Predicting energy expenditure from dynamic body acceleration in a large boreal ungulate species

Background: Energy expenditure is a central topic in ecology because of the importance of overall energy balance on individual fitness, which can affect populations and ecosystems. However, energy expenditure is difficult to measure in wild animals due to logistic constraints and invasiveness of most methods. Here, we developed a method to calculate energy expenditure of moose (Alces alces) from accelerometer data. We implanted heart rate loggers and deployed accelerometer collars on eight captive female moose and simultaneously recorded heart rate (every 30 s) and accelerometer data (continuously at 32 Hz) during three-day-long sampling periods in spring, early and late summer, and autumn (n = 25 sampling periods). We used a previously published random forest model to predict seven common behaviors from the accelerometer data. We fitted a dynamic generalized additive model to predict heart rate from dynamic body acceleration. Because we could not measure energy expenditure directly, we used a previously published calibration equation of moose metabolic rate against heart rate to calculate energy expenditure from heart rate. Finally, we compared behavior-specific energy expenditures to published values. Results: Our model accurately predicted the trends observed in our heart rate measurements: increasing heart rate with increasing dynamic body acceleration and seasonal variation in heart rate, with a median heart rate of 32 bpm in the autumn and 58 bpm in late summer. Calculated energy expenditure was comparable between our measured and predicted heart rates, increased with increasing activity level of the behavior from lying to running and varied by season. Our estimates of behavior-specific energy expenditures were lower than previously reported values in early summer and autumn and higher in spring. Conclusions: The method presented here facilitates the estimation of movement-based energy expenditure in moose from collar-mounted accelerometers, giving vital parameters for bioenergetic models and future studies of energetic consequences of disturbances and changing environmental conditions. Furthermore, the use of collar-mounted accelerometers to estimate energy expenditure circumvents the need for surgical logger implantation. We discuss the use of dynamic generalized additive models to account for varying slopes in the relationship of dynamic body acceleration with heart rate over a range of behaviors.