GIS & Remote Sensing

Contributed Oral Presentations


Contributed paper sessions will be available on-demand for the duration of the conference, then again at the conclusion of the conference.


Multi-Source Variation in the Movement Behavior of African Elephants Across Various Land Tenures in Kenya
Guillaume Bastille-Rousseau; George Wittemyer
Movement is a critical behavior that animal use to balance tradeoffs, including between foraging and risk avoidance. For many species, balancing these tradeoffs involve nuanced responses leading to complex behaviors, especially when evaluated among numerous individuals occupying a broad spatial extent. In these instances, assessing how behaviors vary spatially, temporally, and individually within a population can help identify potential tradeoffs and enhance understanding how animals balance them. We used a long-term tracking dataset of over 150 African elephants inhabiting the Laikipia-Samburu ecosystem in Northern Kenya that were exposed to different types of land tenure (protected areas, community conservancy, private conservancy, and communal land). We used step-selection functions to understand how elephant behavior differed within these land tenures, while also accounting for daily-, seasonal-, and individual-level changes in behavior. After evaluating the resource selection behavior, we calculated metrics (specialization, similarity, temporal consistency, and temporal reversal) that provide additional information regarding the degree and strength of differences in resource specific movement and behavior among land tenure types. We used a random forest classification to assist in identifying the variables that best highlight the differences in behavior. When looking only at the actual behavior, movement properties (speed and directionality) showed bigger differences among land tenure types than resource selection coefficients. However, these variables had a lower ability to explain differences in elephant behavior than the metrics associated to the variation in behavior. Seasonal and daily consistency and specialization were often more informative than the actual selection behavior in capturing differences. Likewise, daily reversal in tortuosity was best at capturing variation in movement properties among land tenure types. Overall, our work illustrates how the nature and structure of the variation around a behavior might offer additional insights into the driver of this behavior.
Beluga Habitat in the Western Hudson Bay
Emma Ausen; Laura Dalman; Marianne Marcoux; David Barber
The Churchill, Seal, and Nelson estuaries are summer habitat for the Western Hudson Bay (WHB) beluga whale (Delphinapterus leucus) population, one of Manitoba’s most valuable natural resources. Knowing what features make habitat suitable for species can improve understanding of key habitat areas for a population. Benefits of estuary habitat are not fully understood, but theories include that shallow waters provide protection from predators, warm estuary water has metabolic benefits for growth and molting, and that estuary habitat is rich in prey. Increased water temperatures in the Hudson Bay will result in increased shipping traffic, changed prey distribution and an influx of predators. A greater understanding of the relationship between WHB belugas and their habitat is required for improved site-specific population management. Beluga locations and age classes were identified using aerial photographs that were collected from western Hudson Bay estuaries in 2018. Home range analysis and the comparison of age class distribution in estuaries was completed using ArcMap. Selected habitat characteristics will be investigated by comparing location with distance from shore, sea surface temperature and surface concentration of chlorophyll a from remotely sensed Aqua and Terra imagery. This research contributes to an improved understanding of beluga distribution within estuaries and the benefits of estuary habitat in Manitoba. Habitat associations obtained from this analysis will be used to support conservation decisions. This presentation will describe methods used, and findings of applying a resource selection function model based on remote sensing derived environmental variables and beluga locations. Differences in beluga group distribution based on age class composition and environmental variables tested using multiple discriminant analysis will also be described. Improving understanding of beluga estuary use and critical habitat will assist in management as warming temperatures result in habitat-based risks to the WHB beluga population.
Tracking the Distribution of a Geographically Isolated Species in a Diminishing Ecosystem
Kira L. Hefty; John Koprowski
Successful implementation of conservation and management actions is inextricably linked to the understanding of species-specific habitat requirements and the distribution of habitat now and into the future. We used occupancy modeling, spatial analyses, remote sensing, climate projections, and random forest models to estimate the current and projected distribution of the state-threatened Big Cypress fox squirrel (Sciurus niger avicennia: BCFS). This species is endemic to geographically isolated wetland forests in southwest Florida–a vegetative community which is threatened by rapid and extensive habitat loss, degradation, and climate change. Current distribution of BCFS is limited by extensive anthropogenic development and low habitat quality within remaining preserves. Projected distribution of available habitat is indirectly and directly impacted by the effects of climate change, including sea level rise, increased temperatures, more intense hurricanes, and changes in seasonal precipitation. Researchers suggest a need for joint interagency-derived adaptive management and scenario-based objectives to promote habitat resilience, increase landscape permeability to movement, and improve habitat quality in patches in the northern region of the species distribution.
