Linking wild turkey hen movement data to nesting behavior

Monitoring of wild turkey (Meleagris gallopavo) females to obtain representative measures of apparent nesting and nest survival rates has long been an important component of wild turkey management. In the interest of improving estimates of apparent nesting and nest survival rates, we assessed the feasibility of inferring reproductive behavior of wild turkeys from remotely acquired movement data. We ascertained detailed nesting stage chronologies using Global Positioning System (GPS) data downloaded from 44 females fitted with transmitters that had both a very high frequency (VHF) beacon and a GPS logger. Then we tracked females using VHF to confirm nest locations and nest fate in north and north-central Florida. Using twice hourly GPS derived locations, we evaluated 3 predictive machine learning models (Chi-squared automatic interactive detection [CHAID], exhaustive CHAID, and multivariate adaptive regression splines [MARSpline]) for their ability to accurately classify nesting behaviors (laying and incubating stages) that were confirmed via visual inspection of all potential nest sites. Incubation stage was readily identifiable by the spatial signatures present in each female's GPS-based movement track across all models (up to 92% true positive and as low as 8% false positives in exhaustive CHAID models), whereas the laying stage was more difficult to identify (up to 65% true positive and as low as 35% false positives in exhaustive CHAID models). Daily averages of movement metrics generally achieved better performance in predictive models than data sets with more frequent sampling (i.e., 30-minute daytime locations). Our results indicated that exhaustive CHAID models performed best in identifying nesting stages, using movement metrics from GPS data. With continued development, these models may facilitate and improve large-scale wild turkey monitoring.