Stephen P. Prisley, and J. Steven Carruth

GPS Speeds Data Collection on GIS Road Networks

ABSTRACT

Many GIS databases provide sufficient detail and accuracy for distance computations in network analysis. However, while travel distances depend only on the route, travel times depend on speed limits, traffic signals, congestion, road conditions, and time-of-day. Collecting data on actual travel times has been prohibitively expensive. However, because Global Positioning Systems (GPS) provide location and time data, they can serve as automated travel time data recorders. An example application is discussed, in which GPS units mounted on log trucks collected information which was loaded into a GIS and used to attribute a road network for truck routing applications. The greatest advantage of this procedure is that data collection can be done without additional personnel, during normal logging activities. An added benefit is that the data can be used to verify the spatial accuracy of the road network, identifying omitted or misaligned road segments. Using GPS for collection of vehicle speed data in this manner has allowed for the incorporation of inexpensive, realistic travel time data and resulted in more reliable network analyses.

INTRODUCTION

As part of Westvaco's Forest Research program, the Harvesting Project is evaluating ways of more efficiently moving forest products from the woods to delivery points. This transportation component of the overall harvesting operation accounts for significant costs and therefore provides opportunities for substantial savings from improvements in efficiency. For several years, the Harvesting Project has been using GIS technology to study the transportation process and identify areas in which GIS network analyses can provide information not readily available elsewhere.

The project has already identified several areas in which GIS has the potential to expand our analysis capabilities. For instance, many of the rural areas in which forest products are harvested and transported have lower-grade roads and bridges. In the South Carolina coastal plain, weight restrictions are being placed on many older bridges. These weight restrictions can limit the possible routes for transporting wood to processing facilities, and can increase logging costs not only by increasing the miles traveled, but also by lengthening the travel time and reducing the number of loads of wood that can be moved daily. Using road databases and NETWORK software, the impact of bridge restrictions can be readily quantified so that alternatives can be evaluated.

Another application for GIS in wood transportation analysis is that of balancing the production effort in harvesting. Logging contractors typically run two operations in the process of harvesting and delivering wood. First, an in-woods operation cuts trees and moves them to log decks adjacent to woods roads where they can be loaded onto log trucks. Second, the log trucks make numerous daily round trips between the log deck and a wood delivery point, such as a mill or woodyard. Significant efficiencies can be gained when the production from the in-woods operation is in balance with the ability of the trucking operation to deliver wood; i.e., neither side of the operation spends time waiting on the other. In order to achieve this balance, contractors must know the production rates for their in-woods operations (tons per hour) and must utilize the appropriate number of log trucks given the distance to the mill. When combined with timber stand volume information, travel times from the stand to the wood delivery point provides the essential data for balancing the in-woods and over-the-road operations.

In order to effectively employ GIS to solve some of these transportation problems, a detailed and accurate road database is essential. First, the road database must be complete, especially in the rural areas where forest operations are concentrated, and which are far away from the urban centers which seem to be the first places where road are built and marketed. Next, the roads that exist in the database must be properly categorized. Transportation costs and times can vary significantly with the road surface, width, and condition. Thus, a road database which has all necessary road segments in their proper geographic location is still not useful until the roads can be reliably classified into meaningful categories. Finally, realistic average speeds for each of these road categories is essential for performing quickest route analyses, where time is more crucial than distance.

For our evaluation projects, we found that by using a combination of public data (TIGER files) and internally digitized road layers (taken from USGS 1:24,000 scale quadrangle maps), the completeness and spatial accuracy of our road databases was adequate. However, the classification of roads did not provide reliable information about travel speeds and times. It was noted early in our project that this would be a primary obstacle to overcome before useful analyses could be performed with the GIS. It takes only one improperly coded road arc to destroy the credibility of a quickest route analysis.

As we evaluated different techniques for attaching attributes to roads in our databases, we were also evaluating the application of Global Positioning Systems (GPS) to GIS data collection. It became immediately obvious that GPS not only offers a means of recording location of roads, but through the time component that is inherent to GPS operation, we could also record travel times and speeds. The advantage offered by such a GPS data collection process was that data could be collected in an unmanned, automated manner while normal transportation activities were taking place, which meant costs would be far lower than other alternatives we were considering.

GPS DATA COLLECTION PROCEDURE

The idea behind a GPS data collection process for road attributes was that GPS units would be installed on log trucks, with the consent of the contractors. Data would be collected during normal business activities. The units would be initialized in the morning and left on during an entire business day. Therefore, the data collected during numerous round-trips between the woods and the delivery point would reflect local traffic conditions, time-of-day, and any alternate routes chosen. By compiling data from several days of multiple round-trips, a realistic sample of travel times and speeds was obtained. For those roads frequented by log trucks, an impressive amount of data could be collected without any additional effort beyond turning on the GPS receivers at the beginning of each day, and uploading data files at day's end.

Several trials of this process have been run, using Trimble Pathfinder BasicPlus receivers installed on log trucks operated by contractors. GPS positions were collected at five-second intervals; with a 10,000-point storage capacity, these receivers could collect nearly fourteen hours of continuous data. These data files were uploaded to PCs, where they were differentially corrected against base station data collected during the same operating times. Using Trimble's PFINDER software, the corrected points were then output into ASCII text files with X and Y coordinates (in the same projection system as the GIS road database), and time-of-week (seconds from the beginning of the week; a number from 0 to about 605,000). Note that while some GPS software allows the recording and uploading of actual velocity data, the velocities recorded are influenced by errors inherent in the GPS data (such as Selective Availability). Wildly inaccurate velocities can be noted in uncorrected data, and most differential correction programs do not correct the velocities, but only adjust the point coordinates.

