Prior research and research interests

An intersection of natural hazards, landscape evolution and machine learning methodology with remote sensing applications

Hazard mitigation grants

Landslide susceptibility and risk

For five counties in eastern Kentucky, modelled susceptibility using physical models and dual machine learning bagged trees + logistic regression analysis. Risk analysis is a map-based representation of R = H x V x C.

Sinkhole and karst hazard

For eight counties in central Kentucky. Analysis of lidar derivatives to remotely map sinkholes and assess karst hazard with geological maps. Neural network assistance in depression delineation.

Mapping geomorphic landforms

As a pilot study, a catchment was isolated and had its surficial expression (i.e., lidar hillshades) classified into distinct geomorphic landforms.

Using slope angle, curvature, flow direction and aerial photography as well as other supplementary maps the catchment's landforms were mapped remotely using GIS. The idealized process is seen in the figure below.

From the landform map details can be drawn about the erosional and depositional histories, hazard occurrences can be correlated with neighboring landforms, and indications of a larger landscape evolution are hinted at.

This map shows the location of the catchment in Kentucky.

Landscape evolution

At this point, the landscape evolution trends are limited to what can be inferred from the landform mapping. I hope to have further analysis into the data I collect that leads to explanations of how the landscape evolved and how it will evolve into the future.

Comprehending the 'Black Box'

How do Machine Learning models use input data? Can data be optimized? What do machine learning model metrics equate to? Are they practical assessments? How can we better understand machine learning tools and esnure proper usage?

Projects

Impacts from data amount

Analyze variation in model performance across number sets through stochastic simulation

Impacts from data collectors

How do individual human data collectors affect the data and subsquent models?

Impacts from data characteristics

Are some data better than others?

Other work

Automated mapping

Using Topographic Position Index or other methods including machine learning to make landform mapping more efficient. Landslide mapping could also see aid through automatic segmentation

Drone applications

Using drones for case specific change detection and monitoring applications.

Home

Publications

Contact

Page updated

Report abuse