I suggest that predator intelligence may be responsible. The hypothesis that predators can learn the locations of artificial nests has been tested previously using arboreal nests. Yahner & Mahan (1999) examined avian and non-avian predation between fixed-location and randomly relocated nests, and found no such learning effexct. Similar analysis of learning behaviour has been performed by Anglestam (1986), who used cryptic nests in edge predation experiments, but was also unable to find a significant effect of learning. However, most learning studies utilise arboreal nests, and a previous study on ground nests using chicken eggs in a heavily fragmented forest indicated the predators had become habituated to the nests, and sought them out \cite{Yahner_1996}. This risk is particularly prevalent in small isolated forest fragments, where populations are small and the rate of migration is low \cite{Krebs_2002,articlea}. In such a small area, mammalian predators can learn to associate olfactory cues with artificial nests, and previous studies have indicated that mammalian predators already use olfactory cues to locate artificial nests containing plasticine eggs \cite{Purger_2012}. While plasticine scent was controlled for in this study, the nests were manufactured from conifer leaves, which are not present in the experimental forest fragment, and their cryptic scent and appearance which may provide routes for habituation. Population densities of predator species in the forest fragments should be examined in future study to examine evidence of possible learning behaviour, and hence the usefulness of artificial nest experiments in this small forest fragment.
Another possible mechanism for this unusually high level of predation is the mesopredator release effect. In the absence of large predators, mid-ranking predators can take over the ecological role of apex predators in the food web \cite{SOULE_1988}. In unfragmented habitats, apex predators can exert top down ecological control of an ecosystem, which has indirect benefits on prey populations by suppressing mid-level predators \cite{Elmhagen_2007}. However, larger species are more vulnerable to habitat fragmentation due to their larger range size, meaning mesopredator release may become a growing problem for an increasingly fragmented U.K. forest habitat \cite{Prugh_2009}. Future study in this area which should aim to examine the populations of apex predators and mesopredators in this woodland, and examine whether the mesopredator release hypothesis is a good fit for these findings. If so, apex predator control measures may be a valid conservation action for ground nesting bird populations through top-down control of rodent populations.

Limitations

Artificial nest experiments have inherent flaws, and the results are not always directly applicable to wild bird populations. Artificial nests can overestimate predation rates, by attracting different predators and may not mimic the seasonal variation in predation rates observed in natural nest experiments \cite{Zanette_2002,WEIDINGER_2008}. However, comparisons between nest height and edge proximity to examine the factors that drive predation are useful model of underlying predator behaviour, and continue to be a useful tool in avian nesting success studies. 

Acknowledgements

I would like to thank Cristina Banks-Leite of Imperial College London for help and advice regarding the pilot study, statistical analysis and invaluable guidance throughout the project. I would also like to thank the technical staff at Silwood Park for their assistance using the GPS unit for spatial analysis. Many thanks to the Department of Life Sciences at Imperial College London, and specifically the Biology Undergraduate department, without whom this project could not be possible.