An open-source method for classification of activity states in small vertebrates using automatic VHF-telemetry- Data and Code

To ensure the complete reproducibility of our research conducted in the publication: "Classifying activity states of small vertebrates using automatic VHF telemetry", all data and the code used are stored here. In addition, the aim of this study is to equip the research community with new tools. In two detailed tutorials, 1. all necessary steps to get from the recorded raw VHF signals to an active/passive classification and 2. all analysis steps of the ecological case study are presented. The total dataset is > 287 GB, which is why we recommend downloading only the necessary partial datasets. In the following, we explain which data set is necessary for which analysis. Trained models: For all work steps involving an actual classification of VHF signals into passive/active, it is necessary to download the trained models and store them in the "extdata" folder of the installed tRackIT R-Package. The models can be downloaded from the collection named "Activity classification models". Model tuning: To understand the machine learning process that led to the trained models, we offer a download of the corresponding training data and the R code in the collection named "Scripts and data for the training of ML-models to classify activity based on VHF-signal-pattern" Since we use an elaborate feature selection procedure, the training process can take several days. Model validation: The models were trained on a very large dataset consisting of observed behaviour of tagged bat individuals. 50% of this dataset was set aside for testing the model. In addition, the transferability was tested on two completely independent datasets. One dataset consists of observations of a tagged woodpecker and the other of simulations of activity levels performed by humans with transmitters. All three datasets including the R-code to validate the method can be found in the collection named "Validation of the performance of the random forest model for activity classification". Ecological case study: The complete data set including the raw data used in the ecological case study as well as the R scripts can be found in the collection named "Full data and code for activity classification and analysis of tracked bats from 2018-2021". The dataset is 287 GB in size. Already compiled datasets that are necessary for the reproduction of our analyses are in the repository for the tutorials (next section). Tutorials: We provide two tutorials to show 1. which sequence of tRackIT R-Package functionalities finally leads to a classification of active/passive (tRackIT-Tutorial-for-activity-classification) and 2. to be able to reproduce all statistical analysis steps from the ecological case study (bat_data_HGAM_tutorial). Data, html and rmd files can be found in the collection named "Activity classification based on VHF-signal pattern -Tutorials".