Use of n-fold Cross-Validation to Evaluate Three Methods to Calculate Heavy Truck Annual Average Daily Traffic and Vehicle Miles Traveled

Transcription

1 TECHNICAL PAPER ISSN J. Air & Waste Manage. Assoc. 57:4 13 Use of n-fold Cross-Validation to Evaluate Three Methods to Calculate Heavy Truck Annual Average Daily Traffic and Vehicle Miles Traveled Shauna L. Hallmark and Reginald Souleyrette Iowa State University, Center for Transportation Research and Education, Ames, IA Stephen Lamptey Canada Ministry of Transportation, Toronto, Ontario, Canada Copyright 2007 Air & Waste Management Association ABSTRACT Reliable estimates of heavy-truck volumes in the United States are important in a number of transportation applications including pavement design and management, traffic safety, and traffic operations. Additionally, because heavy vehicles emit pollutants at much higher rates than passenger vehicles, reliable volume estimates are critical to computing accurate inventories of on-road emissions. Accurate baseline inventories are also necessary to forecast future scenarios. The research presented in this paper evaluated three different methods commonly used by transportation agencies to estimate annual average daily traffic (AADT), which is used to determine vehicle miles traveled (VMT). Traffic data from continuous count stations provided by the Iowa Department of Transportation were used to estimate AADT for single-unit and multiunit trucks for rural freeways and rural primary highways using the three methods. The first method developed general expansion factors, which apply to all vehicles. AADT, representing all vehicles, was estimated for short-term counts and was multiplied by statewide average truck volumes for the corresponding roadway type to obtain AADT for each truck category. The second method also developed general expansion factors and AADT estimates. Truck AADT for the second method was calculated by multiplying the general AADT by truck volumes from the short-term counts. The third method developed expansion factors specific to each truck group. AADT estimates for each truck group were estimated from short-term counts using corresponding expansion factors. Accuracy IMPLICATIONS Heavy trucks have different traffic patterns than passenger vehicles. Because the truck fleet represents a significant amount of vehicular VMT, reliable estimates of truck volume are key to the accurate estimation and prediction of emissions levels. Evaluation of the different methods to estimate heavy truck AADT indicated that use of expansion factors specific to a particular truck category resulted in the most accurate estimates of truck volumes. Inaccurate estimates of truck volumes can result in overestimation or underestimation of the heavy truck contribution to emission inventories. of the three methods was determined by comparing actual AADT from count station data to estimates from the three methods. Accuracy of the three methods was compared using n-fold cross-validation. Mean squared error of prediction was used to estimate the difference between estimated and actual AADT. Prediction error was lowest for the method that developed separate expansion factors for trucks. Implications for emissions estimation using the different methods are also discussed. INTRODUCTION Heavy-duty trucks make up slightly 3% of the on-road vehicle fleet. 1 In contrast, they account for 7% of vehicle miles traveled (VMT) on roadways in the United States. Heavy trucks are estimated to contribute a significant proportion of regulated ambient emissions, which includes particulate matter (PM), CO, oxides of nitrogen (NO x ), and volatile organic compounds (VOCs). The U.S. Environmental Protection Agency 2 estimates that the contribution of highway vehicles is 32% of NO x emissions with heavy trucks responsible for 38% of that amount. Other studies indicate that heavy trucks contribute as much NO x as passenger vehicles. 3 The total estimated highway vehicle contribution to VOCs is 30%, 9% of which comes from heavy trucks. They also contribute 13% of the CO emissions attributed to highway vehicles. Nationally, heavy trucks are also responsible for 65% and 75% of the on-road vehicle contribution to PM 10 and PM 2.5, respectively. 2 EXPERIMENTAL WORK Current Methodologies Used to Estimate Heavy-Truck VMT Information about truck traffic is necessary to meet federal reporting requirements and to assist state and local agencies in assessing system performance and needs. Estimates of truck percentages and volumes are also used to develop emission inventories. Approximately 70% of state departments of transportation (DOTs), including the Iowa DOT, use a traffic count-based method for estimating truck volumes and VMT. 4 For count-based methods, states maintain traffic count programs to collect volume data continuously at permanent count stations sites. Classification counts are also collected at a limited number of 4 Journal of the Air & Waste Management Association Volume 57 January 2007

