Close Window

Colorado State University: "Development and operational use of fire weather support from a high-resolution mesoscale forecast model"

Final Report


This final report discusses our three-year Cooperative Project that began 6/1/00 and ended 5/31/03. Throughout the project, the primary objectives have been to develop and improve the realtime mesoscale forecast modeling system at Colorado State University (CSU) and its application to forecast operations in general, and specifically to fire weather forecasting, at the Weather Forecast Office (WFO) in Grand Junction, Colorado (GJT). While significant progress was made in achieving these broad objectives, we were unable to fully achieve certain important aspects. On the other hand, a number of unexpected opportunities for operationally oriented research became available during the project, and significant accomplishments that were originally unforeseen were achieved. Below are summaries of the objectives and accomplishments relating to each major component of the project. The summaries include new results for the final six-month period, as well as many items that are addressed in more detail in our previous annual and six-month reports.

Realtime modeling system at CSU

Several major advancements were made to the realtime modeling infrastructure at CSU during this project, the first being the full operational transition of the realtime forecast model, the Regional Atmospheric Modeling System (RAMS), from version 4a to version 4.29. Version 4a had been running for several years on a UNIX workstation cluster, supported in part by an earlier COMET Cooperative Project (Award No. UCAR S96-71867) involving NWS partners at the WFOs at GJT and Cheyenne, Wyoming (CYS). A new PC cluster was acquired during the latter stages of that project, funded in part by the NWS and COMET through the WFO GJT. It featured single and dual Pentium II processors, built under the Linux operating system. Besides the changes in cluster and model versions, the transition involved improvements in the automated control of realtime operations, data ingest, model parameterizations, parallel processing optimization, cluster inter-nodal communication efficiency, post-processing, and web-based forecast products.

Among the operational improvements allowed by the 4.29 infrastructure were faster run times by factors of 3-5, increased operational lead-time for NWS forecasters, and two 48h forecast cycles daily as opposed to a single 36h cycle previously. For both 00 and 12 UTC forecast cycles, the parent grid (Grid 1) was expanded to cover the lower 48 states at 48km grid spacing. The 00 UTC cycle was dedicated to Colorado regional forecast interests including those of the WFO GJT under this project, with a nested Grid 2 at 12km spacing covering all of Colorado, most of Wyoming, and portions of adjacent states as in the older 4a version. The 12 UTC cycle's Grid 2 was also often located over Colorado, but was relocatable to other regions within Grid 1 to support Department of Defense interests that helped fund the 4.29 cluster. Significantly, a third nested Grid 3, covering a relocatable local domain of 150x150 km2 at 3km spacing, was implemented in the 4.29 simulations. Vertical resolution was also greatly improved, with vertical grid spacing reduced to 150m in lower levels on all grids.

Further improvements to the realtime RAMS modeling infrastructure continued through the duration of the project. A succession of processor upgrades culminated in a cluster consisting of 13 identical Pentium III dual-processor nodes. With this configuration, up to 24 processors were used for the forecast simulation, with one of the dual-processor nodes utilized for graphics production. This architecture reduced runtimes for a 48h forecast cycle to about 4-5h and simplified system management of the cluster. In addition, increased memory allowed for the expansion of the 3km Grid 3 to 246 x 246 km2, which provided fine-scale guidance to the WFO GJT over a significantly larger area.

The final major change was the transition from RAMS version 4.29 to 4.3. The new version allowed more complex microphysical parameterizations to be utilized, and most significantly for this project, offered a post-processing capability that could facilitate the dissemination of the RAMS output to AWIPS at the WFO GJT. Sue van den Heever, under other funding at CSU, tested and implemented version 4.3 along with an alternate set of web-based forecast products. Version 4.3 became fully operational during July 2002, when it was used for the 00 UTC cycle to support a field project over southern Florida; the 12 UTC cycle continued with version 4.29 and the nested grids over Colorado. Afterward, both cycles were applied to Colorado, with the 12 and 00 UTC cycles using the long-familiar version 4.29 and new version 4.3, respectively. This allowed an extended comparison of the performance of the two versions, and exposed several rather minor deficiencies in the 4.3 simulations and post-processing, which were fixed prior to switching completely to that version. After the final major hardware upgrade described above was completed in November 2002, version 4.3 was used for both 00 UTC and 12 UTC cycles. Through the winter of 2002-2003 and through the completion of this project, both cycles ran most days, with the fine grids focused on Colorado in at least one cycle daily (and usually both) to support the WFO GJT.

