Fig. 2. LAPS analysis of surface wind (long barb = 10 kts) and MSL pressure for 1800 UTC on 24 March 2001. Fig. 4a. MM5hot 24 h forecast of surface winds (background is temperature) valid 1800 UTC 24 March. Fig. 4b. As in Fig. 4a but for the Mesoeta. Fig. 1. Schematic of the DCVZ, along with terrain (m) and METAR sites. Fig. 4c. As in Fig. 4a but for the RUC-20 km run. Note that the background color in this case is wind speed. AN EXAMINATION OF THE OPERATIONAL PREDICTABILITY OF MESOSCALE TERRAIN-INDUCED FEATURES IN EASTERN COLORADO FROM SEVERAL MODELS Edward J. Szoke and Brent L. Shaw* NOAA Forecast Systems Laboratory Boulder, Colorado 80303 1. INTRODUCTION As radar and surface observations have improved, forecasters have become increasingly aware of the importance of mesoscale circulations. Such features can have a significant impact on the sensible weather, but in the past were difficult to observe with real-time data and seldom captured by operational numerical models. However, with the increase in grid resolution of operational models, and the availability of locally run smaller-scale models, the possibility of resolving and predicting features on the mesoscale has become a reality. In this study we will look at the predictability of two well-known mesoscale features that occur in northeastern Colorado by examining forecasts from the latest versions of the Eta and RUC models, as well as from a local model being run quasi-operationally at the NOAA Forecast Systems Laboratory (FSL) for the co-located Boulder Weather Forecast Office (WFO). The features of interest are known locally as the "Denver Cyclone" and the "Longmont Anticyclone," and both have been well-documented at conferences and in the literature, through observational studies and numerical modeling using various research models (see, for example, Szoke et. al., 1984, Szoke, 1991 and the references in that paper, and Wesley, 1995). Both features are induced by the interaction of the synoptic flow with terrain. The resultant weather that can arise in association with these features ranges from dramatic variation in the wind field to mesoscale distribution of precipitation and localized occurrence of severe (winter and convective season) weather. Clearly there is high interest in trying to make operational forecasts of these features. Up to now (before the advent of finer grid resolution models) forecasters generally used forecasts of the synoptic flow and an understanding of the potential mesoscale features that could develop under such flow conditions to predict the possible occurrence of the two flow features. The potential to forecast these features with numerical models gives the possibility for more accurate predictions of the occurrence of these important phenomena. Although local models have been running at FSL (and made available to the Boulder WFO) for a number of years, there has not been any consistent verification effort aimed at these two flow features. Also, a recent change of the local model to a version of the MM5, with somewhat better resolution of the lower levels, has improved the overall ability to forecast the circulations. Meanwhile, the reduction of the grid resolution of the operational Eta to 22 km has resulted in better forecasts of both features with this model. The Rapid Update Cycle (RUC) model (Benjamin et. al., 2000) is undergoing a change to a 20-km grid resolution, and this new version of the RUC has also demonstrated predictability of these features. In this paper we will look at all three models in a subjective examination of their predictions of the Denver Cyclone and Longmont Anticylone. Comparison is made with sensible weather using detailed observations (METAR and local mesonet), coupled with the Local Analysis and Prediction System (LAPS, McGinley et al., 1991) analysis for the verification times. In addition, we will examine the model point forecasts for some of the sites where the Boulder WFO is required to make a Terminal Aviation Forecasts (TAF), as a further test of model performance. We will concentrate in the paper on predictions of the Denver Cyclone, and hope to include additional cases that include the Longmont Anticyclone at the poster session during the conference. 