Challenging some tenets of Regional Climate Modelling

1 downloads 133869 Views 2MB Size Report
affordable means of entering into appealing applications of timely societal ... dling within the RCM domain of the large scales used to drive the RCM at the LBC.
Meteorol Atmos Phys 100, 3–22 (2008) DOI 10.1007/s00703-008-0292-9 Printed in The Netherlands

1 2

Universite du Quebec, Montreal, Quebec, Canada Consortium Ouranos, Montreal, Quebec, Canada

Challenging some tenets of Regional Climate Modelling R. Laprise1 , R. de Elı´a1;2 , D. Caya1;2 , S. Biner2, P. Lucas-Picher1, E. Diaconescu1 , M. Leduc1 , A. Alexandru1 , L. Separovic1 , Canadian Network for Regional Climate Modelling and Diagnostics With 11 Figures Received 19 May 2007; Accepted 25 November 2007 Published online 14 August 2008 # Springer-Verlag 2008

Summary Nested Regional Climate Models (RCMs) are increasingly used for climate-change projections in order to achieve spatial resolutions that would be computationally prohibitive with coupled global climate models. RCMs are commonly thought to behave as a sort of sophisticated magnifying glass to perform dynamical downscaling, which is to add fine-scale details upon the large-scale flow provided as time-dependent lateral boundary condition. Regional climate modelling is a relatively new approach, initiated less than twenty years ago. The interest for the approach has grown rapidly as it offers a computationally affordable means of entering into appealing applications of timely societal relevance, such as high-resolution climatechange projections and seasonal prediction. There exists however a need for basic research aiming at establishing firmly the strengths and limitations of the technique. This paper synthesises the results of a stream of investigations on the merits and weaknesses of the nested approach, initiated almost a decade ago by some members of our team. This short paper revisits some commonly accepted notions amongst practitioners of Regional Climate Modelling, in the form of four tenets that will be challenged: (1) RCMs are capable of generating small-scale features absent in the driving fields supplied as lateral boundary conditions; (2) The generated small scales have the appropriate amplitudes and statistics; (3) The generated small scales accurately represent  tude et la Correspondence: Rene Laprise, Centre ESCER pour l’E  M Ouranos, 550  chelle Regionale, UQA Simulation du Climat a l’E Sherbrooke St West, 19th Floor, West Tower, Montreal, Quebec, H3A 1B9 Canada (E-mail: [email protected])

those that would be present in the driving data if it were not limited by resolution; (4) In performing dynamical downscaling, RCMs operate as a kind of sophisticated magnifying glass, in the sense that the small scales that are generated are uniquely defined for a given set of lateral boundary conditions (LBC). From the partial failure of the last two tenets emerges the notion of internal variability, which has often been thought to be negligible in one-way nested models due to the control exerted by the imposed lateral boundary conditions. A fifth tenet is also discussed, relating to the handling within the RCM domain of the large scales used to drive the RCM at the LBC. We close the article with an appeal to the RCM community to spend more effort in basic research in order to tackle a number of lingering issues that otherwise could jeopardize the credibility of the tool.

1. Introduction Global General Circulation Models (GCMs) constitute the primary tool to study the processes responsible for the quasi-equilibrium of the climate and the natural variability around this equilibrium. GCMs are therefore capable of simulating the evolution of the climate system in response to changes in external forcing (IPCC 2007). The complexity of coupling the atmosphere, land surfaces, oceans and sea-ice, compounded by the long simulations required to reach internal equilibrium between the components and to provide statistical robustness of the

4

R. Laprise et al.

results, impose enormous computational cost for performing useful simulations. The rapid increase of computational cost with increasing horizontal resolution forces to use rather coarse computational grids for GCMs. Horizontal meshes of the atmospheric component of the GCMs in the 4th Assessment Report of the IPCC (2007) range from 400 to 125 km. Such resolution precludes the explicit representation of several physical processes deemed important for climate, and such processes must therefore be parameterised or neglected. The increased awareness that climate changes may result from the anthropogenic release of greenhouse gases in the atmosphere has fostered several studies aimed at anticipating the numerous impacts that will result from such unprecedented rapid climate changes. However most studies of environmental, societal and economic impacts associated with anticipated climate changes resulting from anthropogenic influences demand spatially detailed information at much finer scales than currently affordable with GCMs by most modelling groups. Nested Regional Climate Models (RCMs) offer an appealing complementary avenue. RCMs have the potential for resolving mesoscale processes over a limited region of interest, at a computationally affordable cost. Used for climate-change simulations driven by GCM’s projections, RCMs provide physically consistent information among all of the simulated climate variables. Limited-area models, with suitable parameterisation of subgrid-scale physical processes, constitute a regional (i.e. non global) version of climate models in which the computational effort is directed over a subset of the globe, which allows a large increase in resolution reaching some tens of km for most contemporary RCMs (IPCC 2007). Such high-resolution nested models constitute an initial- and boundary-value problem. RCMs are driven at their lateral boundaries by time-dependent atmospheric variables (winds, temperature, pressure and water vapour) obtained from interpolation of coarse-resolution GCMsimulated data or historical analyses. Such approach is called one-way nesting because the RCM receives information at its lateral boundary but it does not feed back any to the driving model. It is also referred to as dynamical downscaling because the RCM develops small scales that are

believed to be physically consistent, both internally and with the driving large-scale flow through the lateral boundary conditions (LBC). The large-scale flow sometimes also drives within the domain as discussed later. Despite some potential difficulties with the nested approach (e.g., Staniforth 1997; Warner et al. 1997; Laprise 2003), it is by-and-large the most widely used approach in Regional Climate Modelling (e.g., Giorgi and Mearns 1991; McGregor 1997; Wang et al. 2004; B€arring and Laprise 2005; IPCC 2007). A number of global models are currently being developed with stretched grids or variable resolution GCMs (e.g., McGregor et al. 2002; Yeh et al. 2002; Gibelin and Deque 2003; FoxRabinovitz et al. 2005) that offer high resolution locally and feedbacks between regional and global processes that otherwise would only be possible with two-way nesting in limited-area RCMs (Lorenz and Jacob 2005). This paper however will only discuss issues related to one-way nested limited-area RCMs; many of the issues are however common to any regional approach. Regional climate modelling has a relatively short history: the first results were published by Dickinson et al. (1989) and Giorgi and Bates (1989). Several RCMs have since been developed, their performance assessed, and they have been applied to simulate climate and to project future climate changes over several regions of the world, often in the context of coordinated efforts: for example over the United States (PIRCS: Takle et al. 1999), over Europe (PRUDENCE: Frei et al. 2003; Deque et al. 2006; Christensen et al. 2007), over Asia (Wang et al. 2003; RMIP: Fu et al. 2005), over the Arctic (ARCMIP: Curry and Lynch 2002; Rinke et al. 2005) and over the Pacific Ocean (Stowasser et al. 2007). Other coordinated experiments are just being launched, for example over North America (NA RC CAP: Mearns et al. 2005). RCMs are increasingly coupled interactively with other components of the climate system, such as regional oceans and sea ice (e.g., Bailey and Lynch 2000; D€oscher et al. 2002; Rinke et al. 2003; Sasaki et al. 2006; de Szoeke et al. 2006; Seo et al. 2007; Xie et al. 2007), which improves some aspects of the simulation and opens interesting perspective for climate-change studies. Notwithstanding the popularity of RCMs and some of their successes, several difficulties re-

Challenging some tenets of Regional Climate Model

main in order to quantify precisely their ability to simulate some physical processes and their skill at reproducing the current climate, which is a prerequisite for estimating their added value in projections of climate changes. Concerns have been repeatedly raised in the reports of the Working Group on Numerical Experimentation (WGNE) of the World Climate Research Programme (WGNE 1999, 2000) concerning potential indiscriminate use of RCMs. Discussion of some of these concerns has been offered in specialised reports such as Laprise et al. (2002) and B€arring and Laprise (2005), but these have had modest impact in the community due to their limited distribution. The basic assumption behind RCMs is that, initialised and driven by large-scale flow, highresolution RCMs with suitable subgrid-scale parameterisation will develop fine-scale structures, thus generating physically consistent highresolution atmospheric fields. Assuming that RCMs are not intended to affect the large-scale flow, the RCM evaluation process infallibly hinges on the isolation of fine scales from the rest of the flow in order to identify the added value they provide. The separation of scales can be performed either in physical space with digital filters, as in Feser and von Storch (2006) and Feser (2006), or in spectral space with Fourier filters (e.g., Errico 1985; Denis et al. 2002a; Castro et al. 2005). The proper validation of high-resolution RCMs is hampered by uncertainties in reanalysis data (e.g., Annamalai et al. 1999) and by the absence of dense climate observation networks over most of the world except some specific regions. For example the very popular gridded analysis of historical surface data by New et al. (2000), with data over continents on a 0.5 grid mesh, does not contain information at that scale over several regions of the world, including most of Canada for example. Reanalyses from data assimilation with forecast models can also serve as validation basis, but they are of insufficient resolution and their behaviour is strongly influenced by the forecast model in data poor regions. Satellite data offer some interesting validation perspectives, but there are issues related to the short time life of the platforms, drift of the measurements, difficulty of calibration and extraction of desired climate parameters from radiances. The use of

