Challenges in Regional-Scale Climate Modeling

Raymond W. Arritt, Markku Rummukainen
2011 Bulletin of The American Meteorological Society - (BAMS)  
CHALLENGES IN REGIONAL-SCALE CLIMATE MODELING by raymond W. arritt and markku rummukainen terrain. There is also added value associated with higher statistical moments and with extremes rather than mere spatial and temporal means. Added value also depends on spatial scale. Filtering or spatial averaging can improve the agreement of model results with climatological observations, which implies that there is a trade-off between accuracy and spatial detail. How to make this trade-off in terms of
more » ... e-off in terms of specifying a scale of filtering or averaging depends both on the variable of interest and on the location being considered. ENSEMBLE METHODS. Since the first Lund workshop there has been increasing use of ensemble approaches for regional climate modeling. Participants discussed how best to take advantage of ensembles, including how to weight individual ensemble members and measures for establishing ensemble skill. A consensus has emerged that single-model ensembles tend to be underdispersive and that multiple models are needed to better sample the uncertainty space. A comparison of ensembles with different numbers of models shows that most of the additional value of multimodel ensembles comes from the first few P resent global climate models (GCMs) do not represent local-and regional-scale processes that are important to key impacts of climate variability and change. Therefore, adaptation efforts that need local-to regional-scale climate information require tools such as regional climate models (RCMs) to complement global model projections. This workshop 1 created an opportunity to take stock of recent advances in regional climate modeling techniques and discuss how the community can better meet the growing demand for regional-scale climate information. Participants also discussed ways to improve the interaction of regional-scale climate modeling with other disciplines (including the global modeling, observation, and climate change impacts communities) and with stakeholders. V A L U E O F R E G I O N A L C L I M AT E MODELING. Given the cost of regional climate modeling in computational resources and human effort, the question arises whether this cost produces added value. Presentations showed that the answer depends on the circumstances at hand and, in particular, on location. The added value of RCMs tends to be associated with physiographic detail that cannot be resolved in global models, such as coasts or complex 1 The Second Lund Regional-Scale Climate Modelling Workshop was a follow-up to the first such workshop held in Lund, Sweden, in 2004 (Bärring and Laprise 2005) . This workshop gathered a total of about 220 participants from 43 countries, more than twice as many participants as in the 2004 workshop. There were 170 papers on seven broad topics (Rockel et al. 2009), reflecting the maturation of regional climate modeling as a field of research. 365 march 2011 amErIcaN mETEOrOLOGIcaL SOcIETY | models. One presentation showed that in an ensemble of RCMs driven by an ensemble of GCMs the uncertainty for seasonal means was dominated by the choice of GCM, indicating that it is important to use multiple GCMs in climate change downscaling studies. Workshop participants extensively discussed methods for weighting the members of an ensemble. Results presented at the workshop showed that incorrectly weighting the ensemble members can produce much worse results than a simple equally weighted ensemble mean, and that weighting narrows the distribution of outcomes. Optimized weighting schemes (as compared to equal weighting) are most suitable when the models are very different. There may be a need to determine optimum weights for individual grid points rather than for the entire domain. An overarching concern is that optimal weights for performance of ensembles in reproducing current or past climates may not be the optimal weights for future climates. APPLICATIONS AND CLIMATE IMPACTS ASSESSMENT. Several presentations discussed climate impacts and other uses for RCMs. Requirements differ markedly for different applications; some can make use of monthly data at fairly coarse resolution, while others need daily or subdaily data at spatial resolution of the order of 1 km. In such cases RCMs provide a means to produce physically consistent data to drive process-based models of climate impacts. Applications of RCMs often need adjustment for model biases. One presentation showed that temperature biases are not linear but tend to be higher for warmer temperatures. This suggests that biases may be greater in warming climates than at present. Bias correction also tends to reduce model spread. D E V E L O P M E N T S I N N U M E R I C A L METHODS AND PHYSICAL PARAMETER-IZATIONS. Several presentations discussed numerical approaches to regional climate modeling other than the usual nesting of a high-resolution limited-area model within global model data, such as global models with fine resolution in specific target regions or adaptive grid techniques. Much attention was also given to the use of spectral nudging as a method to enforce RCM consistency with large-scale boundary conditions. Differences in approach to spectral nudging reflect two schools of thought: one being that the RCM should add small-scale detail but otherwise adhere to its large-scale driving data, and