2 permanent count stations. Short-term traffic counts are collected at other locations to estimate site-specific volumes. Short-term counts are usually collected for periods 48 hr. To account for temporal variations in short-term counts, data from permanent sites are used to develop expansion factors. Short-term counts are multiplied by expansion factors to estimate annual average daily traffic (AADT) and, subsequently, VMT. Factors are typically generated for each day of the week by month for separate road types. Vehicle classification data are used to estimate AADT and VMT by vehicle class. One count-based method used to estimate truck VMT requires the development of separate expansion factors for specific classes of heavy vehicles. AADT from shorterterm classification counts for a class of heavy vehicles is factored up using the expansion factors. Truck VMT for a highway segment is obtained by multiplying truck AADT by the length (center line mileage) of a roadway section. This method is resource intensive, and most DOTs use a more aggregate method to derive truck VMT, whereby generic expansion factors are developed that apply to all vehicle classes. A limited number of vehicle classification counts are used to estimate truck fraction. For short-term counts, the generic expansion factors are applied, and AADT for all vehicle types is estimated. VMT is calculated by multiplying AADT times the section length. Truck VMT is calculated by multiplying total VMT by the average truck percentages (by truck types) obtained from limited classification counts. Truck percentage may also be determined from short-term counts. Several studies have indicated problems with the use of aggregate expansion factors for estimating truck VMT or volumes. Although truck volumes, like passenger car volumes, vary over time and space, the pattern of temporal variability in truck volumes differs significantly from that of passenger vehicles. Trucks experience more variability between weekdays and weekends than passenger vehicles. As such, adjustment factors derived from aggregate count data (total volume) may fail to adequately explain temporal variations in truck traffic culminating in biased estimates of annual average daily truck traffic. Hu et al. 5 evaluated extrapolated data from permanent count stations and reported that more accurate estimates result for passenger vehicles than for heavy trucks and that estimates are more precise when volumes are high than for low-volume situations. Stamatiadis and Allen 6 reported that trucks experience more seasonal variability than passenger vehicles. They also observed more variability between weekdays and weekends for heavy trucks than for passenger vehicles. Hallenbeck 7 also observed that trucks do not exhibit the same seasonal patterns as passenger vehicles. As a result, seasonal estimates based on aggregate count data may fail to adequately explain seasonal variations in truck flow. Weinblatt 8 also indicated that although extrapolated traffic counts can be quite accurate in estimating VMT for systems of roads, less sophisticated methods are often used to estimate VMT by vehicle class resulting in less satisfactory results. They recommended using seasonal and day-of-week factors developed for several groups of vehicles classes to better reflect heavy-truck patterns and to reduce errors in heavytruck AADT estimates. Research Scope Heavy-truck VMT is a vital input to air quality models and significantly affects regional emission inventories. This paper discusses a project that evaluated three different traffic count-based methods to estimate heavy truck AADT. Because VMT is simply the product of AADT and section length, AADT is the variable used to evaluate the different methods. One year of count data was obtained from the Iowa DOT for all of the rural primary and rural freeway permanent count stations. Actual AADT for single-unit (SU) and multiunit (MU) trucks was calculated for each station. Three count-based methods commonly used by transportation agencies were also used to estimate AADT for each truck group for each station and were compared against actual AADT to evaluate the accuracy of each method. Data Automatic traffic recorder (ATR) station data from the Iowa DOT were used to compare the three different methods. The Iowa DOT maintains ATR data by functional class. Rural interstates and rural primary highways were selected for the analysis, because permanent count stations