Assessment of summer extremes and climate variability over the north-east of North America as simulated by the Canadian Regional Climate Model

INTERNATIONAL JOURNAL OF CLIMATOLOGYInt. J. Climatol. (2011)Published online in Wiley Online Library(wileyonlinelibrary.com) DOI: 10.1002/joc.2382

Assessment of summer extremes and climate variability overthe north-east of North America as simulated by the

Canadian Regional Climate Model

Philippe Roy,a* Philippe Gachona,b and Rene Laprisea

a ESCER (Etude et Simulation du Climat a l’Echelle Regionale) Centre, Universite du Quebec a Montreal, Montreal, Quebec H3C 3P8, Canadab Adaptation & Impacts Research Section, Climate Research Division, Environment Canada, Montreal, Quebec H5A 1L9, Canada

ABSTRACT: The present study focuses on the evaluation and comparison of the ability of two versions of the CanadianRegional Climate Model (CRCM) driven by re-analyses (NCEP–NCAR) to reproduce the observed extremes and climatevariability in summer (1961–1990). The analysed variables are daily precipitation, minimum and maximum temperaturesover three regions located in north-eastern North America that are characterized by different topography and observationdensity. The validation has been performed with multiple climate extreme indices characterizing the frequency, intensityand duration of precipitation and temperature events. The assessment of the ability of the CRCM is done through anin-depth analysis of the statistical distribution, performance scores and interannual variability of extreme indices. Thereference database has been constructed by kriging the daily observed data from local meteorological stations onto theCRCM 45-km grid. The vast majority of results over the three regions show that, with respect to the previous (i.e. 3.7.1)CRCM version, the latest version (4.1.1) improves in general the simulated extreme events. In particular, the intensity ofextreme hot summer temperature, diurnal temperature range, wet days occurrence, seasonal dry spell, and to a lesser extentextreme cold summer temperature and heavy rainfall. The study suggests that improvements in the simulated extremes inthe latest version are due mainly to the introduction of the new land surface scheme (CLASS 2.7), with a more sophisticatedrepresentation of the soil moisture content. This suggests the importance of surface processes parameterization as a potentialcause of errors in simulated extremes. Copyright 2011 Royal Meteorological Society

KEY WORDS regional climate modelling; kriging; extremes; variability; North America

Received 2 February 2010; Revised 17 May 2011; Accepted 21 May 2011

1. Introduction

Main concerns for the vulnerability, impacts and adap-tation (VIA) studies are related to changes in extremesand climate variability at the regional scale. These stud-ies need to take into account fine-scale physical feed-backs affecting regional systems, especially when society,ecosystems or infrastructures are concerned. In particu-lar, frequency and severity of extreme events, perhapsmore so than gradual changes in the average climate(Mearns et al., 1984; Katz and Brown, 1992) play animportant role in socioeconomic impacts. Latest coupledAtmosphere–Ocean Global Climate Models (AOGCM)used in the Fourth Assessment Report from the Inter-governmental Panel on Climate Change (IPCC) have atypical resolution varying from 100 to 400 km (IPCC,2007), which is still too coarse to realistically representclimate variability and extremes at the regional scale. Cli-mate change projections from AOGCM cannot be applieddirectly at the regional scale without further downscalingof the information. Hence, there is an increasing need

* Correspondence to: Philippe Roy, ESCER (Etude et Simulation duClimat a l’Echelle Regionale) Centre, Universite du Quebec a Montreal,Montreal, Quebec H3C 3P8, Canada. E-mail: [email protected]

among the scientific community and policy maker forclimate projections of climate variability and extremes atregional and fine temporal (i.e. daily or sub-daily) scales,as they will likely have significant impacts on the envi-ronment and human activities.

Dynamical downscaling of AOGCM simulations, suchas the use of a Regional Climate Models (RCM), is auseful tool to develop high-resolution climate scenariosat appropriate temporal and spatial scales relevant forVIA studies. The high resolution of RCMs can take intoaccount regional forcing (orographic features or land/seadiscontinuities) not explicitly included in AOGCMs (e.g.Mearns et al., 2003), which is a potential added valuefor the simulation of realistic extreme events. In partic-ular, this includes subgrid-scale processes parameteriza-tions dealing with clouds and precipitation (Meehl et al.,2000), as well as surface and boundary layer parame-terization (see Sushama et al., 2006; Christensen et al.,2007; de Elıa et al., 2008; Karlsson et al., 2008). How-ever, varying degrees of uncertainty remain in many ofthe quantitative aspects of RCM simulations, especiallywith respect to the representativeness of precipitationwhere there is a mismatch of scale between precipitation

Copyright 2011 Royal Meteorological Society

P. ROY et al.

in models and observation stations (see Haylock et al.,2006; Schmidli et al., 2007).

In Europe, the evaluation of downscaling methodsused to simulate reliable high-frequency climate infor-mation linked with climate extreme occurrence andintensity has been recently addressed in a systematicmanner through projects such as STARDEX, PRU-DENCE and ENSEMBLES (see details about theseresearch projects under the Climate Research UnitWeb site, http://www.cru.uea.ac.uk/cru/research/). OverCanada, evaluation of simulated climate have been madewith AOGCMs (e.g. Barrow et al., 2004; Bonsal andProwse, 2006) and the Canadian RCM (CRCM) (Lapriseet al., 1998, 2003; Plummer et al., 2006; Sushama et al.,2006; De Elıa et al., 2008), but both with a focus on meanclimate evaluation over current and future periods. Theuncertainty associated with RCM projections of climateextremes over North America remains under-explored,despite an earlier analysis of rainfall extremes (annualmaximum) through intensity–duration–frequency curvesanalysis over southern Quebec (Mailhot et al. (2007)and a recent study (Mladjic et al., 2011) investigat-ing the projected changes to precipitation extreme char-acteristics over Canada in the CRCM. There is agreat opportunity with the increasing horizontal reso-lution of RCMs and/or the recent available RCM runsfrom the NARCCAP (North American Regional ClimateChange Assessment Program, (Mearns et al. 2009) seehttp://www.narccap.ucar.edu/) project to do a completeassessment of climate extremes.

The aim of this study is to assess the dynamical down-scaling ability of two versions of the CRCM, withoutAOGCM biases introduced when the RCM is drivenby global models, of capturing the high- and low-frequency variability, with respect to observed meteo-rological patterns. To achieve this, both versions of theCRCM were driven by global atmospheric re-analyses(NCEP–NCAR, National Centers for Environmental Pre-diction–National Center for Atmospheric Research, e.g.Kalnay et al., 1996; Kistler et al., 2001) over three dif-ferent areas in north-eastern North America in summer.These areas represent different surface conditions thatmainly affect the climate regime through the presence ofthe Appalachian Mountains, Prairies and the Great Lakes,covering the performance of the CRCM under differentclimatic conditions. As summer season is less affected bylarge-scale influences, the experiment set-up constitute agood exercise to evaluate the effects of the regional-scalefeedbacks captured by the CRCM within the domain (e.g.from well-resolved surface conditions) on the occurrence,intensity and duration of extreme events.

In the following sections, the methodology is presentedwith a description of the main characteristics of thetwo CRCM versions, data used for the validation, theinterpolation procedure, and the area of interest andcriteria of analysis. In Section 3, results from the regionwith higher observation density, and combined resultsover the three regions are presented, followed by adiscussion and conclusion given in Section 4 and 5.