5

individual weather stations is limited to studies in which only local time variability is of interest and no information regarding spatial structure is sought. Some specific field campaigns have produced high-density observations that are invaluable for process studies; their short time scale and the very limited number of regions over which they were carried does not fulfil the required climate timescales. The problem is worsening with time, as RCMs’ resolution continues to increase at a faster rate than climate observation networks. Because of the difficulty of assessing high-resolution RCMs directly against observations or analyses, several studies have approached the issue by designing an idealised experiment in which ‘‘virtual reality’’ verification datasets are constructed by model simulations. One such approach initiated by Laprise et al. (2000) and now referred to as the ‘‘Big-Brother Experiment’’ (BBE), will be described below in Sect. 2. As a result of the difficulty of validating RCMs, initial prejudices and expectations (that have evolved slowly) are based on rather few and limited experiments. Quite contrary to the norm in science, not much effort has been devoted to consolidate or challenge the common wisdom regarding RCMs. In our view, the original paradigm of Regional Climate Modelling that has remained unchanged for many practitioners despite some conflicting evidence, can be summarised in the form of four fundamental tenets as follows: Tenet 1: RCMs are capable of generating smallscale features absent in the driving fields supplied as lateral boundary conditions (LBC); Tenet 2: The small scales that are generated have the appropriate amplitudes and climate statistics; Tenet 3: The generated small scales accurately represent those that would be present in the driving data if it were not limited by resolution; Tenet 4: In performing dynamical downscaling, RCM generated small scales are uniquely defined for a given set of LBC. Beyond these four initially widely accepted points, there are issues for which consensus has never been reached within the RCM community.

6

R. Laprise et al.

One of these relates to the handling of large scales within an RCM domain; ‘‘large scales’’ are meant here to refer to those scales that partake in the driving of an RCM. Several schools of thoughts exist about the handling of large scales by nested models, which we will state as a fifth tenet, in three opposed versions: Tenet 5a: The large scales are unaffected within the RCM domain; Tenet 5b: The large scales may be improved owing to reduced truncation and explicit treatment of some mesoscale processes with increased resolution within the RCM domain; Tenet 5c: The scales larger than or comparable to the RCM domain are degraded because the limited domain is too small to handle these adequately. This last tenet in fact relates to another issue that does not make unanimity within the RCM community, relating to the driving technique. Nudging of the large scales in the interior of the RCM domain (e.g., Biner et al. 2000; von Storch et al. 2000; Miguez-Macho et al. 2004; Castro et al. 2005) has been proposed as an alternative to the traditional driving at the LBC only. There are also nearly a dozen of spectral LAMs using various nesting approaches (Tatsumi 1986; Segami et al. 1989; Kuo and Williams 1992, 1998; Cocke and LaRow 2000; Juang and Hong 2001). This paper however will focus on RCMs when driven in the traditional manner at their lateral boundary only: we consider this study a necessary step before evaluating the effects of large-scale nudging. These five tenets are meant to relate to the impact of nesting on the quality of simulated results by regional models, rather than to the intrinsic skill of any model, whether regional or global. All tenets come with an implicit assumption that the possibility of satisfying the tenets is obviously limited by an RCM’s resolution and sophistication. This paper almost exclusively focuses on results obtained within an idealised framework in which a specific RCM is driven by reanalysis data, and several results are based on the BBE protocol. The aim of this paper is to propose an interpretation of nested Regional Climate Modelling that is consistent with and integrates the findings reported in several papers

that are referred to. This paper is constructed as follows. The next section begins by summarising the BBE that are used several times throughout the paper. Then sub-sections describe specific idealised experiments designed to assess unequivocally each of the tenets in turn, and present the ensuing conclusions. Section 3 finally concludes with a summary of the consequences of the findings discussed in the paper. 2. The idealised experimental framework The difficulty of assessing high-resolution RCMs directly against analyses has motivated the development of an idealised, ‘‘virtual reality’’ experiment in which verification datasets are constructed by simulations of a RCM itself. The approach is now referred to as the ‘‘Big-Brother Experiment’’ (BBE). The approach has been used for several studies by the Canadian group (Denis et al. 2002b, 2003; de Elı´a et al. 2002, 2007; Antic et al. 2005; Dimitrijevic and Laprise 2005; Diaconescu et al. 2007; Leduc and Laprise 2008) and elsewhere (Herceg et al. 2006; Nutter et al. 2004). The approach has been reviewed in detail in Laprise (2008) and will only be briefly described below. The ‘‘Big-Brother Experiment’’ (BBE) has been designed to isolate the errors that are specific to the nesting method, independently of other modelling errors and errors in the large scales used to drive an RCM at its lateral boundary. In its full version the BBE approach would consist first in performing a high-resolution GCM simulation (referred to as the Big Brother, BB) that would serve the dual purposes of (1) serving as reference against which an RCM simulation (named the Little Brother, LB) would be compared to, and (2) defining the LBC needed to drive the LB RCM. An important aspect is that the LBC would be defined by retaining only the large scales of the BB, filtering out fine scales in order to emulate the low resolution typical of the data used to drive a RCM. With the BBE protocol, the differences in the climates simulated by the LB and the reference BB could be attributed unambiguously to the nesting approach of the limited-area RCM. Within the BBE protocol, the BB GCM and the LB RCM would have to use not only the same resolution, but also the same physics, dynamics and numerics. Such an

Challenging some tenets of Regional Climate Model

7

Fig. 1. Flow chart of the Big-Brother Experiment. A reference solution (the Big Brother) is obtained by a large-domain version of the RCM

experiment however is computationally very demanding (in fact prohibitively expensive for most groups) due to the use of high-resolution GCM as BB. An alternative lower cost version of the BBE has hence been used extensively (Fig. 1): the approach consists in replacing the expensive highresolution GCM by a cheaper high-resolution large-domain RCM to serve as BB.

The RCM used for several results reported below is the Canadian RCM (CRCM; Laprise et al. 1998, 2003; Caya and Laprise 1999). In its typical configuration, the CRCM is integrated on a polar-stereographic grid with grid-point spacing of 45 km at 60 N, with 18 or 30 levels in the vertical. The CRCM uses a 15-min timestep and driving data are interpolated linearly in time to provide the LBC at each time step. Figure 2 shows a typical domain, in this case over the eastern part of North America. In this specific instance, the BB simulation is made over a 196 by 196 grid-point domain, driven by six-hourly NCEP reanalyses at the lateral boundary, and the LB domain is 100 by 100 grid-point wide and is driven by the BB-simulated data (with filtering for some experiments, as described below), provided at three-hourly interval. 2.1 Tenet 1: RCMs are capable of generating small-scale features absent in the driving fields supplied as LBC

Fig. 2. Domains of the Big Brother and Little Brother simulations

It has been known for some time that high-resolution nested models, initialised and driven at their lateral boundary by large-scale atmospheric fields that are devoid of small scales, do generate fine-scale structures that appear rather realistic (e.g., Miyakoda and Rosati 1977; Anthes et al.

