis of a 14-month experiment showed that these modifications significantly improved the simulation of summer precipitation over North America without adversely affecting the simulation of winter precipitation. 1. Introduction Since the pioneering works of Dickinson et al. (1989) and Giorgi (1990), regional climate models (RCMs) have become one of the most popular tools to address regional climate impact studies. RCMs are able to provide valuable regional finescale information, especially in regions where the climate variables are strongly regulated by the underlying topography and the surface heterogeneity (e.g., Giorgi 1990; Jones et al. 1995; Laprise et al. 1998; and see Houghton et al. 2001 for a comprehensive review). In atmospheric models, precipitation is one of the Corresponding author address: Yanjun Jiao, Department of Earth and Atmospheric Sciences, Université du Québec à Montréal, 550 Sherbrooke West, 19th floor, West Tower, Montréal QC H3A 1B9, Canada. E-mail: jiao.yanjun@uqam.ca most critical climate variables and end product for climate studies. Some studies using RCMs have shown very skillful simulations, such as the precipitation over the northwest United States (Giorgi et al. 1993; Liang et al. 2004) where the orographic forcing dominates throughout the year. However, the simulation of precipitation is still the most problematic part for some current RCMs. For example, Giorgi et al. (1993) showed that the fourth-generation Pennsylvania State University National Center for Atmospheric Research (PSU NCAR) Mesoscale Model (MM4)-based Regional Climate Model (RegCM) tends to overestimate summer precipitation over the southern and central Rocky Mountains because of the frequent occurrence of strong gridpoint rain events over mountainous terrain. Hong and Leetmaa (1999) found that the National Centers for Environmental Prediction (NCEP) regional spectral model (RSM) had positive biases for both winter and summer precipitation over the United States, 2006 American Meteorological Society

920 M O N T H L Y W E A T H E R R E V I E W VOLUME 134 FIG. 1. Model domain and topographic field of the CRCM over the Pan-Canadian area. Contours at 200-m intervals. The dashed line represents the width of the buffer zone. The solid thick lines roughly outline the six climatic subregions considered in this work: WC west coast, PC prairie Canada, CC central Canada, NW northwest United States, GP Great Plains, and SE southeast United States. which is related to the deficiencies in the moist convective parameterization scheme used in RSM. The initial results from the Project to Intercompare Regional Climate Simulations (PIRCS) also revealed that most participant models showed an obvious positive bias in simulating 1988 extreme drought over the central United States (Takle et al. 1999) but underrepresented the extreme flooding over the same region in 1993, which suggests that the interannual variability of the precipitation might be damped in the RCMs (Anderson et al. 2003). More recently, Gutowski et al. (2004) reported that the RegCM2 tends to produce larger-thanobserved monthly average precipitation in the Mississippi Delta area. This problem has attracted the attention of the climate research community. For example, an international project, the North American Monsoon Experiment (NAME), has been setup to detect the sources and limitations of predictability of warm season precipitation over the North America monsoon region (http://www.joss.ucar.edu/name). Similar shortcomings have also been found in the Canadian Regional Climate Model (CRCM; Laprise et al. 1998; Caya and Laprise 1999). The first- and secondgeneration CRCMs were used to study current climate and climate change over western Canada (Laprise et al. 1998, 2003). These studies demonstrated that on the one hand, the CRCM had reasonable abilities to simulate seasonal mean and temporal variability of the large-scale circulation and had the potential to reproduce enhanced spatial variability in low-level fields. On the other hand, the error of overestimated simulated summer precipitation in western Canada was also noticed. The recently developed third-generation CRCM includes an updated preprocess package to deal with different sources of reanalysis data, so as to provide perfect boundary conditions to the CRCM. A state-ofthe-art spectral nudging technique had also been introduced. The observed monthly mean sea surface temperature (SST) and sea ice concentration from the Atmospheric Model Intercomparison Project (AMIP) are prescribed as the lower boundary conditions. In addition, a more sophisticated mass flux convective parameterization scheme (Bechtold et al. 2001) was implemented into this new version of the CRCM as an attempt to overcome the aforementioned overestimation problem in simulated summer precipitation. This new version of CRCM is intended for climate change studies; as such, its overall performance must be assessed. As suggested by the Intergovernmental Panel on Climate Change (IPCC; Houghton et al. 2001), in this study, we first evaluated the CRCM performance by conducting a 5-yr baseline simulation over a Pan- Canadian domain (Fig. 1) using NCEP NCAR reanalysis data (NNRA; Kalnay et al. 1996) as the proxy of the perfect initial and lateral boundary conditions. The seasonal mean climatologies simulated by the CRCM were then compared with available sources of observations, with particular attention being given to the model-simulated summer precipitation. As will be seen, the CRCM still showed positive biases in simulating summer precipitation despite updating the convective parameterization scheme, prescribing the observed monthly mean SST and sea ice concentrations, and using the so-called perfect lateral boundary conditions. Sensitivity experiments revealed that the major error was caused by excessive soil moisture and the deficiency of the turbulent mixing in the planetary boundary layer (PBL). To overcome these deficiencies, a simple but strong boundary layer mixing process was implemented into a modified version of the CRCM along with an adjustment to the soil water holding capacity, and an additional 14-month-long experiment was conducted to verify the impact of these modifications on the simulated summer precipitation. A brief description of the CRCM and experimental configuration is given in section 2. The improvements introduced in the CRCM are also discussed in this section. Section 3 briefly presents the validation of the 5-yr baseline simulation to identify the strengths and weaknesses of the current CRCM. Section 4 introduces the new boundary layer mixing processes, the adjustment to the soil water holding capacity, and the verification of the results. The paper concludes with the summary in the section 5.

