Marsh birds are important indicators of wetland health, many of which face decline due to wetland loss and degradation. Monitoring is therefore crucial for assessing wetland restoration success, but their secretive nature and inaccessible habitat present challenges. We evaluated whether autonomous recording units (ARUs) combined with deep neural network (DNN) sound recognition software can improve marsh bird detections and occupancy compared to traditional call-broadcast surveys. In collaboration with the National Fish and Wildlife Foundation (NFWF) and Indiana DNR (IN DNR), we collected and analyzed 60 ARU deployments at 16 sites and four states between 2022-2024, each paired with call-broadcast surveys using the Conway protocol. ARUs recorded for 46 days and recordings were classified using BirdNET Analyzer for 16 focal marsh bird species. We validated fifteen days of recordings for each species using classifier-guided listening. ARUs obtained higher richness than point counts in 65% of point-years and detected species of high conservation concern that point counts did not (e.g., King Rail). ARUs also yielded more detections for use in species-specific occupancy models than point counts due to the longer sampling period. Overall, our results indicate that ARUs can considerably improve marsh bird monitoring and strengthen our assessments of wetland restoration.
