Abstract:Current motor vehicle emission modelling efforts in the United States are focusing on forecasting carbon monoxide, hydrocarbon and oxide of nitrogen emissions based on the multivariate distribution of modes of vehicle operations. Examples of variables that describe ''modal'' operations include power (acceleration x velocity²) thresholds, acceleration rates, cruise speeds, and idle duration. Although these models promise to improve the accuracy of emissions forecasts, we currently lack models to forecast joint distributions of speed and acceleration needed as input. This paper presents interim results of a three-year investigation aimed to develop a statistical model capable of forecasting joint distributions of modal activity. Our initial effort has been to develop an experiment in which modal activity distributions can be collected on freeways and related to various traffic, roadway and control conditions. To streamline a potentially huge full-factorial experiment, we employ the FRESIM simulation model to eliminate second-order effects, leaving only main effects in the experiments. We employ a heuristic hierarchical tree-based regression (HTBR) method to identify important factors. We then parameterise the HTBR model to demonstrate how modes of operation could be forecast given ''real'' roadway, traffic, and traffic control conditions.