8

R. Laprise et al.

1982). The presence of small-scale surface forcing (such as high-resolution topography, landsurface characteristics or land–sea contrasts) is known to induce the development of fine-scale structures. Even in the absence of such strong surface forcing, however, high-resolution models spontaneously generate small-scale features through physical processes such as nonlinear interactions (stretching, folding and shearing by meanders in the flow) that can effectively transfer variance from large to small scales, hydrodynamic instabilities (such as barotropic and baroclinic

conversion processes) resulting from thermal and velocity gradients, and parameterised diabatic effects such as condensation and convection. Figure 3 shows an example of the fine scales that develop spontaneously during the integration of the CRCM, after being initialised and driven for a few days by NCEP reanalyses, for a case where localised surface forcing plays a negligible role. At least subjectively, it is fair to say that fine-scale structures are generated in RCMs. This impression may be easily confirmed objectively by performing a spectral study of these

Fig. 3. Instantaneous 850-hPa relative humidity (colours) and mean sea level pressure (contours) in NCEP reanalysis (left) and in 45-km mesh CRCM simulation driven by NCEP data (right)

Fig. 4. Time evolution of the normalised 850-hPa vorticity spectrum in a CRCM simulation initialised without fine scales (reproduced with modifications from de Elı´a et al. (2002) # 2002 American Meteorological Society)

Challenging some tenets of Regional Climate Model

fields (which it will done in the next sections), thus confirming Tenet 1. 2.2 Tenet 2: The generated small scales have the appropriate amplitudes and climate statistics This tenet relates to the impact of nesting on the quality of simulated results, rather than to the intrinsic skill of the model. Hence this tenet is here assessed in the context of the perfect prognosis approach, that is, without taking into account model errors. de Elı´a et al. (2002) presented idealised forecast experiments using the BBE. The Discrete-Cosine Transform (DCT; Denis et al. 2002a) was used to filter the BB-simulated fields used to initialise and drive the LB simulation; length scales smaller than about 450 km (non-dimensional wavenumbers larger than 8) were removed from the BB data. Then the question is whether fine scales will redevelop in such simulation with the proper amplitude, using the BB as reference. The DCT is used to obtain the spectra of the LB- and BB-simulated fields. Figure 4 displays the time evolution of the average spectrum of

9

low-level vorticity in the LB simulations, presented here as the relative root-mean square amplitude, normalised by the corresponding value in the BB simulation. The idealised forecast experiment was repeated 24 times, starting from initial conditions one day apart, so each curve plotted at three-hour intervals represents the average of 24 cases, which provides reasonable statistical robustness. It can be seen that the LB reconstructs almost perfectly the spectral amplitude (i.e. normalised value of unity) after some time. For lowlevel vorticity, equilibrium is reached in less than a day; different fields exhibit different development time scales, but in most instances, fine scales amplitudes reach equilibrium in less than a few days. We now turn our attention to the climate statistics. The BBE protocol has been used for several experiments to assess the dynamical downscaling skill for mid-latitude climate, during winter and summer seasons, and for domains over the East and West Coasts of North America (Denis et al. 2002b, 2003; Antic et al. 2005; Dimitrijevic and Laprise 2005). In general the LB succeeds to reproduce rather well the climate

Fig. 5. Root-mean square of the ratio of the Little Brother small-scale vorticity variance to that of the Big Brother as a function of domain size (expressed by the number of grid points along one axis, with a 45-km grid mesh). The evaluation is made over a 38 by 38 grid-point central portion of the RCM domain

10

R. Laprise et al.

statistics of the BB for all simulated fields, for both the large and small scales, and for stationary and transient eddies. The results obtained with the BBE are comforting in that they indicate dynamical downscaling skill for an RCM driven by low-resolution fields (not shown). It would then appear that Tenet 2 is satisfied for mid-latitude climate statistics, at least for the case in which the large scales used to drive the RCM are perfect as in the BBE. On the other hand, Herceg et al. (2006) made a BBE over a tropical domain to study the skill of the NCEP RSM RCM in seasonal prediction. The reproduction of the small-scale precipitation is less skilful than for mid-latitude domains, possibly as a result of the strong coupling between convection and circulation in topical regions. A recent study by Leduc and Laprise (2008) has shown that the proper development of fine scales from low-resolution LBC requires sufficient domain size to achieve spin-up. Figure 5 shows the amplitude of the Little Brother smallscale vorticity relative to that of the Big Brother, for various domain sizes. While low-level variables recreate the appropriate amount of smallscale variability even for modest domains (80 grid points), the small scales at upper levels are amplitude deficient even for the larger domains tested. It appears that the strong flow in the upper troposphere at mid-latitudes tends to blow the fine scales out of the domain while they develop toward their equilibrium amplitudes. Satisfactory development is only obtained for rather large domain sizes, suggesting that some minimum domain size is required for proper amplitudes to be generated at all heights. Tenet 2 appears to be mostly verified, with the caveat of the results presented by Herceg et al. (2006), at least in the low levels and for suitably large domain size. But it is noteworthy that the small scales can be very amplitude deficient in the upper troposphere for typical domain sizes. This defect may have gone unnoticed by most RCM groups because their interest usually lies in surface processes and precipitation, which is formed in the lowest troposphere. For applications in which the vertical structure of fine scales in the entire troposphere may be important (mountain-induced internal gravity waves come to mind as an example), sufficient lateral spin-up

should be ascertained in order to obtain physically correct results. 2.3. Tenet 3: The generated small scales accurately represent those that would be present in the driving data if it were not limited by resolution The RCM skill in seasonal mean statistics was addressed in the preceding Section. Climatic time-averaged fields are rather devoid of finescale details, except in regions where fine-scale surface forcings are active (coasts, mountains, etc.). In homogeneous forcing areas, weather details are eventually wiped out by taking long enough time averages, leaving little trace of finescale features in the mean. Fine-scale information will however exist in other moments of the distribution; these moments are presumably the richest part of high-resolution modelling. It is interesting to raise the ante by verifying whether spatial phasing is respected in the time variation of fine-scale weather fields. The respect of this condition is sufficient (but not necessary) to respect phasing in seasonal mean statistics. In this Section and the next one, we deviate temporarily from climate statistics to use short-term predictability measures to verify the assertion of tenets that are more demanding than the previous ones. The question is whether the fine scales that develop are at the right geographical location and at the right time. We will investigate this tenet from the perspective of the accuracy of deterministic forecasts, which is more demanding than in a climate context. As for the previous tenet, this one too relates to the impact of nesting on the quality of simulated results, rather than to the intrinsic skill of the model. Hence this tenet is here too assessed in the context of the perfect prognosis approach, that is, without taking into account model errors. This is the only part of this paper in which the emphasis is on forecast skill; all other sections deal with climate. de Elı´a et al. (2002) studied the error growth in an idealised forecast experiment, analysing the skill of the LB forecast using the BB simulation as verification. Figure 6a shows the relative rootmean square error, normalised by the square root of the variance of the corresponding BB. Again the results are for 24 different cases in a month.

Challenging some tenets of Regional Climate Model

11

Fig. 6. Time evolution of the normalised 850-hPa vorticity error spectrum in a CRCM simulation for winter 1993: (a) initialised without fine scales, (b) initialised and nested with data containing fine scales (Reproduced with modifications from de Elı´a et al. (2002) # 2002 American Meteorological Society)

In terms of this non-dimensional error measure, length scales that are absent result in an error of 1 (this is the case at initial time for non-dimensional wavenumbers greater than about 8), and length scales with correct amplitude but random pffiffiffi phase (i.e. uncorrelated) result pffiffiffi in an error of 2. Figure 6a shows that the 2 limit is eventually reached for the shortest scales, as might have been expected from predictability theory and as is well documented to take place in global models. A distinguishing behaviour of RCMs compared to global models, however, is the limited growth rate of the larger scales. In nested models the error in the larger length scales appears to reach an asymptotic limit (which depends on wavenumber) well below the no-skill value of pffiffiffi 2 obtained with global models; this has been referred to as ‘‘extended predictability’’ by Anthes et al. (1989). This feature is of little practical application however since the scales that exhibit extended predictability are those that are prescribed as LBC. Figure 6a shows that the shorter scales that are not prescribed at the LBC, and hence that constitute the added value of nested models, do not benefit from extended predictability. In fact for most length scales the error grows in time from its initial value; an exception is a narrow band between wavenumbers 10 and 20 (corresponding to length scales of 360 and

180 km) where the error initially decreases before it eventually begins to increase like all other length scales. From these results, it would be tempting to conclude that the only scales to retain predictability skills are those that are supplied at the lateral boundary; however this turns out to be wrong, as we shall show next. Given that numerical weather prediction is by definition an initial-value problem, one can wonder to what extent the poor deterministic forecast skill displayed in Fig. 6a is due to the absence of small scales in the definition of the initial state, possibly compounded by the similar absence of small scales in the LBC. To address this issue de Elı´a et al. (2002) performed another experiment in which all scales of the BB simulation were retained in both initial and LBC of the LB; the results are presented in Fig. 6b. A small level of error was induced at initial time by performing a set of interpolations from model vertical coordinate to pressure, and back to model coordinate, in order to emulate the minimal error that is infallibly introduced when initialising a model from reanalyses. It can be seen on Fig. 6b that the errors initially increase with time, and that they asymptote to values that are close to those in the previous experiment (Fig. 6a), after about 4 days for this specific field. It can be noted that scales shorter than about 150 km have lost deter-

