1989–2008 climatology by subdomains
The map of relatively homogenous geoclimatic regions is presented in Fig. 6. These regions are based on those developed by Bukovsky (2011), adapted to the CRCM5 simulation grid. In order to reduce the overall number of regions, the compound Bukovsky subdomains were used, where feasible. In total, 10 subregions will be presented in the article.
The climatology of 2-m air temperature and precipitation for the 1989–2008 period over the Bukovsky subdomains is presented on Tables 2, 3 and 4 for the JJA and DJF seasons and the whole year, respectively. Mean value and interannual standard deviation (IASD) are presented for CRCM5 simulation and for the reference ensemble. The ensemble mean of ERA-Interim, CRU TS3.1 and UDel data were used as reference for the 2-m air temperature, and CRU TS3.1 and UDel ensemble mean values were used as reference for precipitation. The biases and interannual correlation coefficients of annual and seasonal mean values, calculated for every year in the 1989–2008 period, between simulated and reference data are also shown.
Table 2 CRCM5-simulated and observation-based 2-m air temperature and precipitation for summer (JJA), 1989–2008
Table 3 CRCM5-simulated and observation-based 2-m air temperature and precipitation for winter (DJF), 1989–2008
Table 4 CRCM5-simulated and observation-based 2-m air temperature and precipitation for the whole year, 1989–2008
The maximum biases between simulation and reference means occur in the Desert subregion: −1.9 °C for JJA, −3.0 °C for DJF and −2.3 °C for the whole year. Maximum precipitation biases occur in the Pacific NW region: 0.7 mm/day for JJA, 2.3 mm/day for DJF, and 1.8 mm/day for the whole year. It can be seen from Table 4 that the annual averaged temperature for most subdomains is slightly underestimated in the CRCM5 simulation, and the average precipitation rates are overestimated by 5–30 %, with exception of the complex Pacific NW subregion (41 %) and of the Arctic Land (45 %) subregion, where an all-year-round wet bias is produced by the model.
The correlation coefficients both for 2-m air temperature and for precipitation are noticeably lower in summer than in winter. This might be related to convective nature of summer precipitation, difficult to simulate, as noted earlier by other authors (e.g., Plummer et al. 2006; Jiao and Caya 2006).
As a general rule, interannual correlation coefficients exceeding 0.55–0.6 are statistically significant at 95 % CI (shown in bold in Tables 2, 3, 4). Lower correlation values are in most cases statistically insignificant, which means that for these regions and variables the interannual variability remains largely unresolved by the CRCM5 model.
The interannual correlation coefficients between precipitation and 2-m temperature values for CRCM5 data and the reference base of Tables 2, 3 and 4 are presented in Fig. 7 for summer and winter periods. In most cases the signs and values of correlation coefficients in simulations are close to those in observation data. Exceptions are Arctic Land in JJA and Boreal in DJF. The signs of JJA and DJF correlations are in good agreement with similar results, presented by Mearns et al. (2012), and the correspondence between simulated and reference correlations is consistent with that of NARCCAP models presented in that article. The correlation coefficients between the biases of simulated precipitation and 2-m temperature from corresponding reference values are presented as hollow diamonds in Fig. 7. In summertime the bias correlations are positive in the North of the continent (Arctic Land, Boreal), in the Central subdomain and along the Pacific coast, while the correlations between temperature and precipitation values, presented by color bars, are negative in these regions. In the rest of the continent the bias correlations are positive and close to those presented by color bars. In wintertime in most subdomains the signs of bias correlations and of color bars are opposite to each other, so that the 2-m temperature and precipitation biases are negatively correlated in all subdomains, except in Pacific SW.
