BETA-FOR_SP9_Birds_Soundboxes_2022/2023

one-time sampling + rapid assessment 2023 ################ BETA-FOR 2022/2023 Sound data - Protocol * Sources: ** Bavarian Forest (2022,2023) ** University Forest (2022, 2023) ** Hunsrueck (2023) ** Luebeck (2023) ** Saarland (2023) * Data Cleaning * Analyzing ** birdNET V2.4 ** custom model: beta-for model (_bf) (retrained birdNET V2.4) *** Params: Hidden units,Dropout,Batchsize,Learning rate,Crop mode,Crop overlap,Upsamling mode,Upsamling ratio,use mixup,use label smoothing 1024,0.5,8,0.001,center,0,mean,0.25,True,False ** general confidence threshold: 0.5 * Data Filtering ** research period (based on expert knowledge): 24.03. - 15.06. ** target species (based on expert selection, s. 'IndenGebietenmoeglicheArten.csv') ** species-specific confidence thresholds (based on https://doi.org/10.1007/s10336-024-02144-5 / expert knowledge, s. 'birdNET_validation_thresholds_logit_JM.csv') ** replace Picoides_tridactylus by Picoides_tridactylus_bf (based on expert selection) ** consider only one species per 2min-file as (1) detection ** pivot -> species incidences per file; including files without detections * Aggregation ** group by plot_year (sum incidences, count files) ################ We greatly acknowledge the help of: Michael Junginger, Sophia Hochrein, Ruth Pickert, Katharina Kallnik, Manuel Mauermann, Tim Becker, Tamara Rosch, Ralph Martin, Interns of the Bavarian Forest Nationalpark and all other assistants