Comparing analytical protocols for identifying causes of population changes

Understanding the factors driving population dynamics is critical for conservation efforts, enabling e.g. the implementation of effective recovery strategies. In this study, we analysed common methods for their ability to correctly identify factors contributing to population changes.

We compared time series (TS) and species distribution models (SDM), alongside process-oriented models (POM), which integrate mechanistic components with a correlative approach. To achieve this, we used a virtual ecologist approach to generate synthetic data where the true underlying drivers were known. We then fitted all three models to the data, mimicking the TS and SDM by aggregating the data within a specific domain (spatial or temporal).

Our results show that POM outperforms the other methods in all performance measures used, exhibiting the highest accuracy (0.88), sensitivity (0.84), and specificity (0.93) in terms of identifying factors that contribute to population changes. SDM demonstrated medium (0.68), while TS had the lowest accuracy (0.50). Furthermore, our results indicate that variable selection marginally improved the performance of suboptimal models (TS) but significantly reduced the accuracy of alternative methods (SDM and POM).

Our findings highlight the importance of incorporating mechanistic aspects into species distribution modelling to more accurately identify the causes of population changes. These results have the potential to be applicable to a range of ecological and conservation studies and would facilitate the selection of analytical protocol for developing specific indicators to monitor population changes.