12

R. Laprise et al.

ministic predictability (in the sense of exceeding a non-dimensional value of unity) in less than a day and those shorter than about 250 km in less than two days. Hence despite the fact that model errors do not contribute to the growth of error in this BBE framework and that fine scales were included in the initial and LBC of the RCM, short scales do not benefit from the property of extended predictability. It is for this reason that high-resolution nested models are seldom employed for numerical weather prediction beyond a few days. From the deterministic forecast point of view, it would appear that Tenet 3 is not satisfied, except possibly for very short time scales when small scales are included in both initial and LBC. But of course such configuration would contradict the mission of dynamical downscaling. For most climate applications, violation of Tenet 3 may not matter as it is the statistics, not the specific sequence of weather events, that count. Failure of Tenet 3 however has implications for the validation of RCMs with short simulations and process studies. It is also indicative of the presence of internal variability, as will be discussed in the next sub-section. 2.4 Tenet 4: In performing dynamical downscaling, RCM generated small scales are uniquely defined for a given set of LBC RCMs have long been considered something like a sophisticated magnifying glass, unravelling fine scales that cannot be resolved on coarsemesh grids but that are ‘‘latent’’ in the large scales supplied as driving information. From low-resolution LBC, RCMs have been thought to generate sets of dynamically consistent highresolution fields: this is the essence of dynamical downscaling. This tenet in fact contains two related notions: developing small scales are uniquely defined for a given set of LBC, and large scales are unaffected by nested models. The first is discussed below and the second will be discussed in the following Section as part of a fifth tenet. Without stating it explicitly, some RCM users seem to expect that dynamically downscaled fields are uniquely defined for a given set of LBC, implying the absence of ‘‘internal variability’’ (IV) or spread between members in an en-

semble of simulations with identical external forcing. In a global model, IV should on average be equal to natural, transient-eddy variability. In an ensemble of nested RCMs’ simulations driven by identical LBC, the LBC conditions exert (some) control on the evolution of the interior solution, and hence the IV is expected to be smaller than natural variability, on average; but this argument is mute about its magnitude beyond this inequality. The fact that the RCMs’ small scales exhibit realistic climate statistics but not deterministic skill, as discussed in Tenet 3, hints to the existence of some degree of IV and chaotic behaviour in nested models simulations. The presence of IV in nested models simulations has been signalled by several authors (e.g., Giorgi and Bi 2000; Rinke and Dethloff 2000; Weisse et al. 2000; Christensen et al. 2001; de Elı´a et al. 2002; Caya and Biner 2004; Rinke et al. 2004; Wu et al. 2005; Alexandru et al. 2007; Vanvyve et al. 2007; Lucas-Picher et al. 2008; Separovic et al. 2008); but the concept does not seem fully grasped by a large part of the scientific community. In order to gain further insight into IV, Alexandru et al. (2007) studied the spread in an ensemble of twenty simulations with the CRCM for a period of three months, starting from initial conditions one-day apart, driven by NCEP reanalyses for summer 1993. Their 120 by 120 gridpoint computational domain covered the east coast of North America with a 45-km mesh. The top row of Fig. 7 shows the time evolution of the domain-average inter-member spread and the bottom row, the spatial distribution of the time-averaged spread, for precipitation and 850hPa geopotential. Most of the time during the three-month period, the spread between members remains small compared to natural variability. For example, the time-averaged 850-hPa geopotential spread in the middle of the domain only reaches about 4% of the square root of the transient-eddy variance, but larger spread occurs near the exit region of the domain. There are episodes when the spread is much larger than its time mean, such as in the second part of July 1993, when it reaches about 20% on a domainaverage basis, with even larger values locally (not shown). Inspection of the time evolution of IV points toward a mixture of triggering mechanisms, including parameterised diabatic effects such

Challenging some tenets of Regional Climate Model

13

Fig. 7. Time evolution of the domain-average inter-member spread (top row) and spatial distribution of the timeaverage inter-member spread (bottom row), for precipitation (left column, in mm da1 ) and 850-hPa geopotential (right column, in m). The inter-member spread is defined as the square root of the mean of quadratic differences between the individual members and the ensemble mean

as convection and condensation, and hydrodynamic instabilities in regions of strong temperature gradients. The study of Alexandru et al. (2007) also shows that the amplitude of IV generally increases with the size of the regional domain (not shown). de Elı´a et al. (2007) showed that the effects of IV are reduced when the fields are averaged in time (as when defining climate values): as expected the internal variability de-

creases as the averaging period lengthens. Nevertheless differences between the time average of two 20-year long simulations remains as large as 0.5 K for surface air temperature and 0.5 mm da1 for precipitation for specific regions of North America (not shown). Our current understanding of IV in nested models can be summarised by the schematic diagram on Fig. 8 showing the time evolution of

Fig. 8. Cartoon of the time evolution of spread between members of a conceptual ensemble of simulations driven by identical LBC

14

R. Laprise et al.

spread between members of a hypothetical ensemble of simulations driven by identical LBC. For a given season and configuration (geographical location, domain size, grid mesh) of a nested model, there appears to be a well-defined average level of IV, around which the spread varies, from ‘‘quiet’’ episodes with little IV to ‘‘active’’ episodes with larger IV. The average IV appears to be an intrinsic property of the time evolution in a nested model for a given configuration and season. As Giorgi and Bi (2000) and Caya and Biner (2004) have shown, IV is independent of the origin or magnitude of the initial kick used to induce separation of members. In order to gain further understanding of the IV, each field ’ðx; t; mÞ, with x denoting the horizontal coordinates, t the time and m the individual members, can be decomposed into its ensemble mean h’ðx; tÞi and departures thereof ’# ðx; t; mÞ as follows: ’ðx; t; mÞ ¼ h’ðx; tÞi þ ’# ðx; t; mÞ

ð1Þ

The ensemble mean will be referred to as the ‘‘forced’’ component (uniquely defined as a result of the prescribed sea-surface conditions and LBC) and the departures as the ‘‘free’’ compo-

Fig. 9. Spectra of the 925-hPa geopotential for a 20-member ensemble simulation of CRCM. The figure also shows the corresponding spectrum for the NCEP data used to drive the 45-km mesh simulations

nent. The word ‘‘free’’ is taken here rather loosely; it does not imply that perturbations are independent of the forcing, but that the forcing is not sufficient to constrain their evolution in a deterministic fashion. Next the spectra of ’, h’i and ’# are calculated with the DCT as in Denis et al. (2002a), and the corresponding time-averaged spectral amplitudes evaluated by t

t

t

h2’ i ðkÞ ¼ 2h’i ðkÞ þ h2’# i ðkÞ t

ð2Þ

where denotes time averaging and k the wavenumber. Separovic et al. (2008) performed a spectral decomposition of the 20-member ensemble simulations of Alexandru et al. (2007). One of their results is presented on Fig. 9 for the 925-hPa geopotential. The figure also shows the corresponding spectrum for the NCEP data used to drive the simulations. It can be seen that the spectral amplitudes of the NCEP fields are deficient beyond length scales of about 1100 km (corresponding to non-dimensional wavenumber 4); in principle the amplitude should vanish beyond this scale, but the interpolation from spherical coordinates to polar-stereographic projection compounded by some Gibbs’ effects associated to the DCT produces some noise resulting in non-vanishing variance. By comparison the spectra of CRCM members extend to larger wavenumbers (shorter scales) than the NCEP fields. The spectrum of the ensemble mean however has less variance at short scales than individual simulation members; this stems from the larger departures of individual simulations from the ensemble mean at short scales, resulting in destructive interference between the free components. Nevertheless the spectrum of the ensemble mean of the CRCM simulations extends to shorter scales than the NCEP data used to drive the nested model. Hence part of the added value of RCM simulations is contained at intermediate scales in the ensemble mean, i.e. the so-called forced component. Beyond scales of about 250 km (corresponding to non-dimensional wavenumber 18), the free component dominates over the forced one on average. Separovic et al. (2008) found that the transition length scale varies in time, moving to larger length scales during episodes of intense IV. The part of the added value of RCMs’ simulations that is contained in the small length scales where IV is largest must be treated with a probabilistic ap-

Challenging some tenets of Regional Climate Model

proach using ensembles. The free component in an ensemble can be used to define probability distribution functions (PDF) of events in climate simulations (e.g., de Elı´a and Laprise 2003), similarly to what is done in medium-range weather forecasting and seasonal-to-interannual prediction. The forced and free components defined here on the basis on an ensemble appear to correspond respectively to the predictable and unpredictable components in the initial-value problem presented by de Elı´a et al. (2002). The scales subject to large IV in regional climate applications appear to be those bounded by deterministic predictability limits in forecast applications. For climate applications, however, deterministic predictable skill is not required. Ensemble simulations can be used to sample adequately the statistical distribution of those fine scales subject to IV. These fine scales contribute to the added value of high-resolution RCM simulations; they appear not so much in the time-averaged climate but in the time variation, which may be quantified for example in terms of transient-eddy variance. Clearly Tenet 4 is not verified, as the small scales that are generated are not uniquely defined by the driving boundary conditions. This has major consequences in case or process studies, or in forecasting, but much less in climate applications.

