Close Window

Univ. of Wisconsin - La Crosse: "An integrated investigation of multisource rainfall estimates over the Upper Mississippi River Valley"

Final Report

Prepared by Dan A. Baumgardt, Principal Investigator, Science and Operations Officer, NWSO La Crosse, Wisconsin
Ronald A. Weinkauf, Principal Investigator, Professor, Department of Geography & Earth Science, University of Wisconsin- La Crosse, Wisconsin


1.1 Summary of Project Objectives, Participants, and Responsibilities

1.1.1 Introduction

The challenge addressed by this COMET project was to provide timely and locationally specific forecasts of precipitation amounts using a GOES satellite precipitation estimation algorithm integrated into a GIS environment. Even with the advances in precipitation forecasting brought by the new Doppler radar installations at National Weather Service (NWS) forecast offices, accurate quantitative precipitation forecasts to permit reliable flash flood predictions for local drainage basins continue to be problematic.

This COMET project proposed a new approach to the problem of flash flood prediction. Using data provided by the GOES-8 and 9 satellites on 15 minute and 60 minute intervals and reworked mathematically using an algorithm developed by Gilberto Vicente1 of the National Oceanic & Atmospheric Administration (NOAA), one obtains estimates of precipitation over 2.8 by 2.8 km pixel areas of the Upper Mississippi Valley region. This study tested the usefulness of these precipitation estimates over the 28 county National Weather Service La Crosse area of responsibility. Flash flooding is a major concern in the area as local relief in excess of five hundred feet and slope angles exceeding 30 percent provide for rapid runoff of storm water, particularly during periods of spring rains over frozen snow covered ground, and during heavy summer thunderstorms.

GOES radiance data from the 3.9-micron and the 10.7-micron wavelength bands gives a means of mathematically computing precipitation in an area based on satellite-derived IR cloud-top temperatures. The precipitation rate estimates are then adjusted for different moisture regimes using the most current National Center for Environmental Prediction (NCEP) Eta model fields of precipitable water and relative humidity. These raw data files were made available for computer download from NOAA (NESDIS) computers for one, three, six, and twenty-four hour periods and consisted of 14,000 2.8 by 2.8 km pixel data points over the study area. A UNIX computer running a cron computer program downloaded all six and twenty-four hour data sets to an archive machine at UW-La Crosse for further evaluation. The one and three hour GOES algorithm QPF files were assessed qualitatively via RAMSDIS display at the NWS-La Crosse office.

Gauge precipitation data was acquired from the Daily Cooperative Observer Network and Surface Observation Network and collected by the North Central River Forecast Center (NCRFC) for six and twenty-four hour periods in the study area. Unfortunately, the one and three hour reports were too few to produce reliable statistical evaluation. The NCRFC produced data was automatically downloaded by UW-La Crosse using the same UNIX cron program identified for the GOES data.

1.1.2 Summary of Project Objectives

Four objectives were formulated to guide the project:

1. To develop and routinely access GOES satellite-derived estimates and ground-based measurements of rainfall for the Upper Mississippi River Valley.

2. To statistically test, evaluate, and document the usefulness of precipitation estimates derived from GOES 8 and 9 satellite algorithm data using ground-based rain gauge measurements of precipitation acquired by the Daily Cooperative Observer Network collected by the North Central River Forecast Center (NCRFC).

3. To define, develop, and maintain a geographical information system (GIS) optimized for providing precise locational information as to high precipitation events leading to potential flash floods.

4. To provide the means to incorporate satellite precipitation estimates into the National Weather Service forecast operations.

All of the objectives were accomplished, thus not only supporting the research herein described, but laying a foundation for additional local and regional research initiatives. Some avenues for future investigative priority are identified in another section.