These infrastructural upgrades resulted in significantly more reliable model forecast production at CSU. At the beginning of the project, the old RAMS version 4a system produced only one operationally useful forecast daily on about 79% of the days; during the last six months, the new RAMS 4.3 infrastructure produced at least one (and usually two) operationally useful forecasts over Colorado on 94% of the days. Combined with the much greater lead times and the high-resolution Grid 3, these improvements resulted in significantly more reliable and detailed mesoscale forecast guidance to forecasters at WFO GJT.

Fire weather applications at WFO GJT

The primary operational and research objectives of this project were to provide mesoscale forecast guidance as derived from the realtime RAMS simulations to the WFO GJT for fire weather forecast operations, and to identify and develop improved products for fire weather forecasting. The start-up period of the project coincided with a very active 2000 wildfire season in Colorado and the western U.S., thus providing a good opportunity to assess this operational utility. A major objective was to utilize the relocatable Grid 3 by positioning it over active fires or high-threat areas, as requested by GJT. In June, before fires became a problem in western Colorado, Grid 3 was centered on the eastern slope near Estes Park when the Bobcat Gulch fire started west of Loveland. Although not in the GJT forecast area, it was found that the 3km grid was quite good in forecasting the mesoscale flows that affected the fire in the complex terrain.

Grid 3 was still over the eastern slope through mid-August 2000, and thus was not over southwestern Colorado when two fires broke out over Mesa Verde in July and early August. However, in a detailed case study (McAnelly et al. 2000) of the first of these, the Bircher fire of 20-29 July, we re-ran the RAMS model in its 00 UTC forecast cycle configuration, with Grid 3 repositioned over Mesa Verde, for four 48-h forecast cycles from 20-23 July. These forecasts covered the period from the fire's ignition by lightning through its growth to almost maximum extent on the 24th. The model's performance was evaluated using radar, satellite, and Remote Automated Weather System (RAWS) observations, provided by Liz Page at COMET. It was found that the regional scale weather was simulated well, with very useful mesoscale forecast guidance for the day-to-day changes in weather that affected the fire's growth and behavior. However, some of the finer scale subtleties of the local circulations were not simulated particularly well, due partly to overly wet soil moisture initialization and in part due to the relatively poor topographic resolution of Mesa Verde, even on the 3km grid. This is in contrast to the more dominant forcing induced by the more prominent eastern slope topography in Grid 3 in the Bobcat Gulch fire, which led to more accurate fine-scale forecasts.

Wildfire activity was relatively light over western Colorado during the 2001 fire season (Year 2 of the project), and there were no specific requests by the WFO GJT for positioning Grid 3. Instead, Grid 3’s high-resolution guidance was utilized at various locations in Colorado, with the results available online to support fire weather forecasting on the eastern slope (e.g., by the Denver WFO), as well as other interests such as heavy precipitation threat assessment, various field projects, and agricultural and aviation forecasting.

The 2002 fire season (Year 3) in western Colorado was relatively high again, and the WFO GJT requested Grid 3's placement over numerous wildfires. In May, it was located over the Black Mountain Fire near Bailey west of Denver, and then was moved to the west slope northeast of Nucla at the request of Dr. Michael Meyers at GJT. In late June, it was moved to the southern San Juans over the devastating Missionary Ridge Fire northeast of Durango, and then moved in mid-July over the Burn Canyon Fire near Norwood. In August, it was moved first to fires on Mesa Verde, and then re-centered near Steamboat Springs to cover multiple fires over the Park Range and the Flattops. The RAMS forecast products, including specialized fire weather products such as mixed-layer height and winds and high-elevation Haines Index, provided beneficial high-resolution forecast guidance over these specific fires and their unique topography.

Other accomplishments

Besides fire weather forecast applications, the RAMS forecast guidance supports a broader spectrum of forecast applications at the WFO GJT. With the forecast staff at GJT becoming increasingly aware and confident of the RAMS realtime forecasts, they have increasingly utilized its guidance in general forecast operations as well as fire weather applications. The highly resolved topography on Grid 3, and even on the 12km Grid 2, exceeds that in operational NCEP models, and thus weather is simulated in much more detail over the complex terrain of Colorado. This detailed guidance has been very useful for forecasting strong winds, summer convective precipitation, and snowfall in winter storms. Case studies described in Section 2 illustrate some of these broader forecast applications.