Assessing Classification of Groundcover Using Machine-Learning
Charles B. Jacobi; Gad Perry; Robert D. Cox; Samantha S. Kahl
Traditional methods for classification and quantification of groundcover are subject to human error. We assessed an alternative method using ImageJ and Trainable WEKA Segmentation. ImageJ is an open-source image analysis software often used at the cellular level. Trainable WEKA Segmentation is an ImageJ plugin for classifying image attributes by user-trained classification models. We used a cellular device camera to capture 27 standardized images of randomly-placed 50x50cm Daubenmire frames and 9 separate images that consisted of only one category. On-site, we categorized and recorded groundcover within the Daubenmire frames as vegetation, litter/dead, or bare ground. We then developed four protocols to create classification models in WEKA Trainable Segmentation to classify images. Output model images were analyzed using the ImageJ “Histogram” function, creating category pixel values. We compared these values to the on-site estimations to assess the performance of each classification model. Traditional classification model training was accurate across all three groundcover categories (r(25) = 0.897, 0.911, 0.844, p <0.01), though improved when additional smaller images extracted from the image set consisting of one category were also used as training data (r(25) = 0.940, .0931, 0.893, p <0.01). Using only the smaller extracted images was more accurate than traditional training (r(25) = 0.920, 0.911, 0.898, p <0.01). Using the 9 images collected on site consisting of only one category only was reasonably correlated, though had the lowest vegetation correlation (r(25) = 0.858, 0.938, 0.863, p <0.01). Thus, combining ImageJ and WEKA Trainable Segmentation produces accurate groundcover classification amounts and improves standardization compared to traditional methods. This methodology may save time for researchers working in the field because on-site classification is exempt, cumbersome materials aren't required, and research fatigue is limited.
Continuous Monitoring of the United States Using All Available Landsat Data, the Release of the U.S. Geological Survey’s Next Generation Products: LCMAP
Josephine Horton; Christopher Barber; Ryan Reker; Jesslyn Brown; George Xian; Roger Auch
The U.S. Geological Survey (USGS) has implemented a new approach to mapping and monitoring national land cover as part of the Land Change Monitoring, Assessment, and Projection (LCMAP) initiative. This new approach leverages technological advances in computing, the application of state-of-the-art time series algorithms, and the organization of the U.S. Landsat archive into tiled Analysis Ready Data (ARD). Through this approach, LCMAP can provide information on land cover and land change with greater efficiency, frequency, and consistency than previously possible. The LCMAP implementation of the Continuous Change Detection and Classification (CCDC) algorithm analyzes all available clear Landsat observations on a per-pixel basis to develop harmonic time series models. Attributes of these time series models are used as inputs to classify land cover and produce a suite of ten annual land cover and land surface change products for a 33-year record (1985 through 2017) across the conterminous United States. These products can be utilized by the wildlife and natural resources communities to assist in various conservation and management projects. The time series approach enables the monitoring of annual land cover class conversions and detection of disturbances, including more subtle conditional landscape changes, while also mitigating typical challenges associated with land cover mapping efforts such as cloud cover or phenological cycles. A separate collection of 25,000 validation sites across the United States was utilized to provide quantitative measures of accuracy and consistency of the LCMAP land cover and change products. Collectively, LCMAP enables the USGS to better support the land change science community’s need for more frequent land cover information and provides a new science foundation for decisions, assessments, and projections. This presentation will introduce the LCMAP initiative, Version 1 products, and highlight applications of the products for land change assessments and land cover projections.
