Zac
Snyder

For my capstone, I had to find a spot for a new landfill in Florida. A site had to be big enough, close to roads, far enough from people, fairly flat, and not too rainy.

I turned each of those rules into a spatial filter and narrowed the state's parcels down one step at a time, then used kriging to estimate rainfall everywhere. Three sites made it through: Okeechobee, Lake Placid, and Yulee.

A big warehouse district has been going up west of downtown Suffolk, Virginia, and I wanted to measure how much forest and farmland it replaced.

I downloaded Sentinel-2 red and near-infrared bands for August 2018 and July 2026, calculated NDVI for each year, and subtracted them. About 330 acres went from dense vegetation to roofs, pavement, or cleared ground.

To check it, I placed 60 stratified random points and labeled each one by hand against high-resolution imagery from both years. The map came out 92% accurate (kappa 0.83). It's a little conservative: most of its misses were real development it didn't flag, so the acreage is probably a slight undercount.

The Loess Canyons are steep canyon country in southwest Nebraska, where eastern redcedar has been spreading into grassland. The goal was to map how much of one area is covered by trees.

I drew training samples for trees, grass, and bare ground on NAIP aerial imagery, then ran a maximum likelihood classification to sort every pixel into one of the three. I put the imagery and the result side by side so you can check the classification against what's really on the ground.

Trees came out at 63% of the study area, with grass at 25% and bare ground at 11%. Most of the grass sits along the canyon floors, and the bare ground shows up as bright patches on the slopes.

Ruby-throated Hummingbirds spend the winter in Mexico and Central America, then many of them fly straight across the Gulf of Mexico each spring. I wanted to show that whole year in one short clip.

The data is eBird's modeled weekly relative abundance, not my own analysis. My part was getting it to work and making it readable. The 52 weekly rasters came without a usable projection, so they landed in the wrong place on the map. After some troubleshooting, they turned out to be in the Equal Earth projection. Once that was defined and they were reprojected, I built them into a time-enabled raster dataset in ArcGIS Pro.

From there I designed the symbology, framed the map on the full migration range, and animated it week by week in ArcGIS Pro. The titles, date counter, and final timing were added with a Python script.

I wanted to find out exactly which land drains into Brush Valley Creek, a small stream that feeds the Root River in southeast Minnesota.

Starting from an elevation model, I filled the sinks, figured out which way water flows across every cell, pulled out the streams, and traced everything upstream of the outlet. Then I laid it over a faded topo map so you can see the roads and the town of Hokah nearby.

The question was simple: if you needed to build new roads around Duluth, where would it cost the least?

I mapped where you can actually build, how far each spot is from an existing road, and how steep the ground is, then combined them into one cost map. The cheap areas sit close to the roads you already have, and the steep bluffs along Lake Superior get expensive fast.

This one is mostly about design. I wanted Mount Shasta to look like a mountain the moment you see it, without needing the legend.

I made a hillshade from the elevation data, laid a color ramp over it, and kept the title and legend small so the terrain stays the focus.