for urban interstates and arterials are sparse. The DOT Office of Transportation Data maintains 14 permanent ATRs on the Iowa rural interstate network. They also maintain a number of ATR stations on rural primary highways, of which 36 had relatively complete datasets for the analysis period. At each permanent count station, data are collected on an hourly interval, 24 hr/day, 365 days/yr. At some ATR sites, data are collected for three vehicle classes: passenger vehicle, SU truck, and MU truck. At others, all 13 classes of the Federal Highway Administration (FHWA) vehicle classification scheme are collected. However, many of the truck categories have low volumes. Expansion factors developed with low volume counts can result in unstable or unreliable expanded counts. To develop reliable seasonal and day-of-week truck adjustment factors, an aggregation of the FHWA 13 classification scheme into three or four vehicle categories is recommended. 9 To match stations where only three vehicles categories were collected, data were aggregated into the following categories: passenger vehicles (FHWA classes 1 3), SU trucks (classes 4 7), and (classes 8 13). Methodology The three methods were compared by comparing AADT estimated from short-term counts to actual AADT values for each permanent count station. AADT was compared using n-fold cross-validation. In n-fold cross-validation, data are split into n partitions and data from the nth partition used to validate the values estimated from the remaining data. For example, with four partitions, in the first iteration, data from partition n 1 are removed from the sample, and data from partitions n 2, n 3, and n 4 (referred to hereafter as the model dataset) are used to develop the model of interest. Data from partition n 1 (referred to hereafter as the validation dataset) are used to validate the model. For the second iteration, data from n 2 are used as the validation dataset, and data from n Volume 57 January 2007 Journal of the Air & Waste Management Association 5

3 1, n 3, and n 4 are used as the model dataset, and so forth. The 14 ATR rural interstate and the 36 rural primary stations were each divided into four partitions and evaluated separately. Actual AADT was calculated for each validation station using the American Association of State Highway and Transportation Officials (AASHTO) method, which effectively removes most biases from missing data, especially when the missing data are unequally distributed over time. 9 The AASHTO method for computing AADT is given by eq 1: AADT i 1 m 1 n 1 n k-1 VOL imk (1) where AADT is the actual AADT; VOL imk is the daily traffic for day k, of day-of-week i, and month m; i is the day of the week; m is the month of the year; k is 1 when the day is the first occurrence of that day of the week in a month and 4 when it is the fourth day of the week; and n is the number of days of that day of the week during that month (usually between 1 and 5, depending on the number of missing data). Data from the ATR stations in the remaining partitions served as the model dataset. Expansion factors for each day of the week for a given month were developed. The combined seasonal and day-of-week expansion factor was given by the ratio of the AADT to the monthly average day of the week traffic (MADWT) as given in eq 2: Table 1. Average MSEP for rural interstate. Variable Average MSEP for All Days and Stations Method 1 Method 2 Method 3 SU 161,331 61,490 34,028 MU 10,623,191 1,700, ,851 each of the 24-hr short-term counts was calculated for each station in the validation dataset by applying expansion factors using eq 3. AADT v f nim 24-hr_Vol v (3) where ADDT v is the estimated AADT for station v in validation dataset; f nim is the aggregate seasonal and dayof-week factor calculated for partition n for day of week i and month m; and 24-hr_Vol v is the volume for 24-hr short-term count for station v in the validation dataset. Because expansion factors for method 1 represented all of the vehicle classes, the estimated station-specific AADT was for all of the vehicle categories. Truck AADT for both heavy-truck categories was estimated by multiplying each station s estimated AADT by an annual average statewide truck percentage. The annual average statewide percentage of trucks reflects total truck volume over the year divided by total vehicle volume over the year for a specific roadway type for the model datasets using eq 4: f nim AADT j (2) MAWDT j where f nim is the aggregate seasonal and day of week factor calculated for partition n for day of week i and month m; ADDT j is the aggregate annual average daily traffic for station j in model dataset; and MAWDT j is the monthly average day-of-week traffic for station j in model