Reproduction of temperature and precipitation: analysis by subregions
Figures 8, 9, 10, 11, 12, 13, 14, 15, 16 and 17 will show, for each subregion, the annual cycle of 2-m air temperature and precipitation, consisting of monthly means averaged over the 20-year-long 1989–2008 period. Simulated data will be compared with the same observation-based data as in the previous section: ERA-Interim, CRU TS3.1 and UDel for temperature, CRU TS3.1 and UDel for precipitation data. Along with multi-annual averaged monthly temperature and precipitation values (connected with lines), the interannual variability of monthly means is shown (boxes and whiskers). For all subregions, distributions of daily precipitation intensities for all tiles within subdomains, binned over intervals 0, 0.1 and 2n mm/day, where n = −2,−1, 0, 1, 2, etc., are shown for summer (JJA) and winter (DJF) seasons for the 2001–2008 period. In fact, the data should rather be shown as histograms, but they are shown as curves to ease the comparison of different data. The sum of all the bins gives the average precipitation (in mm/day) for the season and the region. For all presented datasets and seasons considered, the following data are printed in the figures: percent number of dry events (i.e. days with precipitation less than 0.1 mm/day), average (for all days, dry and wet) and maximum daily precipitation (mm/day), along with the 99th precipitation percentile for daily precipitations exceeding 1 mm/day (Pq99). CRCM5-simulated precipitation statistics are compared with GPCP-1dd data, and for all subregions located to the south of 50°N and for southern parts of Central, Pacific NW and Mt. West subregions, TRMM data is also used. For ease of comparison, GPCP-1dd and TRMM data were first interpolated on the CRCM5 grid, using the nearest neighbour method. The CRCM5 hourly and TRMM 3-hourly cumulative precipitations were cumulated over 24-h periods, while daily GPCP-1dd data were used directly.
The subregion Arctic Land represents the taiga and tundra regions, and corresponds to the combination of Bukovsky’s subregions East Taiga, West Taiga, East Tundra, Central Tundra and West Tundra. Arctic Land roughly corresponds to the Köppen ET (tundra) and EF (ice cap) climate areas of the North American continent, with polar tundra and taiga being the main vegetation types. The Arctic Land subregion does not include the Canadian Arctic Archipelago and most part of Alaska where substantial differences between observation and simulation data, as well as between different observation datasets are present (see previous Section). The observation and simulation results, presented in Fig. 8, are in accordance with Köppen classification. The average annual temperature is evidently below 0 °C, with averaged annual maximum in July at around 10 °C and minimum in January at around −26 °C. The interannual variability of CRCM5 monthly mean values of 2-m air temperature is higher than in observations in summertime, and comparable with observations in other periods of the year. There is good general agreement between simulation data and observations, with differences smaller than the interannual variability, except in summertime when a weak cold bias (1–2 °C) can be noticed. The precipitation over the Arctic Land subregion is overestimated in simulations by 0.25–1 mm/day. The shape of the annual cycle, however, is well reproduced, with a precipitation minimum in winter and early spring (January–March) and maximum in summertime. The simulated precipitation maximum is reached in September instead of July–August in observations. The interannual variability of CRCM5-simulated precipitation and temperature is comparable with that of observations in wintertime and exceeds them in summertime.
The frequency distributions of daily precipitation in the CRCM5 simulations and in the GPCP-1dd data are notably different in winter (Fig. 8d). The bell-shaped precipitation distribution is wider and lower in the CRCM5 simulation than in the GPCP-1dd data: relatively more low-precipitation and high-precipitation events were produced by the model in comparison with observation-based data. The number of dry events in simulation is however remarkably lower than in GPCP-1dd (30 % in the model vs. 68 % in observations). The 99th percentile in CRCM5 data is shifted towards higher precipitation rates. In summertime, the precipitation frequency distribution has a maximum at the 8–16 mm/day bin in the model and observations (Fig. 8c). The percent number of dry events is also similar: 42 % in CRCM5 simulation and 51 % in GPCP-1dd. The CRCM5 distribution is slightly shifted towards higher daily precipitation rates compared with GPCP-1dd data, which is consistent with higher 99th percentile CRCM5 value.