MARCH 2006 J I A O A N D C A Y A 921 2. Model description and experimental configuration a. The third-generation CRCM The model employed in this study is the thirdgeneration CRCM. The CRCM is a fully elastic, nonhydrostatic model based on the three-dimensional Euler equations Mesoscale Compressible Community (MC2) model. The dynamical kernel of the CRCM is characterized by an advanced semi-implicit semi- Lagrangian (SISL) three-time-level differential scheme in solving the prognostic equations (Bergeron et al. 1994). The prognostic/diagnostic variables are distributed on an Arakawa C-type staggered grid with a uniform horizontal resolution in the polar-stereographic projection. In the vertical, the Gal-Chen terrainfollowing coordinate (Gal-Chen and Somerville 1975) is adopted. As with most of the current RCMs, the CRCM employs a one-way nesting technique to provide large-scale lateral meteorological boundary conditions. A state-of-the-art spectral nudging optionally controls the CRCM large-scale fields toward the values of the driving data in the interior of the model domain (Riette and Caya 2002). At the lateral boundaries, Davies nudging (Davies 1976) is generally applied on the wind field over a nine-gridpoint buffer zone. The CRCM shares most of its physical parameterization schemes with the second generation of the Canadian General Circulation Model (CGCM), and these are described in detail in McFarlane et al. (1992) and summarized in Caya and Laprise (1999). In this version of the CRCM, the preprocessing package was updated to allow the CRCM to be driven either by the CGCM output or by reanalysis data. Therefore, different sources of reanalysis, such as the NNRA or the European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis (ERA) data, can be used as the proxies of observations to verify the model performance. Moreover, to prevent simulated largescale circulation from drifting too far away from the driving conditions when a large domain is used (as in this study), spectral nudging has been applied on the horizontal wind fields. The wavelengths to be nudged (greater than 1400 km) are selected by a spectral filter based on the discrete cosine transform function (Denis et al. 2002). Efforts were also undertaken to improve the treatment of the time-dependent SST and sea ice fields in the CRCM. In the earlier versions of the CRCM, the monthly mean climatologies of SST and sea ice concentrations from the CGCM output were linearly interpolated to obtain time-dependent sea surface conditions, which actually impeded interannual variations and finescale structures of the lower boundary forcing. The new CRCM overcomes this deficiency by introducing observed monthly mean SST and sea ice concentrations from AMIP. Moreover, to preserve the original monthly mean values and interannual variabilities in the interpolated SST and sea ice time series, the midmonth values of observed SST and sea ice concentrations are first computed following Sheng and Zwiers (1998) and then linearly interpolated to each of the CRCM s time step. The Bechtold Kain Fritsch mass flux scheme (Bechtold et al. 2001), which is a modified version of the Kain Fritsch scheme (Kain and Fritsch 1990), was implemented in the CRCM (Paquin and Caya 2000). Although there are some substantial differences with the Kain Fritsch scheme in moist thermodynamics, in constructing entrainment rates and in computing the updraft, the Bechtold Kain Fritsch scheme can be summarized into three major components as most mass flux parameterization schemes used: the trigger function, the mass flux formulation, and the closure assumption. The convection is triggered by an at least 60-hPa deep mixed layer, raised to its lifting condensation level without any entrainment, and is capable of producing a 3000-m thick cloud. The convective updraft is parameterized by a one-dimensional entrainment/detrainment plume-cloud model based on the conservation of enthalpy and total water mixing ratio (see Fritsch and Chappell 1980 for a detailed description). The downdraft is driven by cooling through melting and evaporation of precipitation, and is computed using the equivalent potential temperature. Finally, a closure assumption is used to control the intensity of the convection, which assumes that at least 90% of the convective available potential energy (CAPE) is removed by the overturning mass in updraft and downdraft, and mixing with the surrounding environment within an adjustment period (30 60 min for deep convection and 3 h for the shallow convection). The shallow convection shares the same parameterization strategy as the deep convection in the Bechtold Kain Fritsch scheme except that the cloud radius is set to 50 m and the minimum cloud depth is 500 m, rather than 1500-m radius and 3000-m minimum cloud depth for the deep convection. Furthermore, there is no downdraft, and thus no precipitation generated in the shallow convection. b. Experimental configuration The model domain, as shown in Fig. 1, comprises the whole Canadian territory, most of the United States