Next, a program was written to read the GPS text file and calculate a speed for each point. This was done by computing the distance and time from the previous point and the distance and time to the next point, and averaging to obtain a velocity (in miles per hour) at each point. The program then outputs two files: a GENERATE format file to create an ArcInfo point coverage, and a text file containing point ids, speeds, and times.

Within ArcInfo, an AML is executed to GENERATE the point coverage using the text file created in the preceding step. An INFO table is also created to contain the point number, time, and speed, which are read from the second text file created previously. This INFO table is then joined to the PAT of the point file. Therefore, we then have a point coverage with speed and time as point attributes. What we really want are average speeds by road arc, so a few more steps are needed. Using the ARC NEAR command, each GPS point (within a specified distance to a road in the road coverage) is tagged with the arc-id of the nearest road. By judicious setting of the distance tolerance, we can identify GPS points that are not sufficiently close to a road arc to be tagged. This allows for later examination of untagged GPS points which typically indicate a missing or misaligned road segment in the road coverage.

Now, since each GPS point is labeled with a corresponding road arc-id, we can run STATISTICS on the PAT of the GPS point coverage, and for each arc-id identified, we can obtain the number of GPS points assigned to that arc, and the average speed of the GPS points assigned to the arc. Finally, a simple RELATE of the INFO file created by the STATISTICS command allows us to update the road coverage AAT with the FREQUENCY and SPEED items contained in the statistics file. The result: our road coverage now contains accurate average speeds for all road arcs traveled by the log trucks during the data collection period.

The next step was to designate travel speeds for the roads on which no data had been collected with the GPS units. Given a consistent, comprehensive classification system for roads in the database, it is possible to use the sample of roads traveled during the data collection phase to obtain average travel speeds by road class. All arcs in a road class can then be assigned the average travel speed for the class with some confidence. Thus, even a road which had a recorded average speed of 43.7 mph may be designated with a travel speed of 45 mph based upon its classification. While this means that individual arcs may exhibit disagreement between the GPS-recorded speed and the class average speed, there will be consistency among arcs within the same road class.

We noted several conditions which require correction at this point. First, some untraveled arcs will be attributed improperly. This is caused by some GPS points being attached to side streets rather than the traveled route due to small errors in GPS and/or road positions. These cases are usually easy to identify, since these side streets will have only a few GPS points attached, and only near the ends of the arc. Another error condition is the missing or mislocated road arcs which can be readily identified by observing which GPS points remained untagged after the NEAR command.

ADVANTAGES AND DISADVANTAGES

This procedure offers many advantages, but clearly does not resolve all the difficulties in attributing a road database with travel time information. Several of the advantages have been noted earlier, foremost being the automated manner in which data can be collected. With low-cost GPS units, a single person can coordinate data collection (installing units, uploading and correcting data) for a small fleet of vehicles. The GPS points collected provide evidence useful in updating both road locations and attributes. The frequency of round-trips provides a sample of travel times that can reflect varying traffic conditions across times during the day and days of a week. Collecting data during numerous trips alleviates misinterpretations of travel time due to unusual circumstances such as delays due to accidents.

Additional analyses are possible with the data collected in this procedure. For example, turnaround times (for loading and unloading of trucks) can be evaluated, as can average delays at traffic signals. This is done by designating points where delays are noted, such as busy intersections, stop lights, or delivery points. Then, by creating buffers around these points, we can select GPS points falling within a specified distance of these locations. Using the TIME item included in the GPS point file, we can subtract the minimum time from the maximum time for points within the buffer zone (for a particular trip) and obtain the elapsed time spent at these locations. This information may be useful in evaluating turnaround times for loading and unloading, or even for attributing turntables with delay times.

The main disadvantage in this procedure is that not all road arcs are automatically attributed. However, if road classifications are consistently applied in a database, it is likely that realistic averages can be obtained for each road class, and these averages can then be applied with confidence to road arcs not traveled during the data collection process.

SUMMARY

Many applications of GPS for GIS data collection can be found in the literature. Most of these applications involve a person visiting a location and manually recording attributes. While these applications can be vastly more efficient than any alternatives, they still require one human per GPS unit, and may not provide all the necessary information. For example, if roads are attributed simply by driving them once while recording information such as surfacing material and condition, any observations about travel speed are influenced by variables such as day of the week, time of day, traffic conditions, etc. Multiple trips are required to obtain realistic travel time estimates, and costs escalate with each trip if the GPS units must be manned.

It is fairly easy and quite useful to take advantage of the time component of a GPS signal and thereby obtain a record of not only where the GPS unit was located, but when. Use of the time portion of a point record allows applications of GPS technology in time-and-motion studies, traffic analyses, and efficiency studies such as the one reported here. The result of this effort will be a more realistic database of travel times, which should lead directly to more reliable NETWORK analyses.

Stephen P. Prisley
Forest Resources Information Manager
J. Steven Carruth
Harvesting Project Leader
Westvaco Corporation
180 Westvaco Road
Summerville, SC 29484
Telephone: (803) 851-4709
Fax: (803) 851-4706