The Boreal subregion combines two Bukovsky subdomains, EBoreal and WBoreal, and corresponds mostly to Köppen climates EF (ice cap climate) and Dfb (hemiboreal temperate climate). This region corresponds in general to a wide band of boreal conifer and mixed forests, spread across the northern part of the continent, from northern Rocky Mountains to the Atlantic coast of Québec and Newfoundland and Labrador. The annual cycles of air temperature and precipitation for this subregion are shown in Fig. 9. The simulated air temperature cycle is consistent with observations; the difference between the CRU TS3.1 and other observation datasets exceeds that between simulated data and observation datasets. The typical continental annual temperature cycle with minimum temperature in January (around −18 °C) and maximum in July (around 15 °C) can be seen in Fig. 9a. As in the Arctic Land subregion, the interannual variability of monthly means in simulation data is higher than that of observations in summer and comparable in other seasons. The precipitation annual cycle, with a maximum in summertime and a minimum in winter, is in general reproduced by the CRCM5. In summer and in winter the simulated and observed precipitation rates are comparable. However, during transitional seasons the simulated precipitation is overestimated compared to observations. The excess of precipitation reaches its maximum of around 1 mm/day in April and around 0.5 mm/day in October in autumn. Note the strong interannual variability of both simulation and observation precipitation. The precipitation distribution in the CRCM5 simulation is close to observations both in winter and in summertime. In winter season, higher fraction of CRCM5 precipitation is produced by low-precipitation events than in GPCP-1dd data; in summertime, there is a weak bias in simulated precipitation distribution towards higher precipitation rates.
The Central subregion regroups the CPlains, NPlains, SPlains and Prairies basic subregions of Bukovsky. It encloses the areas with the Köppen climate types Dfb (hemiboreal), Dfa (hot summer continental), BSk (mid-latitude steppe) and CFa (humid subtropical with uniform precipitation distribution). While different climate types are present over this large subregion, it is representing relatively homogenous landscapes of Great Plains and Prairies. The temperature and precipitation patterns for this subregion are presented in Fig. 10. The annual 2-m air temperature cycle shows a cold winter bias of CRCM5, which is also visible in Fig. 3. In summer a weak warm bias can be noted, comparable to differences between observation datasets. The maximum temperature of around 23 °C is reached in July and the minimum (−5 to −7 °C) in January. The interannual variability of monthly means is similar in observations and in simulation data over the whole annual cycle. The precipitation cycle, with winter minimum and summer maximum, is reproduced by the simulation in general, although the precipitation rate is overestimated by nearly 1 mm/day in winter and spring, and is underestimated in summer and in early autumn by 0.5 mm/day. Differences between observation datasets are considerable in summer (Fig. 10c), where GPCP-1dd average summer precipitation exceeds that of CRCM5 and TRMM and its distribution is shifted towards lower daily precipitation values. In wintertime the CRCM5-simulated precipitation distribution is in general similar to observational data, although the CRCM5 distribution is slightly shifted towards lower daily precipitation rates (Fig. 10d). There is noticeable difference between TRMM and GPCP-1dd data both in winter and in summer, concerning the total average daily precipitation, frequency distributions, 99th percentiles and the number of dry events. The possibility of such an important discrepancy between two observation-based datasets makes us to restrict the comparison of simulation data with observations to most general features.
The Great Lakes area, one of the basic Bukovsky subregions, is dominated by Köppen’s continental (Dfa, Dfb) climate types. The Laurentian Great Lakes cover almost 50 % of its area. In CRCM5 simulation, the interactively-coupled 1D Flake model is used to reproduce the water temperature, ice fraction, thickness and temperature over lakes; thus, the quality of reproducing the climate of this subregion depends strongly on the performance of the coupled lake model. As it can be seen in Fig. 11, the averaged 2-m air temperature is in general reproduced over the whole annual cycle; the simulated temperature is warmer than that of the reference datasets in summertime. However, Fig. 3 suggests that this result is actually due to compensating biases over lakes and surrounding land area. It is known (Martynov et al. 2012) that the FLake model overestimates the summer temperature of deep Great Lakes and leaves the water free of ice longer than observed, thus creating a warm bias in winter. There is also a notable distinction of CRU TS3.1 temperature data in wintertime from other datasets. The precipitation cycle, according to observations, has a weak maximum in summertime; average precipitation rate decreases slowly towards the minimum in February. The precipitation rate is well reproduced by the model in summertime, while for the rest of the annual cycle it is overestimated by 0.7–1.2 mm/day. During autumn and winter months excess precipitation can supposedly be explained by enhanced evaporation from the overly warm ice-free surface of the Great Lakes in CRCM5. The interannual variability of precipitation rates is high in simulation and observation-based data, thus the statistical significance of differences is questionable. In summertime, the precipitation distribution in CRCM5 simulation is in between TRMM and GPCP-1dd data (Fig. 11c); its shape resembling mostly that of TRMM data. GPCP-1dd data are shifted towards lower daily precipitation rates. The precipitation distribution in winter (Fig. 11d) demonstrates the relative abundance of low-precipitation events in CRCM5 data, supporting the hypothesis of lake evaporation-enhanced precipitation. This is however not seen in observation-based data.