2. OVERVIEW OF THE MODELS FSL has been testing the potential of running a local scale model at a WFO for a number of years (Shaw et al., 2001), using the Boulder (formerly Denver) WFO as a test site. The key idea behind local modeling is to utilize a local analysis based on a variety of data, some of which may only be available at a WFO (as opposed to a national center), to initialize the model. In this regard, LAPS has been used to initialize several different models that have been run locally at FSL at a grid resolution of 10 km, including the Eta, SFM (Scalable Forecast Model, a version of the Colorado State University RAMS model), and MM5 (NCAR/Penn State University Mesoscale Model-Version 5). Since the mid-1990's all three of the local models were run, usually twice daily, with output to the FSL webpage (Szoke et al., 1998). To better demonstrate the feasibility of local modeling at a WFO, over the last couple of years one local model (the SFM) was run out of the Boulder WFO itself, using a separate multi-processor computer connected within the firewall to their Advanced Weather Interactive Processing System (AWIPS, Wakefield, 1998) workstation. The project successfully demonstrated the capabilities of such an approach, using LAPS analyses to initialize the model for four runs per day out to 18 h. More recently (over the last year) FSL has made a few changes to the local modeling system, including replacing the SFM with the NCAR/PSU MM5 model, employing a "hot-start" through LAPS to initialize the model (Shaw et al., 2001), and running the model over an expanded domain that is considerably larger than the WFO forecast area (still at 10-km horizontal grid resolution) of 125 by 105 points. The vertical grid consists of 41 levels, with the highest resolution contained within the boundary layer. The Schultz explicit microphysics and the Kain-Fritsch convective parameterization are employed. The rapid radiative transfer model (RRTM) scheme is used as the longwave radiation package, and the Blackadar scheme is used for the PBL parameterizations. Although the model conceivably could still run on the same machine that was used for the SFM, with the availability of the new FSL supercomputer and colocation of the Boulder WFO, the model was run by FSL using some nodes of the supercomputer with the results transmitted to the WFO for display on their AWIPS. Four runs are made each day, with output expanded to go out to 24 h. The model output is also available online through the FSL LAPS homepage at http://laps.fsl.noaa.gov. Although they are not run at the same 10-km grid resolution, two other models are applicable for the forecast problem of mesoscale circulations and were used in this study. One is a new 20-km grid resolution version of the RUC model (online documentation and access to this model is available at http://ruc.fsl.noaa.gov/). The purpose of the RUC model is to provide high frequency mesoscale analyses and short-range forecasts for the domain of the continental United States (CONUS). Extensive documentation of the RUC model can be found at the web site and through Benjamin et al. (2000). Essentially, the RUC is quite a different model than the MM5, being an isentropic model with sigma levels closer to the surface (40 levels are used in the new 20-km RUC). Another aspect of the RUC model is its ability to ingest off synoptic time data from sources like ACARS, satellite, and surface data in its analysis scheme. Some of the model characteristics, such as radiation and microphysics schemes, are versions of those used in the MM5, while other schemes are designed for the RUC. For this paper we used output from the 20-km RUC that was available online, concentrating on predictions of the surface wind. The RUC is updated hourly, with forecasts made hourly out to 3h, and out to 12 h at 6 h intervals. In addition, 24 h forecasts (made twice per day) were also available online. At the time of this writing the 20 km RUC was still considered experimental (with the 40 km RUC operational), but is sceduled for implementation soon. The final model that was used for comparison is the eta model, which for the period of comparison (beginning in the Fall of 2000) was being run by the National Center for Environmental Prediction (NCEP) at a resolution of 22 km (Black, 1994). Output from this model is available out to 48 or 60h for runs every 6 h at WFOs nationwide through AWIPS, with the best resolution output distributed for a subsection of the CONUS under the title of "Mesoeta" (Black, 1994). The 22 km output is actually interpolated to a 20-km grid, with surface output transmitted for display on AWIPS at this highest resolution. An online description of the mesoeta can be found at http://nimbo.wrh.noaa.gov/wrhq/96TAs/TA9606/ta96-06.html. Note that the native output of either 10 km for the MM5 or 20 km for the RUC was used for the other two models. 