UC Davis

For reasons related to traffic congestion, emissions, safety, physical activity and health, there has been an increased focus on active transportation modes, including cycling and walking, by transportation planners and policymakers in the United States. In this regard, estimating bicycle and pedestrian volumes is key to evaluating transportation systems, building new infrastructure, safety studies, and understanding the impact of policy changes. Researchers have used various methods to estimate these volumes but most of the studies are limited to a single city or include study locations only in urban areas. This study contributes to the existing literature by including study locations from rural areas and using unique explanatory variables from other tools such as the Strava Fitness app and the Bicycle Network Analysis (BNA) tool from PeopleforBikes’. I built a set of models using the Random Forest algorithm to predict Annual Average Daily Bicycle Traffic (AADBT) at the street level and Annual Average Daily Pedestrian Traffic (AADPT) at the intersection level. The dependent variable in the bicycle models is the AADBT calculated using permanent counts from San Francisco and San Diego and short-term counts from Caltrans District 1 (including Del Norte, Mendocino, Humboldt, and Lake counties). The data from rural locations is limited to four counties in Northern California and thus I built separate models (urban, rural, and generalized: urban + rural) to account for the time and space limitations in the counts. The dependent variable in the pedestrian models the AADPT calculated using the annual average crossing volumes from 1308 intersections in California. Unlike the bicycle count locations, pedestrian count locations are spread across various geographies and thus I developed a generalized pedestrian model that accounts for all neighborhood types.