922 M O N T H L Y W E A T H E R R E V I E W VOLUME 134 and Greenland, northeastern Pacific Ocean, northwestern Atlantic Ocean, and Arctic Ocean. There are 193 145 polar stereographic grid points with a horizontal resolution of 45 km (true at 60 N). The 8600 6500 km 2 domain places most of western, eastern, and northern boundaries over the Pacific Ocean, Atlantic Ocean, and Arctic Ocean, respectively, and the southern boundary is placed over the southernmost part of the complex topography to avoid the effects of unrealistic surface energy budget calculations near the boundaries (Giorgi and Mearns 1999). Meanwhile, the model domain puts our area of interest, the Canadian territory, right in the center of the domain with large enough surrounding spaces to study the sensitivity of the internal forcings (Seth and Giorgi 1998); the 45-km horizontal resolution is fine enough to adequately capture the underlying surface forcing, such as Coastal Ranges, Cascade Mountains, and Sierra Nevada in the western coast and the Appalachian Mountains in the eastern part of continent. There are 29 uneven Gal-Chen levels capped at about 29 km in the vertical (see Table 1 for the complete list of the CRCM levels). To allow the planetary boundary layer and lower troposphere to be well resolved, most of levels are assigned in the lower troposphere. Both the initial and lateral boundary conditions stem from NNRA data. The simulation was initiated on 1 January 1987 and was run continuously for 5 yr, to 31 December 1991, with a 15-min time step. The 6-hourly lateral boundary conditions (large-scale horizontal wind, temperature, specific humidity, and surface pressure) were linearly interpolated to every time step. The large-scale horizontal wind of the driving data was blended with the CRCM over a nine-gridpoint sponge zone. The SST and sea ice concentrations were prescribed from the monthly mean observations of the AMIP. The soil moisture was initialized by the monthly mean climatology of the CGCM output. 3. The simulated seasonal mean climatologies TABLE 1. List of 29 thermodynamic levels used in terrainfollowing Gal-Chen coordinate of the CRCM. The notation is based on the reference atmosphere with flat surface (P S 1000 hpa, Z S 0 m) at level 1 2 and top at level 29 1 2 (P T 5 hpa, Z T 29 042 m). Level index Pressure (hpa) Height (m) 29 18 25 869 28 28 23 024 27 43 20 262 26 66 17 503 25 101 14 763 24 146 12 390 23 200 10 364 22 261 8650 21 328 7178 20 398 5933 19 466 4917 18 531 4076 17 592 3376 16 648 2794 15 698 2315 14 742 1922 13 780 1600 12 813 1333 11 841 1115 10 865 934 9 885 787 8 902 664 7 916 565 6 930 467 5 944 371 4 958 276 3 972 183 2 986 91 1 996 26 In this section, the seasonal mean climatologies based on the 4 yr of the simulation (1988 91) are analyzed and compared to observations or reanalysis data. Figure 2 presents the CRCM simulated 500-hPa seasonal mean geopotential height fields and the difference with the NNRA data for winter (DJF) and summer (JJA). As can be seen, the large-scale circulations simulated by the CRCM are quite close to NNRA for both seasons. The CRCM simulation is only about 1 (2) dam higher than that of the NNRA over the eastern part of the continent, 3 (4) dam lower over the southwest United States in winter (summer). The CRCM also successfully reproduced the basic distribution patterns of the screen-level surface temperature and its seasonal variation over the North American continent. Figure 3 compares the simulated monthly mean screenlevel surface temperature with the Climate Research Unit TS 2.02 observation (hereafter referred as to CRU; Mitchell and Philip 2005), which is a 0.5 0.5 longitude latitude high-resolution monthly climate data constructed from meteorological stations. Basically, the CRCM simulation successfully captured the seasonal variation over North American continent with some negative biases in winter and spring but with some minor positive biases in summer and fall. The seasonal mean biases averaged over all land grid points are only 2 C colder in winter and 1 C warmer in summer (Table 2). Figure 4 presents the seasonal mean precipitation simulated by the CRCM and the CRU observation, which was interpolated from 0.5 0.5 latitude

MARCH 2006 J I A O A N D C A Y A 923 1mmday 1 in summer (Fig. 4d), but the CRCM simulation showed more than 3 mm day 1. Figure 5 compares the simulated mean annual cycle for precipitation with the CRU observation over six subregions, which include west coast (WC), prairie Canada (PC), central Canada (CC), northwest United States (NW), Great Plains (GP), and southeast United States (SE) (see boundaries for the subregions in Fig. 1). Even though the model replicated the basic annual variability of the precipitation, the CRCM overestimated summer precipitation in all six subregions. Table 3 presents seasonal mean precipitation biases and relative biases of simulated precipitation. Clearly, the CRCM overestimated the seasonal precipitation over all six subregions for all seasons except the winter and fall precipitation over central Canada and the southeast United States. The largest bias appears in the simulation of summer precipitation over the Great Plains, where the bias reached 3.32 mm day 1, while the largest relative bias (about 187%) was found in the dry summer over the northwest United States, where the observed seasonal mean precipitation was approximately 1 mm day 1. Comparatively, the most reliable simulation was found in central Canada, where the model underestimated only about 4% and 12% of winter and fall precipitation, respectively, and overestimated about 40% of the summer precipitation. FIG. 2. Seasonal mean 500-hPa geopotential heights simulated by the CRCM (solid lines at 4-dam intervals) and the differences with the NCEP NCAR reanalysis (dashed lines at 1-dam intervals) for (a) winter [December January (DJF)] and (b) summer [June August (JJA)] for the period of 1988 91. The lower right corner of the label boxes indicates the isoline to which it is attached. longitude grid to the CRCM polar stereographic grid for the purpose of comparison. The CRCM realistically reproduced the winter precipitation patterns over the North American continent (Fig. 4a); the maximum precipitation was correctly located along the western coast of the British Colombia and the south shore of Alaska, which was in good agreement with the CRU observations (Fig. 4c). The model also reproduced the dry conditions in the central continent and captured the precipitation maximum over the southeastern United States, with only a slight deficit in the rainfall amount. Unfortunately, the CRCM systematically overestimated the summer precipitation over the whole continent (Fig. 4b). The model was unable to capture the dry situation in southwestern United States (over California, Nevada, Utah, and Idaho), where