3. A BRIEF OVERVIEW OF THE DENVER CYCLONE AND LONGMONT ANTICYCLONE Owing to space limitations, in this paper we will concentrate on the modelling of the Denver Cyclone. We hope to also show cases of the Longmont Anticyclone feature at the conference. 3.1 Denver Cyclone The Denver Cyclone is a mesoscale flow feature that was documented when data became available from a mesonetwork of automated surface stations installed by FSL in 1980 for the purpose of testing the utility of such data. A schematic of the feature is shown in Fig. 1. Also known as the Denver Convergence-Vorticity Zone (DCVZ), because it may appear as an approximate north-south zone of low level convergence and cyclonic vorticity rather than a full fledged circulation, it is a relatively common feature, appearing on 20 to 30 percent of summer days (Szoke et al., 1984). Numerous successful modeling studies with research models have helped establish a likely cause of the feature (many of these studies are summarized in Szoke 1991), which is the response of southerly component flow passing over the terrain feature known as the Palmer Divide (the east-west ridge south of Denver depicted in Fig. 1) under conditions of appropriate stability. One of the reasons the DCVZ is important for local weather in the Boulder WFO is because of its influence on winds, which can be important enough to determine runway configuration at Denver's International Airport (DIA), since this site often lies close to the DCVZ and so at times can either experience 20-25 knot plus southerly flow, or north to northwest flow at around 10 knots (airport LLWAS sensors have even documented cases where a portion of a runway experienced one flow while the opposite end had the other). Also, the DCVZ is often the location of initial convection and later severe storms (particularly non-supercell tornadoes, owing to its association with regions of localized cyclonic vertical vorticity). For this paper we will use the wind forecast problem associated with the Denver Cyclone as a test of the model, verifying the various models against the observed winds at the sites where the Boulder WFO has responsibity for issuing TAFs (see Fig. 1). In addition to the various modelling studies of the DCVZ noted above, a number of years ago the RUC model was used in a nested grid formulation of 80 km (the operational RUC) for the outer domain and 20 km for the inner domain to successfully model a single Denver Cyclone case (Benjamin et al., 1986). Because of the larger grid used for that study, it was the first modeling demonstration of a DCVZ-like feature north of two similar east-west terrain features along the Front Range, the Raton Mesa near the Colorado-New Mexico border, and the Cheyenne Ridge near the Colorado-Wyoming border. 3.2 Longmont Anticyclone The Longmont Anticyclone is another flow feature that results from the interaction of the lower level flow with the terrain in the area. In this case, northwesterly flow across southern Wyoming apparently interacts with some of the higher terrain of the Rockies and the Cheyenne Ridge, resulting in a turning of the flow to north or northeast as it enters Colorado and moves south along the Front Range (Wesley et al., 1995). In more extreme cases the flow will turn all the way to southeast along the Front Range, with the center of the turning sometimes located near the town of Longmont (located northeast of Boulder), hence the name. The feature is sometimes associated with enhanced precipitation near the Front Range, and of course is important for determining low-level wind direction and speed. 