Predicting trail condition using random forest models in urban-proximate nature reserves

Abstract

Monitoring and managing the condition of recreational trails is a time- and resource- intensive process often requiring significant field data. We developed a method for predicting trail condition over 183 km of trails in three urban-proximate nature reserves in Orange County, California using field data from 118 km of trails and random forest (RF) models. Further, we use data from the fitness tracking application Strava to measure recreation use intensity, activity type, and spatial dimensions of visitor use. Our results indicate 30 km of trails in two nature reserves that are at risk of significant trail degradation. Additionally, trail grade, NDVI, and use-related factors such as activity type and use intensity were ranked among the most important variables for predicting trail condition. Variable importance measures produced by RF models can help inform site-specific trail management that takes environmental, managerial, and use-related factors into account. We argue that RF models, in combination with Strava data, are powerful tools for outdoor recreation monitoring and management.