Snowpack information is used make management and planning decisions by many organizations in the West, including the Salt River Project, New Mexico and Colorado state engineers, and the California Department of Water Resources. Originally, data from national Snow Telemetry (SNOTEL) stations provided a simple index of spring runoff. The relatively long historical record of such data has been used to develop empirical models that generate springtime runoff forecasts based on wintertime SNOTEL snow water equivalent (SWE) values. However, since the skill of these models decreases substantially during extreme climatic conditions—such as droughts and floods, which are anticipated to increase due to global climate change—SNOTEL data may become less useful. Therefore, this research focuses on deriving forecasts using hydrologic models that consider the spatial variability of snowpack. The use of these models is especially important for forecasting runoff during unusual climatic conditions and for simulating streamflow under different climate change scenarios.
To provide data for current and future stakeholder needs, CLIMAS researchers developed spatially distributed SWE products that can give estimates of future runoff almost as quickly as snow cover conditions change. The researchers also developed snow-covered area products using remotely sensed data such as aerial photographs and satellite images taken at different wavelengths. By applying a statistical model to the snow water equivalent point measurements from SNOTEL sites, they obtained spatially distributed snow water equivalent estimates for the region. After correcting the results by "masking," or removing from the analysis, areas identified by remote sensing as having no snow, researchers incorporated the estimates of snow water equivalent into hydrologic models to demonstrate the use of the products to stakeholders.
Got snow? This question may seem insignificant in a desert environment, but the answer can mean billions of dollars to agricultural and hydropower industries that rely on accurate estimates of snow cover. Snows that accumulate in mountain ranges (known as snowpack) are the natural water towers of the world. In the western United States, winter snowfall in the mountains provides 50 to 80 percent of our water supply. Therefore, understanding seasonal delivery and distribution of mountain snow, a snowpack's storage and water release potential, and the effects of climate and climate change on such processes is increasingly significant.
Snow distribution is extremely variable in time and space. Climatic patterns dictate snow distribution at the regional scale, while physical terrain features such as mountain ranges control distribution at the local scale.
The variability of snow distribution has a large impact on water resource management decisions. Reservoirs, dams, and other water retention structures need to be managed to maximize water storage for agriculture, hydropower generation, and municipal water consumption. Water managers must also maintain adequate storage capacity to accommodate excess water during floods. This balancing act can result in billion dollar losses for industry if shortages occur, or in losses due to flood damage if supply exceeds storage capacity.
Variability of snow distribution also can affect seasonal weather patterns. For example, the distribution of snow affects the regional and global climate by reflecting solar radiation. This enhanced reflectance has been shown to result in a weakening of monsoon circulation and flow of moisture into New Mexico during the subsequent summer season.
In the mountainous regions of the western United States, snow water equivalent is the primary water storage variable for calculating potential runoff during the winter and spring. Snow water equivalent values tell scientists how much water the snowpack will yield when it melts; heavy snow, for example, will provide more water than will dry powder snow. The rate of snowmelt is largely dependent on elevation. Therefore, estimates of snow water equivalent at various elevations are especially important for runoff forecasting and improved decision making.
Currently, point measurements at individual locations are used as an index of the snow water equivalent at different elevations. The Natural Resource Conservation Service provides these point measurements as part of their national Snow Telemetry (SNOTEL) station network. The SNOTEL stations were, until recently, one of the only ways to estimate snow cover over a wide geographic area, although the accuracy of the data they supply are limited for two main reasons. First, snow does not fall in an even layer over a landscape. The steepness of a mountain, the density of tree cover, the direction a slope faces, and the location of a mountain in relation to the sun all influence where snow falls, how deep the accumulation will be, and how quickly it will melt or evaporate. Researchers need access to SNOTEL sites, so most sit on level slopes, often in forested areas. Thus the site may receive more snow than the steep mountainsides above it, and may be more protected from strong winds and the sun's rays. Using the snow depth and weight at the SNOTEL site as an average cover for an entire region can lead to incorrect estimation of total snow-covered area. Also, the sparse sampling density of the SNOTEL network limits their value in representing conditions across the landscape. Although the data are used as if they were averages, they represent only discrete points between which data must be estimated or generalized.
To provide a more comprehensive inventory of snow water equivalent, which would improve runoff forecasts, and to develop the next generation of operational hydrologic models to forecast streamflow, researchers need data that provide a comprehensive estimation of snow water equivalent across the landscape.
Snow researchers obtained snow-covered area data at a 1-km resolution from Advanced Very High Resolution Radiometer (AVHRR) satellite imagery. Snow is highly reflective in the visible portion of the electromagnetic spectrum but highly absorptive in the infrared portion of the spectrum. Thus, researchers can detect the presence of snow using satellite imagery by calculating how much light reflects off the snow (reflectance) and the temperature of the land surface at different wavelengths (known as brightness temperature).
The surface reflectance for a given wavelength is the ratio of the outgoing radiation and the incoming radiation–basically, how much light bounces back to the satellite as compared to how much sunlight hits the snow surface. Researchers used remotely sensed data from AVHRR to measure the outgoing radiation, and they modeled the incoming solar radiation at the land surface based on seasonal changes and atmospheric conditions. To derive brightness temperatures for the image, researchers selected near-infrared (band 4), used especially for vegetation type identification and health, for outlining bodies of water, and for detecting soil moisture.