dataset. Expansion factors developed from the model dataset were used to estimate AADT for each station in the validation dataset. To do this, three 24-hr counts were randomly selected from each station in the validation dataset to serve as short-term counts. Data were randomly selected during the summer months, because this is when most DOTs carry out their yearly counting program. AADT for the SU and MU truck categories were estimated for each of the three short-term counts using the expansion factors developed from the model dataset for each of the three different methodologies. Actual AADT was then compared against estimated AADT for each. Four iterations were performed with each partition serving once as the validation dataset. The three methodologies are more fully described in the following sections. Method 1. A single set of aggregate expansion factors was developed for all of the vehicle classes for each partition. Daily volume data for the ATR stations in the model dataset were used to develop monthly and daily expansion factors using eqs 1 and 2. An estimated AADT for P cn Vol cj (4) Total_Vol j where P cn is the annual percentage for vehicle category c for partition n; Vol cj is the volume of vehicles in vehicle category c for station j in the model dataset for partition n; and Total_Vol j is the volume for all vehicle classes for station j in the model dataset for partition n. Truck AADT for the SU and MU vehicle categories for each short-term count were calculated by multiplying the annual average statewide truck percentage for the corresponding vehicle class using eq 5: AADT cv1 AADT v P cn (5) where AADT cv is the average daily traffic for vehicle category c for station v in validation dataset for partition n for method 1; AADTv is the extrapolated AADT for station v in validation dataset for partition n; and P cn is the annual percentage for vehicle category c for partition n. Table 2. Average MSEP for rural primary. Variable Average MSEP for All Days and Stations Method 1 Method 2 Method 3 SU 11, MU 98,837 28,773 12,341 6 Journal of the Air & Waste Management Association Volume 57 January 2007

4 Figure 1. Monthly variations for rural interstate stations. (a) Rural Interstate Station 120; (b) Rural Interstate Station 100; (c) Rural Interstate Station 111; (d) Rural Interstate Station 104. Method 2. Aggregate expansion factors from method 1 were used to extrapolate AADT for each station in the validation dataset using eq 3 as for method 1. The difference between methods 1 and 2 is that truck percentages for the two heavy-truck categories were derived from each short-term count rather than using statewide truck percentages. Truck percentages from short-term counts were calculated using eq 6: P 24cvn 24_hr_Vol cvn 24_hr_Vol vn (6) where P 24cvn is the percentage of trucks for category c for station v in validation dataset for partition n for 24-hr short-term count; 24 hr_vol cvn is the 24-hr short-term count volume for vehicle category c for station v in validation dataset for partition n; 24hr_Vol vn is the 24-hr volume for all of the vehicle classes for station v in validation dataset for partition n. Truck AADT for the SU and MU vehicle categories were calculated by multiplying the truck percentage for that vehicle class (SU and MU) from eq 6 by stationspecific AADT using eq 7: AADT cv2 AADT v P 24cvn (7) where AADT cv2 is the average daily traffic for vehicle category c for station v in validation dataset for partition n for method 2; AADT v is the extrapolated AADT for station v in validation dataset for partition n; and P 24cvn is the percentage of trucks for category c for station v in validation dataset for partition n for 24-hr short-term count. Method 3. Daily volume data for the ATR stations in the model dataset were used to develop a separate set of monthly and daily expansion factors for each heavy truck category using eqs 1 and 2. Truck AADT for each 24-hr short-term count for each heavy truck category was extrapolated using vehicle-specific expansion factors as described by eq 8: AADT cv3 f cnm 24_hr_Vol cv (8) Volume 57 January 2007 Journal of the Air & Waste Management Association 7

