Close Window

University of Michigan (Greg Mann): "Improving local and regional forecasts over the Great Lakes region"

Final Report

I. Project Objectives

The COMET fellowship co-sponsored by the University of Michigan (UM) and the National Weather Service Forecast Office (NWSFO), Detroit/Pontiac (DTX) was fulfilled during the academic years of 1995-1996 through 1998-1999. The graduate student fellowship was created to establish a working relationship between the Department of Atmospheric, Oceanic, and Space Sciences (AOSS) at UM and NWSFO DTX. Specifically, the collaboration centered around the development and implementation of a local forecasting system based upon the NCAR/PSU mesoscale model, MM5, to assist in the prediction of mesoscale details within the Great Lakes region. This task was to be accomplished at UM and the output from the various real-time model simulations would then be transferred to DTX for operational use.

Initially, it was hoped that separate with-lake (WL) and no-lake (NL) simulations, run at 10 km, could be performed daily at the UM Center for Parallel Computing. However, CPU competition and the timeliness required to perform such simulations necessitated the implementation of only the WL simulations, run at 30 km, and executed on a local workstation. In conjunction with the real-time mesoscale modeling, evaluation of mesoscale model output in the operational setting was performed on a subjective basis, and to a very limited extent objectively. The evaluation was of the utility of mesoscale model information in the Great Lakes region and was based upon input from the forecasting staff.

Additionally, active participation in office workshops, seminars, staff meetings, and occasionally routine forecasting shifts was expected. The purpose was to gain an understanding of the environment and daily operations of a National Weather Service Office. Furthermore, active involvement in the SOO program and interactions with the forecasting staff were strongly encouraged. The ultimate goal being that with a clear understanding of the current state of operational meteorology, better and more directed applied meteorological research could be developed and hopefully someday implemented to some extent in the forecast office.

II. Operational Activities

The primary objective of the fellowship activities centered around the development and implementation of a real-time mesoscale model for the western Great Lakes region. From September 1996 until June 1998, real-time mesoscale simulations were executed twice daily using the MM5 model system. The simulations utilized a 90 km coarse domain that encompassed the contiguous United States and a 30 km domain encompassing the entire Great Lakes region (MM5-30) and were run twice daily (00GMT and 12GMT initializations). Initially, a 30-hour projection was performed; however, by late 1997, 36-hour projections were executed. Some localization of the modeling system was required to better capture lake-effect storms. Specifically, the convective parameterization was modified to parameterize shallow precipitation convective structures (i.e., lake-effect snowbands). The simulations were initialized using the NCEP Eta-48 model output. The MM5 output was converted to GEMPAK format and distributed to DTX, GRR, APX, MQT, and CLE. The offices could view the output using GARP on the SAC or LINUX PC workstation. Each simulation required approximately 6 hours to complete and an additional 30-60 minutes to transfer to the NWS offices. The transfer speeds were very sluggish due to the inadequate bandwidth to each NWS office. Nevertheless, output was typically available at each office around 5 AM/PM LST. The MM5-30 output was most useful for updating existing forecast packages, given that it arrived much too late each cycle to be used when the scheduled packages were created.

Evaluations of model performance are more anecdotal than objective. Namely, MM5 did demonstrate added skill over the Eta-48 and the NGM in predicting the onset and duration of lake-effect snowfall. In general, the model was able to locate snowfall that was focused by lakeshore geometrical properties (e.g. curved shorelines, upwind bays, etc.). However, MM5 was unable to adequately detect more widespread activity and frequently underestimated the degree of modification to the boundary layer by the lakes. This is an indication that the model relied heavily upon mechanically forced ascent along the lake-land interfaces and that it was still not adequately treating the shallow, but intense, convection, despite the modifications to the convective parameterization.

A limited verification study of model performance in certain lake-effect storm cases was performed. Some of the results of this study are currently being published in the National Weather Digest by Mike Evans, a former forecaster at DTX, and Dick Wagenmaker. Their findings indicated that MM5's representation of near surface properties was inadequate in reproducing certain locally driven mesoscale circulations associated with long lived snowbands forced by differential cooling boundaries near the southern end of Lake Michigan. They concluded, after consultation with myself, that the treatment of snowcover and boundary layer processes associated with arctic airmasses was insufficient. Consequently, even though the model produced a snowband in the general vicinity of that which was observed, significant errors still existed in position and intensity and appeared to be correlated with errors in the surface thermal field and boundary layer structure.