the CRU observation generally suggested precipitation less than 4. Sensitivity experiments and modification to the CRCM parameterizations Based on the results presented previously, this section reports on the sensitivity experiments conducted to diagnose problems in the simulated summer precipitation and the modifications to the CRCM. The verification of the modified CRCM is also presented. a. Sensitivity experiments Generally speaking, overestimation problems related to simulated summer precipitation can easily be attributed to the convective parameterization scheme used in the model, simply because of the convective nature of the summer precipitation (e.g., Hong and Leetmaa 1999). In the CRCM, about 60% (in some specific areas, the proportion is more than 90%) of the summer precipitation is generated by convection (Fig. 6), which is parameterized by the Bechtold Kain Fristch mass flux scheme. This suggests that the Bechtold Kain Fristch scheme is perhaps responsible for the overprediction of summer precipitation. We conducted several sensitivity experiments with special attention focused

924 M O N T H L Y W E A T H E R R E V I E W VOLUME 134 FIG. 3. Monthly mean surface temperature simulated by the CRCM (solid line) and the CRU observation (dashed line) averaged over all land grid points for the period of January 1988 to December 1991. on the adjustable empirical parameters in the Bechtold Kain Fristch scheme, such as the reference cloud radius, the minimum required cloud depth and the minimum mixed layer depth to sustain convection, etc. All sensitivity experiments were conducted for one month only (branched off from 5-yr baseline integration and restarted on 1 July) since the simulation of summer precipitation was our main concern. Detailed results of each sensitivity experiment are not presented because our experiments did not find any adjustable parameters sensitive enough to lower the spurious land convective precipitation to a reasonable level, even when several parameters were adjusted simultaneously. The adjustable parameters in the Bechtold Kain Fristch scheme only changed the intensity and the location of some individual convective events in the daily precipitation variability, but had very limited impact on the amount and overall pattern of the monthly mean precipitation. The sensitivity experiments did not effectively correct the overprediction problem in the CRCM; some physical parameterizations other than the Bechtold Kain Fristch convective scheme were investigated. Figure 7 compares the model-simulated domain-averaged vertical profile of the monthly mean specific humidity with the driving NNRA and ERA-40 data. For comparison purposes, both NNRA and ERA-40 data were interpolated from 2.5 latitude longitude grid to the CRCM s polar stereographic grids in the horizontal direction, and in the vertical, from 17 and 23 standard pressure levels to the 29 Gal-Chen levels. Although there were minor differences between the NNRA and the ERA-40 (generally less than 1 gkg 1 ), the basic shapes of the two analyses were quite similar. However, large differences between the CRCM simulation and the reanalysis data were present from the surface to the lower troposphere. The CRCM retained too much moisture in the near-surface boundary layer below level 8 (below 660 m; see Table 1) but showed an obvious deficit of moisture in the lower troposphere between levels 8 and 18 (between 660 and 4000 m). The largest difference appeared in the lowest model layer, where the specific humidity in the CRCM reached 13.6 g kg 1, whereas the corresponding NNRA and ERA-40 were 8.9 and 8.6 g kg 1, respectively. Can this error in the moisture field be linked to the positive biases in simulated summer precipitation? A possible and reasonable conjecture is the following: the water vapor accumulated in the lower boundary layers provides a potential moisture source and favorable conditions for convection; once the convection is triggered by the daytime surface heating, the abundant moisture in the near surface layer will be transported to the upper layer and then condense and precipitate. To confirm this hypothesis, taking advantage of the spectral nudging, we carried out another one-month-long test with the specific humidity on all CRCM interior grid points spectrally forced (the adjustable spectral nudging weight set to 1.0) to the NNRA values at every time step. The strong nudging kept the large-scale distribution of the model moisture exactly the same as the NNRA driving data. Using this method, the model performance (especially the response of the Bechtold Kain Fristch scheme) was validated under a welldefined large-scale humidity environment. CAPE objectively determines how convective the atmosphere is. Kain and Fritsch (1990) demonstrated a large sensitivity of parameterized mass flux (thus the convective precipitation rate) to the CAPE. Figure 8 compares the simulated monthly mean CAPE before TABLE 2. Seasonal mean surface temperature bias ( C) compared with the CRU observation. Winter Spring Summer Fall 1988 2.11 2.36 0.57 0.28 1989 1.93 1.48 1.03 0.29 1990 2.21 1.64 1.04 0.19 1991 1.64 1.59 0.94 0.58 Mean 1.97 1.77 0.90 0.34

