Assessing acoustic receiver detection efficiency using autocorrelation adjusted machine learning models
Abstract Background Detection efficiency is a key performance metric for acoustic telemetry arrays, providing an estimate of the probability of detecting a passing tagged organism. It is influenced by environmental (e.g., discharge), technological (e.g., transmitter power), and habitat (e.g., noise)...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
BMC
2025-07-01
|
| Series: | Animal Biotelemetry |
| Subjects: | |
| Online Access: | https://doi.org/10.1186/s40317-025-00419-z |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Background Detection efficiency is a key performance metric for acoustic telemetry arrays, providing an estimate of the probability of detecting a passing tagged organism. It is influenced by environmental (e.g., discharge), technological (e.g., transmitter power), and habitat (e.g., noise) factors, making predictions of detection efficiency a challenging task in the field of movement ecology. To predict detection efficiency, we applied regression-based machine learning models in two distinct river systems: a small mountainous and a large regulated river. The models incorporated daily discharge, water temperature and depth, substrate type, a receiver metadata metric indicative of noise, and the distance between receiver and acoustic tag. Results While both spatial and temporal autocorrelation were evaluated, only temporal autocorrelation required adjustment, which was addressed using a rolling cross-validation approach. Optimal cross-validation parameters differed between systems, with 30-day validation windows and 90-day steps for the large river, and 3-day validation windows and 5-day steps for the mountainous stream. Receiver distance and our utilization of receiver metadata as an indication of environmental noise consistently emerged as the most influential predictors, while environmental variables contributed relatively evenly to model performance. The small mountainous river model explained 30.7–89.5% of the variability in detection efficiency while the large regulated river model explained 43.8–90.6% of the variability explained. The model’s accuracy varied across resamples based on short rapid environmental changes during rolling cross-validation temporal binning. Conclusion Our autocorrelation adjusted machine learning model demonstrated adequate estimates of detection efficiency, explaining an average of 68% of the variability across two distinct rivers. Restricted data availability in the mountainous stream and short rapid environmental changes in both systems presented challenges for model accuracy. Accounting for detection efficiency is an important component of describe animal movement using acoustic telemetry and our findings demonstrate machine learning models as an approach to predicting detection efficiency in acoustic receiver arrays across riverine environments with diverse hydrological and geomorphological characteristics. |
|---|---|
| ISSN: | 2050-3385 |