Using Snow Telemetry (SNOTEL) Data to Model Streamflow: A Case Study of Three Small Watersheds in Colorado and Wyoming
M.S. (ESS-Watershed Science) 2013 Colorado State University, Fort Collins, CO, USA 80523-1476
B.A. (Geography and Environmental Studies) 2005 University of Colorado, Colorado Springs, Colorado 80918
The use of operational snow measurements in the Western United States is instrumental in the successful forecasting of water supply outlooks. The focus of this study was to determine if hydro-meteorological variables available from Snow Telemetry (SNOTEL) stations could successfully estimate the annual total runoff (Q100) and components of the hydrograph, in particular, the date of the passage of 20% of the Q100 (tQ20), 50% of Q100 (tQ50), 80% of Q100 (tQ80), and the peak runoff (Qpeak). The objectives are to: (1) determine the correlation between streamflow and hydro-meteorological variables (from SNOTEL station data); (2) create a multivariate model to estimate streamflow runoff, peak streamflow, and the timing of three hydrograph components; (3) run a calibration/verification on the model; and (4) test the transferability to two other locations, differing in catchment area and location.
Snow water equivalent (SWE) data from the Natural Resources Conservation Service (NRCS) Joe Wright Snow Telemetry (SNOTEL) was correlated to streamflow at the United State Geological Survey (USGS) Joe Wright Creek gauging station. This watershed is located between the Rawah and Never Summer Mountains in Northern Colorado and has a drainage area of 8.8 km2. Temperature data were not used due to non-stationarity of this time series, while the SWE data were stationary over the 33-year period of record. From the SNOTEL SWE data, peak SWE, date of peak SWE, and number of consecutive days with snow on the ground up to the date of peak SWE had the strongest correlation to streamflow (R2 = 0.19 to 0.58).
A collection of models runs were tested with various SNOTEL variables to develop optimal models for each of the five hydrograph components (tQ20, tQ50, tQ80, Q100, Qpeak). Five of the six estimates of were made at the date of Peak SWE. A refined estimate was made for the Q100 at Melt-out, when the SWE equaled zero at the SNOTEL station. For the model development, most of the model trials (78%) had a Nash-Sutcliffe coefficient of efficiency (NSCE) value of greater than 0.50. The greatest accurate models from the calibration and verification period were selected as optimal model configuration for each of the hydrograph components. The optimal model configuration in the Joe Wright Creek watershed had strong performance for the tQ20, tQ50, Q100 and Qpeak (NSCE > 0.50). The tQ80 model was the least accurate model (NSCE = 0.32).
Applying the optimal model equation to the two larger watersheds; Shell Creek is located in Big Horn Mountains of Northern Wyoming (with a drainage area of 59.8 km2) and Booth Creek is located north of Vail in Central Colorado (with a drainage area of 16.0 km2). Basin specific coefficients were generated for a calibration period (1980 to 1996), and evaluated for verification period (1997 to 2012). A majority of the model outcomes were considered good, with 72% of the outcomes having NSCE > 0.50. The Q100 at melt-out model performed the best (NSCE = 0.62 to 0.94).
In a final analysis, the Joe Wright Creek coefficients were applied directly to the two larger watersheds to test model transferability. The location specific model coefficients did not perform well for the other two basins. However, for the Shell Creek watershed, results were still good for the following variables: tQ20, Q100 (using data up to peak SWE and using all SWE data including melt-out) and Qpeak, with NSCE values of 0.45, 0.46, 0.47, and 0.37, respectively. The similar results between Joe Wright Creek and Shell Creek watersheds suggest comparable physiographic characteristics between the two watersheds. An earlier observed onset of snowmelt (as indicated by tQ20) at the Booth Creek watershed influenced the overall accuracy of the model transferability. Despite the differences in the transferability of the model, the optimal configured models derived from accessible SNOTEL data and basin specific coefficients serve as a beneficial tool to water managers and water users for the forecasting of hydrograph components.
Advisor: Steven Fassnacht
Melinda Laituri (Watershed Science)
Mazdak Arabi (Civil Engineering)