1.1.3 Project Participants

One of the major accomplishments of this project was to bring persons with a wide range of skills into a network to address the problems encountered as the project unfolded. Participants included:

National Weather Service - WFO La Crosse
Dan Baumgardt, SOO and Principal Investigator
Glenn Lussky, MIC
Randy Bresser, Data Acquisition Program Manager and GIS specialist

University of Wisconsin- La Crosse- Department of Geography & Earth Science
Ronald Weinkauf, Professor and Principal Investigator
Paul Stoelting, Associate Professor and GIS expert
Cory Hanson, Graduate Student and Archive Manager

University of Wisconsin- La Crosse- Department of Mathematics
Abdul Elfessi, Assistant Professor and Statistics Expert

Cooperative Institute for Meteorological Satellite Studies (CIMSS)
University of Wisconsin- Madison
Scott Bachmeier, GOES imager (QPF) Specialist
Gary Wade, McIDAS specialist
Robert Rabin, QPF specialist in SSEC/CIMSS
Thomas Achtor, Program Manager, SSEC/CIMSS

M. Paul Menzel, Advanced Satellite Product Development
Gilberto Vicente, QPF algorithm developer

North Central River Forecast Center
John Halquist, Development and Operations Hydrologist
Brian Connelly, Hydrologist

Additional support provided by:
William Gresens, Grants and Contracts Officer UW-L
Richard Augulis, NWS-Central Region Director
Richard Livingston, NWS-Central Region SSD Chief

1.1.4 Division of Responsibilities University

The University of Wisconsin - La Crosse Department of Geography & Earth Science personnel undertook the work that permitted the GOES algorithm output and NCRFC (gauge data) to be brought into a common framework (GIS/ARCVIEW software platform) for analysis and comparison. Some of the procedures required to create that framework included:

The Mathematics Department, specifically Dr. Abdul Elfessi, ran a series of procedures in Statistics Package for the Social Sciences (SPSS) including:

The statistical analysis included GOES minus gauge mean error and standard deviations of the error. NWS

The NWS was responsible for the following project areas:

1.2 Research Development Accomplishments and Project Findings

1.2.1 University Accomplishments and Findings Data Matrix

The GOES satellite algorithm data were processed using a Visual Basic program (NWS La Crosse - Randy Breeser) which assigned latitude and longitude values to the individual data points provided in the raw data files. This produced a 14,000 pixel data matrix over the Upper Mississippi River Valley study area that could then be incorporated into a GIS for the region. The precipitation gauge data were sent with latitude and longitude values attached to the individual gauge reports and could be incorporated directly into the area's GIS. Geographic Information System (GIS)

Additional locational information added to the study area's GIS database included U.S.G.S. digital line graph (DLG) data on transportation, hydrography, boundary, and public land survey. Elevation and slope/aspect data was obtained using the new 30 m digital elevation model (DEM) data provided by the U.S.G.S. and processed using the Spatial Analyst Extension in ArcView 3.1 GIS program. A portion of the region could also be viewed using the one meter resolution digital ortho quad (DOQ) data provided by the U.S.G.S. All these data layers were then displayed over the 1:24,000 scale digital raster graphic (DRG) topographic maps provided by the U.S.G.S. Nearest Neighbor and Cluster Sampling of GOES Data

To provide a means of testing the viability of using the GOES precipitation estimates in flash flood forecast, the GIS was instructed to sample the nearest GOES pixel to each gauge station recording and output a table of the results. As a means of checking these nearest neighbor estimates of GOES precipitation estimates, a second table was constructed using average precipitation estimates for all GOES pixels that were found within a 3000 m buffer drawn around each gauge station.

Finding: The clustering technique versus nearest neighbor was statistically insignificant to the verification results. Statistical Analysis

Individual three, six, twelve, and twenty-four hour interval periods were analyzed as to the correlation between GOES precipitation estimates and actual gauge recordings. The small number of gauge recordings for less than twenty-four hour periods quickly forced our analysis to shift to the longer twenty-four hour data sets to gain an adequate sample size and to cover the entire study region adequately. Although this longer time period would not be used for near real-time flash flood forecasting, it provided the best possible data for testing the GOES algorithm estimates of precipitation in the area.

