Geology topics

Matthew J Weldy

Publications and source records attributed to Matthew J Weldy.

3 recordsLinked to original sources

USGS2025

Simulated soundscapes and transfer learning boost the performance of acoustic classifiers under data scarcity

1. The biodiversity crisis necessitates spatially extensive methods to monitor multiple taxonomic groups for evidence of change in response to evolving environmental conditions. Programs that combine passive acoustic monitoring and machine learning are increasingly used to meet this need. These methods require large, annotated datasets, which are time-consuming and expensive to produce, creating potential barriers to adoption in data- and funding-poor regions. Recently released pre-trained avian acoustic classification models provide opportunities to reduce the need for manual labelling and accelerate the development of new acoustic classification algorithms through transfer learning. Transfer learning is a strategy for developing algorithms under data scarcity that uses pre-trained models from related tasks to adapt to new tasks. 2. Our primary objective was to develop a transfer learning strategy using the feature embeddings of a pre-trained avian classification model to train custom acoustic classification models in data-scarce contexts. We used three annotated avian acoustic datasets to test whether transfer learning and soundscape simulation-based data augmentation could substantially reduce the annotated training data necessary to develop performant custom acoustic classifiers. We also conducted a sensitivity analysis for hyperparameter choice and model architecture. We then assessed the generalizability of our strategy to increasingly novel non-avian classification tasks. 3. With as few as two training examples per class, our soundscape simulation data augmentation approach consistently yielded new classifiers with improved performance relative to the pre-trained classification model and transfer learning classifiers trained with other augmentation approaches. Performance increases were evident for three avian test datasets, including single-class and multi-label contexts. We observed that the relative performance among our data augmentation approaches varied for the avian datasets and nearly converged for one dataset when we included more training examples. 4. We demonstrate an efficient approach to developing new acoustic classifiers leveraging open-source sound repositories and pre-trained networks to reduce manual labelling. With very few examples, our soundscape simulation approach to data augmentation yielded classifiers with performance equivalent to those trained with many more examples, showing it is possible to reduce manual label-ling while still achieving high-performance classifiers and, in turn, expanding the potential for passive acoustic monitoring to address rising biodiversity monitoring needs.

Methods in Ecology and Evolution↗

USGS2024

Counting the chorus: A bioacoustic indicator of population density

Passive acoustic monitoring has grown in utility for tracking wildlife populations, although challenges remain when using acoustic detections to monitor population size and density. Distance sampling is considered the ‘gold standard’ for estimating animal densities but has several important limitations, especially for rare, cryptic, and high-density species. Here, we test the performance of a simple, quickly derived bioacoustic indicator for monitoring population density: call density—the proportion of recording samples containing vocalizations. Over three years, we collected synchronized bioacoustic and point-transect distance sampling data for eight forest bird species native to the Island of Hawai‘i, including four endangered species, across diverse ecosystems ranging from subalpine dry woodland to montane rainforest. The species studied exhibit varied population structures, from gregarious flocks to small, territorial family groups. Our results revealed significant, strong correlations between call density and distance sampling-based animal density estimates for all species, demonstrating that call density is a reliable indicator of animal density that can be used independently or in combination with traditional monitoring methods. Our analysis uses a fixed amount of manual validation of machine learning classifier output examples, without requiring prohibitively high classifier performance, and is robust to variation in vocal activity rates across time and space, making it both adaptable and scalable. This approach could enhance passive acoustic monitoring by providing a more sensitive population health indicator than commonly used detection/nondetection methods, facilitating prompt conservation and management decisions, particularly for species that are difficult to monitor with distance sampling.

Hawaii↗

USGS2023

Long-term monitoring in transition: Resolving spatial mismatch and integrating multistate occupancy data

The success of long-term wildlife monitoring programs can be influenced by many factors and study designs often represent compromises between spatial scales and costs. Adaptive monitoring programs can iteratively manage this tension by adopting new cost-efficient technologies, which can provide projects the opportunity to reallocate costs to address new hypotheses, adapt to changing ecological conditions, or adjust sampling scale or resolution. If there is interest in longer time series of monitoring data, methodological transitions may necessitate integrated models to link newer data with historical data. However, data integration can be difficult if spatial or temporal scales are mismatched. Here, we develop an integrated multistate site-occupancy model and resolve sample unit spatial mismatch to link datasets from two northern spotted owl ( Strix occidentalis caurina ) monitoring schemes that broadly overlapped during a methodological transition. The first dataset was obtained from a decades-long spotted owl monitoring program using call-playback and mark-resight surveys on historical territories of varying size and shape. This monitoring program has recently transitioned to passive acoustic monitoring of randomly selected 5-km 2 hexagons over larger spatial extents. Both monitoring datasets overlapped with areas in which barred owl ( Strix varia ), an invasive competitor that has played an important role in northern spotted owl declines, were being removed experimentally. Reconciling spatial mismatch substantially increased the representation of the call-playback dataset and integrating the two datasets increased precision of spotted owl use and paired occupancy estimates relative to single dataset estimates. Estimates of spotted owl pair occupancy across the study area were lower than previous territory-based estimates based on call-playback surveys. Our integrated model further showed that a concurrent barred owl removal experiment increased landscape use and site occupancy by pairs of spotted owls. Our empirical application of an integrated modelling approach demonstrates a useful analytical framework for long-term monitoring efforts undergoing methodological transitions (e.g. mark-recapture to non-invasive population monitoring). This framework allows monitoring programs to maintain continuity of monitoring objectives across methodological transitions, rigorously incorporate previous findings, and adaptively respond to changing ecological conditions.

Oregon↗