2. Methodology

2.1. Model and observation data

2.1.1. Model: the Canadian Regional Climate Model

In this study, two versions (3.7.1 and 4.1.1) of theCRCM are used (thereafter V3 and V4); V3 is describedin Plummer et al. (2006), and V4 in Music and Caya(2007) and Brochu and Laprise (2007). Both versions ofthe CRCM use the semi-Lagrangian semi-implicit MC2(compressible community mesoscale model) dynamicalkernel (Laprise et al., 1997) with the Canadian AOGCMphysics parameterization mostly based on the secondversion of the atmospheric GCM (McFarlane et al., 1992)for V3 and the third version (Scinocca and McFarlane,2004) for V4. The horizontal grid cover a large portionof North America (see Figure 1) and have a grid meshof 45 km (true at 60°N).

A major difference between the two versions is theland-surface scheme. V3 uses a bucket land-surfacescheme (McFarlane et al., 1992), modified by Plum-mer et al. (2006), whereas V4 uses the more sophis-ticated Canadian land-surface scheme (CLASS, version2.7; Verseghy, 2000). The modified bucket is a one-layerscheme with a fixed water-holding capacity (10 cm),while CLASS simulates the exchange of heat and mois-ture through a sophisticated three-layer soil scheme.Each grid cell in CLASS can have up to four sub-areas: bare soil, vegetation-covered, snow-covered andsnow–vegetation covered (Verseghy, 2000).

Another difference concerns the parameterization ofthe vertical flux of heat, moisture and momentum. V3uses a vertical mixing in the boundary layer that has beenmodified to include non-local mixing of heat and mois-ture under conditions where the buoyancy flux at the sur-face is upward (Jiao and Caya, 2006). The mixing schemein V3 equally adds fluxes to the entire boundary layer torepresent the vertical profiles of water vapour and poten-tial temperature in a well-mixed planetary boundary layer(PBL), resulting in an improved distribution of mois-ture in the PBL and better summer precipitation simula-tions (Jiao and Caya, 2006). V4 uses modified turbulent

Figure 1. Domain area over North America used for the CRCMsimulations. The topography (in m) is shown in colour scale.

Copyright 2011 Royal Meteorological Society Int. J. Climatol. (2011)

EXTREMES AND CLIMATE VARIABILITY IN SUMMER SIMULATIONS OF THE CRCM

transfer coefficients for surface exchanges of heat, mois-ture and momentum from the third-generation CanadianAOGCM (CGCM3), which allows different roughnesslengths for heat and momentum, and gives transfer coef-ficients that are in agreement with the Monin-Obukhovsimilarity theory (Abdella and McFarlane, 1996).

The atmospheric and oceanic boundary conditionscome from re-analysis of observations. NCEP–NCARglobal re-analysis data set (Kalnay et al., 1996; Kistleret al., 2001) have been used for the atmospheric lateralboundary conditions. Oceanic data are prescribed fromthe Atmospheric Model Intercomparison Project (AMIPII) data set (available at 1° by 1° horizontal grid),consisting of monthly sea surface temperature and sea-icethickness obtained from Fiorino (1997) which are linearlyinterpolated every day from consecutive monthly values.

2.1.2. Observation data (OBS)

For this study, three daily variables have been used:maximum temperature (Tmax), minimum temperature(Tmin) and total daily precipitation (Prec) for the period1961–1990. In the United States, daily data from theNational Climatic Data Center (NCDC, see www.ncdc.noaa.gov) are extracted from approximately 8000 stationsscattered over the eastern part of the country. In Canada,the homogenized temperature data sets from EnvironmentCanada (EC, Vincent et al., 2002) and from Ministry ofEnvironment in Quebec (MENV, Yagouti et al., 2008)are used, as well as the rehabilitated values of Prec fromEC (Mekis and Hogg, 1998).

The number of available stations decreases with time.For example, in region A (see the description of the studyregions in Section 2.1.3 and their locations in Figure 2)the number of stations has decreased gradually with time,i.e. at the end of the 1990s compared to early 1960s, from575 to 394 (−31.5%) precipitation stations and from 322to 250 (−22.4%) temperature stations.

Figure 2. The three climatic regions used in the analysis and observationstations (black dots). Region A covers Pennsylvania, region B coversOhio and parts of Indiana and region C covers south-eastern Ontarioand south-western Quebec. For the observation station networks, thelabel NCDC refers to US data from the National Climate Data Center,EC to Environment Canada and MENV to Ministry of Environment in

Quebec.

Prior to this study, no daily gridded values wereavailable at the resolution of the CRCM grid and forthe period 1961–1990 from observed data over NorthAmerica. Some products exist, but they all have drawbackfor this study. The North American Regional Re-analysis(e.g. Mesinger et al., 2006) is a re-analysis product overNorth America at the 32-km horizontal resolution, but itcovers only the last three decades (i.e. 1979–2009). TheCanadian Precipitation Analysis (CaPA) is available atthe regional scale over Canada, but does not go back to1961. Hence, the need to develop daily gridded analysesof Tmin, Tmax and Prec has emerged.

Ordinary kriging (Matheron, 1962; Matheron, 1963a,1963b) of daily values of minimum temperature, maxi-mum temperature and precipitation has been performed inthis study, using an algorithm developed by Baillargeon(2005). To assess the kriging, a cross-validation processhas been done using a standard jack-knife procedure ontwo semi-variogram models, spherical and exponential.This validation process ensures that we use the best semi-variogram model for each day, giving us a better estimateof predicted values at the grid point of the model. Thedaily kriged values have been used to compute the diag-nostic observed extreme indices over the summer season(covering the 3 months of JJA).

Root square of the kriging variance (defined by Equa-tion (1)) are presented in Table I for the three regions.These values can be seen as an approximation of thekriging error.

2.1.3. Study areas

Three study areas have been chosen to validate theCRCM simulations over north-eastern United Stateswhere a high density of observed stations is present,and south-eastern Canada where the density of observedstations is smaller. Figure 2 shows the three study areasand the distribution of the observation stations from thedata sources for United States and Canada. The densityof stations varies between the three regions: the averagenumber of stations per CRCM 45-km grid cell is 6.5, 4.6and 1.6 for the three regions A, B and C, respectively(see Table II).

Region A covers the state of Pennsylvania, crossed bythe Appalachian Mountains through a south-west/north-east axis (see the topography of region A in Figure 3).The precipitation regime is influenced both by the pres-ence of topography and the vicinity of the Atlantic Ocean,where a high occurrence of synoptic weather systems

Table I. Root square of kriging variance for maximum tem-perature (Tmax), minimum temperature (Tmin) and precipitation

(Prec) for ordinary kriging for the three regions.

A B C

Tmax (°C) 2.0 2.1 3.4Tmin (°C) 2.0 1.7 3.1Prec (mm/day) 5.9 6.1 6.5


P. ROY et al.

Table II. Information about the three study regions defined in Figure 2, in terms of latitude and longitude, number of CRCMgrid points per area, number of observed stations and density of stations per CRCM grid cell, and data sources.

