Browsing by Subject "Logistic Regression"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Landslides in Northeastern Minnesota: Inventory Mapping and Susceptibility Assessment
    (2020-12) Richard, Emilie
    Landslides and other mass-movement events are common geomorphic phenomena in Minnesota that threaten water quality, infrastructure, and public safety. Most published studies on the subject are geographically biased to mountainous regions, and little research has focused on low-relief landscapes like the central lowlands of the United States. This study focuses on slope instability across northeastern Minnesota as part of a collaborative, nearly statewide, landslide inventory and susceptibility mapping project. I developed a database of 2,005 remotely-mapped slope failures from historical records, lidar data, and aerial imagery using GIS software. Field verification of 702 slides determined that remote mapping was approximately 97% accurate. To develop a landslide susceptibility map, I applied a logistic regression (LR) analysis using a set of nine predictive independent variables that may impact slope stability (slope, aspect, elevation, relief, depth to bedrock, soil erodibility, substrate, land cover, and distance to streams). The multivariate LR analyses utilized landslide inventories from two separate study areas that represented different scales and paleogeomorphic settings for comparison: Jay Cooke State Park (JCSP)(32.8 km2) and the Lake Superior South watershed (LSSW)(1,628 km2). The JCSP area along the St. Louis River contains glaciolacustrine sediments and shoreline deposits from pro-glacial lake Duluth, and the LSSW hosts subglacial and ice-marginal moraine deposits from the Superior Lobe. Data sampled from the landslide inventories were subdivided into 80% training and 20% test data in each area. Confusion matrices, comparing model predictions to actual inventory data, were used to assess model accuracy. I found that slope, depth to bedrock, distance to streams, and substrate were statistically significant variables to predict landslides in a multivariate LR analysis in both test areas, though slope alone was a strong enough variable to predict the majority of landslides. Models were more accurate at a scale similar to the resolution of the state datasets used in the analysis (83% in JCSP; 95% in LSSW). The models' transferability was then tested in a third study area, the Mission Creek watershed (28.5 km2) an area adjacent to JCSP with similar surficial material and different bedrock. The JCSP model performed with higher accuracy (92%) than the LSSW model (56%) at predicting landslides in the Mission Creek Watershed. Model comparisons revealed the importance of considering paleogeomorphic settings such as ice-margins, glacial lake-basins, or shoreline environments on landslide susceptibility and occurrence. Outcomes from this research lay the groundwork for future studies across the state and allow stakeholders to reduce risks from future landslides in the face of a changing climate.
  • Thumbnail Image
    Item
    Modeling Distributions of Chromosomal Modifications Using Chromosomal Features
    (2012-02) Baller, Joshua
    Chromatin plays a major role in the regulation and evolution of genomic DNA. The advent of high-throughput sequencing, and the subsequently increasing availability of sequencing data from chromatin immunoprecipitation experiments, is leading to a comprehensive view of the chromatin landscape in key model organisms such as S. cerevisiae. To date, little has been done to exploit the availability of such data. My work develops a logistic regression based framework capable of dissecting the observed distribution of a particular chromosomal modification. This framework models the observed distribution in terms of other known chromosomal features in the organism. I have applied this approach to the distributions of Ty5 and Ty1 retrotransposons, identifying previously unknown integration patterns. For Ty5, I identified integration, independent of the canonical mechanism, at sites of open DNA. For Ty1, I identified precise integration events on a single surface of nucleosomes found near Polymerase III transcribed genes. Additionally, a similar logistic regression approach was developed to predict origins of replication in terms of nucleosome patterning. This resulted in a 200-fold enrichment for origin sites and over 7000-fold enrichment when ORC occupancy data was considered. Together these studies present a general model capable of utilizing the available chromosomal data to provide either mechanistic models or site predictions in a variety of organisms.