In addition to the CSU/GJT collaboration on fire weather forecast operations and research, they collaborated on a number of case studies involving a wide variety of weather scenarios. These events included such phenomena such as a winter windstorm (Jones et al. 2002), a heavy thunderstorm rainfall event in complex terrain (Avery et al. 2003), a tornado near Vernal, Utah, and a major winter storm over Colorado (Meyers et al. 2004). As a result of such case studies, forecasters are able to utilize the mesoscale model guidance in future similar situations more effectively.

The high-resolution, relocatable Grid 3 is sometimes used for other applications when the WFO GJT does not target it for fire weather applications. During the winter seasons 2001-03, Grid 3 was centered over Storm Peak Laboratory (SPL), on the Park Range of north central Colorado near Steamboat Springs. This collaboration was in conjunction with a winter season, mountain weather field project conducted by SPL (under the Desert Research Institute (DRI) of the University of Nevada), the National Science Foundation, WFO GJT and COMET (through a Partners project (2002-03). The Partners Project with partners Dr. Meyers of the WFO GJT and Dr. Melanie Wetzel of DRI examined snowfall QPF in radar-limited terrain. For both 00 and 12 UTC forecast cycles, the RAMS model had the 3km grid over SPL during a field experiment in January and February 2002 and again in December 2002. The RAMS forecasts were found to provide much greater detail than the operational NCEP models and were quite accurate in predicting the onset, duration and magnitude of snow events at SPL. Wetzel et al. (2002; 2003) examined three separate case studies from the experiment. From daily snowfall observations taken at five sites from near ridgeline at 3170m westward into the Yampa Valley at 2030m, the analyses showed that RAMS predicted the high elevation QPF very well, trending to an under-prediction at lower elevations.

CSU has worked with the WFO GJT on several other case studies these past few years in which they have run post-mortem type simulations. These events were sometimes unrelated to fire weather applications involving both winter and other summer severe weather. For instance, a team of forecasters and researchers at GJT and CSU examined a severe springtime windstorm near Steamboat Springs that was associated with strong easterly flow across the Park Range in a strong synoptic system. A RAMS simulation using the realtime operational configuration compared well with the observations, provided much insight into the mechanisms behind the severe wind event, and helped forecasters predict similar events (Jones et al. 2000; 2002). A flash flood event which occurred in a monsoonal environment in the Dallas Creek drainage of western Colorado (Avery et al. 2000a,b; 2002) was also investigated. Other case studies include a tornadic event near Vernal, Utah, and an extreme Colorado snowstorm of March 2003 Meyers et al. 2003).

This project also contributed to collaboration between CSU and NWS entities besides the WFO GJT. For instance, a project with forecasters at the WFO CYS (Mike Weiland and Rich Bann) originated during the previous Cooperative Project, and was continued informally during the early portion of this project. The forecasters had found that severe weather reports tend to be clustered to the north of Pine Ridge in extreme northwestern Nebraska in relatively uniform southerly flow regimes. They hypothesized that a zone of low-level convergence and vorticity, similar to the well documented Denver convergence and vorticity zone (DCVZ), may be responsible for favoring strong convection in that region. Idealized RAMS simulations with uniform flows of various speeds from the south and southeast were run using a nested grid with 5km grid spacing, and such a zone of convergence and cyclonic vorticity was indeed found (Bann et al. 2003).

We have maintained collaboration with Dr. Paul Wolyn, SOO at the WFO in Pueblo (PUB), in his efforts to build an operational/research version of RAMS 4.29 on a Linux system at PUB. He has helped us in our efforts to get the model output into AWIPS.

Other related exchanges:

The NWS partner, Mike Meyers, served as Chairman of Session VIII: Fire Weather, at the Second Southwest Weather Symposium held in Tucson, AZ, on 21-22 September 2000.

Dr. William Cotton of CSU presented invited talks at the COMAP Symposium on Heavy Precipitation and Flash Flooding in Boulder on Sept. 12 and 26, 2000.