Beginning Fall 1998, I was able to test the alpha release of the workstation version of the NCEP Eta model. It was ultimately utilized for regional mesoscale prediction instead of MM5-30. Significant local pre and post processing development was required to implement the modeling system for local use. Much of this work was in conjunction with the efforts of NCEP and FSL in creating a beta version of the code. The beta release is more user friendly, yet still requires some localization. The principle factor for abandoning MM5 in favor of the Eta was the timeliness of output availability. In the same time, and on the same machine, that the MM5-30 simulation (43x34x23 points) completed, a 10-km Eta (Eta-10) simulation (85x85x50 points) covering the western Great Lakes could be executed. Thus, there were two clear advantages to implementing the workstation Eta. First, at comparable resolutions the simulation could finish in substantially less time. Second, the faster simulations allow configurations at higher resolution, thus keeping the local regional modeling effort one step ahead of NCEP operational model resolution. The advent of the Eta-32 rendered the MM5-30 simulations less useful because of the comparable resolution and the lack of timeliness of the local MM5 runs.

Initial indications are that the Eta-10 clearly outperforms the MM5-30 and Eta-32 in simulating the surface wind field, especially in the marine environment. Furthermore, the Eta-10 appears to more adequately reproduce lake-effect precipitation in the correct locations with more realistic intensity compared to the other models. Like the other models, the Eta-10 is still deficient in simulating weak, widespread, small wavelength lake-effect snowbands. However, these observations are based upon subjective evaluation of a hand full of events. The beta release of the workstation Eta model is the cornerstone to an additional COMET project, which focuses upon marine forecasting innovations. That project will likely utilize the model at even higher resolutions, perhaps approaching 5 km. Thus, keeping ahead of the pace set by NCEP and the national modeling effort.

III. Research Activities

The primary mode of research centered around the doctoral dissertation work. This work focused upon multiple lake interactions within the Great Lakes system and their impact on lake-effect storm evolution. The results show that not only are there direct (linear) impacts of adjacent lakes (e.g., upstream moisture influences), but there are synergistic influences (i.e., non-linear interactions) from both upstream and downstream lakes that can have significant influences on the evolution of lake-effect snowbands. In fact, the findings show that under certain circumstances the formation of a dominant snowband, capable of producing snowfall rates of 5 inches/hour, was a result of multiple lake influences and likely would not have occurred with those processes. Much more detailed results are presented within the doctoral thesis, which accompanies this summary.

As a precursor to the thesis work, some work was done to determine whether the lake aggregate disturbance was responsible for changing lake-effect precipitation patterns. This work served to bridge the gap between current lake aggregate studies and my doctoral research. I was also involved with work related to lake aggregate effects during the autumn. The research revolved around the study of an unusual event nicknamed Hurricane Huron. This particular system took on tropical characteristics, including a visual queue of an eye-like feature. The event demonstrates the ability of the Great Lakes to significantly influence slow moving synoptic scale systems outside of the winter season, when the lake surface temperatures are significantly higher than the overlying atmosphere.

Stemming from the thesis work is a current study of dominant bands over the western Great Lakes. Part of the study is to determine the frequency of such events, which is being done by another graduate student at UM. The extension of the thesis work will be to identify modes of dynamics responsible for the formation of these snowbands. As stated in the dissertation, an additional mode of formation was discovered to exist with dominant bands. This will be further investigated and compared to topographically/ geographically forced dominant bands.

Finally, work during the Lake Induced Convective Experiment (Lake-ICE) demonstrated some utility of running with-lake and no-lake simulations in real-time. The WL and NL MM5-90 simulations were performed in real-time to support the lake aggregate disturbance investigations within Lake-ICE. This was done on a limited basis during the two-moth period of the field experiment with some success. Evaluation of lake aggregate effects within the observed cases is still ongoing. Additionally, interaction among the research and operational community proved to vital to the success of the field experiment and its forecast operations. As one of two project lead forecasters, I had the privilege to work with both research and operational meteorologists in the forecast operations of the experiment. In fact, several individuals from the DTX office participated and/or assisted in the forecast operations. The lake-effect forecasts were generated without the use of traditional lake-effect forecasting techniques (e.g., lake-effect decision trees). But rather, use of mesoscale model output and sound conceptual models of lake-effect formation and maintenance were applied with a high degree of success. The accuracy and precision of the forecasts were very important to project operations as a substantial amount of resources, both human and financial, were devoted to each intensive observational period.

