Biology subjects

Zwerts, J. A.

Publications and source records attributed to Zwerts, J. A..

3 recordsLinked to original sources

bioRxiv2024

OPTIMIZING SEGMENTATION IN OCCUPANCY MODELLING OF CAMERA-TRAP DATA

O_LIAccurate estimation of species abundances is a common challenge in conservation biology, particularly when abundances are compared in space or time. Occupancy modelling provides relative abundance estimates from camera-trapping data without the need for individual recognition. This requires segmentation of continuous records into a series of intervals with either detection or non-detection. While the segmentation method may have profound effects on the accuracy of occupancy modelling, no form of segmentation optimization is yet available. C_LIO_LIWe assessed how segmentation, defined by interval length and number, influences the accuracy of predictions by the Royle-Nichols occupancy model and how this relationship depends on species density, study duration, and the number of sampling points. We simulated capture data using an individual-based model in which we varied the species densities between study locations, and then fitted models using different segmentations. Using the simulation results, we developed a simple tool for choosing optimal segmentation and the best minimum number of intervals to use. To provide an example, we used the optimization tool on actual data from a camera-trapping study in Western Equatorial Africa and compared relative wildlife abundances between two forest management types. C_LIO_LIWe found that the optimum interval length for the Royle-Nichols occupancy model varied with species density, study duration, and the number of sampling points. By analyzing the empirical data, we found that optimal segmentation and minimum number of intervals differed substantially between species. Modelling with optimized, species-specific interval numbers and lengths yielded more conservative outcomes (i.e. fewer significant effects) than did modelling with fixed numbers and lengths. Furthermore, the choice of interval length can affect the direction of relationships. C_LIO_LIOur results indicate that the interval length is by no means a parameter to be standardized at a given value but should be carefully chosen based on the properties of the data at hand. This study shows that the arbitrary segmentation that is commonly used in occupancy modelling may not be optimal. Our tool helps to optimize segmentation, increases the accuracy of relative abundance estimations, and thus facilitates the use of camera-trapping studies to evaluate conservation measures. C_LI

ecology↗

bioRxiv2021

Real-time alerts from AI-enabled camera traps using the Iridium satellite network: a case-study in Gabon, Central Africa

O_LIEfforts to preserve, protect, and restore ecosystems are hindered by long delays between data collection and analysis. Threats to ecosystems can go undetected for years or decades as a result. Real-time data can help solve this issue but significant technical barriers exist. For example, automated camera traps are widely used for ecosystem monitoring but it is challenging to transmit images for real-time analysis where there is no reliable cellular or WiFi connectivity. Here, we present our design for a camera trap with integrated artificial intelligence that can send real-time information from anywhere in the world to end-users. C_LIO_LIWe modified an off-the-shelf camera trap (Bushnell) and customised existing open-source hardware to rapidly create a smart camera trap system. Images captured by the camera trap are instantly labelled by an artificial intelligence model and an alert containing the image label and other metadata is then delivered to the end-user within minutes over the Iridium satellite network. We present results from testing in the Netherlands, Europe, and from a pilot test in a closed-canopy forest in Gabon, Central Africa. C_LIO_LIResults show the system can operate for a minimum of three months without intervention when capturing a median of 17.23 images per day. The median time-difference between image capture and receiving an alert was 7.35 minutes. We show that simple approaches such as excluding uncertain labels and labelling consecutive series of images with the most frequent class (vote counting) can be used to improve accuracy and interpretation of alerts. C_LIO_LIWe anticipate significant developments in this field over the next five years and hope that the solutions presented here, and the lessons learned, can be used to inform future advances. New artificial intelligence models and the addition of other sensors such as microphones will expand the systems potential for other, real-time use cases. Potential applications include, but are not limited to, wildlife tourism, real-time biodiversity monitoring, wild resource management and detecting illegal human activities in protected areas. C_LI

ecology↗

bioRxiv2020

High performance machine learning models can fully automate labeling of camera trap images for ecological analyses

O_LIEcological data are increasingly collected over vast geographic areas using arrays of digital sensors. Camera trap arrays have become the gold standard method for surveying many terrestrial mammals and birds, but these arrays often generate millions of images that are challenging to process. This causes significant latency between data collection and subsequent inference, which can impede conservation at a time of ecological crisis. Machine learning algorithms have been developed to improve camera trap data processing speeds, but these models are not considered accurate enough for fully automated labeling of images. C_LIO_LIHere, we present a new approach to building and testing a high performance machine learning model for fully automated labeling of camera trap images. As a case-study, the model classifies 26 Central African forest mammal and bird species (or groups). The model was trained on a relatively small dataset (c.300,000 images) but generalizes to fully independent data and outperforms humans in several respects (e.g. detecting invisible animals). We show how the models precision and accuracy can be evaluated in an ecological modeling context by comparing species richness, activity patterns (n = 4 species tested) and occupancy (n = 4 species tested) derived from machine learning labels with the same estimates derived from expert labels. C_LIO_LIResults show that fully automated labels can be equivalent to expert labels when calculating species richness, activity patterns (n = 4 species tested) and estimating occupancy (n = 3 of 4 species tested) in completely out-of-sample test data (n = 227 camera stations, n = 23868 images). Simple thresholding (discarding uncertain labels) improved the models performance when calculating activity patterns and estimating occupancy, but did not improve estimates of species richness. C_LIO_LIWe provide the user-community with a multi-platform, multi-language user interface for running the model offline, and conclude that high performance machine learning models can fully automate labeling of camera trap data. C_LI

ecology↗