|This report is an archived publication and may contain dated technical, contact, and link information|
Publication Number: FHWA-HRT-04-109
Date: September 2004
From 2001 to 2003, the Road Weather Management Program of the Federal Highway Administration (FHWA) Office of Operations partnered with the National Weather Service (NWS) to sponsor five research projects to improve the sensing, prediction, and use of weather-related road conditions in road maintenance and operations. These projects are unique because they each involved collaborative partnerships between NWS Weather Forecast Offices (WFOs), State departments of transportation (DOTs), and universities. Participants and their project titles are listed in the table below:
|Pennsylvania State University, Pennsylvania Department of Transportation (PennDOT), and the State College, PA NWS WFO||"Developing an Interactive Mesonet for PennDOT"|
|Iowa State University, Iowa DOT, and the Des Moines, IA NWS WFO||"Improved Frost Forecasting through Artificial Neural Networks"|
|University of Nevada (Desert Research Institute), Nevada DOT, and the Reno, NV NWS WFO||"Use of Road Weather Information Systems in the Improvement of Transportation Operations in the Complex Terrain of Western Nevada"|
|State University of New York at Albany, New York State DOT (NYSDOT), and the Albany, NY NWS WFO||"The New York Integrated Weather Data Network"|
|University of Utah, Utah DOT, and the Salt Lake City, UT NWS WFO||"Applications of Local Data Assimilation in Complex Terrain"|
The primary purpose for these projects was to evaluate the use of weather observations and modeling systems to improve highway safety and to support effective decisions made by the various jurisdictions that manage the highway system. In particular, the research evaluated how environmental sensor station data, particularly Road Weather Information System (RWIS) data, could best be used for both road condition forecasting and weather forecasting. The collaborative efforts also included building better relations for training and sharing information between the meteorological and transportation agencies.
The Cooperative Program for Operational Meteorology, Education and Training (COMET®), an organization funded primarily by the NWS, received project funding from the Road Weather Management Program. The awards, made in early 2001, were for 2-year projects and primarily supported university researchers, with matching efforts from local NWS WFOs and State DOTs.
COMET solicited proposals, conducted the review process, and managed the project contracts. This was the first time that the COMET Program had extended its usual university-NWS collaborations to the transportation community, an important step in building institutional relations for further research and operations.
The project reports, written by the project participants, make up the main body of this report. Below is a summary of each project and its accomplishments.
Both the NWS and FHWA are interested in evaluating use of RWIS data for road condition forecasting and broader weather forecasting. For example, the NWS is interested in creating mesoscale networks (mesonets) that include RWIS sites, while FHWA would like to establish guidelines that define the optimal density for a network of RWIS sites in different seasons and various locations. For both parties, the usefulness of RWIS observations depends on several factors including site considerations and data quality, which varies according to performance specifications for equipment and maintenance practices.
One of the main goals of this project was to construct a mesonet of hourly weather reporting sites across Pennsylvania. These sites include 32 Automated Surface Observing Systems (ASOS) and a few Automated Weather Observing Systems (AWOS), both operated by the Federal Aviation Administration (FAA), 82 RWIS operated by PennDOT, and 47 Commonwealth of Pennsylvania Air Monitoring System (COPAMS) sites maintained by the Pennsylvania Department of Environmental Protection (PADEP). In addition, the State's approximately 200 hourly precipitation gauges of the Integrated Flood Observing and Warning Systems (I-FLOWS) are also being incorporated in the network. More than 300 daily observations will be available when the NWS's cooperative weather stations (about 125 sites) submit their data on a daily basis, including their reports of maximum temperature, minimum temperature, and 24-hour precipitation.
The data from these sites are stored in a database and processed each hour for real-time information (http://pasc.met.psu.edu/MESONET) of weather-derived parameters and time series displays from each site. Streamlines are overlaid on a topographic map to help identify patterns over the complex terrain in the State.
Developing the mesonet was particularly challenging because PennDOT contracted with three different manufacturers to provide, install, and maintain the sensors in various regions of the State. The result is data in three different formats within this single network. In addition, ground-based network connections and limited server capacities posed a serious challenge to including the RWIS data in the mesonet, and irregular polling times added difficulty to developing an automated quality control system. Delays in reporting the observations also hampered production of real-time products that are useful in "nowcasting" hazardous weather and highway conditions. Another problem in managing the data flow resulted from lower reliability in RWIS data during the seasons when the DOT has a reduced need for the observations.