Zone A Zone B Zone C

Latitude (°) 39.5–42 °N 39.2–41.2 °N 45.5–48 °NLongitude (°) 279.2–285.2 °W 273–279.5 °W 280–285 °WNumber of stations (max, min) P : (575,394) P : (360,264) P : (78,65)

T : (322,250) T : (236,201) T : (71,59)

Density(

stationsgrid cells

)6.48 4.60 1.57

Data NCDC NCDC NCDC, EC and MENV

P and T refer to precipitation and temperature, respectively, and the associated number in parentheses to the maximum and minimum numberof available stations over the 1961–1990 period, respectively

Figure 3. The topography (in m) of region A is given from (a) observed meteorological stations (564 stations) and (b) CRCM 45-km grid(interpolated from US Navy with 1/6° × 1/6° gridded topography).

are present for most of the year. Region B, covering thestates of Indiana and Ohio, is more continental and influ-enced by the Great Lakes. Region B is mainly coveredwith plains, bare soil and low relief. Region C covers thesouth-eastern part of Ontario and the south-western partof Quebec, and is located north of the Great Lakes. In thesouth of Region C part of the plain of the St-Lawrencevalley is present while the north is mainly covered byforest with low relief.

2.2. Validation methodology

2.2.1. Diagnostic: extreme indices

Kriged daily Tmin, Tmax and Prec have been used to com-pute six seasonal extreme indices (see Table III), chosenfrom a selection of indices developed by Expert Team onClimate Change Detection Monitoring and Indices (ETC-CDMI) and Statistical and Regional dynamical Down-scaling of Extremes for European regions (STARDEX)(Goodess, 2005). Main criteria for extremes indices arethe intensity, duration and frequency of an event. Theindices have been selected to analyse the intensity ofwarm and cold extremes (through the 90th and 10th per-centiles of Tmax and Tmin, respectively), the mean seasonaldiurnal temperature range (DTR, i.e. differences betweenTmax and Tmin during the day), as well as frequency of

wet days (using a threshold of 1 mm/day, see Hennessyet al., 1999), maximum duration of dry sequences andextremes of daily precipitation (through 90th percentilevalues).

Table III. List of the six extreme indices used in the studyto analyse cold and warm extremes, diurnal amplitude oftemperature, wet days, maximum duration of dry sequences,

and heavy rainfall.

Name Type Definition (unit)

Tx90 Intensity 90th percentile of Tmax (°C)Tn10 Intensity 10th percentile of Tmin (°C)DTR Variability Mean diurnal temperature range, i.e.

difference between Tmax and Tmin

during the day (°C)Prcp1 Frequency Days with precipitation ≥1 mm

(days)CDD Duration Maximum number of consecutive dry

days (days)P90 Intensity 90th percentile of daily precipitation

(mm/day)

For more details, please refer to STARDEX, and to ETCCDI (ExpertTeam on Climate Change Detection and Indices, see http://www.clivar.org/organization/etccdi/etccdi.php)



2.2.2. Performance score

The validation is done through a set of performance score:BIAS, variance ratio (VR), spatial correlation (SC) andinterannual anomalies (IA). BIAS, defined as the meandifference between the model values and observationvalues, concerns mainly the extreme indices since thebias-correction scheme has been applied (see Section2.2.3) on each basic variable (Tmin, Tmax and Prec). VR isdefined as the ratio of the model variance divided by theobservation variance, averaged over all grid points. SCis defined as the Spearman’s correlation of each spatialpattern. SC is calculated for each year and averaged overthe 1961–1990 period, and only statistically significant(at the 95% level) results are used. IA used in this study isa standardized temporal anomaly (see Equation 1) and iscomputed at each grid point. The grid-point standardizedanomalies are then spatially averaged for each year toprovide insight on the CRCM performance over thewhole region during time:

Anoi = 1

K

K∑j=1

Iij − I (61–90)j

σ(61–90)j

(1)

where Anoi represents the anomaly for year i, Iij is thevalue of the index, j the grid point, I (61–90)j is the meanand σ(61–90)j is the standard deviation, for the 30-yearperiod for grid point j . Equation 1 is used for all threedata sets (observation, V3 and V4), which means thatthe mean and standard deviation is relative to the dataset used. By dividing by the standard deviation over theregion, we minimize the effects of regional variabilityof the studied variables and we ensure that we canintercompare the three regions without considering theinherent variability of each index over the study region.

These performance scores are complementary, providea strong assessment of the performance of each model,and the last three (VR, SC and IA) are not subject tosystematic bias. A model that reproduces well extremeindices must have (1) a good VR so that extremescan be reproduced in their amplitude and frequency;(2) a good SC so that the regional characteristics of theextreme indices is well reproduced; and (3) a good IAso that the model is able to simulate the main seasonalsynoptic anomalies that force the extreme indices. IAis useful in evaluating the synchronicity of summerextremes and indicates whether or not the models arecorrectly reproducing atmospheric feedbacks at largescale, and their potential links with the regional and localforcings resolved at the RCM grid-scale. This increasesthe confidence that changes in regional forcing will becorrectly incorporated in the RCM under climate-changeconditions.

Confidence intervals (CI) of performance scores crite-ria (i.e. BIAS, VR and SC) have been calculated througha standard bootstrapping method (Efron, 1979) with 2000iterations. This number of iterations is ample enough togive a good estimate of CI (i.e. stability of CI upon rep-etition of the bootstrap process; see Dibike et al., 2008).

2.2.3. Systematic bias correction

To correctly assess the capacity of models to simulateextreme indices, we work under an ‘unbiased mean state’,where the systematic bias (SB) of each simulated variableis corrected. The SB is grid-point dependant, meaningthat a different correction is applied for each grid point.For Tmin and Tmax, we subtract the respective mean SB oneach simulated daily value of the period 1961–1990, sothat simulated and observed grid-point distributions havethe same mean value. For precipitation, a multiplicationfactor is calculated and applied on the daily value sothat the two distributions have the same average value.Once the variables are corrected, we calculate the extremeindices from the corrected daily values.

3. Results

3.1. Analysis over the region A

3.1.1. Spatial pattern of Bias

Figure 4 shows the bias for the three variables (Tmax,Tmin and Prec) on the simulated summer values withrespect to kriged daily time series. For Tmax and Tmin, thebias is defined as the difference between the simulatedand observed values; for Prec, the bias is defined hereas relative bias of precipitation, expressed in percent ofobserved value. The main outcomes shown from thisfigure are as follows:

3.1.1.1. Tmax: Both versions have a strong positive bias(warm bias), with similar amplitude and spatial pattern,except for a negative bias on the northern part for V3.There is a stronger positive bias in V4 over south-centralpart of the region;

3.1.1.2. Tmin: Different behaviours are shown from thetwo versions for the Tmin variable: V3 exhibits a strongerbias, with largest warm bias in the western part and V4exhibits a slightly positive bias over the western part andnegative bias over the eastern part. This suggests a mixof a warm and cold bias over higher and lower altitude(see topography of region A in Figure 3), respectively;

3.1.1.3. Prec: In V3, the precipitation bias is exclusivelynegative, suggesting a systematic underestimation of theaccumulated summer precipitation. In V4, bias is smaller,negative over the mountains and positive over loweraltitudes in western part and south-eastern part of thedomain (see topography of region A in Figure 3).