Three basic Bukovsky subdomains, Appalacia, Mid Atlantic and North Atlantic, were grouped to make the East subregion, where continental (Dfa, Dfb) and humid subtropical (Cfa) Köppen climate zones are prevailing. The temperature and precipitation cycles for this subdomain are presented in Fig. 12. A small negative bias of simulated 2-m air temperature is present in autumn; it is comparable with interannual variability of both observation and simulation data. In other seasons, the average simulated 2-m air temperature is very close to observation values. The annual precipitation cycle is flat, without pronounced difference between seasons, both in observation and simulation data. The simulation overestimates the precipitation by 0.5–1.0 mm/day, except in summertime, where the wet bias almost vanishes. The interannual variability is strong both in the observation-based datasets and in the CRCM5 simulation data. In summer the CRCM5 precipitation frequency distribution is close to that of GPCP-1dd, which is however slightly shifted towards low precipitation rates (Fig. 12c). TRMM average daily precipitation is low in relation to other datasets; however, in summer the precipitation frequency distribution shapes of CRCM5 and TRMM are close, including the 99th percentile values, with GPCP data shifted towards lower precipitation rates. In winter the precipitation frequency distributions of CRCM5 and GPCP-1dd are very close, and the TRMM data are slightly biased towards higher precipitation intensities.
The South subregion consists of Southeast and Deep South basic Bukovsky subregions. It is entirely covered by the Cfa (humid subtropic) Köppen climate type. As shown in Fig. 13, there is a small negative bias of 2-m air temperature in winter, while during the rest of the year, the simulated temperature is fairly close to observation-based data. The flat precipitation cycle is well reproduced and the differences between observed and simulated precipitation multi-annual averages are small in comparison with strong interannual variability. The precipitation distribution in summertime demonstrates good coincidence with observation-based data (Fig. 13c), especially with the TRMM database. In winter, the 99th percentile value of CRCM5 is considerably lower than those of both observation datasets: the model produces less high precipitation rate events than observed in this region, because the simulated precipitations are biased towards lower precipitation ratios (Fig. 13d). The daily precipitation rates are high in comparison to all other subregions of this study; there are relatively few precipitation events with daily precipitation rates lower than 1 mm/day. However, the percentage number of dry events is relatively high, 50–78 % in winter and 44–58 % in summertime.
The Mt. West is constituted of N Rockies, S Rockies and Great Basin basic Bukovsky subregions, thus covering the most part of mountainous areas of the Western North America. Because of complex orography and vast geographical extent, various climate types are presented in this subdomain: Bsk (arid steppe climate, prevailing), Bwk, Bwh (arid desert), Dfb, Dsa (continental), and Cwb (temperate with dry winters). As shown in Fig. 14, the strong annual cycle of 2-m air temperature is reproduced by the CRCM5 with notable cold bias in wintertime, reaching approximately 4 °C. This is the largest temperature bias for all subdomains of the continent presented in this article. The cold bias reaches its maximum in December and vanishes between April and September. The interannual variability of temperature data is relatively small. The precipitation annual cycle is weakly pronounced, according to observations. However, in CRCM5 simulation there is a strong precipitation minimum in summertime (July–September), when it drops 0.2–0.3 mm/day below observation data; during the rest of the annual cycle the precipitation is overestimated by 0.3–0.5 mm/day. The simulated precipitation annual cycle resembles rather that of the Pacific NW region, shown below. Both in summer and in winter the shapes of simulated and observed precipitation distribution are quite similar; the average daily precipitation however is underestimated in summer (Fig. 14c) and overestimated in wintertime (Fig. 14d) in comparison with both reference datasets.
