Assessment of three dynamical climate downscaling methodsusing the Weather Research and Forecasting (WRF) modelJeff Chun-Fung Lo,1Zong-Liang Yang,1and Roger A. Pielke Sr.2Received 24 July 2007; revised 17 November 2007; accepted 7 January 2008; published 10 May 2008.[1] The common methodology in dynamical regional climate downscaling employsa continuous integration of a limited-area model with a single initialization of theatmospheric fields and frequent updates of lateral boundary conditions based ongeneral circulation model outputs or reanalysis data sets. This study suggestsalternative methods that can be more skillful than the traditional one in obtaininghigh-resolution climate information. We use the Weather Research and Forecasting(WRF) model with a grid spacing at 36 km over the conterminous U.S. to dynamicallydownscale the 1-degree NCEP Global Final Analysis (FNL). We perform three typesof experiments for the entire year of 2000: (1) continuous integrations with a singleinitialization as usually done, (2) consecutive integrations with frequent re-initializations,and (3) as (1) but with a 3-D nudging being applied. The simulations are evaluatedin a high temporal scale (6-hourly) by comparison with the 32-km NCEP North AmericanRegional Reanalysis (NARR). Compared to NARR, the downscaling simulation usingthe 3-D nudging shows the highest skill, and the continuous run produces the lowestskill. While the re-initialization runs give an intermediate skill, a run with a more frequent(e.g., weekly) re-initialization outperforms that with the less frequent re-initialization(e.g., monthly). Dynamical downscaling outperforms bi-linear interpolation, especially formeteorological fields near the surface over the mountainous regions. The 3-D nudginggenerates realistic regional-scale patterns that are not resolved by simply updatingthe lateral boundary conditions as done traditionally, therefore significantly improvingthe accuracy of generating regional climate information.Citation: Lo, J. C.-F., Z.-L. Yang, and R. A. Pielke Sr. (2008), Assessment of three dynamical climate downscaling methods usingthe Weather Research and Forecasting (WRF) model, J. Geophys. Res., 113, D09112, doi:10.1029/2007JD009216.1. Introduction[2] Coupled atmosphere-ocean general circulation models(AOGCMs) are numerical tools that are used to performclimate simulations and predict future climate change[IPCC, 2007], although skill at this spatial scale has notyet been achieved [Trenberth, 2007]. Because of limitationsin computational resources, existing AOGCMs typicallyrun at horizontal grid intervals on the order of 200 km,which is far too coarse for applications at regional or localscale regimes at scales of 10 –50 km [Leung et al., 2003;Wang et al., 2004; Giorgi, 2006]. Thus the nested regionalclimate modeling technique, also referred to as dynamicaldownscaling, was developed to mitigate this problem, andcurrently it has become a common approach to obtain high-resolution regional climate information from A OGCMs[Giorgi, 2006].[3] The starting point of dynamical downscaling istypically a set of coarse-resolution large-scale fields eitherfrom AOGCMs or from global reanalysis, which are usedto provide the initial (ICs), and lateral meteorological andsurface boundary conditions (LBCs) to the nested regionalclimate model (RCM). The RCM is not intended to modify/correct the large-scale circulation of the AOGCM but isintended to add regional detail in response to regionalscale forcing (e.g., topography, coastlines, and land use/land cover) as it interacts with the larger-scale atmosphericcirculations [Giorgi, 2006]. The purpose of downscaling isto obtain high-resolution detail as accurately as possibleover the region of interest.[4] During the past 20 years, the approaches to thesimulation in nested RCM, along with their value-added,have often been debated. Castro et al. [2005], for example,concluded that the RCM cannot add skill to simulations oflarge-scale weather features beyond what is already in theparent global model or reanalysis, since the RCM is sostrongly influenced by the parent model or r eanaly sis.B. Rockel et al. (Dynamical downscalling: Assessment ofmodel system dependent retained and added variability fortwo different regional climate models, submitted to Journalof Geophysical Research, 2008) using a separate modelJOURNAL OF GEOPHYSICAL RESEARCH, VOL. 113, D09112, doi:10.1029/2007JD009216, 2008ClickHereforFullArticle1Department of Geological Sciences, The Ja ckson School ofGeosciences, The University of Texas at Austin, Austin, Texas, USA.2Cooperative Institute for Research in Environmental Sciences,University of Colorado, Boulder, Colorado, USA.Copyright 2008 by the American Geophysical Union.0148-0227/08/2007JD009216$09.00D09112 1of16from that of Castro et al. [2005] confirmed this conclusion.The RCM simulated climate is necessarily strongly domi-nated by the parent model or reanalysis.[5] Castro et al. [2005] also identified three types ofregional climate modeling (defined as when the ICs areforgotten) in which skill becomes progressively moredifficult as the parent global input goes from a reanalysis,to a global model with some aspects of the system prescribed(e.g., sea surface temperature) to a global prediction model inwhich all aspects of the climate system are predicted.[6] From the experience of numerical weather prediction(NWP), the skill of limited area models diminish veryrapidly with time, becoming useless for the simulationperiod beyond a week. Therefore the very first regionalclimate simulation was conducted in a short-term re-initial-ized weather forecasting mode [Dickinson et al., 1989]. Theregional climatology was obtained from the statistics ofmultiple short runs. Over the years, significant developmenthas been achieved in the representation of physical processesin RCM, e.g., atmospheric radiation, cloud microphysics,shallow and deep convection, and land-surface processes[e.g., Leung and Ghan, 1998; Liang et al., 2001], to improvethe representation of these processes within RCMs in longsimulations. The simulation approach of the RCM switchesfrom re-initialization mode into a long-term continuousclimate prediction mode, and the simulation length ismuch longer than few days as is the case in weatherforecasting mode. Previous studies with nested RCMs havebeen run for integration times from a monthlong [Giorgiand Bates, 1989; Giorgi, 1990] to multidecadal simulations[Machenhauer et