3.1.2. Probability distribution function of variables

Figure 5 shows the probability density functions (PDFs)for Tmax and Tmin, as well as the quantile–quantile (Q–Q)plot for Prec, all calculated from all grid point and allsummer daily values for raw and bias corrected (BC)variables. For Tmax and Tmin, a kernel-smoothing method


P. ROY et al.

Figure 4. Bias in region A for V3 and V4 runs with respect to kriged values of (a) Tmax and, (b) Tmin (absolute values in°C) and (c) precipitation(relative values in %). All values are computed for the summer season (JJA) and over the 1961–1990 period.

Figure 5. Summertime PDFs for region A of (a) Tmax and (b) Tmin (in °C) and (c) Q–Q plot (in mm/day) of raw and BC variables. For Tmax

and Tmin, the kernel-smoothing algorithm was used for the computation of probability estimate.



Figure 6. Boxplot graphs for (a) Tx90, (b) Tn10, (c) DTR, (d) Prcp1, (e) CDD and (f) P90 for region A. Middle red line indicates the median,blue rectangle is the Inter-Quartile Range (IQR), the whisker corresponds to the values of 1.5 × IQR and red dots to outliers.

was used for the computation of probability densityestimate. From the analysis of PDFs and Q–Q plots, wecan argue that:

3.1.2.1. Tmax: V4 and V3 share similar features, withcomparable median value in both versions, and with astrong overestimation of the observed variability, espe-cially in the upper tail of the distribution. BC distributionsslightly reduce the overestimation of the hottest days, buta warm bias of around 3 to 4 °C still persists during theseevents;

3.1.2.2. Tmin: With the SB correction V3 is stronglyimproved, whereas V4 remains practically unchanged;

3.1.2.3. Prec: All versions show a poor simulation ofextreme values with higher quantiles largely underesti-mated. The departure from Observation data (OBS) isclear above 40 mm/day for V3, V4 and V4 BC. ForV3 BC, the ability to simulate extreme values is bet-ter in spite of the remaining underestimation, but thereis an overestimation of smaller amount of precipitationbetween 20 and 60 mm/day.

3.1.3. Boxplot of extreme indices

Figure 6 shows the boxplot graphs, for temperature andprecipitation extreme indices, all calculated from all gridpoints and all seasonal values of extreme indices oversummer season (1961–1990 period) for region A. Inthese figures, the inter-quartile range (IQR) represents ameasure of the variability induced by the spatial forcings(such as the topography and/or regional subgrid scaleconditions) and the temporal forcings (i.e. interannualvariability).

3.1.3.1. Tx90: The non-corrected values (V3 and V4)confirm the overestimation of the 90th percentile of Tmax,i.e. of around 3–4 °C in average for V3 and V4, consistentwith a general overestimation of the upper tail of thestatistical distribution analysed through the whole PDFof Tmax. There is a remaining warm bias of about 2 °Cin the median values for both versions once the SB iscorrected (V3 BC and V4 BC). V4 slightly reduces theexcessive spread (as shown by the IQR) present in V3.The lowest values are better simulated in V4 whereas V3have too many low outliers.

3.1.3.2. Tn10: As shown in the whole PDF of Tmin,V4 displays a net improvement compared to V3; hencethe uncorrected boxplot shows a weaker bias for V4.However, once the SB is corrected, V3 have a bettermedian value than V4. The IQR remain similar for bothversions, i.e. similar performance between versions forvariability. V4 apparent overall score is mostly causedby a combination of slight positive and negative biasesover the whole region (not shown), as suggested for meanvalues of Tmin in Figure 4(b).

3.1.3.3. Diurnal temperature range: V3 underestimatesthe median while V4 overestimates it with respect to theOBS. There is a small overestimation of the IQR in V3,while V4 underestimates it. Once the SB is corrected forboth Tmax and Tmin, median values and IQR are stronglyimproved, especially for V4 BC.

3.1.3.4. Prcp1: V3 shows an underestimation of thenumber of wet days. There is an improvement withV4 related to V3, with a better median value withrespect to OBS. However, there is an underestimationof the observed variability in V4, while V3 shows anoverestimation but has better agreement. The correction


P. ROY et al.

of the SB improves the median value of both versionsand further reduces the variability of V4.

3.1.3.5. CDD: V3 presents a strong overestimation ofthe variability, which is reduced with SB correction. V4shows an improvement of this index in terms of medianand IQR, and the low bias of V4 results in no perceivedimprovement between V4 and V4 BC values. In V4, theskewness (OBS: 1.18; V3: 0.82; V4: 1.28) of the OBSCDD is well simulated, which is not the case with V3.

3.1.3.6. P90: Both versions of the model reveal a fairlygood performance. This good behaviour is explained bythe combination of two different factors, (1) the effect ofthe interpolation scheme that smoothes the extreme val-ues of the observed local field on the 45-km grid, and (2)the representativeness of precipitation in the models. Theformer reduces the high values of the observations, bring-ing them closer to the model values; while the latter pro-duces smaller simulated extreme values (see Figure 5(c)).The mismatch in the scale between point observation andmodel grid point can be quite important in summer, asconvective processes, smaller than the model grid reso-lution, take places for intense and localized storms. Themodel is not expected to reproduce this index well, butinterpolation and representativeness make it possible toobtain these apparent good results. The topography doesnot seem to have a crucial impact as results for regionB, where the topography is smaller and less variable,are similar with those of region A (see Section 3.2).Putting those considerations aside, Figure 6(f) shows thatSB correction of precipitation decrease the performanceof V3 with a higher median value and IQR in V3 BC.For V4, the small correction factor of this version slightlyimproved the performance by lowering the median value,without affecting the IQR.

In summary, the analysis of Figure 6 highlights therelative importance of defining performance scores thatare not subject to SB. Indeed, as shown for Tn10 inFigure 6(b), the improved performance of a given versionof the model can be caused solely on a better simulationof average conditions, which are somewhat irrelevantfrom an extreme events perspective. In principle, oncethe SB is corrected, we can more easily assess theresidual bias of extreme indices. However, as explained inSection 4, there are some issues related with SB scheme(i.e. non-homogeneity of the bias through the statisticaldistribution).

3.1.4. Interannual variability

Figure 7 shows the standardized anomalies (Equation (1))over the 1961–1990 period for temperature and precipi-tation extreme indices for region A.

3.1.4.1. Tx90: There is a better correlation with observedvalues in V4 than in V3, where there is a systematicoverestimation of the magnitude of the anomalies during

the 1980s. Note that Tmax time correlation (not shown) isbetter in V3 than V4. This clearly illustrates that averagevalues does not guaranty a similar behaviour in extremevalues. Variance of IA is strongly improved with V4.

3.1.4.2. Tn10: The temporal correlation is improved withV4 compared to V3. This is however lower than forvalues of Tx90; despite better results for Tmin in termsof VR and SC (see Section 3.2). The annual sequencesof cold and warm days/nights after 1985 are quite wellreproduced. However, representations of some anomalies(i.e. 1967, 1973 and 1984) are not reproduced in theiramplitude. Variance of IA is also slightly improved inV4 with respect to observed values.

3.1.4.3. Diurnal temperature range: Variability of DTR istoo high with V3, but with a better time correlation withOBS than for V4 (Figure 7(c)). This better correlationcomes directly from the better time correlation of Tmax

in V3 (not shown). During the 1980s, the Tmax largeoscillation (not shown) can be seen in DTR’s IA.The amplitude is better reproduced by V4, but thesynchronicity is better is V3. The decrease of DTR from1961 to 1990 can be explained by a positive trend inTmin (not shown) through the same time period and theabsence of a trend for Tmax (not shown).

