Energy governs species' life histories and pace of living, requiring individuals to make trade-offs. However, measuring energetic parameters in the wild is challenging, often resulting in data collected from heterogeneous sources. This complicates comprehensive analysis and hampers transferability within and across case studies. We present a novel framework, combining information obtained from eco-physiology and biologging techniques, to estimate both energy expended and acquired on 48 Adélie penguins (Pygoscelis adeliae) during the chick-rearing stage. We employ the machine learning algorithm random forest (RF) to predict accelerometry-derived metrics for feeding behaviour using depth data (our proxy for energy acquisition). We also build a time-activity model calibrated with doubly labelled water data to estimate energy expenditure. Using depth-derived time spent diving and amount of vertical movement in the sub-surface phase, we accurately predict energy expenditure (R2=0.68, RMSE=344.67). Movement metrics derived from the RF algorithm deployed on depth data were able to accurately (accuracy=0.82) detect the same feeding behaviour predicted from accelerometry. The RF predicted accelerometry-estimated time spent feeding more accurately (R2=0.81) compared to historical proxies like number of undulations (R2=0.51) or dive bottom duration (R2=0.31). The proposed framework is accurate, reliable, and simple to implement on data from biologging technology widely-used on marine species. It enables coupling energy intake and expenditure, which is crucial to further assess individual trade-offs. Our work allows us to revisit historical data, to study how long-term environmental changes affect animals' energetics.
Read full abstract