Clouds and bodies of water are often similar to snow in their reflective and brightness temperature characteristics. To eliminate erroneous readings for cloud-covered pixels, researchers masked cloud cover by manually interpreting and editing pixels, then using a computer image analysis. They also masked water and highly reflective land features to prevent the computer from interpreting them as snow. Finally, to identify pixels that were only partially snow covered, researchers put brightness temperature and surface reflectance variables into probability models.
Once researchers had a reliable method for estimating snow cover at SNOTEL sites (see Figure 3 for sites in Arizona and New Mexico), they developed models to predict snow cover in areas that did not contain sampling sites. Researchers used data from 231 SNOTEL stations surrounding the Colorado and Rio Grande River basins to estimate snow cover at unknown sites based on elevation, topography, slope angle, and other factors. The technique required developing a linear equation between snow water equivalent and elevation and applying this equation to a digital elevation model to calculate snow water equivalent for 1-square-kilometer grid cells across the entire basin. (Each grid cell represents about 0.4 square miles.)
By doing this, they obtained daily snow water equivalent estimates at a 1-km resolution without the information gaps that had previously existed.
To test the accuracy of this formula calculation, they removed known snow water equivalent values at SNOTEL sites and compared the estimates derived from the formula for the sites with the actual values. They subtracted the actual from the estimated snow water equivalent values, and applied this correction to the entire grid. The final product provides researchers and water managers with snowpack and snow water equivalent estimates across a larger land area than was previously available.
The researchers chose to adapt the snow water equivalent products to fit into the Precipitation-Runoff Modeling System (PRMS) because the U.S. Geological Survey and other agencies use it operationally worldwide. At every grid point within the study watersheds, researchers adjusted the PRMS-modeled snowpack condition to match their snow water equivalent product condition (Figure 4). They then simulated snowmelt and runoff over the landscape throughout the snow accumulation and snowmelt seasons, and compared their results to the observed patterns of streamflow fluctuations.
Snow water equivalent estimates from individual states, as well totals from all states that share water supplies or rivers, provide vital information to water resource managers and users. Managers in Arizona will look at data not only from their state, but from Utah and Colorado as well, since the Colorado River Basin runs through all three states and each states’ actions affect the others’ water supplies.
Researchers produced 1-km resolution snow-covered area products from 1995–2002 for the Colorado and Rio Grande River basins for select cloud-free time periods. They also produced daily un-masked 1-km resolution snow water equivalent (SWE) products for the same time period. Masked snow water equivalent products were produced for all dates when snow-covered area was available. Due to cloud cover and poor satellite viewing, researchers were able to process only 3 to 17 remotely sensed scenes for snow-covered area each month. Each snow-covered area scene covers approximately 1.3 million km2.
In addition to preparing products, researchers analyzed the time series of snow-covered area and masked snow water equivalent data to assess seasonal and annual variability in snow cover. In the upper Rio Grande watershed, snow-covered area was seen historically as being most persistent during February. Analysis of the masked snow water equivalent time series in the upper Rio Grande showed that maximum basin-wide snow water equivalent occurs in March, when the higher elevations of the watershed are still accumulating snow and snowmelt is not yet substantial at the lower elevations. Moreover, the researchers found that ground-based snowpack telemetry (SNOTEL) stations preferentially represent densely forested areas and are located relatively close to mountain barriers; thus, snow cover, as measured by SNOTEL sites, appears to persist longer than the watershed average. Their research provides key insights to the development of observation networks, using both remotely-sensed and ground-based data, which are ideally suited for evaluating spatially distributed estimates of snowpack properties.
Researchers also developed new methodologies to estimate SWE in alpine basins using digital elevation models (DEMs) and regression tree models. This research shows that differences in DEMs make significant differences in modeled snow distribution. New techniques were also developed to estimate SWE from remotely sensed satellite data during periods when cloud cover interferes with observations. This research demonstrates that surface temperature data can be useful in determining snow cover beneath clouds.
Dressler, K., G. Leavesley, R. Bales, and S. Fassnacht. 2004. Streamflow estimation from hydrologic model assimilation of remotely sensed snow information in snowmelt dominated basins. Eos Trans. AGU, 85(47).
Molotch, N. and R. Bales. 2006. SNOTEL representativeness in the Rio Grande headwaters on the basis of physiographics and remotely sensed snow cover persistence Hydrological Processes 20:723–739. DOI: 10.1002/hyp.6128
Molotch, N. and R. Bales. 2005. Scaling snow observations from the point to the grid element: implications for observation network design. Water Resources Research 41. DOI: 10.1029/2005WR0004229
Molotch, N., M. Colee, R. Bales, and J. Dozier. 2005. Estimating the spatial distribution of snow water equivalent in an alpine basin using binary regression tree models: the impact of digital elevation data and independent variable selection. Hydrological Processes 19:1459–1479. DOI: 10.1002/hyp.5586.
Molotch, N., S. Fassnacht, R. Bales, and S. Helfrich. 2004. Estimating the distribution of snow water equivalent and snow extent beneath cloud-cover in the Salt-Verde River basin, Arizona. Hydrological Processes 18(9):1595-1611.
Rosenthal, W. 1996. Automated snow mapping at subpixel resolution from NOAA- AVHRR data; Final report: Programs for estimating subpixel snow-covered area. Technical Report, U.S. Army Cold Regions Research and Engineering Laboratory, Hanover, N.H.
Serreze, M., M. Clark, D. McGinnis, R. Pulwarty, and R. Armstrong. 1999. Characteristics of the western United States snowpack from snowpack telemetry (SNOTEL) data. Water Resources Research 35(7):2145-2160.