The United States Environmental Protection Agency’s (EPA) MOVES emission factor model requires speed distributions by roadway type. Emissions are sensitive to speed and EPA guidance recommends the use of local speed data. Sources including cell phones, GPS devices, and ITS infrastructure are providing increasingly rich options for determining present-day speed distributions. However, emissions estimates are often needed for future years as well, and EPA guidance indicates that speed estimation methods for base and forecast years should be consistent. Travel demand forecasting models are generally the only option for providing consistent base and forecast year speed inputs, but such models have been optimized for demand estimation rather than speed estimation.
Speed post-processing methods can improve the estimation of speeds to more closely match observed speed distributions. As part of a National Cooperative Highway Research Program Project to develop MOVES input guidelines, alternative speed post-processing methods were compared with real-world sources in two cities, Atlanta and Jacksonville. Some key findings include: (1) Post-processing improved the speed distributions, and brought emissions estimates to within 5 percent or less of estimates based on observed speed distributions. (2) Modeled and real-world distributions still did not match well, reflecting theoretical and practical limitations of today’s travel demand models. (3) Modeled average speeds tend to be lower than real-world speeds, and the use of a single model “free-flow” speed fails to account for high-speed vehicle travel. (4) The Bureau of Public Roads equation is generally improved through the use of coefficients other than the standard values, but shows not enough travel in the lower speed ranges and too much in the 60-65 mph range. (5) The Akçelik equation provides the best match in the upper speed ranges, but overpredicts travel in the lowest speed ranges, especially under congested conditions.
If the analyst is in a situation where congestion is limited or is not likely to change substantially in the future, use of observed rather than modeled speeds may be preferable. Substantial improvements in speed prediction capabilities are likely to occur only with more widespread use of new-generation simulation models with dynamic traffic assignment.