3.1.4.4. Prcp1: IAs of Prcp1 are better reproducedin V4 with a better correlation score. However, V4underestimates the variance while V3 reproduce quitewell the amplitude of the IA. Occurrence of strongdeviations between V3 and OBS (i.e. 1962, 1981 and1982) is reduced in V4.

3.1.4.5. CDD: V4 reproduces well the IA, despite a smalldegradation of the variance of the IA with respect to V3.V4 is better at reproducing IA of dry spells duration indexthan wet day’s occurrence index, as shown through thecorrelation score and variance of both indices.

3.1.4.6. P90: Both versions cannot capture the IAs ofobserved heavy precipitation, as shown by the absenceof a significant correlation between OBS and simulatedvalues. V4 is able to reproduce the occurrence of wet days(Prcp1) and dry series (CDD), but suffers from problemsconcerning the temporal evolution of intense precipita-tion events with respect to observed values. These areinfluenced by local or regional effects such as subgrid-scale physical processes (e.g., topography, convection)not fully resolved at the CRCM scale. Variance of V4 iscloser to OBS, while V3 is too high.

3.2. Mean spatial errors for the three regions

In Figure 8, the spatially and temporally averaged bias(BIAS), variance ratio (VR) and spatial correlation (SC)are shown for the three regions. BIAS for the variable



Figure 7. Standardized anomaly for (a) Tx90, (b) Tn10, (c) DTR, (d) Prcp1, (e) CDD and (f) P90 for region A. VAR is the variance of the timeseries and R is the Spearman correlation score. Only statistically significant correlations are retained (at the 95% level). ‘Not SS’ means that the

correlation is not statistically significant.

Figure 8. Mean performance score of V3 and V4 for (a) BIAS, (b) variance ratio (VR) and (c) spatial correlation (SC) with respect to krigedobservation values. Only statistically significant results (at the 95% level) are used for the computation of performance scores. Vertical lines

represent the confidence intervals (CI), as calculated through a standard bootstrapping method.

(Tmax, Tmin and Prec) and DTR are not discussed heresince the SB correction scheme eliminated the biasof these variables and indices. As stated earlier, SBcorrection is performed prior evaluating the extremeindices.

In summary, through the analysis of statistical scoresacross the three regions, the main results show that for:

BIAS:

• For Tx90 and Tn10, the residual bias is similar forboth versions and all three regions;

• For Prcp1 and P90, region A performs better thanregion B which performs better than region C, espe-cially true for Prcp1;

• There is a slight improvement with V4 for Prcp1, CDDand P90.

Variance ratio (VR):

• There is an improvement of performance with V4 forall variable and extreme indices with the exceptions ofDTR in region B and Tmax, Prcp1 and P90 in region C;

• There is a major improvement in simulated varianceof Tx90 with V4;

• Simulated variance of Tn10 is better than Tmin in regionA and C;

• For Prec and DTR, region A performs better than Bwhich performs better than C, while for Tmax, Tn10,Prcp1 and CDD, region B is worse than regions Aand C.

Spatial correlation (SC):

• For temperatures and all temperature indices, SC isbetter in region A;

• For precipitation and all precipitation indices, SC ispoor for both versions and all three regions, especiallytrue for Prcp1 in region A.

Note that with V4, simulated variance of most precipi-tation indices is good, but that SC is poor. This means thatthe model is able to reproduce the statistics of observedprecipitation indices, but that the spatial features are notcoherent with the observed precipitation indices.


P. ROY et al.

As shown in the CI values in all statistical scores andindices (Figure 8), the CIs for the BIAS scores are lowerfor temperature indices than for precipitation indices,especially for wet day’s occurrence which constitutes theworst performance BIAS score with respect to all otherindices. This last strongly increases from region A toregion C, and suggest a systematic underestimation acrossthe different areas. There is no obvious effect of stationdensity between the three regions on the uncertainty ofPrec and precipitation indices.

4. Discussion

Tmax and Tx90 are overestimated in both CRCM versionsin links with the overestimation of the incoming short-wave for V4 as suggested in Markovic et al. (2009) anddue to the small amount of available soil moisture in V3(Plummer et al., 2006; Music and Caya, 2007) leading toan underestimation of evaporation. Since the shortwaveradiation scheme used in V3 is the same in V4, we canargue that the overestimation of incoming shortwave radi-ation is also present in V3. This means for V3 under highinsolation conditions, too much drying conditions appearin summer for the single soil layer. As there is not enoughwater to dampen the incoming radiation into latent heat,a direct conversion of radiative balance and fluxes intosensible heat induces exacerbated Tx90 values, as shownin Figures 5(a), 6(a) and 8(a). For V4, the performanceis closer to OBS, with a lower spread in the distribution(Figure 6(a)), a consequence of the higher soil moisturecontent, with the use of three layers within the surfacescheme.

IA results suggest that the main subgrid-scale processesresponsible for a large part of summer daily hot/coldextremes are better simulated in V4. This suggests thatthe CRCM is responding to large-scale atmosphericforcing in a more realistic and physical manner withthe improved land-surface scheme. This is evidencedby the better time correlation and magnitude of theIA signals for all of the temperature extreme indices,with the notable exception of DTR’s time correlation.Both versions overestimate Tx90 and Tmax variability(Figures 5(a), 6(a), 7(a) and 8(b)), as also found in otherstudies over other temperate regions from AOGCMs,mainly for Tmax interannual values (Collins et al., 2001;Giorgi, 2002; Raisanen, 2002), which is likely linked tothe representation of land-surface processes in general(Vidale et al., 2007). The small soil water capacity ofV3 results in a high variability of the surface conditionswhich cause a low thermal inertia, as evidenced by thelarge amplitude of the IA of Tx90. In V4, CLASS 2.7reproduces a better IA of Tx90, through a more realisticwater storage capacity resulting in a higher thermal inertiaof the soil. This also results for V4 in a smaller IQR(Figure 6(c)), lower IA variance (Figure 7(c)) and betterVR for DTR (Figure 8(b)). However, this improvementis not systematic (i.e. region B), in part due to theperformance of CLASS 2.7 over bare soil, as discussedbelow.

There is some evidence that the CRCM performsdifferently for average and extreme temperature values.For Tmin, as suggested in Figure 8(b), VR in regions Aand C is better simulated for Tn10 than for Tmin, whilethe performance is worse in region B. In all cases, theminimum or night-time temperatures is mainly forced bythe outgoing long-wave radiation, which is modulatedby the cloud cover (Easterling et al., 1997). The reasonbehind the differential performance of Tmin versus Tn10could be related to systematic biases in cloud cover,as extreme cold nights are often achieved in clear skyconditions (see further details in Markovic et al., 2009).

As precipitation is a complex phenomenon, producedby large and mesoscale features as well as local convec-tive processes, the effects of more appropriate surfaceconditions are less obvious than the results found for sum-mer temperatures. Summer precipitation is more locallyinfluenced (i.e. soil moisture-precipitation feedbacks, e.g.Hohenegger et al., 2009) than any other season (Giorgiet al., 1994; Jiao and Caya, 2006). In the CRCM, about60% (more than 90% in some area) of summer precipi-tation is generated by convection (Jiao and Caya, 2006).In that respect and considering that the main differencesbetween V3 and V4 versions are mostly encompassed inthe new land surface scheme and from the parameteriza-tion of the vertical flux of heat, moisture and momentum,we can expect that most of the differences in results comefrom these changes. Earlier studies found that discrepan-cies in summer precipitation are linked to biases in theshortwave radiation scheme used in V3 and to the simpleland surface scheme (Plummer et al., 2006). For regionA, the additional humidity available in the soil from theaddition of CLASS 2.7 (V4) produces an overestimation(9.6%) of summer precipitation amount for 1961–1990(see Figure 4(c)), while the simple bucket model pro-duces an underestimation of precipitation (−15.9%).

