Geology topics

Sheng Li

Publications and source records attributed to Sheng Li.

4 recordsLinked to original sources

USGS2024

Using global remote camera data of a solitary species complex to evaluate the drivers of group formation

The social system of animals involves a complex interplay between physiology, natural history, and the environment. Long relied upon discrete categorizations of “social” and “solitary” inhibit our capacity to understand species and their interactions with the world around them. Here, we use a globally distributed camera trapping dataset to test the drivers of aggregating into groups in a species complex (martens and relatives, family Mustelidae , Order Carnivora ) assumed to be obligately solitary. We use a simple quantification, the probability of being detected in a group, that was applied across our globally derived camera trap dataset. Using a series of binomial generalized mixed-effects models applied to a dataset of 16,483 independent detections across 17 countries on four continents we test explicit hypotheses about potential drivers of group formation. We observe a wide range of probabilities of being detected in groups within the solitary model system, with the probability of aggregating in groups varying by more than an order of magnitude. We demonstrate that a species’ context-dependent proclivity toward aggregating in groups is underpinned by a range of resource-related factors, primarily the distribution of resources, with increasing patchiness of resources facilitating group formation, as well as interactions between environmental conditions (resource constancy/winter severity) and physiology (energy storage capabilities). The wide variation in propensities to aggregate with conspecifics observed here highlights how continued failure to recognize complexities in the social behaviors of apparently solitary species limits our understanding not only of the individual species but also the causes and consequences of group formation.

PNAS↗

USGS2023

Aging contrast: A contrastive learning framework for fish re-identification across seasons and years.

The fields of biology, ecology, and fisheries management are witnessing a growing demand for distinguishing individual fish. In recent years, deep learning methods have emerged as a promising tool for image-based fish recognition. Our study is focused on the re-identification of masu salmon from Japan, wherein fish were individually marked and photographed to evaluate discriminative body characteristics. Unlike previous studies where fish were sampled during the same time period, we evaluated individual re-identification across seasons and years to address challenges due to aging, seasonal variation, and other factors. In this paper, we propose a new contrastive learning framework called Ag ing Co ntrast (AgCo) and evaluate its performance on the masu salmon dataset. Our analysis indicates that, unlike large changes in body size over time, the pattern of parr marks on the lateral line of the fish body remains relatively stable, despite some change in coloration across seasons. AgCo accounts for such seasonally-invariant features and performs re-identification based on the cosine similarity of these features. Extensive experiments show that our AgCo method outperforms other state-of-the-art methods.

Conference Paper↗

USGS2022

Review of field methods for monitoring Asian bears

Efficient and effective monitoring methods are required to assess population status and gauge efficacy of conservation actions for threatened species. Here we review the spectrum of field methods useful for monitoring distribution, occupancy, abundance, and population trend for the five species of Asian terrestrial bears. Methods reviewed include expert opinion, local knowledge, bear sign, visual observations, camera traps, DNA-based methods (hair and scat derived), and radio telemetry. We examine the application of each method in terms of realizing specific monitoring objectives, their assumptions, challenges, and advantages. Our goal is to assist researchers in matching appropriate field methods with sought-after project objectives and to highlight shortfalls and trade-offs. Methods vary greatly in terms of cost, logistics, required number and expertise of staff, and the reliability of the data they provide. Many Asian bear population assessments have relied on expert opinion, local interviews, and sign surveys to provide estimates of distribution, abundance, and trend, in part because these are inexpensive and relatively easy to employ. However, increasing use of camera traps and DNA-based methods now allow for better monitoring via occupancy or rigorous capture–recapture population estimation, with the caveat that these methods may be restricted by inadequate budgets or logistical constraints. For distribution monitoring, camera traps and DNA yield the most definitive records of presence, but in low density bear populations, sign and local knowledge may be more effective. For occupancy, camera traps and DNA are advantageous in providing definitive detections in known time periods. For abundance/density or population trend monitoring in relatively small areas (<10,000 km 2 ), bears must be individually identified and used in a mark-recapture design. This requires DNA from collections of hair or scat, or a camera-based survey in which natural chest marks are clearly visible and individually distinguishable. DNA-methods or camera traps within individual identification is best for population trend when sufficient funding is available. Alternatively, careful use of local knowledge or expert opinion may be viable options, but come with greater uncertainty. For the foreseeable future, we believe that expert opinion will likely continue to play a large part in monitoring Asian bears, but these opinions should be informed by more rigorous data from the other methods we discuss.

Global Ecology and Conservation↗

USGS2022

Physics-guided machine learning from simulation data: An application in modeling lake and river systems

This paper proposes a new physics-guided machine learning approach that incorporates the scientific knowledge in physics-based models into machine learning models. Physics-based models are widely used to study dynamical systems in a variety of scientific and engineering problems. Although they are built based on general physical laws that govern the relations from input to output variables, these models often produce biased simulations due to inaccurate parameterizations or approximations used to represent the true physics. In this paper, we aim to build a new data-driven framework to monitor dynamical systems by extracting general scientific knowledge embodied in simulation data generated by the physics-based model. To handle the bias in simulation data caused by imperfect parameterization, we propose to extract general physical relations jointly from multiple sets of simulations generated by a physics-based model under different physical parameters. In particular, we develop a spatio-temporal network architecture that uses its gating variables to capture the variation of physical parameters. We initialize this model using a pre-training strategy that helps discover common physical patterns shared by different sets of simulation data. Then we fine-tune it using limited observation data via a contrastive learning process. By leveraging the complementary strength of machine learning and domain knowledge, our method has been shown to produce accurate predictions, use less training samples and generalize to out-of-sample scenarios. We further show that the method can provide insights about the variation of physical parameters over space and time in two domain applications: predicting temperature in streams and predicting temperature in lakes.

Conference Paper↗