15

2.5 Tenet 5: The large scales are (a) unaffected, (b) improved or (c) degraded within the RCM domain Since the purported aim of RCMs is to perform dynamical downscaling, that is to add valuable fine-scale details to, but to maintain, the largescale flow used to drive the nested model, one would not expect large-scale features to be significantly modified by a regional model (Tenet 5a). But there is a possibility that LBC errors could be either partly corrected (Tenet 5b) or magnified (Tenet 5c) by a nested model. Within the BBE framework an RCM appears to be ‘‘neutral’’ to the driving large scales, reproducing them without noticeable change (Denis et al. 2002b, 2003; de Elı´a et al. 2002; Antic et al. 2005; Dimitrijevic and Laprise 2005), thus verifying Tenet 5a. This result based on the idealised BBE framework is a necessary but not sufficient test for RCMs to pass in order to be considered useful dynamical downscaling tools of GCM projections. In dynamical downscaling studies for climate-change projections and for seasonal prediction, the errors of RCM and GCM combine to the overall uncertainty. A systematic error of an RCM would add to the uncertainties in the GCMs, and the GCM errors directly affect the RCM lateral boundary conditions. A key issue in Regional Climate Modelling relates to the

Fig. 10. Flow chart of an ‘‘Imperfect Big-Brother Experiment’’. Controlled errors are introduced in the Imperfect Big Brother used to drive the Little Brother in order to study the impact of lateral boundary errors on the climate simulated by the Little Brother. Verification is made against the perfect Big Brother simulation

16

R. Laprise et al.

Fig. 11. Error in stationary (a) and transient (b) large-scale components of Little Brother (LB) simulations, displayed as a function of corresponding error in Imperfect Big Brother (IBB) simulations used to drive the LB. The domain-mean error is computed from the spatial pattern of time mean (a) and transient-eddy variance (b), using the Perfect Big Brother (PBB) simulation as reference. The departure of the spatial correlation coefficient from 100% is used as measure of error (Reproduced from Diaconescu et al. (2007) # 2002 American Meteorological Society)

skill of the climate simulated by a high-resolution nested RCM driven by low-resolution, somewhat imperfect, GCM-simulated large-scale data. In order to investigate the sensitivity of nested models to large-scale errors in the LBC used to drive them, a variant on the BBE has been designed by Diaconescu et al. (2007): the ‘‘Imperfect Big-Brother Experiment’’ (IBBE, Fig. 10). In the IBBE protocol, the LBC are obtained from simulations of a coarse-resolution RCM integrated over different domains; hence by varying resolution and domain size, controllable level of errors ensue, and these errors are dynamically coherent and are typical of those of coarse-mesh CGCMs. The IBBE’s parameters in Diaconescu et al. (2007) were chosen such as to produce errors of comparable amplitude to those of CGCMs; the idealised experimental framework may not represent all the types of CGCM errors and hence the conclusions will remain to be verified by further studies. Figure 11 shows a result of Diaconescu et al. (2007), obtained with the IBBE for the summer season over an Eastern North American domain on a LB domain with 100 by 100 grid points. It indicates that the large-scale errors present in the driving data are reproduced by the RCM almost with the same amplitude, without being reduced nor amplified, lending support to Tenet 5a.

The idealised IBBE study was made using domains with some 100 by 100 grid points. It has been documented that the control exerted by LBC on the RCM solution appears to vary with the size of the computational domain (e.g., Rinke and Dethloff 2000; Alexandru et al. 2007) as well as location and season (e.g., Caya and Biner 2004; Lucas-Picher et al. 2008). In some applications, the flow developing within the RCM domain episodically becomes incoherent with the driving LBC; such occurrence may (Jones et al. 1997, Alexandru et al. 2007, Separovic et al. 2008) or may not (Caya and Biner 2004) affect large-scale climate statistics, the former case invalidating Tenet 5a. When a coarse-resolution GCM is providing the LBC to a nested model, the large-scale information will inevitably contain errors. It is plausible that a high-resolution RCM may tend to produce a large-scale flow differing from that of the low-resolution global model. Indeed an RCM may resolve physical processes that are not resolvable by the coarse-grid of a GCM, but which may have an impact on the average climate (such as the ensemble effect of mesoscale circulations or the precipitation shadow downstream of elevated narrow topography), leading to better agreement with observations. But it remains to be demonstrated whether a localised

Challenging some tenets of Regional Climate Model

increase of resolution can really improve the large-scale flow, as purported by Tenet 5b. Mesinger et al. (2002) have argued that high-resolution models can do much more than downscale, they could actually ‘‘upscale’’, improving the large-scale flow when LBC contain errors. In numerical weather prediction experiments made at NCEP with the nested model ETA with LBC provided by the global model AVN, these authors have reported that the ETA improves the large scales during the first three days of forecast, lending support to Tenet 5b. These results were obtained using a large regional domain (11,500 km by 8,500 km), at least four times larger than is commonly used for RCMs. It is worth noting however that the recommended domain size for the North American Regional Climate Change Assessment Program (NARCCAP; http:==www. narccap.ucar.edu=; Mearns et al. 2005) has a surface area about half this size. A number of studies appear to point to a degradation of large scale within the regional domain, supporting Tenet 5c. For example Castro et al. (2005) noted that the large scales were attenuated in their simulation with the CSU RAMS RCM compared to the global reanalyses used as LBC. Separovic et al. (2008) also noted some reduction of the amplitude of the large scales in their simulations with CRCM driven by NCEP reanalyses, and Juang and Hong (2001) noted different strength of the upper jet simulated by different RCMs. No explanation has yet been offered to explain why such behaviour occurs in some nested models when driven by reanalyses, unlike the case with the idealised BBE in which case no systematic bias of the large scales is noted. It is worth noting that the idealised framework of this study – where the information driving the nested model ultimately comes from observations – ignores additional sources of uncertainty that would be present in practical applications such as downscaling of seasonal to interannual prediction or climate-change projections, where the information driving the nested model comes from an imperfect GCM. 2.6 Nudging of the large scales within the RCM domain Various ways of nudging large scales in the interior of the RCM domain have been proposed

17

(e.g., Biner et al. 2000; von Storch et al. 2000; Riette and Caya 2002; Miguez-Macho et al. 2004; Castro et al. 2005) as an alternative to the traditional driving of nested models at the LBC only. While nudging of the large scales is used operationally in few RCMs (e.g., Weisse et al. 2000; Laprise et al. 2003; Plummer et al. 2006), LBC driving remains the norm for most RCMs. Nudging of the large scales within the domain has been used to prevent the intermittent occurrence of large deviations of the large scales within the RCM domain from those used to drive the RCM. But it has been noted that it is effective in reducing and even preventing the occasional development of large differences at the exit region of the domain, downstream of the dominant flow. Nudging of the large scale also appears to be efficient in suppressing internal variability of the smaller scales, even if nudging is not applied to these scales; this is a topic of current investigations. Because nudging of the large scales ensures by design a good agreement of the large scales through the regional domain, Tenet 5 is not called in when large-scale nudging is applied. But the sheer fact that large-scale nudging has come to be considered an alternative strategy to LBC nesting signals that the large scales simulated by a nested model are not always identical to those driving it, which clearly questions the validity of Tenet 5a and even Tenet 5b. 3. Discussion and conclusions RCMs have been designed to downscale largescale fields, that is, add small scales upon the large scales that are used to drive RCMs at their lateral boundary. RCMs share a number of sources of model errors with global models, such as numerical approximations and limited resolution, parameterisation of subgrid-scale physical effects, limited description of geophysical fields defining the virtual Earth. But in addition nested models have specific sources of errors, such as limited-area computational domain, nesting technique, resolution jump between the RCM and its driving data, update frequency of the lateral boundary conditions (LBC), and imperfections in LBC data. Despite the enormous progress made by Regional Climate Models in the recent years, there

