Publication Date

12-2012

Type of Culminating Activity

Thesis

Degree Title

Master of Science in Hydrologic Sciences

Department

Geosciences

Major Advisor

James P. McNamara

Abstract

An algorithm is constructed to use snow-depth estimates, derived from repeat airborne LiDAR (Light Detection and Ranging), to identify the sampling strategy that requires the fewest total measurements to estimate the total snow volume in the Dry Creek Experimental Watershed (DCEW) Idaho. LiDAR is used to map snow cover by differencing the digital elevation models (DEMs) obtained from a snow-covered overflight and a snow-free overflight. Sixteen independent variables known to influence snow distribution are derived from a LiDAR digital elevation dataset, obtained during snow-free conditions, and used to predict snow distribution via binary regression trees. Variable ranges leading to the terminal nodes are used to segment the watershed. The algorithm calculates the minimum total number of samples needed to meet pre-defined accuracy thresholds for estimating total basin snow volume. It uses an iterative process to incrementally assign depth measurements to the region that maximizes the reduction in the mean absolute deviation (MAD) of snow volume. It identifies the best combination of regions by constraining the size ofthe terminal nodes in the binary regression tree and repeating the sampling code for each set of discrete regions. The combination of regions that requires the fewest samples to achieve the desired level of accuracy is suggested for use. Thus, during future field campaigns, an optimal number of point measurements of snow depth can be gathered, averaged, and distributed throughout each region. Each unit's snow volume can then be summed to estimate the total basin snow volume. This method assumes that the snow distribution measured on one day in early April 2009 is representative ofthe snow distribution on a given date during a future field campaign. Snow-volume estimates can be combined with snow density measurements to estimate snow-water equivalent (SWE). This method should decrease field time and improve the accuracy of basin estimates of SWE by optimizing snow-depth sampling, which is significantly more variable than snow density. To validate this approach, the relative snow distribution measured during two intensive basin-wide field campaigns is compared to the relative distribution identified by the most accurate binary regression tree. Results show that: 1) elevation, total solar radiation, and local roughness exert the strongest controls on the spatial distribution of snow in the DCEW, 2) tree complexity is directly related to the maximum attainable accuracy, 3) the combination of regions that minimizes sample requirements depends on the desired level of accuracy, and 4) the relative distribution of snow seems to persist in time.

Share

COinS