IV. Training Activities

In addition to the localization and real-time implementation of a mesoscale model in real-time, training materials related to use of numerical model output, particularly mesoscale models, was presented to the NWS staff. These seminars were designed to inform the forecasting staff about some of the particulars of mesoscale models and their output. Details of model output for various cases were presented to demonstrate the model's strengths and weaknesses with regard to particular situations. Additionally, forecasting workshop exercises were devised to demonstrate the added value provided by the regional mesoscale model and to demonstrate the pitfalls of the high resolution output. The main point of the training sessions was to call attention to the realistic nature of the results and caution the forecasters not to believe it at face value, regardless of how believable it may appear. These seminars and workshops were not only conducted at DTX, but also at NWFO GRR, APX, and MQT.

Additionally, I assisted in the development and presentation of several other in-house training exercises, seminars, and workshops covering topics which included lake-effect snow, conditional/symmetric instability, fronts and jetstreaks, supercell dynamics, and current theories of tornadogenesis. Some of the material presented to the DTX staff also was used in a series of workshops provided by myself and Dick Wagenmaker to undergraduate and graduate students at UM each winter semester. The workshops at UM were designed to give exposure

V. Additional Activities

In addition to the research, operational, and training activities, I was able to engage in several other activities such as conferences and workshops. I had the opportunity to be involved with the planning and facilitation of the fifth annual U.S./Canada Great Lakes Operational Meteorology Workshop co-sponsored by UM, DTX, and COMET in September 1996. In particular, I was responsible for technical support and the integration of case study exercises provided by COMET and DTX into the lab setting environment at UM. I also attended several other conferences and workshops during my stay as a graduate student. These included the 15th AMS Weather Analysis and Forecasting Conference, the 8th Annual MM5 Users Workshop, and the Great Lakes Regional Climate Impacts Assessment Workshop. The latter workshop addressed a regional evaluation of climate change and its potential impacts on the Great Lakes region.

During my graduate studies, I guest lectured several courses both within AOSS and outside of the University, including Calvin College and Grand Valley State University. Courses ranged from introductory meteorology to senior/graduate level mesoscale dynamics. Topics of instruction were related to Great Lakes Meteorology, lake-effect snow dynamics, severe convection, and gravity wave dynamics. I also presented department seminars on topics related to utilizing real-time meteorological data, lake-effect snow, and multiple lake interactions.

Finally, I was a member of the forecasting team for the Bayview-Mackinaw Yachting Regatta in 1996, 1997, and 1998. The sailboat race is one of the largest single sailing events in North America each year. Typically, over 250 boats participate in the weekend race from Port Huron to Mackinaw Island. Participants are given a weather briefing on the morning of the start of the race by the forecasting team. They are provided with a detailed marine forecast for Lake Huron for the expect duration of the race (typically 3 days). Each vessel receives a briefing packet in addition to the formal presentation by the forecasting team.

VI. Final Comments

Here a few final comments that I would like to leave regarding the main purpose of this fellowship, namely the implementation of a local mesoscale modeling system. It has been apparent after several years of mesoscale modeling efforts that significant localization of the model is required to attain the most effective results. This localization is not simply a function of the geographical region. In essence it is a function of the mesoscale feature or process, which is of interest at that time. As an example, in the Great Lakes region there are various mesoscale features that are important in the local aspects of the regional weather. The simulation of lake-effect snow convection is quite different than the simulation of deep summer-time convection that interacts with localized lake breeze fronts. This is true from an explicit and implicit point of view as well. Convective parameterization designed for deep convection can be modified to a certain degree to account for shallow, intense, precipitating convection like lake-effect storms. That modified setup will cause premature termination of a simulation if deep summer-like convection is present. Additionally, the grid resolution, both horizontal and vertical can play a role in resolving these mesoscale features. Inadequate vertical resolution can introduce significant errors in the placement of a capping inversion, especially in the winter scenario. Thus, vertical resolution should be adaptable based upon the season in question. So, not only does the specific region dictate the model configuration, but the season, or specific phenomenon may also dictate the configuration. These are items that will need to be addressed as the meteorological community trends toward national scale mesoscale models.