18

R. Laprise et al.

is still discernable scepticism in their regard in part of the scientific community (e.g., WGNE 1999, 2000; B€arring and Laprise 2005). Although its impact is difficult to quantify, there is a generalised impression for many that Regional Climate Modelling has been unable to really prove its worth. Part of this failure, we believe, results from an excessive eagerness to proceed to ‘‘useful’’ applications, investing most of the efforts in downscaling climate-change projections. Although a commendable effort, this should not be done to the detriment of basic research. It is true that basic research has less appeal for public funding than projects with applications of timely societal relevance. But at the same time, the evaluation of these projects rests with science managers who often remain to be convinced of the worthiness of Regional Climate Modelling. This paper attempted to provide some information on the current state of knowledge of the strengths and limitations of nested models when applied to climate simulations. In this paper, some tenets of nested Regional Climate Modelling have been verified, while others have been qualified. The results presented in this paper demonstrate the following properties of nested model simulations. Initialised and driven by data without small-scale information, nested models develop small-scale variance even in the absence of strong surface forcing, through nonlinear cascade of variance from large scales to small scales. In idealised experiments, it is seen that fine scales develop within a few days, and they have the right amplitude and the right climate statistics even if these small scales are absent in LBC. It has been shown however that the full spin-up of small scales within the regional domain requires rather large domains, particularly in the upper troposphere in mid-latitudes where the flow is strong. Within a few days in the simulation, variance spectra and forecast error spectra become independent of the time since the simulation’s initiation and of the presence or absence of small scales in the initial and LBC. In Big-Brother Experiments for mid-latitude domains with 100 by 100 grid points on a 45-km mesh, scales larger than about 800 km appear to retain extended predictability, while scales smaller than about 400 km loose predictability in a fashion similar to global models. For the small scales, the shorter

the length scale, the shorter the predictable time. Hence small scales are generated by nested models, but not with deterministic skill. Climate statistics of small scales appear to be skilful though, lending some confidence in the potential usefulness of RCMs in climate simulations and climate-change projections. Even when fine scales are present initially and in the LBC, they do not maintain deterministic temporal coherence (at the right place at the right time) beyond a day or so. Hence we are forced to conclude that part of the downscaling is not deterministic, i.e. not entirely determined by LBC. A non-deterministic, free component exists which is also reflected in the presence of some level of internal variability (IV) in nested models. IV affects forecasting by imposing a deterministic predictability time limit. It puts however severe limitations on the usefulness of single-season, single-simulation experiments, such as the earlier PIRCS experiment (Takle et al. 1999; Anderson et al. 2003;). Such IV is in addition to the differences between members of an ensemble of simulations that might results from changes in model parameters or model formulation. IV may not be a major problem for climate applications as far as low-order statistics are concerned, as long as the user of an RCM is aware of its presence in the interpretation of results. Ensemble simulations can in fact prove invaluable as a framework for providing probabilistic estimates (e.g., de Elı´a and Laprise 2003). The results presented in this paper have also confirmed that RCMs are quite sensitive to circulation errors present in the fields serving to drive them. Even in the case of an RCM driven by reanalyses, there may be inconsistencies relating to differences in physical parameterisations between the RCM and the analysis model. The inconsistency issue is even more critical for dynamical downscaling of GCM-simulated data in climate-change projections and seasonal predictions, which points for the need of good-quality GCMs simulations to drive RCMs for credible results. The results presented in this study were obtained with a specific model with its nesting technique; there is a need to confirm to what extent the conclusions that are reached also apply to other models and nesting techniques. For example Juang and Hong (2001) noted different sensitivity to domain size with different models.

Challenging some tenets of Regional Climate Model

Several issues of Regional Climate Modelling call for further investigation. The following is a partial list of remaining RCM issues, as we perceive them. There is a need to improve our knowledge of the sensitivity of RCM to a number of arbitrary parameters related to the nesting approach, such as domain size (both in terms of dimension and number of grid points), location of domain (climate regime, ventilation flow), spatial resolution jump, nesting time interval, nesting technique (traditional LBC Versus large-scale nudging). Little is known on the influence of the nesting technique on processes such as nonlinear scale interactions and the cascade of variance, the presence of artificial domain circulations, the ability to capture interannual variability, the impact of errors in LBC driving data and the ability of nested RCM to correct them, and for that matter the ability of nested model to maintain the integrity of large scales used to drive the RCM. What are the dynamical mechanisms responsible for the time variations of IV? What is the spectral signature of IV during episodes of high and low variability? What is the influence of LBC on the free component in IV? How is IV associated to drift between the large scales of the driving field and those of the RCM? Are there secondary effects to the application of the nesting technique of large-scale nudging, besides reducing or even suppressing IV? These are still open questions that should be topics of active investigation. We close this paper by soliciting interest of Regional Climate Modellers for a coordinated effort to address the remaining open questions, following idealised but strict protocols such as the Big-Brother Experiment (BBE) and ensemble simulations described in this paper. Such ‘‘super BBE ensemble’’ is required to confirm some preliminary results obtained to date with a specific RCM. The potential dependence of the results on the model used is a typical example of how difficult is to produce basic research in numerical modelling with widespread application. This is an additional reason why the kind of research presented in this article deserves a coordinated effort, which would allow covering adequately the vast number of remaining issues exceeding the capacity of any single group. There is also a pressing need to expand on existing diagnostic methodology to identify and extract the added value of RCMs.

19

Acknowledgements This research was achieved as part of the scientific research programmes of the Canadian Regional Climate Modelling (CRCM) and Canadian Climate Variability (CLIVAR) Networks. These Networks were funded by the Canadian Foundation for Climate and Atmospheric Sciences (CFCAS), the National Science and Engineering Research Council (NSERC) and the Ouranos Consortium on Regional Climatology and Adaptation to Climate Change. The authors thank Claude Desrochers and Mourad Labassi for maintaining an efficient and user-friendly local computing facility.

References Alexandru A, de Elı´a R, Laprise R (2007) Internal variability in regional climate downscaling at the seasonal time scale. Mon Wea Rev 135(9): 3221–38 Anderson CJ, Arritt RW, Takle ES, Pan Z, Gutowski WJ, da Silva Jr R, Caya D, Christensen JH, Luthi D, Gaertenr MA, Gallardo C, Giorgi F, Laprise R, Hong S-Y, Jones C, Juang H-MH, Katzfey JJ, McGregor JL, Lapenta WM, Larson JW, Taylor JA, Liston GE, Pielke RA Sr, Roads JO (2003) Hydrological processes in regional climate model Simulations of the central United States flood of June– July 1993. J Hydromet 4: 584–98 Annamalai H, Slingo JM, Sperber KR, Hodges K (1999) The mean evolution and variability of the Asian summer monsoon: comparison of ECMWF and NCEP-NCAR reanalyses. Mon Wea Rev 127(6): 1157–86 Anthes RA, Kuo Y-H, Benjamin SG, Li YF (1982) The evolution of the mesoscale environment of severe local storms: preliminary modeling results. Mon Wea Rev 110: 1187–213 Anthes RA, Kuo Y-H, Baumhefner DP, Errico RM, Bettge TW (1985) Predictability of mesoscale atmospheric motions. Adv Geophys 28b: 159–202 Anthes RA, Kuo Y-H, Hsie E-Y, Low-Nam S, Bettge TW (1989) Estimation of skill and uncertainty in regional numerical models. Quart J Roy Meteor Soc 115: 763–806 Antic S, Laprise R, Denis B, de Elı´a R (2005) Testing the downscaling ability of a one-way nested regional climate model in regions of complex topography. Clim Dyn 23: 473–93 Bailey DA, Lynch AH (2000) Development of an Antarctic regional climate system model: Part 2. Station validation and surface energy balance. J Climate 13: 1351–61 B€arring L, Laprise R (eds) (2005) High-resolution climate modelling: assessment, added value and applications. Extended Abstracts of a WMO=WCRP-sponsored regional-scale climate modelling Workshop, 29 March–2 April 2004, Lund (Sweden). Lund University electronic reports in physical geography, 132 pp (http:==www. nateko.lu.se=ELibrary=Lerpg=5=Lerpg5Article.pdf) Biner S, Caya D, Laprise R, Spacek L (2000) Nesting of RCMs by imposing large scales. In: Research Activities in Atmospheric and Oceanic Modelling, WMO=TD – No. 987, Report No. 30, pp. 7.3–7.4 Castro CL, Pielke RA Sr, Leoncini G (2005) Dynamical downscaling: an assessment of value added using a re-