5 Figure 2. Daily variations for rural interstate stations. (a) Rural Interstate Station 120; (b) Rural Interstate Station 100; (c) Rural Interstate Station 111; (d) Rural Interstate Station 104. where ADDT cv3 is the average daily traffic for vehicle category c for station v in validation dataset for partition n for method 3; f cim is the aggregate seasonal and day-ofweek factor for vehicle category c for partition n for day of week i and month m; and 24 hr_vol cv is the 24-hr shortterm count volume for vehicle category c for station v in validation dataset for partition n. RESULTS n-fold Cross-Validation AADT was computed for each station for each of the two heavy-truck categories using the three methods described in the previous section. Actual AADT was also calculated for each station and compared against the values estimated using the three methods. Accuracy of the different methods was compared using mean squared error of prediction (MSEP). MSEP was calculated by averaging the squared error between the estimated and actual AADT for each vehicle category as given in eq 9: MSEP AADT cv AADT actual 2 n (9) where AADT cv is the estimated AADT for vehicle category c for station v; AADT actual is the actual AADT for vehicle category c for station v; and n is the number of stations. MSEP values for the three methods are provided in Table 1 for rural interstates and in Table 2 for rural primary roads for the two heavy-truck categories. The prediction error for method 2, which used generic expansion factors but specific truck percentages, is significantly lower than method 1. However, method 3, with its direct and site specific estimation of truck AADT, had the lowest error. Also note that error terms are most significantly reduced by succeeding methods in the MU truck category. Hourly, Weekly, and Monthly Variations In addition to evaluating the different methods using n-fold cross validation for the two heavy-truck categories, differences between passenger vehicle and heavy-truck travel were also evaluated graphically. Figure 1 illustrates the fraction of total yearly volume by vehicle class that occurs during a specific month of the year for four of the rural interstate stations. For example, at station 120, 8% of the yearly volume for occurs in January, 8% occurs in February, 8.5% occurs in March, and so forth. As 8 Journal of the Air & Waste Management Association Volume 57 January 2007

6 Figure 3. Hourly variations for rural interstate stations (Monday in July). (a) Rural Interstate Station 120; (b) Rural Interstate Station 100; (c) Rural Interstate Station 111; (d) Rural Interstate Station 104. shown, passenger vehicle and SU truck patterns are more similar than. Passenger and SU truck volumes peak in the summer months, reflecting discretionary travel patterns, whereas MU volumes are more constant over the year, reflecting more inelastic business patterns. Figure 2 illustrates daily variation for the same four rural interstate stations. In general, higher truck volumes occur during the weekdays (Monday through Friday) with much lower volumes on weekends for both truck groups. Passenger vehicles peak on Friday and have higher weekend volumes. Figure 3 shows diurnal volume variations for the same stations for a typical Monday in July. Passenger vehicle and SU truck volume appear to follow similarly varying hourly trends, whereas MU truck volumes exhibit less variation. Similar information is presented for four of the rural primary stations with Figure 4 illustrating monthly (or seasonal) variation. Variation by day of week is shown in Figure 5. As observed on the interstate system, rural primary SU truck volumes are similar to passenger vehicle volumes and less like, which peak early in the week. Figure 6 displays diurnal variation for a Monday in July at the four rural primary stations. Although varying somewhat by station, the hourly volumes are more consistent than their interstate counterparts. In particular, MU truck volumes peak at different times than do those of SU or passenger vehicles. Implications for Emissions As discussed, heavy trucks contribute disproportionately to pollutant emissions. Because the truck fleet represents a significant amount of vehicular VMT, reliable estimates of truck volume are key to the accurate estimation and prediction of emissions levels. This paper showed that the method chosen to estimate heavy-truck VMT can significantly affect resulting output. Expansion factors that represent all of the vehicle categories mask spatial and temporal variations of the difference between passenger vehicle and heavy-truck flows. Use of generic expansion factors may overestimate or underestimate emissions depending on when short-term counts are taken. These errors result in corresponding errors in emissions inventories. Accurate estimates of truck volume are also important to forecasted pavement conditions and travel conditions affecting both highway capacity and safety. Similar accuracy needs may be anticipated for these and other important applications areas. Volume 57 January 2007 Journal of the Air & Waste Management Association 9

