Page 70 - A bird’s-eye view of recreation - Rogier Pouwels

A bird's-eye view of recreation
4.4.2 Local impact of recreation on bird densities
For all three species the base model with habitat variables explained most of the deviance. Adding the recreation variables to these models showed significant negative impacts of recreation on the breeding densities of all three species (Fig. 2). For Woodlark the best model showed good performance, explaining almost 86% of the deviance and included the kernel density method as a variable for recreation pressure. For Stonechat the best model also included the kernel density method and explained 71% of the deviance. The best model for Nightjar explained 64% of the deviance and included the trail network method (Appendix 7). The Nightjar shows the strongest effects of recreational pressure as it declines to 50% of its density at visitor densities of 50,000 visitor groups per ha per year. Under high recreation densities Woodlark densities drop by 70% (Fig. 2).
Woodlark
Stonechat
Nightjar
Breeding bird density (index)
0 20 40 60 80 100
0 20 40 60 80 100
0 20 40 60 80 100
0 20000 40000 60000 80000 Recreation pressure (visitor groups per ha per year)
0 20000 40000 60000 80000 Recreation pressure (visitor groups per ha per year)
0 10000 30000 50000 Recreation pressure (visitor groups per ha per year)
Figure 2. Relationship between recreational use and densities of Woodlark, Stonechat and Nightjar (shown as an index where the density in the absence of recreation is set to 100). Standard errors are shown in grey. Recreational use is given in visitor groups per ha and is based on the results from the scientific tools FORVISITS and MASOOR-SCAN.
In order to test the predictive power of the models a 5-fold cross validation was performed for the Woodlark model by using 80% of the observations for the model and 20% for the validation of the model. The average correlation between the predictions of the model and the 20% independent observations was 75.3% with a minimum of 67.6% and a maximum of 81.8%. The correlation of the full model between predictions and observations was 84.7%.
68