A data-integration approach to correct sampling bias in species distribution models using multiple datasets of breeding birds in the Swiss Alps

It is essential to accurately model species distributions and biodiversity in response to many ecological and conservation challenges. The primary means of reliable decision-making on conservation priority are the data on the distributions and abundance of species. However, finding data that is accurate and reliable for predicting species distribution could be challenging. Data could come from different sources, with different designs, coverage, and potential sampling biases. In this study, we examined the emerging methods of modelling species distribution that integrate data from multiple sources such as systematic or standardized and casual or occasional surveys. We applied two modelling approaches, data-pooling and model-based data integration that each involves combining various datasets to measure environmental interactions and clarify the distribution of spe-cies. Our paper demonstrates a reliable data integration workflow that includes gathering information on model -based data integration, creating a sub-model of each dataset independently, and finally, combining it into a single final model. We have shown that this is a more reliable way of developing a model than a data pooling strategy that combines multiple data sources to fit a single model. Moreover, data integration approaches could improve the poor predictive performance of systematic small datasets, through model-based data integration techniques that enhance the predictive accuracy of Species Distribution Models. We also identified, consistent with previous research, that machine learning algorithms are the most accurate techniques to predict bird species distribution in our heterogeneous study area in the western Swiss Alps. In particular, tree-dependent ensembles of Random Forest (RF) contribute to a better understanding of the interactions between species and the environment.

Automated summary

Accurately modeling species distributions and biodiversity is crucial for ecological and conservation challenges. Data on species distributions and abundance are essential for decision-making on conservation priority. Integrating data from multiple sources using model-based data integration approaches can improve the predictive accuracy of species distribution models. Machine learning algorithms, particularly tree-dependent ensembles like Random Forest, are the most accurate techniques for predicting bird species distribution.