1.2.2 NWS Accomplishments and Findings Display of GOES-Rainfall Estimation Algorithm Data

An extensive amount of NWS work was placed into acquiring and manipulating the one-hour and three-hour GOES-Rainfall algorithm output for the WFO La Crosse meteorologists. Since insufficient rain gauge data was available to verify these short-fuse periods, only qualitative assessment could be done by the WFO La Crosse staff. Unfortunately, it took an exhaustive amount of work to simply display the data on the RAMSDIS machine in the WFO La Crosse office.

The GOES estimate AREA files were created via McIDAS (at NESDIS) and placed on a server. CIRA (at Colorado State) then wrote executable McIDAS code for the local RAMSDIS machine to grab the one-hour and three-hour algorithm output. CIRA also assisted in setting up a function key to view the data loops at the touch of a button. Enhancement tables were also made by the NWS for the algorithm data. The time allotted for this part of the project was not sufficient as working with many groups required more time than expected and thus, a reasonable time period for assessment was not available.

Lesson: If the NWS is working on a project with AREA file data and McIDAS data suppliers, the NWS should have a contact or knowledge of McIDAS/RAMSDIS to ensure time lines are kept in the project.

Lesson: If multi-group work is involved in a COMET project, allow more time for goals to be accomplished.

Lesson: The GOES Rainfall Algorithm area files theoretically should have been ported directly into GEMPAK for display. However, and not uncommon from GOES sounder data as well, GEMPAK would not accept the resolutions in the area file headers. Non-standard (experimental) McIDAS area files may need to be remapped into another header (i.e., VISIBLE) on RAMSDIS/McIDAS before porting to Unix/GEMPAK workstations. This may also hold for AWIPS/Net-CDF format. GOES Rainfall Algorithm Qualitative Assessment

Due to the amount of time required to make the data available to the WFO La Crosse staff in near real-time, findings in this area were limited.

Finding: The GOES rainfall algorithm data needs to be very timely, especially for one-hour rainfall totals, to hold operational forecasting usefulness. Emphasis should be placed on making the data available no later than t+30 minutes where "t" is the ending time of the sample period.

Observation: Most staff comments alluded to the over-estimation of QPF amounts in the region of thunderstorm anvil enhancements. This is typically down shear of the storms.

Observation: Under the anvil where the convective towers were actually located, the satellite algorithm performs well with estimates within an acceptable range.

Observation: The algorithm did not seem to add much skill over the WSR-88D rainfall algorithms but did provide another data source for comparison. It would be more useful in cases where the WSR-88D suffered from extensive hail contamination.

Recommendation: The algorithm provides useful data especially at times when radar estimates are in error due to known limitations (i.e., bright-banding). This GOES QPF algorithm should be included into the AWIPS SBN data stream. NCRFC Rainfall Verification Data

A working relationship developed between the NWS La Crosse and the NCRFC. The NCRFC agreed to work on supplying verification data to the project over the given WFO La Crosse domain.

Finding: NWS RFC's can support these studies by supplying rain gauge verification data quite easily. The NCRFC's staff and DOH, John Halquist were very willing to help.

Lesson: NCRFC originally supplied contoured data to the COMET project over the area of the grid. However, due to the configuration of the gauge locales over our local domain, the contouring added error into the verification grid.

Lesson: When verifying satellite rainfall data over a certain domain, it is best to match the GOES satellite subpoint to the actual gauge latitude and longitude location for verification. This limits the error in the verification dataset. Downloading Script

A downloading script, originally listed as an UW-L objective, was completed by the NWS La Crosse to acquire the GOES rainfall algorithm ASCII data and the NCRFC verification ASCII gauge data. The script was written in C-shell language and run via the cron.

Lesson: Insure objectives set in the main project proposal identifies specific people with the ability needed to accomplish that objective. Project time lines and overall outcome can possibly suffer otherwise.

Recommendation: Insure a robust naming scheme is developed for all datasets involved in the study which adds to the ease of automation of data download, analysis, and databasing. This was done early by the NWS and saved countless hours of file management time.