7 Figure 4. Monthly variations for rural primary stations. (a) Rural Primary Station 205; (b) Rural Primary Station 208; (c) Rural Primary Station 221; (d) Rural Primary Station 238. Differences in actual truck volumes that would result using the three different methods are presented for one of the 24-hr counts for Rural Interstate Station 102 to illustrate the impact on truck volumes. Station 102 was used in the validation dataset for partition N 3. The 24-hr count selected for illustration was for a Thursday in June. The expansion factors developed for partition N 3 for a Thursday in June were 0.88 for all vehicles, 0.73 for SU trucks, and 0.81 for. The total 24-hr volume for Station 102 was 13,173 vehicles, the 24-hr volume for SU trucks was 666 vehicles, and the 24-hr volume for MU trucks was 2658 vehicles. As discussed in the previous sections, for method 1, an aggregate expansion factor was used to estimate AADT for all of the vehicles from the 24-hr count, and then annual truck percentages were applied to estimate single and MU truck AADT. The annual truck percentages for partition N 3 were 4.1% SU and 31.1%. The estimated AADT for single and using method 1 was calculated using the following steps: all vehicle types SU trucks 24-hr total count aggregate expansion factor 13,173 vehicles ,592 vehicles aggregate adjusted AADT annual percentage SU 11,952 vehicles 4.1% 469 SU trucks aggregate adjusted AADT annual percentage MU 11,952 vehicles 31.1% 3603 For method 2, an aggregate expansion factor was used to estimate AADT for all vehicles, and then truck percentages from the 24-hr count were used to estimate truck AADT using the following: SU trucks 24-hr total count for SU expansion factor for SU 666 SU SU trucks 10 Journal of the Air & Waste Management Association Volume 57 January 2007

8 Figure 5. Daily variations for rural primary stations. (a) Rural Primary Station 205; (b) Rural Primary Station 208; (c) Rural Primary Station 221; (d) Rural Primary Station 238. all vehicle types Percent SU trucks for 24-hr count Percent for 24-hr count SU trucks 24-hr total count aggregate expansion factor 13,173 vehicles ,592 vehicles 666 SU 13,173 vehicles total 5.1% 2658 MU 13,173 vehicles total 20.2% aggregate adjusted AADT 24-hr percentage SU 11,952 vehicles 5.1% 586 SU trucks aggregate adjusted AADT annual percentage MU 11,952 vehicles 20.2% 2339 For method 3, expansion factors specific to each vehicle category were used to estimate truck AADT using the following: aggregate adjusted AADT annual percentage MU 11,952 vehicles The AADT estimated from each method was compared against the actual AADT for station 102 as shown in Table 3. Total AADT is the same for all methods, because it is calculated using an expansion factor that represents all of the vehicle classes. Total AADT was underestimated by 169 vehicles (1.5%). As indicated, method 3 overestimated SU trucks by only 1 vehicle (0.2%), whereas method 1 overestimated by 16 vehicles (3.2%), and method 2 overestimated by 101 vehicles ( 21%). All of the three methods overestimated MU trucks with method 3 overestimating the least by 101 vehicles ( 5%). Method 2 underestimated by 287 vehicles (14%), and method 1 significantly underestimated by 1551 vehicles ( 76%). Volume 57 January 2007 Journal of the Air & Waste Management Association 11

9 Figure 6. Hourly variations for rural primary stations (Monday in July). (a) Rural Primary Station 205; (b) Rural Primary Station 208; (c) Rural Primary Station 221; (d) Rural Primary Station 238. Use of the different methods may result in overestimation or underestimation of heavy-truck volumes and, subsequently, VMT. If link speeds were not affected by estimates of AADT, a straight-line relationship between estimated volumes and emissions would suggest that method 1 would have overestimated heavy-truck volumes and, subsequently, estimates of VMT and emissions by 76%. Additionally, overestimation or underestimation of link volumes may affect link speeds output by travel demand forecasting models. A common speed-volume relationship used in travel demand models is shown in eq As indicated, link travel time is a function of the link volume and link capacity. If overestimation of the number of heavy trucks on a link also results in overestimation of total link volumes, speed estimates would decrease. If underestimation of heavy truck volumes also resulted in underestimation of total link volume, calculated speeds would be higher than for scenarios where truck volumes were more accurately estimated. Because emission rates are correlated to average speeds, overestimation or underestimation of emissions could also affect calculation of emission rates as well. Table 3. Results for comparison of methods for station 102. Variable Actual AADT (vehicle/day) Estimated AADT (vehicle/day) Difference (vehicle/day and % difference) Method 1 Method 2 Method 3 Method 1 Method 2 Method 3 SU (3.2%) 101 ( 20.8%) 1( 0.2%) MU ( 75.6%) 287 ( 14.0%) 101 ( 4.9%) Total vehicle 11,423 11,592 estimated AADT for all vehicles is the same for all methods 169 ( 1.5%) 12 Journal of the Air & Waste Management Association Volume 57 January 2007