MARCH 2006 J I A O A N D C A Y A 925 FIG. 4. (top) Simulated and (bottom) CRU observation of the seasonal mean precipitation for (left) winter (DJF) and (right) summer (JJA) for 1988 91. Contour intervals are 2.0 mm day 1 ; rectangular boxes with numbers indicate the contour values. and after the specific humidity was nudged. The monthly mean CAPE was reduced significantly, which implies that the simulated convective precipitation decreased over the whole continent (not shown). Correspondingly, the overall patterns of the simulated total precipitation over the North American continent (Fig. 9b) was improved substantially toward the CRU observations (Fig. 10d). In particular, the excessive precipitation over the southwestern United States was strongly reduced and in some areas disappeared altogether. The results imply that the vertical distribution of moisture in the atmospheric column strongly influences the convective activities and precipitation amounts, and the Bechtold Kain Fristch mass flux scheme could work well, provided the large-scale moisture environment is accurate. Based on these findings, in the next section, we present some efforts made to correct the error in the moisture field by introducing new boundary layer mixing processes and adjusting the soil water holding capacity. b. Adjustment of the soil water holding capacity The feedback processes between the soil and the atmosphere are of importance in the hydrological cycle of the climate system. The soil water holding capacity, which affects evaporation and release of latent heat from the underlying surface to the atmosphere plays a critical role in the water and energy balance of the land surface (Milly and Dunne 1994; Betts et al. 1996). For example, as shown by an earlier study (Vidale et al. 2003), the underestimation of the soil water flux resulted in a significant dryness in the central European regions. There is also a positive feedback between soil moisture and the deep convection (Clark and Arritt 1995; Seth and Giorgi 1998). In the CRCM, soil moisture parameterization is adopted from the second-generation CGCM and is represented in terms of a beautified bucket model as described in McFarlane et al. (1992). This simple land surface scheme crudely accounts for the effect of the vegetative evaporation by setting the maximum soil water holding capacity as a function of the porosity and depth of the soil; both of them are weighted functions of the primary and secondary vegetations. For the purpose of accounting for the transpiration effect of vegetation canopy from the deep root zones, the maximum water holding capacity had been assigned to a large value (the maximum value in some tropical area could reach 100 cm) in the original scheme, which has been

926 M O N T H L Y W E A T H E R R E V I E W VOLUME 134 FIG. 5. Monthly mean precipitation (mm day 1 ) variations averaged over six subregions for the CRCM simulations (solid lines with circles) and the CRU observation (dashed lines with squares) for the period 1988 91. proven to be a reasonable setting in the CGCM without the occurrence of significant overestimation of evaporation and precipitation. Contrary to this, our 5-yr baseline experiment revealed that in the CRCM, this large water holding capacity setting provided favorable conditions for evaporation; the excessive evaporation generated abundant precipitation over land, which in turn replenished the soil before any surface runoff could occur. To overcome the excessive evaporation from the soil, in the modified CRCM, we set the maximum water holding capacity back to 15 cm, which is the conventional value (Manabe 1969) and has been widely used in many bucket models. c. New boundary layer mixing scheme In the CRCM, as shown in Fig. 7, a large amount of moisture remains in the near-surface layer without effectively mixing with the upper atmosphere. The description of the original vertical mixing procedure used in the CRCM can be found in McFarlane et al. (1992). The vertical turbulent fluxes in the free atmosphere were parameterized using a stability-dependent eddy diffusivity formulation. The surface fluxes were evaluated by the drag coefficients, which are the functions of the bulk Richardson number in the surface layer. As used in most of boundary layer parameterization schemes, the influence of the surface fluxes was imposed only on the lowermost model layer. In the modified CRCM, we implemented a stronger mixing process, which has already been used in the third-generation CGCM. Instead of exchanging latent and sensible heat fluxes only with the lowest model layer, the new mixing process has an effect on the whole boundary layer so as to mimic a well-mixed vertical distribution of water vapor and potential temperature in the boundary layer.

MARCH 2006 J I A O A N D C A Y A 927 TABLE 3. Seasonal mean CRU precipitation (mm day 1 ), model-simulated precipitation bias (mm day 1 ), and relative bias (%) averaged over six climatic subregions for the period 1988 91. Winter Spring Summer Fall West coast CRU mean 3.99 2.15 2.53 4.22 Bias 0.98 1.07 2.05 1.03 Relative bias 24.56 49.65 80.95 24.33 Prairie Canada CRU mean 0.63 1.07 2.07 1.11 Bias 0.19 0.46 2.43 0.51 Relative bias 30.60 42.92 117.45 45.96 Central Canada CRU mean 1.90 2.14 2.82 3.18 Bias 0.07 0.31 1.08 0.37 Relative bias 3.48 14.46 38.31 11.77 Northwest United States CRU mean 3.40 2.64 1.03 2.69 Bias 0.55 1.28 1.93 0.46 Relative bias 16.28 48.50 187.28 17.13 Great Plains CRU mean 0.41 1.81 2.22 1.09 Bias 0.44 1.32 3.32 1.24 Relative bias 106.52 72.72 150.05 112.93 Southeast United States CRU mean 3.59 3.68 4.35 3.02 Bias 1.28 0.36 2.49 0.42 Relative bias 35.72 9.71 57.36 14.00 As in the original CRCM, the sensible heat (SH) and latent heat (LH) fluxes are computed as SH L 1 2 C p C H V L T L 1 2 T g L 1 2 LH L 1 2 e L C H V L q L 1 2 q g. Here R/C p with R the gas constant for dry air; C p is the specific heat at constant pressure; C H is the drag coefficient; V L represents the wind in the lowest momentum layer; (, T, q, and ) L (1/2) represent the air density, air temperature, specific humidity, and the thickness of the lowest thermodynamic layer, respectively; T g and q g indicate the ground temperature and soil moisture; L is the latent heat; and e is the potential evapotranspiration factor. Starting from the lowest model layer, a new specific humidity q PBL and potential temperature PBL for the whole PBL can be derived by evenly adding surface fluxes to the entire PBL as follows: FIG. 6. Fraction of convective (black) and stratiform (hatched) precipitation averaged over all land grid points. FIG. 7. Vertical profile of the monthly mean specific humidity averaged over all land grid points for July 1991.