Measuring Height Characteristics of Sagebrush Using Imagery Derived from Small Unmanned Aerial Systems (sUAS)
Ryan G. Howell; Ryan R. Jensen; Steven L. Petersen; Randy T. Larsen
In situ measurements of sagebrush have traditionally been expensive and time consuming. Currently, improvements in small Unmanned Aerial Systems (sUAS) technology can be used to quantify sagebrush morphology and community structure with high resolution imagery on western rangelands, especially in sensitive habitat of the Greater sage-grouse (Centrocercus urophasianus). The emergence of photogrammetry algorithms to generate 3D point clouds from true color imagery can potentially increase the efficiency and accuracy of measuring shrub height in sage-grouse habitat. Our objective was to determine optimal parameters for measuring sagebrush height including flight altitude, single- vs. double- pass, and continuous vs. pause features. We acquired imagery using a DJI Mavic Pro 2 multi-rotor Unmanned Aerial Vehicle (UAV) equipped with an RGB camera, flown at 30.5, 45, 75, and 120 m and implementing single-pass and double-pass methods, using continuous flight and paused flight for each photo method. We generated a Digital Surface Model (DSM) from which we derived plant height, and then performed an accuracy assessment using on the ground measurements taken at the time of flight. We found high correlation between field measured heights and estimated heights, with a mean difference of approximately 10 cm (SE = 0.4 cm) and little variability in accuracy between flights with different heights and other parameters after statistical correction using linear regression. We conclude that higher altitude flights using a single-pass method are optimal to measure sagebrush height due to lower requirements in data storage and processing time.
Continuous Monitoring of Vegetation Disturbance Around Wildlife Protected Areas Using Landsat Time-Series
Atupelye Weston Komba
Understanding vegetation cover dynamics such as the spatial extent and pattern of disturbance is critical given demands for quality habitat to support the sustainability of wildlife protected areas. The acquisition and further analysis of these dynamics are hampered due to that vegetation cover around protected areas changes both spatially and temporally due to increased anthropogenic activities such as agriculture expansion. This study aimed at assessing the capability of Landsat time-series and LandTrendr algorithm implemented on Google Earth Engine to detect vegetation disturbance and characterize the historical dynamics. The Rungwa ecosystem in central Tanzania was taken as an illustrative application. The spatial extent, change pattern and attributes of vegetation disturbance during 1999-2019 were detected and mapped. The overall accuracy for disturbance classification is 86.12%. The result shows that more than 16363 km² out of 54918 km² of disturbed land surrounding the protected area and 1853 km² of lost vegetation. Our results revealed that newly detected annual disturbance was more dominant closer to the park up to 2.3% per year as distinct from ongoing detected disturbance further from the borders that were, however, significant 1. 2% per year. This suggests that habitat fragmentation and landscape modification are highly pronounced near the borders of the Rungwa ecosystem. Our approach based on Landsat archive, captured vegetation disturbance efficiently and providing data to improve the understanding of vegetation disturbance over long periods and in large areas and has the potential to support monitoring the conservation of protected areas.
Walking the Tightrope in Farm County: Resolving Landscape Connectivity in the Prairie Pothole Region of North Dakota
Robert Newman; Taylor Holm
Conversion of terrestrial ecosystems to agriculture has transformed vast landscapes and altered ecological processes around the world. For conservation applications, advances in remote sensing have made monitoring landscapes more feasible, but data volume and interpretation create computational challenges, particularly when high resolution data are necessary. This is likely for small, nonflying organisms such as amphibians who interact with the landscape at a fine scale. Amphibian persistence depends on availability of wetlands, habitat in the surrounding landscape, and sites that provide refugia during droughts. Demographic connectivity across a landscape requires habitat suitable for dispersal, which may be limited and discontinuous on agricultural landscapes. We used remote sensing data to determine landscape structure and change, particularly in features that provide habitat for amphibians. Our objectives were to measure losses over the last 10+ years of critical habitat and estimate the potential impact on connectivity. Importantly, we wished to compare estimates obtained through coarser resolution (30m) classified satellite imagery from National Land Cover Data (NLCD) versus higher resolution (0.6 – 1 m) maps derived from annual USDA National Agricultural Imagery Program (NAIP) aerial photography. Our hypothesis was that higher resolution maps would more correctly reveal the landscape for small, non-flying animals. We used two supervised classification methods, pixel- and object-based image analysis, to classify NAIP imagery. We compared accuracy of classified NAIP with NLCD and estimated cost surfaces for potential movement based on simple assumptions about habitat suitability to compare connectivity estimated from NLCD versus NAIP. Classified NAIP was more accurate (> 90%) than NLCD (<75%). Smaller features visible on NAIP maps were not detected on NLCD, including known breeding sites of amphibians, riparian, and tree rows. Both sources revealed changing land cover consistent with USDA statistics, with declining grassland and increasing cropland. Changes in landscape structure predict reduced potential connectivity.