The small (constant) water-holding capacity of V3favours more run-off, thus reducing the available waterfor evaporation (Music and Caya, 2007) and convection.This explains why wet days (Prcp1) tend to be system-atically higher in V4 with shorter dry sequences (CDD)than in V3 over all regions, as shown in Figures 6(d, e)and 8(a). However, this is not systematically transferredto higher heavy precipitation (P90 or higher quantiles)values in V4. This behaviour can be explained by theexcessive heating of the dry soil in V3. The strongerdaytime heating caused by the dry soil can cause strongerthermals that can break through the stable air barrier moreeasily, which leads to deep convection and to subsequentevents of heavy rain (Hohenegger et al., 2009).

Our results confirm the conclusions made by Schmidliet al. (2007) and Haylock et al. (2006) over Europe, thatin general the RCMs correctly simulate the temporal vari-ability of wet/dry day occurrence (e.g. Prcp1, CDD),more so than those characterizing intense events (e.g.P90), as seen in Figure 7. However, under our perfor-mance scores shown in Figure 8, this difference does notappear, as evidenced by the stronger BIAS of Prcp1 than



P90, while VR and SC are similar for CDD and P90,with high uncertainty of these performance scores.

Perhaps surprisingly, our results show that the topog-raphy is not a prime factor in the simulation of most ofthe extreme events. Indeed, most variable and extremeindices have better performance score in region A, withhigher relief, than in other regions. Station density aside,the type of soil and how the diabatic processes are accu-rately simulated in the land surface scheme plays a keyrole. Results from Desborough et al. (1996) suggest thatover bare soil under conditions of high insolation, there isa tendency to overestimate evaporation (Verseghy, 2000).This could explain why the VR of Prcp1 and CDD (albeitunder more complex conditions related to the absenceof convective precipitation as soon as the excess of soilmoisture has been lost through evaporation) in regionB, where more bare soils are present than over otherregions, are worse than that over regions A and C. SCresults shows that Prcp1 have lower SC in region A thanin the other two regions, meaning that insufficient resolu-tion of the topography in the model affects neverthelessthe spatial characteristics of these indices.

Finally, we used a simple correction scheme to subtractthe SB from the simulated mean values. There is evidence(see ENSEMBLES results, e.g. Deque, 2007; Christensenet al., 2008) that the bias is non-homogeneous throughouttemperature and precipitation distribution. The extentto which this nonlinearity and the homogeneous biascorrection over all quantiles of the statistical distributionaffect the extreme indices are not easy to assess butcould be substantial, and could introduce some benefitsor drawbacks according to the considered variables andregions. This is of particular importance as the variableis not normally distributed like precipitation. Therefore,threshold indices such as Prcp1, CDD and P90, where weonly consider values over or equal 1 mm/day, are moreheavily affected by the SB correction than temperatureindices. As our study focuses over the summer season,the normality of the daily temperatures distribution isapproximately encountered but can be not meet over otherseasons as during frost and thaw periods of the year or inthe presence of snow on the ground which induces somenonlinear feedbacks in the air temperature behaviour.Over mountainous or heterogeneous topographic area (asin region A), in spite of the fact that the summer seasonmeans generally no snow or frost conditions over or in theground, the non-normality can be encountered as shownin the Tmax statistical distribution (see Figure 5(a)). Insuch case, the bias correction in mean values (V4 BC vs.V4 in Figure 5(a)) induces some small underestimationin the median values, as a slight asymmetrical behaviourin the Tmax PDF is present (i.e. see the heavy tail in thelower part of the distribution with respect to the highertail; this is also true for Tmin, see Figure 5(b)). Hence,caution is needed when applying homogeneous or linearbias correction over the whole statistical distribution.However, subtracting the SB does not preclude the studyof extreme events in the context of SB correction, asthe bias of a given percentile can be seen as a linear

decomposition between a SB component and a residualbias component.

5. Conclusion

In this study, an assessment of summer extremes andvariability of daily minimum and maximum temperaturesand precipitation amount as simulated by two versions(3.7.1 and 4.1.1) of the CRCM has been comparedand validated using kriged observed daily data coveringthe period 1961–1990, over three areas of the north-eastern US and south-eastern Canada. The assessmentis made through the statistical distribution of variables(PDFs of Tmin and Tmax and Q–Q plot of Prec), andboxplot of extreme indices, as well as BIAS, VR, SCand IA of climate extreme indices (warm and coldtemperature extremes, the mean DTR, frequency ofwet days, maximum duration of dry spell and dailyprecipitation extremes. Results lead to think that:

• The improvement of the new CRCM version ismainly due to a more sophisticated land-surfacescheme (CLASS 2.7) allowing more realistic simula-tions of surface conditions and the associated physi-cal processes, such as the soil moisture-precipitationfeedback;

• The higher thermal inertia and higher soil moisturememory of CLASS 2.7 allows a better simulationof daily temperature variability and consequently thesimulation of high/low extreme air temperatures andthe associated variability. In particular, a more realisticIA for Tx90 and Tn10 is obtained from the new CRCMversion;

• This study confirms the role of the soil and bound-ary–layer parameterization as a potential cause ofbiases in precipitation occurrence, and not only for theextremes of rainfall. As suggested in the last IPCC2007, these parameterizations (i.e. clouds, soil, andboundary–layer processes) continue to be a majorsource of climate models uncertainty. In particular, theincoming shortwave radiation and soil moisture contentinfluence the accuracy of simulated precipitation occur-rence, seasonal drought and heavy rainfall. However,the new land surface scheme seems to underestimatethe magnitude of the IA of precipitation occurrenceand seasonal drought;

• The two CRCM versions perform differently for aver-age and extreme values. Surprisingly, variance of the10th percentile of minimum temperature (Tn10) isslightly better simulated than the variance of minimumtemperature (in regions A and C). Less surprising,variance of maximum temperature is better simulatedthan the 90th percentile of maximum temperature forall regions. Physical mechanisms that generate thoseextremes should be explored;

• Overall, the new CRCM version (i.e. 4.1.1) generallyperform better at simulating the observed variance (i.e.VR), especially for warm temperature extremes butwith the exception of minimum temperatures;


P. ROY et al.

• For SC, there is no significant difference betweenthe two versions. SC of precipitation and of extremeindices is low, whereas maximum consecutive drydays is the best simulated among precipitation indices.Temperature extreme indices have better SC scoreand lower uncertainty. Minimum temperature and thetenth percentile of minimum temperature are betterreproduced than other temperature indices;

• Simulated IA of most extreme indices (Tx90, Tn10,Prcp1, and CDD) are improved by the latest versionof the CRCM, suggesting that this model simulates theregional surface conditions and their associated effectson temperature extremes and wet/dry sequences in amore consistent and physical manner than the previousCRCM version.