All the RWIS data were converted to network Common Data Form (netCDF), a machine-independent format for representing scientific data. Once decisions were made regarding how to handle the data collection issues, a quality control system was designed using the Oklahoma mesonet standards. Concerns about instrument siting and calibration still exist,which can add further errors to the quality control process. Once the data quality is assured, the plan is to make the observations available to all involved parties, including the PennDOT Technology Division and the District offices.
Another major goal of the project was to include near real-time RWIS and other mesonet data sets into the NWS Advanced Weather Interactive Processing System (AWIPS) data display and the Pennsylvania State Climate Office database. This initially required both management and quality control of the data feed. Parameters to collect from the different networks were determined. To be chosen, a parameter had to contribute to augmenting or improving general surface observations, climatology, forecast verification, modeling, or the welfare of the general public. Nineteen parameters met these criteria and are collected in the database.
An additional task arose when researchers found there was little or no station history—records of movement due to construction or other reasons—for the RWIS and COPAMS sites. Assimilating the data into NWS system and the State Climate database requires precise geospatial information that was not readily available from PennDOT. As a result, State Climate personnel visited each of the 82 sites with a global positioning system (GPS) device and digital camera and created comprehensive metadata files for all RWIS sites. Images of the station (and GPS records) are a relatively new but increasingly important part of station history and metadata.
Preparing and delivering 3 training sessions for more than 160 roadway weather managers was the last task in this project. The NWS Warning Coordination Meteorologist (WCM) taught the fundamentals of weather and products available from the NWS WFOs. The State climatologist (the principal investigator on this project) taught forecasting techniques using RWIS data and winter weather and conducted a desktop exercise for all participants. The training occurred during the week of October 7, 2002.
The following summarizes the findings of this project:
Frost is one of the least predictable weather parameters requiring anti-icing or deicing treatment of road surfaces. It often occurs in localized regions under generally quiescent weather conditions. Ice deposition onto road surfaces can occur if road surface temperatures drop below the dewpoint temperature of the nearby air. A previous collaboration between the Iowa DOT and Iowa State University (ISU) produced a frost prediction model, which is currently being tested. Inputs required for the model include predictions of road surface temperature, air temperature, dewpoint temperature, and wind speeds. These predictions could come from raw or statistically post-processed numerical weather model output, such as Model Output Statistics (MOS) derived from the NWS Nested Grid Model (NGM). However, output from these weather models is typically limited to every 3 or 6 hours, and therefore is not available at the high temporal frequency required for good frost predictions. In addition, these models do not normally predict road temperatures. Statistical methods, such as artificial intelligence, can be used to overcome these problems.
Artificial Neural Networks (ANN) are one type of artificial intelligence technique, patterned loosely after the human brain. Different weightings are assigned to different relationships in a training set of data to create a system that can predict one variable based on a range of other variables. These techniques are potentially better than the sole use of forecast model data because the training technique should allow biases in the model data to be corrected. For instance, if the NGM model is always too cold in its forecasts of air temperature (compared to RWIS observations), the neural network will likely determine that the air temperature is going to be equal to the model prediction plus a small correction term.
This project used ANN data to develop a time series prediction system that could be coupled with the frost prediction model to improve roadway frost forecasting. NGM MOS data and RWIS data were collected for 3 years from 1996–98 and used to train and test the system. Results of the project indicated that in general, ANN-derived weather data compared better with RWIS observations than with the NGM-MOS forecasts alone. However, in some circumstances, the ANN predictions were only better than MOS for the first 6 to 9 hours of the prediction period. Nonetheless, the use of ANN data to predict roadway temperature, air temperature, dewpoint, and wind speed appears promising.
A second part of the study used the ANN data in the ISU frost deposition model; "yes" frost predicted and "no" frost predicted results were compared with observations of roadway frost. For the winters of 1996–98, the ANN-based predictions were not particularly good. The poor results may have been due in part to frost observations obtained by maintenance personnel looking out the windows of their vehicles on the way to work, perhaps not a very accurate method. Poor agreement was also attributed to ANN system's physically impossible predictions of a dewpoint exceeding the air temperature.
In the winter of 2001–2002, ISU students closely monitored conditions on one bridge during the early morning hours, likely a more accurate verification technique. During this period, frost predictions using the ANN data were very accurate, nearly matching the forecasts made after the fact when RWIS data were fed into the frost deposition model. However, during thewinter of 2002–2003 when students monitored three bridges, the performance was worse with relatively high false alarm rates and low probabilities of detection.
