Search for: All records

Abstract The general public and scientific community alike are abuzz over the release of ChatGPT and GPT-4. Among many concerns being raised about the emergence and widespread use of tools based on large language models (LLMs) is the potential for them to propagate biases and inequities. We hope to open a conversation within the environmental data science community to encourage the circumspect and responsible use of LLMs. Here, we pose a series of questions aimed at fostering discussion and initiating a larger dialogue. To improve literacy on these tools, we provide background information on the LLMs that underpin tools like ChatGPT. We identify key areas in research and teaching in environmental data science where these tools may be applied, and discuss limitations to their use and points of concern. We also discuss ethical considerations surrounding the use of LLMs to ensure that as environmental data scientists, researchers, and instructors, we can make well-considered and informed choices about engagement with these tools. Our goal is to spark forward-looking discussion and research on how as a community we can responsibly integrate generative AI technologies into our work. 

Bark beetle outbreaks have impacted millions of acres of conifer forests in Colorado and across western North America, which has altered forest structure and function. These alterations may have significant impacts on the ability of forests to fulfill resource requirements for species that utilize them or modify how they use forests to meet these needs. Ungulates in the southern Rocky Mountains utilize conifer forests to fulfill different aspects of their ecologies. However, there is currently a limited understanding of the unique impacts bark beetle outbreaks have on each species. Using GPS location data for mule deer (Odocoilus hemionus), elk (Cervus canadensis), and bighorn sheep (Ovis canadensis) we created third and fourth order resource selection functions using a use-availability framework. Measures of the beetle kill, along with topographic covariates, were derived from remotely sensed data. Preliminary results suggest that measures of outbreak age, size, severity, and heterogeneity all impact ungulate use of forested areas, however these patterns vary across scale and by species. General trends indicate that older and more severe patches of beetle kill forests with moderate outbreak heterogeneity are selected for by ungulates. Understanding the dynamics of ungulate resource selection under these circumstances is critical to inform both wildlife and forest management strategies, as the impacts of beetle-kill continue to evolve and expand across the west. 

Extreme single-day fire spread events are associated with warmer and drier fire seasons and are expected to increase in the future. However, our understanding of the post-fire landscape outcomes of such events is limited. Here, we ask whether extreme events burn more severely or produce landscape patterns that are less conducive to forest recovery. To identify extreme single-day fire spread events, we used satellite fire detections to create day of burning (DOB) maps for 633 fire events >400 ha within forested ecoregions of the southwestern United States. We categorized daily rates of spread as extreme (top 2.5% of events; >4901 ha/day) and non-extreme (bottom 97.5%, <4901 ha/day). We contrasted satellite-measured burn severity and a suite of high severity patch landscape metrics between extreme and non-extreme spread events. We found that extreme single-day fire spread events were associated with increased severity, a greater proportion of area burned severely, and a higher percentage of like adjacencies between high severity pixels. Average distances from high severity patch interiors to edges were also greater in extreme single-day spread events. Additionally, area-weighted mean patch size and total core area of high severity patches were higher for fires containing one or more extreme single-day spread events. Larger and more homogenous high-severity patches produced during extreme events have been shown to limit tree seedling establishment, inhibiting forest recovery and facilitating vegetation type conversion. These landscape outcomes are expected to be magnified under future climate as extreme fire spread becomes more prevalent, accelerating fire-driven forest loss across the southwestern US. 

Global biodiversity is diminishing rapidly. The number of accepted “background” extinction rates range from 0.1 to 4 species per million per year. However, some authors have noted an increase in extinctions to 150–260 species per million per year since 1500. As the largely anthropogenic-induced global biodiversity crisis progresses, an abundance of taxa are faced with changing environments, decreasing population sizes, and/or extinction. Concerning amphibians, at least 43% of all species are experiencing population declines, with the 2008 Global Amphibian Assessment showing that 32% of all amphibian species are threatened or extinct. Populations of high-alpine amphibians in the southern Rocky Mountain region (i.e., Boreal toads [Anaxyrus boreas boreas], Tiger salamanders [Ambystoma mavortium], and Boreal chorus frogs [Pseudacris maculata]) have swiftly declined in recent decades due to high 649 susceptibility to diseases (e.g., Batrachochytrium dendrobatidis and ranaviruses) and changing environmental conditions. This study will focus on the effects of climate change by examining the habitat of these three amphibious species. A Maximum Entropy (MaxEnt) model is a commonly used machine-learning method of creating SDMs by accounting for the most uniform distribution and the provided predictors. Well-fitted models can be produced for different climatic scenarios by including climatic and environmental factors, or predictors, relevant to the species of interest, appropriately adjusting the regularization parameter and active feature classes for each model, and properly evaluating the models. By adopting this method of SDMs along with a decadal time-series approach, we demonstrate a larger-scale pattern of potential habitat shifts over time in this alpine habitat caused by climate change. 

Accurate estimates of mule deer survival are needed to model population dynamics and develop optimal management plans. Survival rates are often estimated using data from radio-collared animals but capture techniques to deploy these collars can risk mortality and injury to the animal. Our objective was to estimate age- and sex-specific survival rates of mule deer in the state of Utah, USA, while also investigating how these rates were affected by capture and handling. We captured 2,977 mule deer throughout the state and fitted them with radio-collars. Using Cox proportional hazard regression, we then estimated survival rates from the collected GPS data. We also examined the effects of age and sex on survival, while accounting for the influence of a variety of other covariates. Finally, we used a model selection framework to evaluate how long survival rates of captured animals were different from those of animals that were not captured. Fawn survival rates were 0.52 (0.45 – 0.60) for females and 0.66 (0.55– 0.79) for males, and overall adult survival rates were 0.76 (0.75 – 0.78) for females and 0.73 (0.70 – 0.77) for males. Survival rates for both sexes varied by age, latitude, and body condition. The negative effect of capture was small and initially disappeared after 3 weeks, but seemed to reappear later, perhaps during periods of high mortality. 

Ecologists have increasingly recognized opportunities to adapt and adopt methodologies and information originally intended for other purposes in a “data fusion” approach. Recently, there has been an influx of studies and training focused on using unmanned aerial vehicles (UAV’s) and remote sensing in wildlife research. Leveraging these technologies could supplement the often resource-intensive field approaches used to monitor population and habitat dynamics for forest dwelling species such as the snowshoe hare (Lepus americanus). Barriers remain, however, especially as agencies lacking the resources to collect data using UAV’s are restricted to freely available, not wildlife-specific, products. Furthermore, technologies may not be advanced enough to “see through” the canopy to the understory, relevant for species that rely on vegetation cover. We thereby conducted a case study to determine whether an approach outlined by previous authors could be successful, wherein the remote sensing products were accessible and originally collected for broader purposes. Our models did not adequately predict snowshoe hare fecal pellet numbers, pointing to deficiencies in the scale and type of available data derived from remote sensing. We also note potential shortcomings in non-invasive field techniques. Regardless, we maintain that open-access remotely sensed imagery is valuable when ground-truthed and combined with supplemental information, adding to knowledge within and beyond the fields of forestry and wildlife biology. 

Please see conference program: https://www.mammalsociety.org/index.php/meetings