Aggregated time-series features boost species-specific differentiation of true and false positives in passive acoustic monitoring of bird assemblages
Data and R code related to the paper Singer et al. 2024 (https://doi.org/10.1002/rse2.385)
Please cite any use of this work:
Singer, D., Hagge, J., Kamp, J., Hondong, H. & Schuldt, A. (2024) Aggregated time-series features boost species-specific differentiation of true and false positives in passive acoustic monitoring of bird assemblages, Remote Sensing in Ecology and Conservation, https://doi.org/10.1002/rse2.385
The provided R-code can be used to reproduce the results of the paper, however it can also be adapted for use with own data.
This script selects the random sample of BirdNET detections per species for human validation and writes the output to a csv-file. The csv-file can be opened and edited throughout the human validation in a spreadsheet application. Additionally, the script also extracts the audio-snippets belonging to the random samples for human validation.
This script calculates the aggregated time-series features for each detection of the human validated random sample. The function calculate_atf() can be found in the script "functions.R"
This script models the species-specific thresholds by using conditional inference trees (CIT) of type 1 and 2 and saves all models as rds.-files.
This script was adapted from the scripts provided by Barré et al. 2019 (https://github.com/KevBarre/Semi-automated-method-to-account-for-identification-errors-in-biological-acoustic-surveys) to calculate thresholds from logistic regression as comparision to the thresholds from CIT.
This script merges all CIT models and filters models with maximum performance as candidate models. Furthermore, the basic models (BirdNET confidence score only) are filtered.
This script identifies the minimal set of model to receive at least one model with maximum performance per species through a set cover optimisation algorithm.
This script compares the performance of the used threshold approaches (universal thresholds, logistic thresholds, CIT thresholds) graphically and by statistical tests.
This script evaluates the importance of the aggreated time-series features for the average model performance by bootstrapping.
This script evaluates the effects of a step-wise reduction of the included aggregated time-series features on the average model performance.