The third part of the project determined if data from the NWS and FAA observation systems, RWIS and AWOS/ASOS respectively, are comparable. The comparisons suggest that RWIS temperature readings have a high bias when wind speeds are light. The behavior is attributed to the fact that, unlike AWOS/ASOS sensors, RWIS temperature and dewpoint sensors are not aspirated. In addition, wind speed data from the RWIS stations was typically a few knots lower than the AWOS/ASOS sensors, probably due to siting differences. Other parameters (such as dewpoint) showed systematic differences that could be quantified and taken into account when using the data. A Web site was developed to allow direct comparisons between the different measuring systems in real time (http://mesonet.agron.iastate.edu/compare/). The NWS found the results particularly useful because data from these sensors have been combined into a mesonet (as part of another, earlier project), providing valuable information for monitoring conditions and validating forecasts.
The final task extended the ANN application to an NWS problem forecasting precipitation amounts during the warm season. As was the case with the ANN data for the frost predictions, results for this study were mixed, but generally suggest that ANN data can provide an improvement compared to model forecasts. Further work needs to be done in this area.
The conclusions of the principal investigators are as follows:
The complex terrain of western Nevada presents a challenge to both weather forecasters and transportation managers, particularly when winter snows cause road closures and delays. The economic impacts of winter storms can be significant. On a typical winter weekend, as many as 1.5 million people travel into the Sierra Nevada to Lake Tahoe and Reno, NV, bringing about $2 million into the economy. A winter storm warning by the NWS on a Thursday for a weekend in the Sierra Nevada can result in an estimated $800,000 decrease in potential revenue due to cancelled room reservations alone. Snow removal operations for road management and public safety are also costly. These impacts highlight the need for improved road-related forecasts that to date are hampered by limited weather observations and imperfect model guidance.
Predicting local weather features, such as high winds and heavy snow, is extremely difficult due to small-scale terrain features and hence large variations in weather over small distances. Forecasters and transportation managers need a dense network of weather observations to better "see" these small-scale features. Critical data is provided by additional State DOT weather stations to the NWS network and weather stations already available for forecasting.
The first task in this project was to develop programs to automatically access and download the data from Nevada Department of Transportation (NDOT) computers and to integrate and archive the data in the database of the Western Regional Climate Center at the Desert Research Institute (DRI). At the same time, NDOT was upgrading 11 mesonet sites and installing 5 new sites. A Web site on the DRI server was established to provide interactive graphical displays of the data once it was acquired (http://www.ndot.dri.edu). The site is still active, but some of the NDOT stations are not currently operational, transmission software in some have changed, and some stations simply do not transfer data to DRI.
A related task was to establish a data assimilation system to transfer data from NDOT to the NWS. The task was difficult due to slow models and frequent computer crashes at the beginning of the project. By the second year, better file transfer procedures were established, and faster computers were installed at both the NWS and NDOT. The collection of this data continues and has been expanded to other areas of the State outside of the local Reno, NV mesonet area. The collection of these additional sites was not part of the original scope of work for this project.
The second goal was to improve NWS operational forecasts through better model guidance and knowledge. The NWS is in the process of implementing a National Digital Forecast Database (NDFD) to give the NWS forecaster the ability to predict weather at spatial and temporal resolutions far greater than previously possible. The high resolution forecast model used by the Reno NWS WFO and the NDFD rely on a dense network of surface data, including the NDOT sites. Verification of high wind warnings and improvement to fire weather forecasts occurred as a direct result of having these data.
One of the main goals of this study was to establish a data assimilation system for the MM5 at DRI. The model was used both with and without the enhanced data set that included NDOT sites for two winter storm case studies. In the first case, the model successfully simulated a frontal system passing through Reno, NV, on March 8, 2002. One finding was that it is important to use an optimum set of data stations. However, each of the surface stations has micro location characteristics and generally is not representative for the larger area that needs to be captured by the model. Nonetheless, additional data from the NDOT stations significantly improved model results.
The second winter storm study looked at a storm that occurred in March of the next year. Again, the results indicated that the model runs that used assimilated NDOT data compared better with measurements for both temperature and wind speed when compared to model runs that did not include NDOT data.
