Research
I measure the shape of the land surface and how it changes, using satellites, drones, cameras on poles, and a lot of code. Most of my fieldwork has been in the eastern Central Andes of Argentina.
Elevation models from satellite stereo
At UW we generate high-resolution digital elevation models (DEMs) from commercial satellite stereo imagery with NASA's Ames Stereo Pipeline, then validate them. When you process hundreds of stereo pairs, checking each one by hand can get messy.
So we built asp-plot, an open-source Python package that turns a finished stereo run into a diagnostic report. It covers acquisition geometry, bundle adjustment residuals, disparity maps, hillshades zoomed into the least and most uncertain areas, and validation against ICESat-2 laser altimetry.
To help newcomers get started with ASP, we also wrote a quickstart guide that explains each step of the stereo workflow, with tutorials that run in a ready-made GitHub Codespace.
Measuring topography from space
Near-global DEMs like SRTM-C, TanDEM-X, ALOS World 3D, and ASTER GDEM2 underpin a huge amount of earth science, but how good are they in steep, arid mountains? Over three field seasons in northwestern Argentina I collected a huge dataset of differential GNSS measurements and used it to validate their elevation accuracy and their geomorphic potential.
Point accuracy only tells part of the story, though. A DEM can match GNSS points well and still carry pixel-to-pixel noise from the sensor, orbit, or processing, and that noise distorts slope, curvature, and flow routing. So we developed metrics that measure this inter-pixel consistency without any reference data: a high-pass hillshade filter strips out the long-wavelength topography, and a Fourier transform picks out artifacts and noise in the frequency domain. Comparing five 30 m global DEMs (SRTM, ASTER, ALOS, TanDEM-X, and Copernicus) over the bare, steep Central Andes, we found that the Copernicus DEM gives the most faithful picture of the landscape and should be the default choice where better local data don't exist.
I also produced DEMs myself, photogrammetrically from raw optical scenes like ASTER L1A, RapidEye, SPOT6 (with these Ames Stereo Pipeline scripts), and ALOS PRISM, and interferometrically from TerraSAR-X/TanDEM-X CoSSC scenes. We also looked at image texture as a quality indicator for optical DEM generation.
Fifteen years of land-level change
Akin to DEM differencing for measuring snow and ice change, we applied careful correction and differencing to the SRTM-C (collected 2000) and TanDEM-X (collected 2011 to 2015) DEMs to measure about 15 years of land-level change in large catchments draining the eastern flank of the Andes in northwestern Argentina. Most land-level changes are small, so only a few signals stood out from the remaining vertical uncertainty, which comes mainly from the lower-quality SRTM-C DEM. Still, we gained insight into riverbed aggradation and incision in the Río Toro and Río Grande, particularly in their downstream sections, which are steeper (due to knickpoints), wetter, and more modified by human activity.
Riverbeds, pebble by pebble
Moving from entire catchments seen with meter-scale spaceborne DEMs, I collected 1,000 to 5,000 m² riverbed surveys at sub-centimeter scales: more than 50,000 photos taken with a camera-on-mast setup (see the videos below), supplemented by some drone surveys. Point clouds, micro-DEMs, and orthomosaics came from structure-from-motion and multi-view stereo (SfM-MVS) processing. Existing grain-sizing tools weren't built for our study site, so we wrote our own Python software, PebbleCounts, and published it.
With PebbleCounts we examined downstream fining in the Quebrada del Toro and found some interesting variability at the coarse tail of the distribution. More recently, we explored ways to segment pebbles directly on 3D surface meshes using curvature.
From space, we explored Synthetic Aperture Radar (SAR) measurements of channel-bed surface roughness from TerraSAR-X, Sentinel-1, and ALOS PALSAR. These reliably pick out gravel-to-sand transitions over huge areas, but absolute grain sizes in the channel stay out of reach.
Landscapes after wildfire
My first research project, for my B.A. at Wesleyan University, was a Keck Geology Consortium study in the Colorado Front Range. In summer 2012, after geology field camp in Wyoming, Idaho, and Montana, I moved into a cabin at the University of Colorado's Mountain Research Station for four weeks of fieldwork in Fourmile Canyon outside Boulder, which had burned in 2010. I studied the relaxation time of extreme erosion events brought on by moderate rainfall after the fire, and the influence of historical mine tailings piles on the geochemistry of the sediment.
