Page 29 - A bird’s-eye view of recreation - Rogier Pouwels

Developing tools and rules of thumb for managers
2.3.4 Visitor density analyses
A random forest model was constructed to estimate which landscape and environmental features account for spatial variation in visitor densities in the area.
For the analysis all maps were converted into a 10 x 10 m grid. This resolution was
chosen to avoid information on different paths being assigned to one cell. We used
the implementation by Wright and Ziegler (2017). Their implementation of random 2 forests follows that of Breiman (2001) and is also suitable for large data sets. The
frequency of use of a path by visitors from a specific car park was used as the response
variable (‘y’ variable).
For practical reasons (data reduction to make the calculations feasible) only locations within 5 km of the car parks, as the crow flies, were taken into account; these amounted to more than 99% of the single data points of the GPS tracks. The model we constructed consists of 500 regression trees, each of which is based on a bootstrap sample from the original data. Each bootstrap sample has the same size as the original data and was obtained by simple random sampling with replacement. This means that some records of the original data set occur more than once, and some never. Data that were not in the bootstrap sample were used for ‘out-of-bag’ validation. The importance of the explanatory variables used (see Table 1) was computed in three steps. First, the out-of-bag mean squared error was computed for each tree. Then, this statistic was also computed for each tree after permuting each predictor variable. Finally, the difference between the two mean squared errors was averaged over all trees (Liaw and Wiener 2002).
2.3.5 Assessment of potential management interventions as an illustration of a practical application
To illustrate how the data and tools could be applied to support decision making we designed three potential management interventions and estimated the visitor densities for the whole area. The visitor densities were used to assess the impact of the interventions on the Nightjar population by comparing it with the current situation. We chose management interventions that restrict visitors by temporary or permanent closures of car parks as these are one of the most commonly used methods of reducing visitor densities in sensitive parts of nature areas (Hammitt et al. 2015). The three possible interventions assessed are: 1) closure of small car parks, 2) closure of relatively isolated car parks that are located near areas with many Nightjars and 3) closure of all but 20 car parks to concentrate visitors near the border of the area or near villages, for example Lymington and Lyndhurst (see Appendix 3 for more details on the chosen interventions).
27