20

R. Laprise et al.

gional climate model. J Geophys Res (Atmos) 110: D05108; DOI: 10.1029=2004JD004721 Caya D, Laprise R, Giguere M, Bergeron G, Blanchet JP, Stocks BJ, Boer GJ, McFarlane NA (1995) Description of the Canadian RCM. Water Air Soil Poll 82(1=2): 477–82 Caya D, Laprise R (1999) A semi-Lagrangian semi-implicit regional climate model: the Canadian RCM. Mon Wea Rev 127(3): 341–62 Caya D, Biner S (2004) Internal variability of RCM simulations over an annual cycle. Clim Dyn 22(1): 33–46 Christensen OB, Gaertner MA, Prego JA, Polcher J (2001) Internal variability of regional climate models. Clim Dyn 17: 875–87 Christensen JH, Carter TR, Rummukainen M (2007) Evaluating the performance and utility of regional climate models: the PRUDENCE project. Climatic Change; DOI: 10.1007=s10584-006-9211-6 Cocke S, LaRow TE (2000) Seasonal predictions using a regional spectral model embedded within a coupled ocean-atmosphere model. Mon Wea Rev 128: 689–708 Curry JA, Lynch AH (2002) Comparing Arctic regional climate models. EOS, Trans Amer Geophys Union 83: 87 de Elı´a R, Laprise R, Denis B (2002) Forecasting skill limits of nested, limited-area models: a perfect-model approach. Mon Wea Rev 130: 2006–23 de Elı´a R, Laprise R (2003) Distribution-oriented verification of limited-area models forecast in a perfect-model framework. Mon Wea Rev 131: 2492–509 de Elı´a R, Caya D, C^ ote H, Frigon A, Biner S, Giguere M, Paquin D, Harvey R, Plummer D (2007) Evaluation of uncertainties in the CRCM-simulated North American climate. Clim Dyn; DOI: 10.1007=s00382-007-0288-z Denis B, C^ote J, Laprise R (2002a) Spectral decomposition of two-dimensional atmospheric fields on limited-area domains using discrete cosine transforms (DFT). Mon Wea Rev 130(7): 1812–29 Denis B, Laprise R, Caya D, C^ ote J (2002b) Downscaling ability of one-way-nested regional climate models: the Big-Brother experiment. Clim Dyn 18: 627–46 Denis B, Laprise R, Caya D (2003) Sensitivity of a regional climate model to the spatial resolution and temporal updating frequency of the lateral boundary conditions. Clim Dyn 20: 107–26 Deque M, Rowell DP, L€ uthi D, Giorgi F, Christensen JH, Rockel B, Jacob D, Kjellstr€ om E, de Castro M, van den Hurk B (2006) An intercomparison of regional climate simulations for Europe: assessing uncertainties in model projections. Climate Change (in press) de Szoeke SP, Wang Y, Xie S-P, Miyama T (2006) Effect of shallow cumulus convection on the eastern Pacific climate in a coupled model. Geophys Res Lett 33: L17713; DOI: 10.1029=2006GL026715 Diaconescu EP, Laprise R, Sushama L (2007) The impact of lateral boundary data errors on the simulated climate of a nested Regional Climate Model. Clim Dyn 28(4): 333–50 Dickinson RE, Errico RM, Giorgi F, Bates GT (1989) A regional climate model for the western United States. Climatic Change 15: 383–422

Dimitrijevic M, Laprise R (2005) Validation of the nesting technique in an RCM and sensitivity tests to the resolution of the lateral boundary conditions during summer. Clim Dyn 25: 555–80 D€ oscher R, Willen U, Jones C, Rutgersson A, Meier HE, Hansson U, Graham PL (2002) The development of the coupled ocean-atmosphere model RCAO. Boreal Environ Res 7: 183–92 Errico RM (1985) Spectra computed from a limited area grid. Mon Wea Rev 113: 1554–62 Feser F, von Storch H (2006) A spatial two-dimensional discrete filter for limited area model evaluation purposes. Mon Wea Rev 133(6): 1774–86 Feser F (2006) Enhanced detectability of added value in limited area model results separated into different spatial scales. Mon Wea Rev 134(8): 2180–90 Fox-Rabinovitz MS, C^ ote J, Dugas B, Deque M, McGregor J, Gleckler P (2005) The international Stretched-Grid Model Intercomparison Project (SGMIP). Amer Meteor Soc (http:==ams.confex.com=ams=Annual2005=techprogram= paper_83463.htm) Frei C, Christensen JH, Deque M, Jacob D, Jones RG, Vidale PL (2003) Daily precipitation statistics in regional climate models: evaluation and intercomparison for the European Alps. J Geophys Res (Atmos) 108(D3): 4124; DOI: 10.1029=2002JD002287 Fu C, Wang S, Xiong Z, Gutowski WJ, Lee D-K, McGregor JL, Sato Y, Kato H, Kim J-W, Suh M-S (2005) Regional climate model intercomparison project for Asia. Bull Amer Meteor Soc 86(2): 257–66 Gibelin AL, Deque M (2003) Anthropogenic climate change over the Mediterranean region simulated by a global variable resolution model. Clim Dyn 20: 327–39 Giorgi F, Bates GT (1989) The climatological skill of a regional model over complex terrain. Mon Wea Rev 117: 2325–47 Giorgi F, Bi X (2000) A study of internal variability of a regional climate model. J Geophys Res 105: 29503–21 Giorgi F, Mearns LO (1991) Approaches to the simulation of regional climate change: a review. Rev Geophys 29: 191–216 Herceg D, Sobel AH, Sun L, Zebiak SE (2006) The big brother experiment and seasonal predictability in the NCEP regional spectral model. Clim Dyn 26(4): 1–14 (http:==dx.doi.org=10.1007=s00382-006-0130-z) IPCC (2007) Climate change 2007: the physical science basis. In: Solomon S, Qin D, Manning M, Chen Z, Marquis M, Averyt KB, Tignor M, Miller HL (eds) Contribution of the Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge and New York Jones RG, Murphy JM, Noguer M (1995) Simulation of climate change over Europe using a nested regionalclimate model. I: Assessment of control climate, including sensitivity to location of lateral boundaries. Quart J Roy Meteor Soc 121: 1413–49 Jones RG, Murphy JM, Noguer M, Keen AB (1997) Simulation of climate change over Europe using a nested regional climate model. Part II: Comparison of driving

Challenging some tenets of Regional Climate Model and regional model responses to a doubling of carbon dioxide. Quart J Roy Met Soc 123: 265–92 Juang H-MH, Hong S-Y (2001) Sensitivity of the NCEP regional spectral model to domain size and nesting strategy. Mon Wea Rev 129: 2904–22 Kuo H-C, Williams RT (1992) Boundary effects in regional spectral models. Mon Wea Rev 120: 2986–92 Kuo H-C, Williams RT (1998) Scale-dependent accuracy in regional spectral models. Mon Wea Rev 126: 2640–7 Laprise R, Caya D, Giguere M, Bergeron G, C^ ote H, Blanchet J-P, Boer GJ, McFarlane NA (1998) Climate and climate change in Western Canada as simulated by the Canadian regional climate model. Atmos Ocean XXXVI(2): 119–67 Laprise R, RaviVarma M, Denis B, Caya D, Zawadzki I (2000) Predictability in a nested limited-area model. Mon Wea Rev 128(12): 4149–54 Laprise R, Jones R, Kirtman B, von Storch H, Wergen W (2002) Atmospheric regional climate models (RCMs): a multiple purpose tool? Report of the Joint WGNE= WGCM ad hoc panel on regional climate modelling, 19 pp (available from the corresponding author) Laprise R, Caya D, Frigon A, Paquin D (2003) Current and perturbed climate as simulated by the second-generation Canadian Regional Climate Model (CRCM-II) over northwestern North America. Clim Dyn 21: 405–21 Laprise R (2003) Resolved scales and nonlinear interactions in limited-area models. J Atmos Sci 60(5): 768–79 Laprise R (2008) Regional climate modelling. J Comp Phys, 227, Special issue on Predicting Weather, Climate and Extreme Events (by invitation), 3641–3666 Leduc M, Laprise R (2008) Regional Climate Model sensitivity to domain size. Clim Dyn (accepted) Lorenz P, Jacob D (2005) Influence of regional scale information on the global circulation: a two-way nesting climate simulation. Geophys Res Lett 32: L18706; DOI: 10.1029=2005GLO23351 Lucas-Picher P, Caya D, de Elı´a R, Laprise R (2008) Investigation of regional climate models’ internal variability with a ten-member ensemble of ten years over a large domain. Clim Dyn (accepted); DOI: 10.1007= s00382-003-0384-8 McGregor JL (1997) Regional climate modelling. Meteorol Atmos Phys 63: 105–17 McGregor JL, Nguyen KC, Katzfey JJ (2002) Regional climate simulations using a stretched-grid global model. In: Ritchie H (ed) Research activities in atmospheric and oceanic modelling. Report No. 32, WMO=TD – No. 1105, pp. 3.15–16 Mearns LO, Arritt R, Boer G, Caya D, Duffy P, Giorgi F, Gutowski WJ, Held IM, Jones R, Laprise R, Leung LR, Pal J, Roads R, Sloan L, Stouffer R, Takle G, Washington W (2005) NARCCAP – North American Regional Climate Change Assessment Program: A Multiple AOGCM and RCM Climate Scenario Project over North America. Preprints of the Amer. Meteor. Soc. 16th Conf. on Climate Variations and Change. 9–13 January 2005. Paper J6.10, pp. 235–8 Meehl GA, Boer GJ, Covey C, Latif M, Stouffer RJ (2000) The coupled model intercomparison project (CMIP). Bull Amer Meteor Soc 81(2): 313–18

