First Advisor

Kelly Gleason

Term of Graduation

Summer 2023

Date of Publication


Document Type


Degree Name

Master of Science (M.S.) in Environmental Science and Management


Environmental Science and Management




photogrammetry, snow, snow cover, stereopair, Structure-from-Motion

Physical Description

1 online resource (ix, 64 pages)


Forest fire occurrence in the western US has increased rapidly since the 1980s, and most western US fires occur in the seasonal snow zone. Burned forests influence snow accumulation and melt patterns for years following fire, and understanding drivers of variability in snow cover across a burned landscape at the basin-scale is necessary for accurate hazard prediction and water resource forecasting. Basin-scale surveys of snowpack are possible with remote sensing, but accurate sensing methods such as Light Detection and Ranging (LiDAR) are often cost-prohibitive. In the last decade, structure-from-motion (SfM), an optical remote sensing technique, has emerged as an affordable alternative to LiDAR for high resolution snow depth mapping. While SfM technique has been used to survey snow in unforested regions, this method is not suitable in forested regions due to the inability of RGB cameras to penetrate the forest canopy. Yet the reduced canopy cover of burned forests may offer a unique opportunity to employ this method in regions otherwise not suitable for SfM surveys prior to burn occurrence. To understand the potential and limitations of SfM-derived snow depth and extent maps in burned forests, we collected aerial stereopair imagery over a 27 km2 region of the burned Breitenbush Watershed in the Oregon Cascades in September of 2022 and February of 2023. We surveyed a smaller region that overlaps the initial survey region in April of 2023. With SfM techniques, we created digital elevation models (DEMs) for each survey. The September DEM was subtracted from February and April DEMs to isolate February and April snowpack. Coincident with the April survey, 200 depth measurements were taken across five 0.8 km transects along a burn severity gradient. We compared modeled snow depth to measured snow depth at point locations through simple regression to understand how variability in modeled snow was driven by actual snow, and this regression was used to adjust SfM snow depth estimates. We used multilinear regressions to assess how variability in adjusted modeled snow was driven by burn severity, pre-fire vegetation, and topography. We then compared binary snow extent maps to Landsat fSCA through confusion matrices to assess how well SfM snow maps predicted snow extent. Lastly, we limited snow depth maps to an ideal region -high or moderate severity burned forest and snow-covered- and assessed how variability in modeled snow constrained by these conditions was driven by burn severity, topography, and vegetation. Multilinear regression showed that in the sampling region, variability in modeled snow was driven by only burn severity. We observed striking differences in the way terrain was modeled in low severity burn versus moderate and high severity burn. SfM modeled snow in low severity burn was significantly different from modeled snow in moderate and high severity burn. Modeled snow variability was far greater in low severity burn than in high and moderate severity burn, and modeled snow was greater in moderate and high severity burn. Our work indicates that SfM snow modeling in high and moderate severity burn regions is distinct from and likely more reliable than snow modeling in low severity burn regions where canopy cover obscures snow from the sensor.


In Copyright. URI: This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).


Funding support from the US Army Corps of Engineers, Research and Development Center #W912HZ2220004, as well as support from the USACE Portland Office for Civil Air Patrol.

Persistent Identifier