Modelling visitor attendance levels subject to day of the week and weather: a comparison between linear regression models and regression trees

Abstract

We have used two different statistical methods to model the dependence of visitor attendance levels for various user types in nature conservation areas on the day of the week and weather conditions. Both approaches – linear regression and regression trees – have been applied to data from a case study in an Austrian national park. The resulting models have been compared in regard to their descriptive quality, their interpretability, and their relevance for management practice. Reliable models based on linear regression were obtained for the daily totals of visitors as well as for specific user groups with high visitor loads, i.e. hikers and bikers. Both the day of the week and various meteorological variables were found to have significant influence on the visitor numbers. Linear regressions for the usage patterns of joggers and dog walkers were much less successful, suggesting only weak dependence on day of the week and weather related factors for these user groups. The regression trees based on meteorological data and the day of the week also work well for daily totals and bikers, slightly less so for hikers. The model quality was again much worse for smaller visitor groups as dog walkers and joggers. The most elaborate relationship between meteorological data and visitor numbers was found for hikers, whereas joggers appear to be completely oblivious to weather. A direct comparison between linear regression and regression trees shows comparable predictive power, with slight advantages for the linear regression, whereas the regression trees allow a simple visual assessment of the results and their relevance for managerial decisions, especially with regard to the identification of typical weather scenarios and the associated visitor attendance levels.