21

Mesinger F, Brill K, Chuang H, DiMego G, Rogers E (2002) Limited area predictability: can upscaling also take place? Research activities in atmospheric and oceanic modelling. Report No. 32, WMO=TD – No. 1105, pp. 5.30–1 Miguez-Macho G, Stenchikov GL, Robock A (2004) Spectral nudging to eliminate the effects of domain position and geometry in regional climate model simulations. J Geophys Res 109(D13): D13104; DOI: 10.1029=2003JD004495 Miyakoda K, Rosati A (1977) One-way nested grid models: the interface conditions and the numerical accuracy. Mon Wea Rev 105: 1092–107 New M, Hulme M, Jones P (2000) Representing twentiethcentury space-time climate variability. Part II: Development of 1901–1996 monthly grids of terrestrial surface climate. J Clim 13: 2217–38 Nutter P, Stensrud D, Xue M (2004) Effects of coarsely resolved and temporally interpolated lateral boundary conditions on the dispersion of limited-area ensemble forecasts. Mon Wea Rev 132(10): 2358–77 Riette S, Caya D (2002) Sensitivity of short simulations to the various parameters in the new CRCM spectral nudging. In: Ritchie H (ed) Research activities in atmospheric and oceanic modelling. WMO=TD – No 1105, Report No. 32, pp. 7.39–40 Rinke A, Dethloff K (2000) On the sensitivity of a regional Arctic climate model to initial and boundary conditions. Clim Res 14(2): 101–13 Rinke A, Gerdes R, Dethloff K, Kandlbinder T, Karcher M, Kauker F, Frickenhaus S, K€ oberle C, Hiller W (2003) A case study of the anomalous Arctic sea ice conditions during 1990: insights from coupled and uncoupled regional climate model simulations. J Geophys Res 108(D9): 4275; DOI: 10.1029=2002JD003146 Rinke A, Marbaix P, Dethloff K (2004) Internal variability in Arctic regional climate simulations: case study for the SHEBA year. Clim Res 27: 197–209 Rinke A, Dethloff K, Cassano JJ, Christensen JH, Curry JA, Du P, Girard E, Haugen J-E, Jacob D, Jones CG, Køltzow M, Laprise R, Lynch AH, Pfeifer S, Serreze MC, Shaw MJ, Tjernstr€ om, M, Wyser K, Zagar M (2005) Evaluation of an ensemble of Arctic regional climate models: spatiotemporal fields during the SHEBA year. Clim Dyn 26(5): 459–72; DOI: 10.1007=s00382-005-0095-3 Sasaki H, Kurihara K, Takayabu I, Murazaki K, Sato Y, Tsujino H (2006) Preliminary results from the coupled atmosphere-ocean regional climate model developed at Meteorological Research Institute. J Meteor Soc Japan 84: 389–403 Segami A, Kurihara K, Nakamura H, Ueno M, Takano I, Tatsumi Y (1989) Operational mesoscale weather prediction with Japan spectral model. J Meteor Soc Japan 67: 907–23 Seo KH, Schemm JKE, Wang W, Kumar A (2007) The boreal summer intraseasonal oscillation simulated in the NCEP climate forecast system: the effect of sea surface temperature. Mon Wea Rev 135: 1807–27 Separovic L, de Elı´a R, Laprise R (2008) Reproducible and irreproducible components in ensemble simulations of a regional climate model. Clim Dyn (accepted)

22

R. Laprise et al.: Challenging some tenets of Regional Climate Model

Staniforth A (1997) Regional modelling: a theoretical discussion. Meteorol Atmos Phys 63(1–2): 15–29 Stowasser M, Wang Y, Hamilton K (2007) Tropical cyclone changes in the Western North Pacific in a global warming scenario. J Climate 20: 2378–96 Takle ES, Gutowski WJ Jr, Arritt RW, Pan Z, Anderson CJ, Silva R, Caya D, Chen S-C, Christensen JH, Hong S-Y, Juang H-MH, Katzfey JJ, Lapenta WM, Laprise R, Lopez P, McGregor J, Roads JO (1999) Project to intercompare regional climate simulations (PIRCS): description and initial results. J Geophys Res 104: 19, 443–62 Tatsumi Y (1986) A spectral limited-area model with time dependent lateral boundary conditions and its application to a multi-level primitive equation model. J Meteor Soc Japan 64: 637–63 Vanvyve E, Hall N, Messager C, Leroux S, van Ypersele J-P (2007) Internal variability in a regional model over West Africa. Clim Dyn; DOI: 10.1007=s00382-007-0281-6 von Storch H, Langenberg H, Feser F (2000) A spectral nudging technique for dynamical downscaling purposes. Mon Wea Rev 128: 3664–73 Wang Y, Sen OL, Wang B (2003) A highly resolved regional climate model (IPRC-RegCM) and its simulation of the 1998 severe precipitation event over China. Part I: Model description and verification of simulation. J Climate 16: 1721–38 Wang Y, Leung LR, McGregor JL, Lee D-K, Wang W-C, Ding Y, Kimura F (2004) Regional climate modelling: progress, challenges, and prospects. J Meteor Soc Japan 82(6): 1599–628

Warner TT, Peterson RA, Treadon RE (1997) A tutorial on lateral conditions as a basic and potentially serious limitation to regional numerical weather prediction. Bull Amer Meteor Soc 78(11): 2599–617 WGNE (1999) Report of Fourteenth Session of the CAS=JSC Working Group on Numerical Experimentation (Recherche en Prevision Numerique, Environment Canada, Dorval Quebec, Canada, 2–6 November 1998), Report No. 14, WMO=TD – No. 964, World Meteor. Org., 28 pp WGNE (2000) Report of Fifteenth Session of the CAS=JSC Working Group on Numerical Experimentation (Naval Research Laboratory, Monterey, CA, USA, 25–29 October 1999), Report No. 15, WMO=TD – No. 1024, World Meteor. Org., 29 pp Weisse R, Heyen H, von Storch H (2000) Sensitivity of a regional atmospheric model to a sea state-dependent roughness and the need for ensemble calculations. Mon Wea Rev 128(10): 3631–42 Wu W, Lynch AH, Rivers A (2005) Estimating the uncertainty in a regional climate model related to initial and lateral boundary conditions. J Climate 18(7): 917–33 Xie SP, Miyama T, Wang Y, Xu H, de Szoeke SP, Small RJO, Richards KJ, Mochizuki T, Awaji T (2007) A regional ocean-atmosphere model for eastern pacific climate: toward reducing tropical biases. J Climate 20: 1504–22 Yeh KS, C^ ote J, Gravel S, Methot A, Patoine A, Roch M, Staniforth A (2002) The CMC–MRB global environmental multiscale (GEM) model. Part III: Nonhydrostatic formulation. Mon Wea Rev 130(2): 339–56