Targeted behavioural and microclimate data collection substantially improves biophysical model predictions of energy expenditure in wild birds

Energy is fundamental for life and accurately predicting animals' energy expenditure under changing environments is central to ecology and global change biology. Energy expenditure in free‐living animals can be directly quantified using the doubly labelled water (DLW) technique or estimated by extrapolating respirometric measurements of metabolic rates or modelled using biophysical principles. However, the accuracy of these latter approaches depends on data quality and model parameterisation. We evaluated respirometry extrapolations and biophysical models for estimating daily energy expenditure (DEE) in a free‐living bird, the Southern Pied Babbler ( Turdoides bicolor ) against DLW measurements. Respirometry extrapolations and biophysical models were parameterised using: (a) assumed animal behaviour and assumed operative temperature ( T e ), (b) assumed animal behaviour and measured T e , (c) fine scale (focal) observations of behaviour under natural conditions and measured T e , or (d) coarse scale (scan) behaviour observations and measured T e . We then quantified the minimum field‐collected data needed to accurately predict measured DEE across a range of daily maximum air temperatures ( T air ). Models parameterised with assumed behaviours and T e failed to predict DEE with 95% confidence across all T air values. Biophysical models using focal observations and measured T e closely matched DLW measurements (within 94.1%–95.3% of measured DEE), while scan observations reduced accuracy (within 81.7%–88.1% of measured DEE). Respirometry extrapolations using focal or scan observations and measured T e predicted DEE within 82.1%–88.9% and 69.0%–87.1% of observed values, respectively. Varying behavioural data inputs showed that improving the accuracy of DEE predictions does not require exhaustive sampling. A targeted approach prioritising behavioural observations and T e measurements across representative T air can minimise effort in the field, while maintaining model accuracy. Biophysical models, increasingly used to predict species interactions, distributions, survival and fitness under environmental change, must be biologically relevant, well‐informed and accurate to reliably inform ecological and conservation research. Read the free Plain Language Summary for this article on the Journal blog.