When and how data integration improves occupancy estimates for low-density and elusive species

Abstract Understanding the distributions and habitat associations of species of conservation concern is critical for informing effective management. However, monitoring elusive species, such as secretive marsh birds, presents significant challenges. Integrating multiple data sources, particularly participatory-science data, is increasingly recognized as a valuable approach for enhancing estimates of occupancy and abundance. Our findings indicate that the integration of diverse data sources can reduce uncertainty; nevertheless, structured data sources remain essential. In Pennsylvania, USA, numerous marsh bird species are classified as species of greatest conservation need and are primarily monitored through systematic, repeated detection/non-detection surveys. Furthermore, supplementary data on marsh birds are available through semi-structured surveys conducted by volunteers adhering to Pennsylvania Game Commission guidelines, as well as semi-structured participatory-science data (e.g. eBird). This study aimed to determine whether supplementing structured surveys with participatory-science or fine-scale intensive data could enhance inference on occupancy and detection probabilities by increasing the precision of occupancy estimates, while balancing the cost and effort of data collection. We employed Bayesian multi-species integrated data occupancy models to estimate state-wide marsh-bird occupancy. Our results demonstrated that combining broad-scale, participatory data with at least one broad-scale structured survey performed similarly to fully integrated models, suggesting that this combination may offer a cost-effective foundation for monitoring species of greatest conservation need. For wildlife and habitat managers working with limited resources, our findings suggest that prioritizing broad-scale structured surveys, supplemented with participatory data, may provide the most reliable and cost-effective foundation for inference.