As far as utilization of high resolution information is concerned, there is an inherent danger to accept the output as ground truth for the simple reason that it looks very realistic. Thus, in order to interrogate the information sufficiently, forecasters must be armed with knowledge of the modeling system and its underlying assumptions and have a firm grasp of sound conceptual models related to the feature in question. Mesoscale forecasting, both spatial and temporal, will be the next great step that operational meteorology faces. At this time, forecast are still made from the synoptic-scale perspective and not ultimately translated down to the mesoscale. Mesoscale models introduce an aspect of precision that has not been seen in the operational community. However, a gain in precision does not imply an increase in accuracy. The forecast process will go through a morphological change in the next decade, as mesoscale detail becomes a reality in each forecast. It is our challenge to approach this task with good judgment and sound scientific methods.

VII. Publications

Greg E. Mann and Peter J. Sousounis, 1999: The Role of Multi-lake Interactions in the Modulation of Lake-Effect Snowstorms. Preprints of the 8th AMS Mesoscale Processes Conference, Boulder, Colorado, June 28-July 1, 1999.

Peter J. Sousounis, James Wallman and Greg E. Mann, 1999: Hurricane Huron: An Example of an Intense Lake Aggregate Effect. Preprints of the 8th AMS Mesoscale Processes Conference, Boulder, Colorado, June 28-July 1, 1999.

Todd Miner, Peter J. Sousounis, James Wallman and Greg E. Mann, 1999: Hurricane Huron. Accepted for publication Bull. of Amer. Met. Soc.

Peter J. Sousounis, James Wallman, Greg E. Mann, and Todd Miner, 1999: Hurricane Huron: An example of an extreme lake aggregate effect in Autumn, submitted to Mon. Wea. Rev.

Peter J. Sousounis, Greg E. Mann, George Young, Richard B. Wagenmaker, Bradley D. Hoggatt, and William J. Badini, 1999: Forecasting during the Lake-ICE/SNOWBANDS Field Experiment, accepted for publication Wea and Forecasting.

Peter J. Sousounis and Greg E. Mann, 1998: Lake aggregate mesoscale disturbances. Part V: Impacts on lake-effect precipitation. accepted for publication Mon. Wea. Rev.

Greg E. Mann and Peter J. Sousounis, 1998: Real-time MM5 simulations in support of Lake-ICE/operations. Preprints of the 8th PSU/NCAR Mesoscale Model User' Workshop, Boulder, Colorado, June 15-16, 1998.

Peter J. Sousounis, Hui-ya Chuang, and Greg E. Mann, 1996: Idealized numerical simulations of mesoscale aggregate vortices. Preprints of the 7th Conference on Mesoscale Processes, Reading, United Kingdom, September 9-13, 1996.

Greg E. Mann and Peter J. Sousounis, 1996: Direct and Indirect Impacts of the Great Lakes upon Cyclogenesis. Preprints of the 15th Conference on Weather Analysis and Forecasting, Norfolk, Virginia, August 19-23, 1996.

Peter J. Sousounis and Greg E. Mann, 1996: Great Lakes Aggregate Effects on Lake-Effect Snowstorms. Preprints of the 15th Conference on Weather Analysis and Forecasting, Norfolk, Virginia, August 19-23, 1996.

Greg E. Mann, 1994: Diagnosing the Impacts of the Great Lakes on an Alberta Clipper. First Place- Father James B. Macelwane Award in Meteorology, American Meteorological Society.

Greg E. Mann, 1994: Diagnosing the Impacts of the Great Lakes on an Alberta Clipper-A Preliminary Investigation. Preprints of the 51st Annual Eastern Snow Conference, Dearborn, Michigan, June 15-16, 1994 (First Place- Student Paper Competition).