the other being that the RCM should be allowed to influence larger scales (as through upscale growth). The optimal scale of spectral nudging is not yet well defined and likely depends on the purpose. Some general characteristics are known; for example, it was shown that if nudging is too weak the RCM can still become inconsistent with its large-scale driving data, while if nudging is too strong the RCM solution is overconstrained and development of fine-scale information can be suppressed. RCMs now include representations of a wide range of physical processes, such as tiled land surface models, detailed treatment of ocean waves, chemical transformations, aerosol microphysics, and urban effects. An issue raised in several presentations was the interaction of physical parameterizations with numerical aspects of RCMs, such as the resolution dependence of physical parameterizations and the interplay of microphysics with nonhydrostatic dynamics. Similar problems arise in high-resolution numerical weather prediction (NWP), so that greater collaboration between the RCM and NWP communities would be mutually beneficial. COLLABORATIVE PROJECTS. The past five years have seen the development of several large projects that include collaboration on regionalscale climate modeling by several groups, such as ENSEMBLES (see sidebar for project acronym expansions; Europe), NARCCAP and MRED (North America), CLARIS-LPB (South America), CIRCE (Mediterranean region), the S-5-3 project (Japan), and ICTS (numerous regions). [See Takle et al. (2007) for an annotated list of collaborative regional modeling projects.] Coordinated projects in regional climate modeling have evolved from an early focus on model intercomparison toward greater emphasis on ensemble applications with more advanced analysis of probabilistic regional climate change projections. They also have contributed to model development efforts. Some of these projects have been conducted in an end-to-end research setting, | involving collaboration of global and regional climate modeling, observations and impact studies. The projects also have begun using standard data formats and protocols, with NARCCAP, MRED, and CORDEX using a common format based on that used by global models in the Coupled Model Intercomparison Project, phase 3/Intergovernmental Panel on Climate Change Fourth Assessment Report (CMIP3/ IPCC AR4) data archive. CHALLENGES AND OUTLOOK. Future challenges for regional-scale climate modeling include the creation of regional Earth system models by incorporating coupled ocean models, dynamic vegetation, biogeochemical cycles, and other components. Another is the use of very high resolution RCMs, that is, 10 km or finer. Such high resolution requires careful attention both to numerical aspects (such as consistency with driving data) and to physical parameterizations appropriate at fine resolution. It also places new demands on evaluation data derived from available observations. The regional climate modeling community has not yet developed the same level of coordination as the global modeling community. For example, there have not been true counterparts to such coordinated global climate modeling experiments as Atmospheric Model Intercomparison Project (AMIP; Gates et al. 1999) and CMIP (Meehl et al. 2000). Coordinated regional-scale climate modeling experiments have been conducted for some regions, although with smaller international participation than in coordinated global modeling experiments. Among the reasons are the inherent focus of regional modeling on specific locales and the difficulty in obtaining GCM results suitable for use as boundary conditions. Regional climate modeling also is computationally demanding, which can prohibit groups from addressing more than their "home" region. Recently, coordinated experiments covering multiple regions have begun to be explored. The Inter-CSE Transferability Study (ICTS; Rockel et al. 2006 ) used an ensemble of models to simulate seven regions around the world. The new collaborative experiment CORDEX (Giorgi et al. 2009) will include simulations for numerous regions covering most of the inhabited parts of the globe and thus should spur cooperation among regional modeling groups. SUMMARY. The development of regional-scale climate modeling has been especially lively during the past five years. The community has grown into a global one, techniques have become more versatile, and applications more wide ranging. The models themselves have improved through advances in theoretical understanding, numerical methods, and parameterization of important processes, aided by links with global climate modeling and NWP communities. Progress has also been made in model evaluation, including model intercomparison and ensemblebased methods, and in links between regional climate modeling and applications. Despite these advances the utility of regional climate modeling has not been fully exploited. For example, most of the regional-scale information in the latest IPCC assessment report (AR4; Solomon et al. 2007 ) was taken from global rather than regional-scale climate modeling. Closer links of the regional climate modeling community with the global climate modeling and NWP communities, other research disciplines, and those needing regional-scale information for climate impacts assessments and other applications are necessary if regional climate modeling is to realize its potential.
doi:10.1175/2010bams2971.1 fatcat:c3czwiyxhvbv5eoqm4gdbfrdpq