Findings from Large-Scale Passive Acoustic Monitoring in Oregon and Washington
Zachary J. Ruff; Damon B. Lesmeister; Cara L. Appel
As part of a long-term monitoring effort focused on northern spotted owls (Strix occidentalis caurina) and barred owls (S. varia) in the Pacific Northwest, we deployed autonomous recording units (ARUs) at 208 sites in the Coast Range in Oregon and the Olympic Peninsula in Washington between March and September 2018. ARUs recorded for 8 hours a day in two bouts centered on sunset and sunrise and were deployed for 6 weeks at a time. We deployed 5 ARUs at each site and during the 2018 field season we collected approximately 350,000 hours of audio. We processed these data by generating spectrograms of each non-overlapping 12-second segment of audio and then using a deep convolutional neural network (CNN) to automatically detect sounds produced by 14 target species within each image. Our target species include spotted owl and barred owl as well as four other owl species, six non-owl bird species, and two mammalian species. For two species, barred owl and Douglas’ squirrel (Tamiasciurus douglasii), we trained the CNN on multiple distinct call types, each comprising a separate class. There was also a catch-all Noise class to cover any sounds not produced by target species. We considered each 12-second audio segment as a potential target species detection if the CNN assigned a high class score (≥ 0.95) to at least one class representing a target species. We manually verified a subset of these potential detections, sufficient to quickly confirm the presence of target species at each site on a weekly basis and generate weekly encounter histories for use in occupancy analyses. We have confirmed over 500,000 detections of our target species from the 2018 season and are using the same methods to process and review a comparable volume of data collected in 2019, and will present findings from both years.
Mapping Groundwater Dependent Ecosystems in the Northeastern United States with the Maximum Entropy Algorithm (Maxent)
Shawn Snyder; Cynthia Loftin; Andrew Reeve
Globally, groundwater dependent ecosystems (GDEs) are increasingly vulnerable to water extraction and land use practices. Groundwater supports these ecosystems by providing inflow, which can maintain water levels, water temperature, and chemistry necessary to sustain the biodiversity that they support. Many aquatic systems receive groundwater as a portion of baseflow, and in some systems (e.g., springs, seepages, subterranean streams, fens) the connection with groundwater is significant and important to the system’s integrity and persistence. Groundwater management decisions for human use often do not consider ecological effects of those actions on GDEs, which rely on groundwater to maintain ecological function. This disconnect between management and ecological needs potentially results in damage to these resources that have repercussions for both the GDEs and human populations that rely on them. This disparity can be attributed in part to a lack of information about where these systems are found and relationships with the surrounding landscape that may influence the environmental conditions and associated biodiversity. Knowledge of occurrence of GDEs in the northeastern United States is incomplete; as expanding urban areas alter the regional hydrology, threats to groundwater resources are expected to increase. Despite the importance of these resources to both human and wildlife populations, GDEs in the region are largely unmapped and poorly studied. An objective of our research is to identify and characterize GDEs across the northeastern United States. We are applying geographically referenced information about known GDEs in the region to the Maximum Entropy Algorithm (MaxENT) to produce a logistic-scale distribution map of GDE habitat suitability across the northeastern states. Model results suggest that areas that lie in topographically low landscape positions, have highly permeable soil, and have highly variable surrounding topography contain suitable conditions for GDE occurrence.


Location: Virtual Date: Time: -