4. A DCVZ FORECAST EXAMPLE A Denver Cyclone case from 24 March 2001 will be used to illustrate the ability of the three models considered to predict the feature. This date happened to be when FSL was running the RUC20 out to at least 24 h in support of the experiment PACJET (Pacific Land-falling Jets Experiment), so we will take advantage of this opportunity to compare 24 h forecasts from the three models, all initialized on 23 March 2001 at 1800 UTC. It is useful to recall that among the model differences is the fact that each has its own initialization scheme, with the LAPS analysis used for the MM5-hot run, with boundary conditions provided by the Eta model (in this case the 1200 UTC run). The verifying LAPS analysis for 1800 UTC on 24 March is shown in Fig. 2. Comparison of the analysis to surface observations (shown later in Table 1) reveal the LAPS analysis does a good job of depicting the location of what is a full Denver Cyclone circulation for this case, centered about 15 to 20 km to the east of Denver. The different 24 h forecasts are shown in Figs 3a to 3c, with all the runs valid for the time of the LAPS analysis in Fig.2. As an overview, perusal of the three predictions in Fig. 3 indicates that all three models were able to forecast a Denver Cyclone. Considering that these are 24 h forecasts, this result alone is considered to be quite impressive, and nicely shows the capabilities of modelling at finer grid resolution. Although all three models generally show the Denver Cyclone in approximately the same area, there are some differences. The MM5hot solution shows a circulation that is elongated in the east-west direction rather than more circular, and appears to be centered about 30 km too far east. While there is some possibility that the actual circulation could be a little farther east than the LAPS analyses indicates (since observations become more spotty east of DIA), we believe the LAPS analysis is a pretty close representation of where the Denver Cyclone was located in this case. The Mesoeta forecast in Fig. 3b has a more circular Denver Cyclone and its position is farther west than the MM5hot forecast, actually positioned a bit too far west and south but really quite a good forecast. The RUC20 forecast shown in Fig. 3c is presented without the county background map that the other figures have, as it was captured from a web presentation at a larger scale. However, it is possible to get an indication of where its forecast of the Denver Cyclone is by comparing to Fig. 3b, since the wind barbs in both figures are displayed at 20 km intervals and positioned in approximately the same location. Such a comparison indicates that the forecast position of the center of the Denver Cyclone from the RUC20 is about 20 km northwest of the Mesoeta position in Fig. 3b, with a similar circular shape. Comparison of each forecast with the LAPS analysis in Fig. 2 again indicates all are good forecasts, with the best forecast perhaps a consensus location from the 3 models for this time. Examination of other times (not shown) using the LAPS analyses indicates that at 1200 UTC on 24 March the Denver Cyclone was centered more over southern portions of Denver, and then moved slowly east-northeast through the mid-afternoon (2100 UTC). The model forecasts varied on this evolution; the MM5hot had a fairly close forecast to what was observed at 1200 UTC, and moved the circulation off correctly to the east-northeast, but was somewhat fast compared to what was observed. The Mesoeta tended to anchor the circulation close to the position indicated in Fig. 3b, while the RUC20 moves the circulation like the MM5 between 1200 and 1500 UTC, but then strengthens with some retrogression to the position in Fig. 3c, before moving it northeastward somewhat by 2100 UTC. In the simulation done experimentally by Crook et al., (1990) the circulation was found to move north to north-northeast with time, but observations have suggested that while this may be true in some cases (Szoke, 1991), there are many variations that include a relatively stationary circulation. For this case it appears the MM5hot may have been closest to simulating the observed motion of the circulation. To get a better idea how the different models actually verified for point wind forecasts, we examined some forecast hours for this and other cases for the METAR site DEN, which is located at DIA, and is therefore of interest to operational forecasters who are required to issue TAFs for this and other locations. The results are shown in Table 1. Table 1: Verification of point wind forecasts Date (all 2001) & Time (UTC) & station Metar Obs Model Forecast Wind MM5 RUC M-eta DCVZ case, for models initialized at 1800 UTC 23 Mar 24 Mar/12/DEN 10011 10005 12010 09010 24 Mar/15/DEN 08009 calm 19010 08005 24 Mar/18/DEN 05008 calm 18005 06005 DCVZ case, for models initialized at 1200 UTC 17 Apr 17 Apr/18/DEN 14006 12005 16005 