CSU and COMET were among the co-sponsors of a symposium held in Boulder on May 29, 2003 that addressed many aspects of the extreme Colorado regional snowstorm of March 16-20, 2003. Dr. Cotton gave an observational account of the storm at his residence in the mountains west of Fort Collins and compared it to previous extreme snowstorm events during his 26-year residency there. Ray McAnelly presented how the RAMS forecast model preformed for the storm, based on a dual set of 1-moment and 2-moment microphysical simulations initialized every 12h through the event.


Avery, B.A., C.N. Jones, J.D. Colton, and M.P. Meyers, 2000a: A southwest Colorado mountain flash flood in an enhanced monsoonal environment. Preprints, 9th Conf. on Mountain Meteorology (Aspen, CO), Amer. Meteor. Soc., 207-212.

Avery, B.A., C.N. Jones, J.D. Colton, and M.P. Meyers, 2000b: A southwest Colorado mountain flash flood in an enhanced monsoonal environment. Poster presentation, Second Southwest Weather Symposium (Tucson, AZ), sponsored by NWS, COMET, and U. Arizona.

Avery, B.A., C. N. Jones, J. D. Colton, and M. P. Meyers, 2003: A southwest Colorado mountain flash flood in an enhanced monsoonal environment. Nat. Wea. Dig., In Press

Bann, R., M. Weiland, and R. McAnelly, 2003: Terrain influences on severe convective storms along the Pine Ridge from east-central Wyoming to northwest Nebraska. Applied Research Paper 27-03, Central Region NWS, Kansas City, MO. 16 pp.

Jones, C.N., J.D. Colton, R. McAnelly, and M.P. Meyers, 2000: A mountain wave event west of the Colorado Park Range. 9th Conference on Mountain Meteorology (Aspen, CO), Amer. Meteor. Soc., 99-104.

Jones, C.N., J.D. Colton, R. McAnelly, and M.P. Meyers, 2002: An examination of a severe downslope windstorm west of the Colorado Park Range. Accepted for publication, National Weather Digest.

McAnelly, R.L., M.P. Meyers, E.M. Page, and W.R. Cotton, 2000: Application of a mesoscale model to fire weather forecasting. Postprints, Second Southwest Weather Symposium (Tucson, AZ), sponsored by NWS, COMET, and U. Arizona, 6pp. (also oral presentation)

Meyers, M.P., J.D. Colton, R.L. McAnelly, W.R. Cotton, D.A. Wesley, J.S. Snook, and G.S. Poulos, 2004: The operational implications of forecasting a heavy snow event over the central Rockies in an atypical flow regime. Abstract, 20th Conference on Weather Analysis and Forecasting (11-15 Jan 2004, Seattle), Amer. Meteor. Soc.

Meyers, M.P, E.M. Page, R.L. McAnelly and W.R. Cotton, 2001: Operational fire weather support through the use of a mesoscale forecast model. Ninth Conf. on Mesoscale Processes (Ft. Lauderdale, FL), Amer. Meteor. Soc., Boston, 234-235.

Meyers M.P., J.S. Snook, D.A. Wesley, and G.S. Poulos, 2003: A Rocky Mountain
storm. Part II: The forest blowdown over the West Slope of the northern
Colorado mountains - Observations, analysis, and modeling. Wea. Forecasting. 18, 662-674.

Page, E.M., M.P. Meyers, M. Chamberlain and R. McAnelly, 2000: Operational use of mesoscale models in fire weather forecasting. Preprints, Third Symposium on Fire and Forest Meteorology, Amer. Meteor. Soc., Boston, MA, Paper 4.1 (2 pp). (also oral presentation)

Poulos, G. S., D. A. Wesley, J. S. Snook, M. P. Meyers, 2002: A Rocky Mountain Storm. Part I: The Blizzard—kinematic evolution and the potential for high-resolution numerical forecasting of snowfall. Wea. Forecasting. 17, 955–970.

Wetzel, M., M. Meyers, R. Borys, A. Rossi, R. McAnelly, W. Cotton, P. Frisbie, D. Lowenthal, W. Brown, S. Cohn, and D. Nadler, 2002: Verification of snowfall forecasts in the Park Range of Colorado. Poster presentation, Ninth Annual Workshop on Weather Prediction in the Intermountain West, November 7, 2002, University of Utah.