928 M O N T H L Y W E A T H E R R E V I E W VOLUME 134 FIG. 9. Same as Fig. 8 but for total precipitation. Contour intervals are 2.0 mm day 1. FIG. 8. Monthly mean CAPE simulated by the CRCM (a) without and (b) with specific humidity spectral nudging for July 1991. Contour intervals are 200 J kg 1. q PBL PBL old LH PBL q l 1 2 l 1 2 PBL old SH PBL l 1 2 l 1 2 PBL The corresponding virtual potential temperature( ) PBL in the PBL is then updated and compared with the old value at the current level. If this newly obtained virtual potential temperature ( ) PBL is greater than the old one ( ) old l (1/2) (implying a statically unstable profile), the specific humidity q PBL and potential temperature PBL are allowed to mix upward with an additional layer and the PBL top is raised accordingly. This procedure is repeated until the newly derived mixed virtual potential temperature becomes less than the old one at the top of the PBL. It is then assumed that the PBL has. reached its full extent and all levels in the PBL share the same values of newly derived specific humidity and potential temperature. In this way, the surface fluxes are imposed evenly in the whole PBL to assure a wellmixed profile in the PBL. d. Influence of the soil water holding capacity and boundary layer mixing process To test the impact of the adjustment in the soil water holding capacity and new boundary layer mixing scheme on the summer precipitation, we conducted a 14-month simulation using this modified CRCM. The experiment was initialized on 1 May 1991 and ended on 30 June 1992. Outputs from the first two months (May and June of 1991) were discarded as spinup since the modification in the soil water holding capacity needs time to reach its equilibrium. For the comparison purpose, we also extended our baseline experiment for another six months to June 1992, so that both the modified experiment and the baseline experiment covered the common period from July 1991 to June 1992. Comparison of the vertical profile of the humidity

MARCH 2006 J I A O A N D C A Y A 929 FIG. 10. Monthly mean precipitation simulated by the (top) modified CRCM and (bottom) CRU observation for (right) July and (left) December 1991. Contour intervals are 2.0 mm day 1. showed the effects of the modified boundary layer mixing scheme and the soil water holding capacity (Fig. 7). Evidently, these two modifications greatly improved the vertical distribution of moisture in the PBL. The profile simulated by the modified CRCM followed the NNRA and ERA reanalyses very well in most of the model layers, although there were still some minor but less than 0.5gkg 1 differences. Figure 10 shows the monthly mean precipitation simulated by the modified CRCM and the CRU observation for July and December 1991. Compared with the original simulation (Fig. 9a), the monthly mean precipitation simulated by the modified CRCM (Fig. 10b) has been reduced over most parts of the continent in July; the overall patterns are similar to the CRU observation (Fig. 10d). In particular, the modified CRCM correctly simulated the dry conditions over the western mountainous areas of the United States, where the simulated precipitation over California, Nevada, Idaho, and Washington states were less than 1 mm day 1 ; these were very close to the CRU observations. Figure 11a summarizes the summer precipitation amounts averaged over six subregions as simulated by the original CRCM, the modified CRCM, and the CRU observations. Clearly, the summer precipitation simulated by the modified CRCM was considerably reduced in most subareas. For example, the relative bias over the southeast United States was reduced from 57% in the original CRCM to about 30% in the modified CRCM. More significantly, great improvements appeared in prairie Canada, the northwest United States, and the Great Plains, where the relative biases were reduced from 139%, 179%, and 121% to 44%, 23%, and 22%, respectively. It should be pointed out that some problems still exist in the modified CRCM. For example, over the West Coast, where the relative biases, in the order of 60%, remained almost the same as in the original CRCM (Fig. 11a). Also, as can be seen from Fig. 10b, over some high-latitude regions, such as Alaska, Yukon Territory, northeastern Quebec, and Labrador, the summer precipitation simulated by the modified CRCM remained overestimated. Although we can partially attribute these errors to the coarser observational data over these regions, our further analysis indicates that these overestimations are still largely related to the excessive wetness of the soil over these regions during the summertime (not shown). This illustrates that simply adjusting the soil water holding capacity cannot solve all problems of the bucket model present in many