Other interpolation schemes can be evaluated in futurework, to incorporate their effects into the uncertainty ofthe CRCM evaluation procedure (as suggested in recentworks in ENSEMBLES, see Haylock et al., 2008). Otherapproaches could be tested through the use of high-level kriging methods (see Boer et al., 2001; Hofstraet al., 2008) or conditional interpolation of precipitation(Hewitson and Crane, 2005). Whatever the interpolationscheme is used, a high density of observation station isan important asset for the validation of climate models, inparticular for wet day’s occurrence (Osborn and Hulme,1997).

Our study suggests that RCM, driven in re-analysismode and with different configurations (i.e. surfaceparameterizations in our case) may be used to analysethe processes that cause the occurrence and intensityof the extreme events. Future work should focus on amore physical oriented analysis and the development ofstatistical tools aimed to understand the model physicsand its behaviours over other regions of Canada andduring the mild and cold seasons.

Acknowledgements

We acknowledge Dr. Andrew Harding for his usefulcomments, and the three anonymous reviewers for theirconstructive comments to improve the final version ofthis manuscript. We also acknowledge Ouranos for theaccess to the CRCM data (generated by the climatesimulation team), and the support from EnvironmentCanada and the Ministry of Environment in Quebecfor the access of observed datasets. Finally, we thankthe financial support from the National Sciences andEngineering Research Council (NSERC) of Canada

References

Abdella K, McFarlane N. 1996. Parameterization of surface-layerexchange coefficients for atmospheric models. Boundary LayerMeteorology 80: 223–248.

Baillargeon S. 2005. Le krigeage: revue de la theorie et application al’interpolation spatiale de donnees de precipitations , Master thesis,Universite Laval, 128 p.

Barrow E, Maxwell B, Gachon P. 2004. Climate variability andchange in Canada: past, present and future, climate change

impacts scenarios project. National Report, Environment Canada,Meteorological Service of Canada, Adaptation Impacts ResearchGroup, Atmospheric and Climate Sciences Directorate Publication,Canada, 114 pp.

Boer EP, de Beurs KM, Hartkamp AD. 2001. Kriging and thin platesplines for mapping climate variables. Journal of Applied Geophysics3: 146–154.

Bonsal BR, Prowse TD. 2006. Regional assessment of GCM-simulatedclimate over Northern Canada. Arctic 59: 115–128.

Brochu R, Laprise R. 2007. Surface water and energy budgets overthe Mississippi and Columbia river basins as simulated by twogenerations of the Canadian Regional Climate Model. Atmosphere-Ocean 45(1): 19–35.

Christensen JH, Carter TR, Rummukainen M, Amanatidis G. 2007.Evaluating the performance and utility of regional climate models:the PRUDENCE project. Climate Change. DOI: 10.1007/s10584-006-9211-6.

Christensen JH, Boberg F, Christensen OB, Lucas-Picher P. 2008. Onthe need for bias correction of regional climate change projectionsof temperature and precipitation. Geophysical Research Letters 35:L20709.

Collins M, Tett SFB, Cooper C. 2001. The internal climate variabilityof HadCM3, a version of the Hadley Centre coupled model withoutflux adjustments. Climate Dynamics 17: 61–81.

De Elıa R, Caya D, Cote H, Frigon A, Biner S, Giguere M, Paquin D,Harvey R, Plummer D. 2008. Evaluation of uncertainties in theCRCM-simulated North American climate. Climate Dynamics 30:113–132.

Deque M. 2007. Frequency of precipitation and temperature extremesover France in an anthropogenic scenario: model results andstatistical correction according to observed values. Global andPlanetary Change 57: 16–26.

Desborough CE, Pitman AJ, Iranneiad P. 1996. Analysis of therelationship between bare soil evaporation and soil moisturesimulated by 13 land surface schemes for a simple non-vegetatedsite. Global and Planetary Change 13: 47–56.

Dibike YB, Gachon P, St-Hilaire A, Ouarda TBMJ, Nguyen VTV.2008. Uncertainty analysis of statistically downscaled temperatureand precipitation regimes in Northern Canada. Theoretical andApplied Climatology 91: 149–170.

Easterling DR, Horton B, Jones PD, Peterson TC, Karl TR, Parker DE,Salinger MJ, Razuvayev V, Plummer N, Jamason P, Folland CK.1997. Maximum and minimum temperature trends for the globe.Science 277: 364–367.

Efron B. 1979. Bootstrap methods: another look at the jackknife. TheAnnals of Statistics 7: 1–26.

ENSEMBLES. Climate change and its impacts at seasonal, decadaland centennial timescales. 2009. Retrieved date: 2009-11-26.http://ensembles-eu.metoffice.com/

Fiorino M. 1997. AMIP II sea surface temperature and sea ice concen-tration observations. http://www-pcmdi.llnl.gov/projects/amip/AMIP2EXPDSN/BCS OBS/amip2 bcs.htm. Retrived date: 2008-04-1

Giorgi F. 2002. Variability and trends of sub-continental scale surfaceclimate in the twentieth century. Part II: AOGCM simulations.Climate Dynamics 18: 693–708.

Giorgi F, Shields C, Bates GT. 1994. Regional climate changescenarios over the United States produced with a nested regionalclimate model. Journal of Climate 7: 375–399.

Goodess C, Wilby R, Ribalaygua J, Schuepbach JE, Plaut G, Cac-ciamani C, Schmith T, Caspary H, Bardossy A, Maheras P. 2005.STARDEX Project, Downscaling climate extremes. ClimaticResearch Unit , Scholl of Environmental Sciences. University of EastAnglia. United Kingdom. Contract no: EVK2-CT-2001-00115

Haylock MR, Cawley GC, Harpham C, Wilby RL, Goodess C. 2006.Downscaling heavy precipitation over the UK: a comparisonof dynamical and statistical methods and their future scenarios.International Journal of Climatology 26: 1397–1415.

Haylock MR, Hofstra N, Klein Tank AMG, Klok EJ, Jones PD,Newe M. 2008. A European daily high-resolution gridded data setof surface temperature and precipitation for 1950–2006. Journal ofGeophysical Research 113: D20119, DOI: 10.1029/2008JD010201.

Hennessy K, Suppiah R, Page CM. 1999. Australian rainfall changes,1910–1995. Australian Meteorological Magazine 48: 1–13.

Hewitson B, Crane R. 2005. Gridded area-averaged daily precipitationvia conditional interpolation. American Meteorological Society 18:41–57.

Hofstra N, Haylock MR, New M, Jones PD, Frei C. 2008. Comparisonof six methods for the interpolation of daily European climate data.Journal of Geophysical Research 113: DOI: 10.1029/2008JD010100.



Hohenegger C, Brockhaus P, Bretherton CS, Schar C. 2009. The soilmoisture-precipitation feedback in simulations with explicit andparameterized convection. Journal of Climate 22: 5003–5020.

IPCC. 2007. Climate Change 2007 – The Physical Science Basis.Contribution of Working Group I to the Fourth Assessment Reportof the IPCC.

Jiao Y, Caya D. 2006. An investigation of summer precipitationsimulated by the Canadian Regional Climate Model. MonthlyWeather Review 134(3): 919–932.

