FORECASTER’S FORUM Toward Improved Prediction: High-Resolution and Ensemble Modeling Systems in Operations

A large gap in skill between forecasts of the atmospheric circulation (relatively high skill) and quantitative precipitation (low skill) has emerged over the past three decades. One common approach toward closing this gap has been to try to simulate precipitation features directly by decreasing the horizontal grid spacing of the numerical weather prediction models. Also at this time, research has begun to explore the benefits of shortrange ensemble forecast methods. The authors argue that each approach has benefits: high-resolution models assist in the development of a forecaster’s conceptual model of various mesoscale phenomena, whereas ensembles help quantify forecast uncertainty. A thoughtful implementation of both approaches, in which this complementary nature is recognized, will improve the forecast process, empower human forecasters, and consequently add value relative to current trends. The science and policy issues that must be addressed in order to maximize this forecast potential are discussed. 1. The forecast skill gap Because of the relative lack of skill in forecasts of precipitation [see the National Center for Atmospheric Research (NCAR) Web site for recent verification statistics for its models online at http://sgi62.wwb.noaa.gov: 808/STATS/STATS.html (height and wind) and at http: www.hpc.ncep.noaa.gov/html/hpcverif.html (precipitation)], both the U.S. Weather Research Program (Fritsch et al. 1998) and the National Research Council Board on Atmospheric Sciences and Climate (National Research Council 1998, p. 174) have declared a principal research goal to be improved forecasts of precipitation occurrence Corresponding author address: Paul J. Roebber, Department of Mathematical Sciences, University of Wisconsin—Milwaukee, 3200 N. Cramer Ave., Milwaukee, WI 53211. E-mail: [email protected] and amount. Numerical weather prediction (NWP) is the main vehicle for forecasting at mesoscale spatial and temporal scales. Improving NWP forecasts can involve increasing the amount of data and advancing data assimilation, improving model physical parameterizations, resolution, and model postprocessing techniques, and addressing inherent forecast uncertainties using ensembles or other appropriate probabilistic methods. The purpose of this paper is to discuss how to improve operational NWP through high-resolution (section 2) and ensemble (section 3) forecasting. Nevertheless, advances in data, data assimilation, model formulation, parameterization, and model postprocessing are intimately connected and are also discussed, as are the limitations imposed by operational realities (section 4). Discussion of the complementary nature of high-resolution and ensemble approaches is provided in section 5, and the paper concludes with our recommendations (section 6). OCTOBER 2004 937 F O R E C A S T E R ’ S F O R U M 2. High-resolution models The National Centers for Environmental Prediction (NCEP) has decreased model grid spacing in their models from 190.5-km grid spacing during the 1970s [Limited-area Fine-mesh Model (LFM)]; Petersen and Stackpole 1989) to 12 km presently in the Eta model (Black 1994) to 8 km for the Nonhydrostatic Mesoscale Model (Janjic et al. 2001). Increasing resolution has resulted in improved model simulations and predictions of key atmospheric phenomena, such as rapidly developing extratropical cyclones in the western Atlantic (e.g., Kuo and Low-Nam 1990; Uccellini et al. 1999), Rocky Mountain lee cyclogenesis (Schultz and Doswell 2000, p. 153), cold surges east of the Rockies (Mesinger 1996), Appalachian cold-air damming (e.g., Weygandt and Seaman 1994), freezing precipitation (e.g., Roebber and Gyakum 2003), orographic winds and precipitation (e.g., Mass et al. 2002 and references within), sea/lakebreeze circulations (e.g., Manobianco and Nutter 1999; Roebber and Gehring 2000), lake-effect snowstorms (e.g., Ballentine et al. 1998; Steenburgh and Onton 2001), and convective systems (e.g., Weisman et al. 1997; Bernadet et al. 2000; Nielsen-Gammon and Strack 2000; Roebber et al. 2002).