930 M O N T H L Y W E A T H E R R E V I E W VOLUME 134 North America continent without substantially deteriorating the simulation of winter precipitation. 5. Summary and conclusions FIG. 11. Seasonal mean precipitation (mm day 1 ) averaged over six subregions for (a) summer (July and August 1991, June 1992) and (b) winter (December 1991, January and February 1992). areas; an imminently more sophisticated biospheric land surface model is required in the CRCM. As expected, the already well-simulated winter precipitation was not affected by the modified CRCM as shown in Figs. 10a and 11b. The simulated maximum precipitation in December 1991 was correctly concentrated over the west coast of British Columbia, and agreed well with the CRU observation (Fig. 10c). The location of the rain belt in the southeastern United States also matched the observations very well, despite a slight deficit in rainfall amount. In addition, the overall patterns and the seasonal mean rainfall amounts over prairie Canada, central Canada, the northwest United States, and the Great Plains were in good agreement with the CRU observations (Fig. 11b). In summary, the improved CRCM with modified water holding capacity and modified boundary layer mixing schemes effectively improved the model performance in simulating summer precipitation over the We first presented the seasonal mean climatology simulated by the third-generation CRCM. The model used the basic physical parameterization package of the second-generation CGCM with the exception of the convective parameterization scheme, which had been updated using the Bechtold Kain Fritsch mass flux scheme. The simulation was performed at 45-km resolution over a Pan-Canadian area. The model was driven by the so-called perfect lateral boundary conditions from the NNRA data and was forced by observed monthly mean SST and sea ice concentration from AMIP. The simulation covered 5 yr, from January 1987 to December 1991. In general, the CRCM reproduced the basic patterns and the seasonal cycle of the large-scale circulation. The simulated surface temperature biases (the difference between the simulation and the CRU observation) were 2 C colder in winter and 1 C warmer in summer. The model reproduced the basic pattern of winter precipitation over the model domain; however, overestimation was present in simulating summer precipitation. The CRCM systematically overestimated the summer precipitation over the continent. In particular, the model generated too much precipitation over the southwestern United States where the observed summer precipitation suggested less than 1 mm day 1 but the model simulation was generally larger than 3 mm day 1. The problem prompted us to conduct sensitivity experiments to pinpoint the fundamental reason for this behavior in the CRCM. Sensitivity experiments were conducted to investigate the empirical parameters in the Bechtold Kain Fritsch mass flux scheme since most simulated summer precipitation originates from convection. However, our sensitivity experiments revealed that the scheme itself was not directly responsible for the overprediction of summer precipitation. An experiment with the specific humidity being nudged toward the values of the NNRA data further confirmed that the Bechtold Kain Fritsch mass flux scheme could perform well as long as the large-scale moisture field is accurate. Further investigation found that too much accumulated moisture in the PBL and too much moisture stored in the soil are directly responsible for the precipitation overestimation in the CRCM. A stronger boundary layer mixing process along with the adjustment to the soil water holding capacity was implemented into the modified CRCM. Instead of mix-

MARCH 2006 J I A O A N D C A Y A 931 ing surface fluxes only with the lowest model layer, the new mixing scheme evenly added fluxes to the whole boundary layer so as to mimic the vertical profiles of water vapor and potential temperature in a well-mixed PBL. These modifications were effective at improving the distribution of moisture in the PBL and consequently improved the summer precipitation simulations considerably. In addition, the quality of the simulated winter precipitation appeared to be unaffected by these modifications. The CRCM underlies the strategy of using the same physics schemes as its parent model, the CGCM, to achieve maximum compatibility between the nested and driving models. The results of this study illustrate that the physic schemes appropriated for the coarserresolution CGCM may not be adequate for the highresolution CRCM. All physical parameterization schemes suitable for the CGCM must be carefully calibrated before they are applied in the CRCM. This conclusion, we believe, is universally applicable to all RCMs underlying the same strategy as the CRCM (Giorgi and Mearns 1999). In the newly developed fourth-generation CRCM, a more advanced and more comprehensive biospheric land surface model, the Canadian Land Surface Scheme (CLASS), has replaced the original bucket model. Some preliminary simulations have shown more realistic evapotranspiration from the underlying surface, and thus more accurate simulated precipitation. Finally, it will be quite interesting to use different sources of driving data, for example, ERA-40, to detect the impact of the lateral boundary conditions on the performance of the CRCM. The perfect lateral boundary conditions constructed from the NNRA contain large uncertainties that result mainly from formulation deficiencies in the assimilation model system. These uncertainties have been shown to contribute significant errors to the RCM simulations (Liang et al. 2001, 2004) and may likely mask the identification of the CRCM formulation deficiencies as documented in this study. Sensitivity experiments using different reanalyses will address this issue. Tests with a domain expanding its south boundary equatorward so that the Gulf of Mexico is included will also be conducted. Acknowledgments. The first author is grateful to the Ouranos Climate Simulations team for their helpful support and to Ouranos for accessing its facilities. We thank Mr. Richard Harvey for providing the required portion of the GCM code and useful discussions, and Dr. Laxmi Sushama for reviewing the manuscript. We acknowledge Dr. Xin-Zhong Liang and another anonymous reviewer for their constructive comments on the manuscript. This research was funded by the Canadian Foundation for Climate and Atmospheric Sciences (CFCAS) through the CRCM network. REFERENCES Anderson, C. J., and Coauthors, 2003: Hydrological processes in regional climate model simulations of the central United States flood of June July 1993. J. Hydrometeor., 4, 584 598. 