14005 17 Apr/21/DEN 14007 12010 19005 13005 18 Apr/00/DEN 18007 14010 16005 15005 DCVZ case, for models initialized at 1200 UTC 17 Apr 17 Apr/18/BJC 27005 09005 06005 calm 17 Apr/21/BJC 03003 11005 12005 calm 18 Apr/00/BJC 34006 05005 10005 18003 LGM Anticyclone; models initialized at 1800 UTC 30 Apr 30 Apr/21/DEN 30013 32015 31520 32005 1 May/00/DEN 09006 34010 34010 26005 30 Apr/21/BJC 30012 34005 30015 29005 1 May/00/BJC 18006 11005 30010 24005 The trends discussed for the first case (24 March) are reflected in the point wind forecasts for DEN. The second case was more of DCVZ rather than a circulation, with the convergence zone starting out near DIA then gradually moving slightly west with time. There are no major differences in the wind forecasts for DEN listed in Table 1, though examination of the individual forecasts showed that the MM5hot best captured the position of the DCVZ, with the RUC20 pushing the southeast flow too far west and the Mesoeta being too weak without much turning of the wind. This is generally reflected in the verification for station BJC (Broomfield-Jeffco Airport) which is south of Boulder and remained on the west side of the DCVZ. The final case shown is for a Longmont Anticyclone which as seen by the wind obs created a turning of the flow to south/east from the prevailing northwest flow between 2100 and 0000 UTC. For this case only the MM5hot captured this turning, perhaps because this is somewhat of a weak case. 5. SUMMARY AND CONCLUSIONS In general all 3 models showed skill in resolving the Denver Cyclone, with an edge to the finer resolution MM5hot for the weaker cases. These limited results are encouraging, and suggest that local models can provide significant support to forecasting even at the 24 h range. 6. ACKNOWLEDGMENTS We wish to thank Nita Fullerton and Joe Golden of FSL for reviews of this paper. 7. REFERENCES Benjamin, S.G., R. Brummer, E. Hsie, EJ. Szoke, and J.M. Brown, 1986: Numerical simulations of a local topographically induced circulation using a nested grid model. Preprints, 11th Conf. on Weather Forecasting and Analysis. Kansas City, MO, Amer. Meteor. Society, 370-374. Benjamin, S.G., G.A. Grell, J.M. Brown, K.J. Brundage, D. Devenyi, D. Kim, B. Schwartz, T.G. Smirnova, T.L. Smith, S.S. Weygandt, and G.A. Manikin, 2000: The 20-km version of the Rapid Update Cycle. Preprints, 9th Conf. on Aviation, Range, and Aerospace Meteorology, Ame. Meteor. Soc., Orlando, FL, 421-423. Black, T.L., 1994: The new NMC mesoscale eta model: Description and forecast examples. Weather and Forecasting, 9, 265-278. Crook, N.A., T.L. Clark and M.W. Moncrieff, 1990: The Denver Cyclone. Part I: Generation in low Froude number flow. J. Atmos. Sci., 47, 2725-2742. McGinley, J.A., S.C. Albers, and P.A. Stamus, 1991: Validation of a composite convective index as defined by a real-time local analysis system. Weather and Forecasting, 6, 337-356. Shaw, B.L., E.R. Thaler, and E.J. Szoke, 2001: Operational evaluation of the LAPS-MM5 "hot start" local foreast model. Preprints, 14th Conf. on Numerical Weather Prediction. Ft. Lauderdale, FL, Amer. Meteor. Society, xxx-xxx. Szoke, E.J., M.L. Weisman, J.M. Brown, F. Caracena and T.W. Schlatter, 1984: A subsynoptic analysis of the Denver tornadoes of 3 June 1981. Mon. Wea. Rev., 112, 790-808. Szoke, E.J., 1991: Eye of the Denver Cyclone. Mon. Wea. Rev., 119, 1283-1292. Szoke, E.J., J.A. McGinley, P. Schultz, and J.S. Snook, 1998: Near operational short-term forecasts from two mesoscale models. Preprints, 12th Conf. on Numerical Weather Prediction. Phoenix, Arizona, American Meteorological Society, 320-323. Wakefield, J.S., 1998: Operational risk reduction: Easing AWIPS into the field. Preprints, 14th IIPS Conf. Phoenix, Arizona, American Meteorological Society, 389-391. Wesley, D.A., R.M. Rasmussen, and B.C. Bernstein, 1995: Snowfall associated with a terrain-generated convergence zone during the Winter Icing and Storms Project. Mon. Wea. Rev., 123, 2957-2977. Table 2: Verification of point wind forecasts Date (all 2001) & Time (UTC) Metar Obs Model Forecast Wind q MM5 RUC Meta 24 Mar/18 30012 1 2 4-5 Nov 97 1 2 3 27-28 Nov 97 2 1 25-26 Feb 98 1 2 3 12 Apr 98 1 3 2s 15 Apr 98 1s 2s 3s 4-6 Oct 98 3 2 1 16-17 Oct 98 1 3 2 4 26-28 Oct 98 3-4 Nov 98 3 2 1 4 6-7 Nov 98 1 2 L 3 9-11 Nov 98 4-8 Dec 98 1 1 1 2 P4.21 4 9th Conference on Mesoscale Meteorology 30 July-2 August 2001, Ft. Lauderdale, Florida