10 t t o V/C 4 (10) where t is the travel time on link; t o is the free-flow travel time; V is the volume on link; and C is the capacity of link. DISCUSSION This research evaluated three different methods to calculate heavy-truck AADT, which is used to estimate VMT and, subsequently, emissions. Traffic data from continuous count stations provided by the Iowa DOT were used to estimate AADT for two different truck groups (SU and MU) using the three methods. The first and second methods developed general expansion factors for all of the vehicles. Truck AADT was calculated by multiplying short-term counts by generic expansion factors and truck percentages. Truck percentages for the first method were based on the overall statewide annual percentage of trucks. The second method used daily truck percentages from short-term counts. The third method developed monthly and daily expansion factors for each truck group, thereby directly estimating truck AADT. Truck AADT was calculated by applying expansion factors to short-term counts using each method. Accuracy of the three methods was compared using n-fold crossvalidation whereby data are divided into n partitions. Data from the each partition were then used to validate remaining data, which were used to develop models for each road type. Short-term counts were sampled from the validation dataset and used to estimate AADT using each of the three methods. Estimated AADT was then compared with actual AADT by heavy-truck category for each count station for each method. MSEP was used to compare each AADT estimation method. Overall, method 3 produced the best results by far. Monthly, daily, and hourly traffic patterns for both heavy-truck categories, as well as passenger vehicles, were also evaluated. Significant variation exists in the temporal patterns of heavy trucks as compared with passenger vehicles. This suggests that although site-specific truck AADT estimates in this study were found to contain far less error than aggregate method-based estimations, caution should be exercised with the use of short-term counts where spatial variation exceeds temporal variation. ACKNOWLEDGMENTS The authors thank the Iowa Department of Transportation and the Midwest Transportation Consortium for funding this project and providing the necessary data. REFERENCES 1. U.S. Department of Transportation, Federal Highway Administration. Annual Vehicle Miles of Travel and Related Data; FHWA-PL ; Federal Highway Administration: Washington, DC, U.S. Environmental Protection Agency, Office of Air Quality Planning and Standards. National Air Pollutant Emission Trends, ; EPA- 454/R ; U.S. Environmental Protection Agency: Washington, DC, Sawyer, R.F.; Harley, R.A.; Cadle, S.H.; Norbeck, J.M.; Slott, R.; Bravo, H.A. Mobile Sources Critical Review: 1998 NARSTO Assessment; Atmos. Environ. 2000, 34, Benekohal, R.F.; Girianna, M. Evaluation of Methodology for Determining Truck Vehicle Miles Traveled in Illinois; Illinois Transportation Research Center, Illinois Department of Transportation: Chicago, IL, Hu, P.S.; Wright, T.; Esteve, T. Traffic Count Estimates for Short-Term Traffic Monitoring Sites: Simulation Study; J. Transport. Res. Board 1998, 1625, Stamatiadis, N.; Allen, D.L. Seasonal Factors Using Vehicle Classification Data; J. Transport. Res. Board 1997, 1593, Hallenbeck, M. Seasonal Truck Volume Patterns in Washington State; J. Transport. Res. Board 1993, 1397, Weinblatt, H. Using Seasonal and Day-of-Week Factoring to Improve Estimates of Truck Vehicle Miles Traveled; J. Transport. Res. Board 1996, 1522, Federal Highway Administration, U.S. Department of Transportation. Traffic Monitoring Guide. Federal Highway Administration: Garber, N.J.; Hoel, L.A. Traffic and Highway Engineering, 3rd ed.; Brooks/Cole: About the Authors Shauna Hallmark is an associate professor and Reginald Souleyrette is a full professor in the Civil, Construction, and Environmental Engineering Department at Iowa State University. They both additionally are affiliated with the Center for Transportation Research and Education in Ames, IA. Stephen Lamptey is currently employed as Project Manager, Central Region Traffic Office, Canada Ministry of Transportation, Toronto, Ontario, Canada. Address correspondence to: Shauna L. Hallmark, Iowa State University, Center for Transportation Research and Education, 2711 South Loop Drive, Suite 4700, Ames, IA ; phone: ; fax: ; shallmar@iastate.edu. Volume 57 January 2007 Journal of the Air & Waste Management Association 13