USGS Science⌕ Search

Geology topics

Brian R. Lubinski

Publications and source records attributed to Brian R. Lubinski.

5 recordsLinked to original sources
USGS2024

Deep learning workflow to support in-flight processing of digital aerial imagery for wildlife population surveys

Deep learning shows promise for automating detection and classification of wildlife from digital aerial imagery to support cost-efficient remote sensing solutions for wildlife population monitoring. To support in-flight orthorectification and machine learning processing to detect and classify wildlife from imagery in near real-time, we evaluated deep learning methods that address hardware limitations and the need for processing efficiencies to support the envisioned in-flight workflow. We developed an annotated dataset for a suite of marine birds from high-resolution digital aerial imagery collected over open water environments to train the models. The proposed 3-stage workflow for automated, in-flight data processing includes: 1) image filtering based on the probability of any bird occurrence, 2) bird instance detection, and 3) bird instance classification. For image filtering, we compared the performance of a binary classifier with Mask Region-based Convolutional Neural Network (Mask R-CNN) as a means of sub-setting large volumes of imagery based on the probability of at least one bird occurrence in an image. On both the validation and test datasets, the binary classifier achieved higher performance than Mask R-CNN for predicting bird occurrence at the image-level. We recommend the binary classifier over Mask R-CNN for workflow first-stage filtering. For bird instance detection, we leveraged Mask R-CNN as our detection framework and proposed an iterative refinement method to bootstrap our predicted detections from loose ground-truth annotations. We also discuss future work to address the taxonomic classification phase of the envisioned workflow.

Massachusetts, Wisconsin↗
USGS2020

Mapping the thermal landscape of the Upper Mississippi River

Temperature has a fundamental influence on physical, chemical and biological processing in aquatic ecosystems. River temperatures respond to a diverse array of drivers including air temperature, streamflow, and thermal inputs, but the physical template has been shown to play a significant role in structuring spatial and temporal variation in water temperature. How these factors interact to affect water temperature in complex floodplain river habitats such as those present in the Upper Mississippi River System (UMRS) is not well-studied. We used a combination of airborne thermal imagery and continuous temperature loggers deployed across aquatic area types to evaluate spatial and temporal patterns in water temperature in Navigation Pool 8 during the summer and fall of 2017. The mid-wave infrared thermal camera available for this study is not commonly used for thermal imagery acquisition over water, so we discuss accommodations that were made to account for potential interferences and describe considerations for future users interested in using the technology. We quantified thermal metrics from imagery and continuous loggers (e.g., mean, coefficient of variation, range) and compared those to hydrogeomorphic variability across aquatic areas using a Geographic Information System (GIS) dataset. Our findings showed that both temporal and spatial temperature patterns were linked to variation in depth and connectivity of aquatic areas across the pool. Despite some of the technical challenges associated with acquiring this imagery, the method shows promise for characterizing spatial variation in surface temperatures in the UMRS associated with complex physical features such as habitat rehabilitation and enhancement projects.

Iowa, Illinois, Minnesota, Wisconsin↗
USGS2018

Distribution and foraging patterns of common loons on Lake Michigan with implications for exposure to type E avian botulism

Common loons ( Gavia immer ) staging on the Great Lakes during fall migration are at risk to episodic outbreaks of type E botulism. Information on distribution, foraging patterns, and exposure routes of loons are needed for understanding the physical and ecological factors that contribute to avian botulism outbreaks. Aerial surveys were conducted to document the spatiotemporal distribution of common loons on Lake Michigan during falls 2011–2013. In addition, satellite telemetry and archival geolocator tags were used to determine the distribution and foraging patterns of individual common loons while using Lake Michigan during fall migration. Common loon distribution observed during aerial surveys and movements of individual radiomarked and/or geotagged loons suggest a seasonal pattern of use, with early fall use of Green Bay and northern Lake Michigan followed by a shift in distribution to southern Lake Michigan before moving on to wintering areas. Common loons tended to occupy offshore areas of Lake Michigan and, on average, spent the majority of daylight hours foraging. Dive depths were as deep as 60 m and dive characteristics suggested that loons were primarily foraging on benthic prey. A recent study concluded that round gobies ( Neogobius melanostomus ) are an important prey item of common loons and may be involved in transmission of botulinum neurotoxin type E. Loon distribution coincides with the distribution of dreissenid mussel biomass, an important food resource for round gobies. Our observations support speculation that energy transfer to higher trophic levels via gobies may occur in deep-water habitats, along with transfer of botulinum neurotoxin.

Michigan, Minnesota, Wisconsin↗
USGS2016

Identifying the origin of waterbird carcasses in Lake Michigan using a neural network source tracking model

Avian botulism type E is responsible for extensive waterbird mortality on the Great Lakes, yet the actual site of toxin exposure remains unclear. Beached carcasses are often used to describe the spatial aspects of botulism mortality outbreaks, but lack specificity of offshore toxin source locations. We detail methodology for developing a neural network model used for predicting waterbird carcass motions in response to wind, wave, and current forcing, in lieu of a complex analytical relationship. This empirically trained model uses current velocity, wind velocity, significant wave height, and wave peak period in Lake Michigan simulated by the Great Lakes Coastal Forecasting System. A detailed procedure is further developed to use the model for back-tracing waterbird carcasses found on beaches in various parts of Lake Michigan, which was validated using drift data for radiomarked common loon ( Gavia immer ) carcasses deployed at a variety of locations in northern Lake Michigan during September and October of 2013. The back-tracing model was further used on 22 non-radiomarked common loon carcasses found along the shoreline of northern Lake Michigan in October and November of 2012. The model-estimated origins of those cases pointed to some common source locations offshore that coincide with concentrations of common loons observed during aerial surveys. The neural network source tracking model provides a promising approach for identifying locations of botulinum neurotoxin type E intoxication and, in turn, contributes to developing an understanding of the dynamics of toxin production and possible trophic transfer pathways.

Lake Michigan↗