From an NDOT perspective, improving weather forecasts is a benefit, but another important challenge is forecasting whether ice will form on roads. To evaluate how sensitive minimum pavement temperature forecasts are to meteorological inputs, DRI researchers tested NDOT's IceCastTMpavement model using the MM5 model results from the first case study as the baseline simulation. They then changed each of the input meteorological parameters by a certain value and ran the pavement model again for each case. These tests resulted in a large variation in the minimum pavement temperature, ranging from –7 to –12°C. Researchers learned that predicted pavement minimum temperature appears to be mostly sensitive to air temperature changes and to the total cloud cover and precipitation. Inaccuracies in the air temperature of 1 and 2°C cause a change in the minimum pavement temperature of more than 0.5 and 1°C, respectively. Significant underestimation of precipitation can yield a change in the minimum pavement temperature of about 2°C; inaccurate predictions of cloudiness can yield a change in the predicted minimum pavement temperature of more than 3°C.
The last project goal—to create a travelers' forecast Web site as one source of all information on road and traffic conditions—was not realized. The researchers plan to continue to work with NDOT to achieve this goal.
The conclusions of the principal investigators can be summarized as follows:
As was the case with the Nevada and Pennsylvania projects, the main objective of the State University of New York at Albany (SUNYA) project was to create a mesonet by integrating the data from the New York State DOT (NYSDOT) RWIS systems with other data sources. Other project goals included:
Assessing the quality of the RWIS data.
Good progress was made on some of the objectives, but others had to be abandoned, partly because of the loss of personnel at NYSDOT and partly because of difficulties that arose in dealing with the RWIS data. These difficulties included delays in the installation of new RWIS sites and problems downloading data from sites that use proprietary software. As a result, the major focus of the project shifted to an effort to download data from a few sample NYSDOT RWIS sites and to audit data quality from the RWIS sites by comparing them to nearby NWS sites.
Data quality for the RWIS stations proved to be a large problem. For example, data from one station indicated an apparent change in calibration and resolution in the wind sensor occurred midway through 2002, but the contractor indicated that there was no reason to question the data. Nonetheless, the data appears to be suspect, at least for part of the year. These kinds of issues still need to be resolved.
The program to establish a network of secondary road monitoring sites was delayed indefinitely. SUNYA tested one system, but this general objective was not achievable during the span of this project.
SUNYA completed a one-dimensional energy balance model for curing concrete bridge decks (the SUNY/Local Atmosphere Bridge Simulation (SLABS) model), which they intended to adapt to predict the state of road surfaces. However, for this task and the one to evaluate local obstacles, a data archive from the stations was needed. Because the data acquisition step required more time than expected, not enough time was available to develop a database for a large number of stations.
Three of the projects described above sought to build mesonets. The University of Utah project could be characterized as constructing a "mega-mesonetwork." The efforts funded by FHWA built on a project that began several years ago when the University of Utah began developing one of the first mesonets in the country. At the time this project began, the mesonet (called MesoWest) accessed data from approximately 2500 observing stations in the western United States. In 2001, most of the RWIS stations in the database were located in Montana, with a few along major interstates in Wyoming, Utah, and the Lake Tahoe area of California and Nevada. Over the course of this project, the network expanded considerably to more than 6000 stations with many sites (including RWIS stations) added from Colorado, Idaho, Oregon, and Washington.
Another important addition to the MesoWest database is weather information at 264 locations along the major Union Pacific rail corridors in the West. Most of the stations report temperature only; however, wind conditions are reported at many critical locations.
A specialized Web interface to the MesoWest database (www.met.utah.edu/mesowest) was developed and maintained for use by Utah Department of Transportation (UDOT) personnel. This interface has been used extensively for winter road maintenance as well as road construction and summer paving projects.
High resolution forecast models used in complex terrain will resolve the weather in adjacent valleys independently. However, observation corrections to the initial background field obtained from a model can bleed laterally through the terrain. To deal with this problem, which has affected the suitability of the analyses along some transportation corridors, a graduate student developed anisotropic weighting functions for the Advanced Regional Prediction System (ARPS) Data Analysis System (ADAS), a data assimilation system. These anisotropic weights are critical to resolving a systematic source of error in the analyses where weather observations in one valley affect the analysis in adjacent valleys.
An additional benefit of the relationships between the NWS, UDOT, and University of Utah was that all three groups worked closely to provide weather and road state information, as well as forecasts, to Olympic organizers, public safety personnel, and the public for the 2002 Winter Olympics and Paralympics. These efforts were judged to be highly successful by representatives of many different organizations involved in the Olympics.
Previous | Table of Contents | Next
Keywords: Mesonet, Artificial Neural Network, IceCast pavement model, Integrated Weather Data Network, MesoWest meso-meterological network
TRT Terms: Weather forecasting Research United States, Highway communications Research United States, Automobile driving Information services Research United States, Road weather information systems, Weather conditions, Research projects