The mountainous regions are particularly difficult for climate models because of high elevations and complex orography, presence of steep slopes, etc. Complex land-surface parameterization schemes are used in CRCM5 (see Sect. 2); however, the quality of reproducing the climate of mountainous regions still represents a serious challenge. Improvement of simulation results can be expected with better horizontal resolution and correspondingly improved topography. It is also important to mention that the adequacy of the observing network in such complex and inhomogeneous areas, and its gridding with meshes of 0.5° and coarser is questionable and it is reasonable to expect that observation-based datasets are more prone to biases in such complex regions. McPhee and Margulis (2005) have shown that the GPCP-1dd data correlation with the North American Land Data Assimilation System (LDAS), based on high density (12,000–15,000) daily precipitation gauge readings and on Doppler radar precipitation measurements (Cosgrove et al. 2003), is at lowest among four large subdomains of continental USA (r = 0.56 for the annual cycle) and that the winter precipitation data are more scattered than those for other seasons.
The Pacific NW subregion is a basic subregion of Bukovsky, corresponding to Köppen oceanic climate types Csb and Cfb. Indeed, Fig. 15 shows mild temperature variations over the annual cycle, reproduced by the CRCM5 model with a cold bias in autumn and winter, and characteristic precipitation annual cycle with dry summer and abundant winter precipitation. The shape of annual precipitation cycle is well reproduced by the model, although there is a wet bias, reaching a maximum in November–January (~3 mm/day) and almost vanishing in summertime. The precipitation distribution of the simulation data is close to GPCP-1dd observations in summer (Fig. 15c) with GPCP-1dd data slightly shifted towards low precipitation rates both in summer and in winter (Fig. 15d). This is consistent with the findings of McPhee and Margulis (2005) that high intensity precipitation events (3 mm/day and higher) are partially missing in this dataset in comparison with LDAS daily precipitation values over Pacific Coast north of 40°N.
Pacific NW is a complex subregion, where the Pacific Ocean meets the steep and high Rocky Mountains, with complex coastline, rich with islands and straits. The ability of the CRCM5 model to reproduce correctly the annual precipitation cycle, despite a wet bias in winter, is noteworthy. It can be expected that better results will be obtained at higher horizontal resolution. As the winter precipitations are brought to the region by westerly winds blowing from the ocean, the winter precipitations strongly depend on the correctness of reproduction of these winds by the model. Because of closeness of the domain limit, it is strongly influenced by the boundary driving conditions; biased driving data could drastically deteriorate the quality of climate simulation along the Pacific Coast. As in the case of Mt. West subregion, the observation-based datasets are prone to errors over the Northern Pacific Coast: McPhee and Margulis (2005) have shown that the correlations between GPCP-1dd and “ground truth” LDAS daily precipitation data are very low (r = 0.21 for the annual cycle) over this region.
Over the Pacific SW region, Csb and Csa (dry-summer subtropical) as well as BWk and BWh (arid desert) Köppen climate types are present. As shown in Fig. 16, the relatively mild 2-m air temperature annual cycle is reproduced by the CRCM5 model with a small cold bias in wintertime: the shape of the precipitation cycle with a very dry summer and relatively wet winter is also well reproduced, with the maximum bias of around 1 mm/day in January–February. The interannual variability is strong, compared with multi-annual average values, both in observation and simulation data. In summer (Fig. 16c), when the typical daily precipitation amounts are very small (~0.03 mm per day), there is substantial difference between simulation data and the observation-based datasets. Dry events are very frequent, 87–94 %, and low precipitation events prevail; on the other hand, the CRCM5 model produces a larger number of higher precipitation events. It is possible, however, that the observation datasets are prone to biases in such extremely dry conditions. In winter (Fig. 16d) the shapes of the precipitation frequency distributions are similar. The average daily precipitation amounts are however quite different: CRCM5 overestimates the average daily precipitation in comparison with observation datasets, which also differ between themselves.