Lesson: Automatic e-mail notification of missing datasets was written into the download script, which made data monitoring much easier during the six-month data collection phase. GIS

A large positive from this project was the two-way learning interaction in the Geographic Information System (GIS) data. The NWS office is using GIS data for plotting various elements in the WFO responsibility area because UW-La Crosse merged multi-state and multi-county GIS datasets for us. We have ported that data into the office and use it for various studies.

One GIS spin-off study is plotting the historical tornado occurrences over the slope variation in the WFO La Crosse area. The WFO La Crosse area is unglaciated and therefore has varying terrain - we are looking into the possible role of this on tornado development.

Our DAPM assisted the university in learning the ESRI ARCVIEW and ARCINFO software. He wrote all needed Visual Basic code for the university to ingest the NCRFC and GOES algorithm output into ARCVIEW.

A student volunteer from UW-La Crosse continues to work at the NWS office to acquire more datasets for use (land-use, etc.).

ARCVIEW/ARCINFO was used to assess the daily QPF data from the satellite, which allowed us to find 4 days of poor algorithm processing. Multiple Group Participation

This project brought an UW-La Crosse statistician and other geography faculty into contact with the NWS La Crosse personnel. This provided an exchange of meteorology applications into their fields, which could be used in the classroom. Having these contacts is useful for future projects in the area.

Lesson: Although having a number of groups involved in a collaborative effort greatly increases the project scientifically, problems arise in leadership and duties. A single, strong leader needs to be identified in the proposal and steadfast and dedicated throughout the project.

Lesson: All participants should be gathered in a "proposal meeting" (distant member via conference call) in order to have all parties understand their responsibilities to the project. This also can focus the proposal and project into achievable dimensions. Including parties into a project simply based on a positive reply to email or phone call is not recommended.

Recommendation: COMET should initially state the responsibilities of the University and NWS. It was not completely clear that the University holds the funds and is, most likely, responsible for steering the project. This needed to be clearly stated so that Universities can be aware of this responsibility before becoming involved.

1.2.3 Preliminary Meteorological Findings

1. Using twenty-four hour samples, GOES generally overestimates low precipitation events (less than 0.25"). This can be a function of the zero bound acting on the gauge and GOES data.

2. As observed precipitation class increases from less than 0.25", to .25-1.00", to greater than 1.00", the mean error decreases. However, the standard deviation error increases substantially.

3. As observed precipitation amount increases, the GOES algorithm tends to underestimate precipitation for warm cloud top cases at an increasing rate. Cold cloud-top rainfall cases do not show this tendency.

4. Wind shear does have an effect on the accuracy of the GOES algorithm. High wind shear cases produced a GOES algorithm overestimation of around 0.40" with a standard deviation of just over one inch. Low wind shear cases had the worst error (0.60" underestimate) in the GOES algorithm for rainfall of greater than 1.00" events with a standard deviation error of 0.60".

5. GOES becomes much better at predicting high precipitation events during the summer months, particularly ones that involved rapid thunderstorm development and dissipation often most dangerous to flash flooding in the region.

Additional findings will be presented at meetings and workshops over the coming year. Additional publications are planned as well.


Due to the nature of this project, the findings over the following year based on the data collected will provide more exchanges. A good portion of the study was data collection.

Volunteer college students from the UW-La Crosse often work at the NWS office. With the acquisition of GIS datasets, they are often helpful and willing to use this to analyze meteorological data.

Randy Breeser, NWS La Crosse DAPM, assisted UW-L Geography personnel with dataset manipulation. He wrote various programs in Visual Basic to manipulate data, which could then be used in teaching classes at the university. This illustrated the extension of application of GIS to the students.

A large database was assembled for the Upper Midwest that includes environmental layers organized by 1:100,000 quadrangles. Digital layers now in the database encompass all available data on hydrology, rail network, roads, contours (digital elevation models), boundaries, public land survey system, and topography. Beyond this immediate study, these resources should be of great benefit in support of many future resource investigations sponsored by NWS and the University.


Four efforts were made to provide ongoing progress results to the university and the wider La Crosse community as well as to those involved in professional research as Geographers.