Wetzel, M., M. Meyers, R. Borys, R. McAnelly, W. Cotton, A. Rossi, P. Frisbie, D. Nadler, D. Lowenthal, S. Cohn, and W. Brown, 2002: Mesoscale snowfall prediction and verification in mountainous terrain. Submitted to Wea. Forecasting.


From University Perspective

The collaboration has been very important to the ongoing improvement in mesoscale model operations and applications at CSU. In particular, Dr. Sue van den Heever, a Research Scientist at CSU, was instrumental in the transition from RAMS 4.29 to 4.3. The first realtime application of the 4.3 version was for a NASA field experiment in July 2003, and it was used for operational support for a bow-echo experiment over the central U.S. from late-May to early July 2003. In addition, several atmospheric science graduate students and computer science undergraduates have been heavily involved in maintaining and upgrading the system and in improving the forecast web-page. The operational production and examination of the realtime forecasts has resulted in further improvements by fine-tuning model parameters. The archived operational forecast simulations are occasionally used for case studies in graduate level classes, and by other researchers in other projects. The realtime forecasts are an ongoing benefit to a number of non-NWS users in the university community and general public.

A major objective of the project was to disseminate the RAMS output directly to AWIPS at GJT, so that forecasters can utilize the forecasts more easily and thoroughly. The analysis code developed for RAMS version 4.3 includes a feature that facilitates this process, namely the conversion of the native RAMS output to grib format. The biggest disappointment in the project was our failure to accomplish this long-sought objective. We were making significant headway in finishing this task, and had hoped that the implementation of the RAMS forecasts in AWIPS would be a vital aspect of a new COMET project proposed by the WFO GJT, CSU and DRI/SPL. However, we were unable to complete this implementation before this project ended, and the new proposal was not funded.

One model problem with RAMS 4.29 has been that it often over predicts precipitation. One source of problem was traced to an overly long time-step in the sedimentation routines. Improved algorithms were made to RAMS 4.3, and the magnitude of the problem was reduced with the operational implementation of that version. In addition, with the twice daily forecast cycles performed over Colorado beginning in December 2002, an improved two-moment microphysical parameterization was utilized in the 00 UTC cycle, while the single-moment scheme continued to be used in the 12 UTC cycle. It was hoped that the two-moment scheme might reduce the over-prediction problem, but there has been little apparent difference in QPF between the two schemes.

From NWS Perspective

During the past several summer seasons (2000-2002) the western United States has been in a prolonged drought which has resulted in some of the worst fire weather seasons on record. This environment created a unique opportunity to employ the RAMS model in a operational fire weather setting. After developing specific fire weather algorithms at the onset of the project, the model was tested in a research mode during the 2000 season. Results from the Mesa Verde Bircher fire were encouraging and demonstrated some of the advantages of using a high resolution mesoscale model in a fire weather application. After coordination between CSU and GJT, the 3 km grid in RAMS would be relocated over wildfire locations to give the fire weather meteorologists a high resolution operational forecast. This joint application allowed us to test the model in an operational fire weather setting. The RAMS forecasts were used during the 2001-02 seasons by the GJT forecast staff as well as IMET's who were deployed to wildfires in the area. With the improved topographical resolution and physics, RAMS was a valuable aid in forecasting fire weather situations especially over the complex terrain of the Intermountain-West.

From a forecaster perspective, the RAMS model output forced the forecaster to examine vertical velocity explicitly rather than through QG diagnostics. This paradigm shift prepared the forecasts as grid spacing in operational models decreases with increased technology. The forecasters also found other parameters quite useful, such as slope/valley winds and spatial distribution of precipitation. These fields were better represented by the fine scale grid space and sophisticated microphysics. This finer scale perspective is needed for fire weather forecasting, as well as other forecasting applications, in the complex terrain of the Intermountain West.

This collaboration has also facilitated the introduction of WES into the forecast environment. The GJT staff has archived numerous cases (both summer and winter), and the system has proven to be an invaluable tool for post-analysis of a particular storm.

The major problem encountered by the NWS during this project was the inability to get the RAMS output into the AWIPS platform. Having RAMS in AWIPS would have further facilitated the integration of the mesoscale model into the daily forecast process. A second problem with the RAMS forecasts was the over-prediction of the precipitation fields, even during the cold season. CSU worked on this problem during the spring 2003 and it appears that the problem has been reduced using version 4.3.