The Desert subregion regroups Bukovsky’s South West and Mezquital basic subregions. As the region name suggests, it is covering mostly the regions with predominant arid desert (BWh, BWk) and steppe (BSh, BSk) Köppen climate types. Indeed, Fig. 17 demonstrates hot and mostly arid climate, with notable summer precipitation maximum, evidencing the presence of the NAM. There is a relatively important negative bias of simulated 2-m air temperature, which is present during the whole annual cycle. This cold bias is also evident in Fig. 3. On the other hand, the annual precipitation cycle is remarkably well reproduced by the model simulation. The interannual variability is relatively small in temperature data, but is very strong in precipitation, in particular during the monsoon season, reflecting high variability of NAM, which will be addressed in more details in Sect. 5. In summer (Fig. 17c) the CRCM5 precipitation distribution is closer to that of TRMM, while in winter (Fig. 17d) both average daily precipitation and distribution demonstrate excellent coincidence with GPCP-1dd data.
Summer diurnal cycle in subregions
For the regions located to the South of 50°N, where TRMM data are available, and for the corresponding parts of Central, Pacific NW and Mt. West subregions, multi-annual (2001–2008) summer (JJA) mean diurnal precipitation cycles are shown in Fig. 18. CRCM5 hourly precipitation data are compared with the 3-hourly values from the TRMM dataset.
In the Central subregion (Fig. 18a) a strong night-time precipitation maximum is present in TRMM data at around 6 GMT, which roughly corresponds to the local midnight. This precipitation maximum is not reproduced by the model. The rest of the diurnal precipitation cycle is well reproduced by the model, including the pronounced minimum at around 18GMT (local midday) and the late afternoon convective precipitation maximum. The nightly precipitation maximum in TRMM is related to the influence of the GPLLJ, which will be discussed in more details in Sect. 6.
In the vicinity of Great Lakes (Fig. 18b) the shape of the diurnal cycle, including nightly precipitation maximum, morning/midday minimum and afternoon rise is well reproduced by the model; however the absolute intensity of precipitation is underestimated by CRCM5 in comparison with TRMM.
In the East subdomain (Fig. 18c) the TRMM diurnal precipitation cycle has a pronounced nightly minimum, which is absent in the CRCM5 data. The daytime part of the diurnal cycle (18–24 GMT) is well reproduced by the model.
Similar TRMM diurnal cycle shape is present in the South subdomain (Fig. 18c) and in the Mt. West subdomain (Fig. 18e). However, in these subregions the CRCM5 model is able to reproduce both the timing of nightly precipitation minimum and the daytime part of the cycle.
The diurnal cycle of the northern part of the Pacific NW subregion (Fig. 18f) is predominantly flat according to both TRMM and to CRCM5. The CRCM5 data contain a broad maximum in the morning hours, characteristic the diurnal cycle of precipitation over the oceans (Tian et al. 2005).
In the Pacific SW subregion (Fig. 18g) the simulated and observation-based diurnal cycles are generally similar: the night-time minimum and afternoon maximum are present on both curves. The structure of CRCM5-simulated daily cycle is more complex, with additional maximum at 15 GMT (~7 LST), early in the morning, which as in the case of Pacific NW might testify the influence of the Pacific Ocean.
In the Desert subregion (Fig. 18h) the simulated and observation-based diurnal cycles almost coincide, with the simulated afternoon maximum stronger and occurring slightly earlier than according to the TRMM data.
In general, we can conclude that the CRCM5 model is able to reproduce adequately the thermally-driven atmospheric convection, responsible for the afternoon rise of precipitation. The processes responsible for the night-time precipitation peculiarities are not yet reproduced entirely satisfactorily by the model.