Kalnay E, Kanamitsu M, Kistler R, Collins W, Deaven D, Gandin L,Iredell M, Saha S, White G, Woollen J, Zhu Y, Chelliah M,Ebisuzaki W, Higgins W, Janowiak J, Mo KC, Ropelewski C,Wang J, Leetmaa A, Reynolds R, Jenne R, Joseph D. 1996. TheNCEP/NCAR 40-Year reanalysis project. Bulletin of the AmericanMeteorological Society 77(3): 437–471.

Karlsson K-G, Willen U, Jones C, Wyser K. 2008. Evaluation ofregional cloud climate simulations over Scandinavia using a 10-year NOAA advanced very high resolution radiometer cloudclimatology. Journal of Geophysical Research 113: D01203, DOI:10.1029/2007JD008658.

Katz R, Brown B. 1992. Extreme events in a changing climate:variability is more important than averages. Climatic Change 21:289–302.

Kistler R, Kalnay E, Collins W, Saha S, White G, Woollen J,Chelliah M, Ebisuzaki W, Kanamitsu M, Kousky V, van denDool H, Jenne R, Fiorino M. 2001. The NCEP-NCAR 50-yearreanalysis: monthly means CD-ROM and documentation. Bulletinof American Meteorology Society 82: 247.

Laprise R, Caya D, Bergeron G, Giguere M. 1997. The formulationof Andre robert MC2 (mesoscale compressible community) model.Atmosphere-Ocean 35: 195–220.

Laprise R, Caya D, Giguere M, Bergeron G, Cote H, Blanchet J-P,Boer GJ, McFarlane N. 1998. Climate and climate change in westernCanada as simulated by the Canadian Regional Climate Model.Atmosphere-Ocean 36(2): 119–167.

Laprise R, Caya D, Frigon A, Paquin D. 2003. Current and perturbedclimate as simulated by the second-generation Canadian RegionalClimate Model (CRCM-II) over north-western North America.Climate Dynamics 21: 405–421.

Mailhot A, Duchesne S, Caya D, Talbot G. 2007. Assessment of futurechange in intensity-duration-frequency (IDF) curves for SouthernQuebec using the Canadian Regional Climate Model (CRCM).Journal of Hydrology 347: 197–210.

Markovic M, Jones CG, Winger K, Paquin D. 2009. The surfaceradiation budget over North America: gridded data assessment andevaluation of regional climate models. International Journal ofClimatology 29: 2226–2240.

Matheron G. 1962. Traite de geostatistique appliquee, Tome I.Memoires du Bureau de Recherches Geologiques et Minieres, no14, Editions Technip, Paris.

Matheron G. 1963a. Principles of geostatistics. Economic Geology 58:1246–1266.

Matheron G. 1963b. Traite de geostatistique appliquee, II : LeKrigeage. Memoires du Bureau de Recherches Geologiques etMinieres, no 24. Editions B. R. G. M., Paris.

McFarlane N, Boer G, Blanchet J-P, Lazare M. 1992. The canadianclimate centre second-generation general circulation model and itsequilibrium climate. Journal of Climate 5: 1013–1044.

Mearns L, Katz R, Schneider S. 1984. Extreme high-temperatureevents: changes in their probabilities with changes in meantemperature. Journal of Climate and Applied Meteorology 23:1601–1613.

Mearns LO, Giorgi F, Shields C, McDaniel L. 2003. Climate scenariosfor the southeastern US based on GCM and regional modelingsimulations. Climatic Change 60: 7–36.

Mearns LO, Gutowski WJ, Jones R, Leung LY, McGinnis S, NunesAMB, Qian Y. A regional climate change assessment program forNorth America. EOS, Vol. 90, No. 36, 8 September 2009, pp.311–312.

Meehl GA, Zwiers F, Evans J, Knutson T, Mearns L, Whetton P.2000. Trends in extreme weather and climate events: Issues related tomodeling extremes in projections of future climate change. Bulletinof the American Meteorological Society 81: 427–436.

Mekis E, Hogg W. 1998. Rehabilitation and analysis of canadian dailyprecipitation time series. Atmosphere-Ocean 37: 53–85.

Mesinger F, DiMego G, Kalnay E, Mitchell K, Shafran PC,Ebisuzaki W, Jovic D, Woollen J, Rogers E, Berbery EH, Ek MB,Fan Y, Grumbine R, Higgins W, Li H, Lin Y, Manikin G, Par-rish D, Shi W. 2006. North American regional reanalysis. Bulletinof the American Meteorological Society 87: 343–360.

Mladjic B, Sushama L, Khaliq MN, Laprise R, Caya D, Roy R.2011. Canadian RCM projected changes to extreme precipitationcharacteristics over Canada. Journal of Climate (In press). DOI:10.1175/2010JCLI3937.

Music B, Caya D. 2007. Evaluation of the hydrological cycleover the mississippi river basin as simulated by the CanadianRegional Climate Model (CRCM). Journal of Hydrometeorology8(5): 969–988.

NARCCAP. North American Regional Climate Change AssessmentProgram. http://www.narccap.ucar.edu/

Osborn TJ, Hulme M. 1997. Development of a relationship betweenstation and grid-box rainday frequencies for climate modelevaluation. Journal of Climate 10: 1885–1908.

Plummer D, Caya D, Frigon A, Cote H, Giguere M, Paquin D,Biner S, Harvey R, de Elia R. 2006. Climate and climate changeover North America as simulated by the Canadian RCM. Journal ofClimate 19: 3112–3132.

PRUDENCE. Prediction of Regional scenarios and Uncertaintiesfor Defining EuropeaN Climate change risks and Effects. 2005.Retrieved date: 2005-09-01. http://prudence.dmi.dk/main.html

Raisanen J 2002. CO2-induced changes in interannual temperatureand precipitation variability in 19 CMIP2 experiments. Journal ofClimate 15: 2395–2411.

Schmidli J, Goodess CM, Frei C, Haylock MR, Hundecha Y, Ribal-aygua J, Schmith T. 2007. Statistical and dynamical downscalingof precipitation: an evaluation and comparison of scenarios for theEuropean Alps. Journal of Geophysical Research 112: D04105, DOI:10.1029/2005JD007026.

Scinocca J, McFarlane N. 2004. The variability of modelled tropicalprecipitation. Journal of Atmospheric Science 61: 1993–2015.

STARDEX: Downscaling climate extremes. 2005. Retrieved date:2007-11-01. (http://www.cru.uea.ac.uk/projects/stardex/)

Sushama L, Laprise R, Allard M. 2006. Modeled current and futuresoil thermal regime for northeast Canada. Journal of GeophysicalResearch 111: DOI:10.1029/2005JD007027.

Verseghy D. 2000. The canadian land surface scheme (CLASS): itshistory and future. Atmosphere-Ocean 38: 1–13.

Vidale P, Luthi D, Wegmann R, Schar C. 2007. European summerclimate variability in a heterogeneous multi-model ensemble.Climatic Change 81: 209–232.

Vincent L, Zhang X, Bonsal B, Hogg W. 2002. Homogenization ofdaily temperatures over Canada. Journal of Climate 15: 1322–1334.

Yagouti A, Boulet G, Vincent LA, Vescovi L, Mekis E. 2008.Observed changes in daily temperature and precipitation indices forSouthern Quebec, 1960–2005. Atmosphere-Ocean 46(2): 243–256.


Documents

Assessment of summer extremes and climate variability over the north-east of North America as simulated by the Canadian Regional Climate Model