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. Bergeron, G., R. Laprise, and D. Caya, 1994: Formulation of the Mesoscale Compressible Community (MC2) model. Internal Report of Cooperative Centre for Research in Mesometeorology, 165 pp. Betts, A. K., J. H. Ball, A. C. M. Beljaars, M. J. Miller, and P. Viterbo, 1996: The land surface atmosphere interaction: A review based on observational and global modeling perspectives. J. Geophys. Res., 101, 7209 7225. Caya, D., and R. Laprise, 1999: A semi-implicit semi-lagrangian regional climate model: The Canadian RCM. Mon. Wea. Rev., 127, 341 362. Clark, C. A., and P. W. Arritt, 1995: Numerical simulations of the effect of soil moisture and vegetation cover on the development of deep convection. J. Appl. Meteor., 34, 2029 2045. Davies, H. C., 1976: A lateral boundary formulation for multilevel prediction models. Quart. J. Roy. Meteor. Soc., 102, 405 418. Denis, B., J. Côté, and R. Laprise, 2002: Spectral decomposition of two-dimensional atmospheric fields on limited-area domains using the discrete cosine transform (DCT). Mon. Wea. Rev., 130, 1812 1829. Dickinson, R. E., R. M. Errico, F. Giorgi, and G. T. Bates, 1989: A regional climate model for western United States. Climatic Change, 15, 383 422. Fritsch, J. M., and C. F. Chappell, 1980: Numerical prediction of convectively driven mesoscale pressure systems. Part I: Convective parameterization. J. Atmos. Sci., 37, 1722 1733. Gal-Chen, T., and R. C. Somerville, 1975: On the use of a coordinate transformation for the solution of the Navier Stokes equations. J. Comput. Phys., 17, 209 228. Giorgi, F., 1990: Simulation of regional climate using a limited area model nested in a general circulation model. J. Climate, 3, 941 963., and L. O. Mearns, 1999: Introduction to special section: Regional climate modeling revisited. J. Geophys. Res., 104, 6335 6352., G. T. Bates, and S. J. Nieman, 1993: The multiyear surface climatology of a regional atmospheric model over the western United States. J. Climate, 6, 75 95. Gutowski, W. J., F. O. Otieno, R. W. Arritt, E. S. Takle, and Z. Pan, 2004: Diagnosis and attribution of a seasonal precipitation deficit in a U.S. regional climate simulation. J. Hydrometeor., 5, 230 242. Hong, S.-Y., and A. Leetmaa, 1999: An evaluation of the NCEP RCM for regional climate modeling. J. Climate, 12, 592 609. Houghton, J. T., Y. Ding, D. J. Griggs, M. Noguer, P. J. van der Linden, D. Xiaosu, K. Maskell, and C. A. Johnson, Eds., 2001: Climate Change 2001: The Scientific Basis: Third Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, 944 pp.

932 M O N T H L Y W E A T H E R R E V I E W VOLUME 134 Jones, R. G., J. M. Murphy, and M. Noguer, 1995: Simulations of climate change over Europe using a nested regional climate model. I: Assessment of control climate, including sensitivity to location of lateral boundaries. Quart. J. Roy. Meteor. Soc., 121, 1413 1449. Kain, J. S., and J. M. Fritsch, 1990: A one-dimensional entraining/ detraining plume model and its 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. Laprise, R., D. Caya, M. Giguère, G. Bergeron, H. Coté, J.P. Blanchet, G. J. Boer, and N. McFarlane, 1998: Climate and climate change in western Canada as simulated by the Canadian Regional Climate Model. Atmos. Ocean, 36, 119 167.,, 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. Liang, X.-Z., K. E. Kunkel, and A. N. Samel, 2001: Development of a regional climate model for U.S. Midwest applications. Part I: Sensitivity to buffer zone treatment. J. Climate, 14, 4363 4378., L. Li, K. E. Kunkel, M. Ting, and J. X. L. Wang, 2004: Regional climate model simulation of U.S. precipitation during 1982 2002. Part I: Annual cycle. J. Climate, 17, 3510 3529. Manabe, S., 1969: Climate and the ocean circulation. I. The atmospheric circulation and the hydrology of the earth s surface. Mon. Wea. Rev., 97, 739 774. 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. Milly, P. C. D., and K. A. Dunne, 1994: Sensitivity of the global water cycle to the water-holding capacity of land. J. Climate, 7, 506 526. Mitchell, T. D., and D. J. Philip, 2005: An improved method of constructing a database of monthly climate observations and associated high-resolution grids. Int. J. Climatol., 25, 693 712. Paquin, D., and D. Caya, 2000: New convection scheme in the Canadian Regional Climate Model. Research Activities in Atmospheric and Oceanic Modelling, H. Ritchie, Ed., WMO Tech. Doc. 987, Rep. 30, 7.14 7.15. 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 Tech. Doc. 1105, Rep. 32, 7.39 7.40. Seth, A., and F. Giorgi, 1998: The effects of domain choice on summer precipitation simulation and sensitivity in a regional climate model. J. Climate, 11, 2698 2712. Sheng, J., and F. Zwiers, 1998: An improved scheme for timedependent boundary conditions in atmospheric general circulation models. Climate Dyn., 14, 609 613. Takle, E. S., and Coauthors, 1999: Project to intercompare regional climate simulation (PIRCS): Description and initial results. J. Geophys. Res., 104, 19 443 19 461. Vidale, P. L., D. Lüthi, C. Frei, S. I. Seneviratne, and C. Schär, 2003: Predictability and uncertainty in a regional climate model. J. Geophys. Res., 108, 4586, doi:10.1029/2002jd002810.