Article by Terry Rindfleisch in the La Crosse Tribune Discovery Section News of health and science: UW-L, weather service join forces in project to better predict FLASH FLOODS Wednesday, September 17, 1997, pp. C-1 and C-2.

ALUMUS University of Wisconsin- La Crosse Article: Predicting Flash Floods, UW-L's Ron Weinkauf Receives Grant to Study Flash Floods, by Bob Seaquist, Fall 1997, Vol.23, No.3, pp.1-3 plus cover.

1997 Geography Awareness Week Program - Thursday, November 20, 1997. Professors Paul Stoelting and Ronald Weinkauf, Geography and Earth Science, "Gully Washers, GOES and GIS: New Technologies for Predicating Flash Floods". Held in Rooms 100 and 245 that featured an oral slide presentation by Weinkauf and a computer demonstration by Stoelting. Approximately 6o persons were in attendance.

Stoelting, P., D. Baumgardt, G. Lussky, A. Elfessi: Gully Washers, GOES, and GIS: New Technologies for Forecasting Flash Floods. The Association of American Geographers 95th Annual Meeting 23-27 March 1999 Honolulu, Hawaii.


4.1 University

4.1.1 Benefits

The project brought many benefits to all involved. In particular greatly improved communication was developed among scientists working on similar projects, but who had not communicated routinely before the project was initiated. Additionally, the GIS development at UW-L provided the means to conduct the research and to tie GOES and ground precipitation information to specific locations. It is a follow on plan to make the full archive available on a WEB site accompanied by sufficient information to enable others to replicate our research. Project funds made possible the purchase of several GIS software packages that greatly improved our ability to manage and analyze satellite and ground observations.

SPSS was also tested and proved to be useful for some but not all of the required procedures. Travel to the NCRFC in Minnesota, to the SSEC at UW-Madison was also supported by the project. Several publications are planned with the manuscripts now in draft form. Major journals are now being considered as suitable for publishing our project results.

4.1.2 Problems

Problems arose in terms of communication required to resolve technical issues. For example, the rather straight foreword task of defining and agreeing on the size and character of the Upper Midwest GOES data matrix proved to be a daunting task. Eventually the issues were resolved satisfactorily for all parties. It would seem that greater effort should be forthcoming from the agency supplying the GOES data to improve this aspect of data evaluation. Other than that problem, few other difficulties were encountered.

4.2 NWS

4.2.1 Benefits

Many of the findings from this project are still being worked on at this time. Future papers of these findings are planned. A large amount of quantitative results including the affects of the GOES algorithm from wind shear and cloud top profiles continue to be researched at this time. This work will include a detailed statistical analysis of how the algorithm performed which will affect many sites east of the Continental Divide. Unfortunately, those findings are not yet completed. Qualitative assessments were stated in Section 1.2.2.

One large benefit of the project was identifying collaborative contacts in various areas of NOAA and the UW-La Crosse. Contacts were made and cultivated at various levels, which will allow for future interaction.

A second large benefit of the project was the inclusion and use of GIS data in the NWS office environment. If not used directly in operations immediately, these data have enormous uses for research and future operations in the AWIPS era. Having the GIS dataset for the WFO La Crosse county warning and forecast area allows for any number of future studies to incorporate land-use, slope personalities, populations, and soil types.

4.2.2 Problems

An area of improvement for this and other COMET COOP projects would be for COMET to clearly state what is expected of the university involved. There is a strong leadership role needed from the university side and not so much the NWS side. The university must understand the commitment and leadership role needed. It was felt a good number of the University project proposal items were accomplished by the NWS that impeded the forward momentum of the project. Identifying specific individuals with the skills to accomplish needed project tasks is a must. Placing project responsibilities for each group according to these specifically identified tasks is also recommended.

Other problems and "lessons learned" have been identified in section 1.2.


Vicente, Gilberto A., Roderick A. Scofield, and W. Paul Menzel, "The Operational GOES Infrared Rainfall Estimation Technique," Bulletin of the American Meteorological Society, Vol. 79, No. 9, September 1988, pp. 1883-1898.