This article is a part of the NASA DEVELOP’s Spring 2017 Article Session. For more articles like these, click here
Using NASA Earth observations and Google Earth Engine, NASA DEVELOP helps to provide National Park Service with a new tool to assess snow cover in Saguaro National Park.
Charles Barrow III
The bedrock pools and streams in the park and the surrounding sky islands are reliant on the accumulation of snow at higher elevations to help supply water during the dry season. Recently, park managers suspected there may be decreases in the amount of snow present compared to historic data and decreases in the amount of water present for supplying bedrock pools and park streams later into the dry season. However, a lack of synoptic, geospatial historical data has left park managers without a way to synoptically quantify the changes they are observing at specific field locations. Additionally, many of the locations where snow occurs are either remote or in rugged terrain, leaving park managers unable to field visit such locations in most instances. To better manage the water resources during the dry season, park managers look to fill a critical data gap on snow occurrence patterns by using satellite imagery to help determine the amounts of seasonal snow cover on a per area basis across the calendar year. With information on the amounts of snow cover for individual watersheds, park managers will be better equipped to manage water resources during the dry season.
Recently, NASA DEVELOP’s Mobile County Health Department team partnered with the National Park Service (NPS) in the Intermountain Region and the Saguaro National Park to help develop needed tools and data sets to assist park managers in assessing snow cover trends for individual watersheds. For the purposes of this study, the Upper Rincon Creek Watershed was chosen as it directly feeds Rincon Creek, which runs through Saguaro National Park (Image B). Landsat 5, 7, and 8 were chosen to observe snow cover due to their increased spatial resolution relative to coarser satellite imagery such as those from MODIS sensors. Landsat data was also chosen as the main data source for snow monitoring because it is available as early as 1984, allowing for historical snow cover trends to be observed and compared to current snow cover trends. With this set of satellites available, the NPS will be able to observe the historical seasonal trends of snow cover.
To ensure that the NPS will be able to address future years’ snow cover, the team created a Google Earth Engine (GEE) script that can be used to analyze the current and future trends with the most recent data available on GEE. In addition to utilizing the script for future data, the script can also be used to assess the amounts of snow cover in surrounding watersheds in the sky island region. Aided with this additional information, park managers as well as other local groups such as the Sky Island Alliance may be able to compare how changes in snow patterns differ for individual watersheds.
To assess the amount of snow cover, a combined snow index was used that utilized the Normalized Difference Snow Index (NDSI)  and the S3  snow index. NDSI utilizes the green and shortwave infrared bands whereas the S3 index utilizes the red, near-infrared, and short-wave infrared bands. Both indices have particular snow classification errors. such as errors due to topographical shadow, cloud shadow, and vegetation.
The S3 index was particularly useful for classifying snow in topographic and cloud shadow as well as in vegetated areas such as those on the higher elevations of the mountains. Using both snow detection indices, thresholds for each index were identified by the team based on a literature review  to only classify snow that met the pre-determined thresholds in both the NDSI and S3 indices (Image C). The combined use of the two snow indices allowed lower snow detection thresholds compared to mapping snow with a singular index. This combined index approach improved the snow identification accuracy. After classifying snow cover in the region, the percentage of an entire watershed that was classified as snow was calculated and then graphed.
When comparing the seasonal snow cover to the seasonal stream discharge, the peaks of both appeared to have a relationship. The highest amount of stream discharge occurred at nearly the same time as the highest amount of snow cover in the watershed. The lag between when the peak stream discharge was observed and when the peak snow cover was observed may be due to the difference between when the Landsat image was taken. The peak amount of snow cover may have occurred before the Landsat image was taken and thus resulted in the peak observed snow cover occurring after the peak stream discharge was observed.
Use of Landsat data for snow mapping resulted in beneficial information on the phenology of snow in the study area’s mountains. Additionally, the USGS stream gauge data and Landsat-based snow persistence data in the mountains generally agreed in terms of showing snow trends. However, this correspondence can also be partially explained by rain contributions at lower elevations than where snow typically occurs. When comparing snow trends for wetter years versus drier years, there was indication that the stream gauge data corresponded to snow cover. In years where there was more snow, the stream gauge registered water flow longer into the dry season even without other precipitation. In contrast, in drier years the stream gauge registered little to no water flow. The project provided evidence to the NPS that Landsat data can be processed into meaningful synoptic maps of snow cover for the study area. However, more work is needed to further process and analyze Landsat-based snow maps with respect to specific watersheds of concern. Overall, the combined use of the NDSI and S3 index for snow mapping appeared to be fairly accurate (92 percent overall for one tested date of Landsat data).
 D.K. Hall et al., “Algorithm theoretical basis document (ATBD) for the MODIS snow and sea ice-mapping algorithms,” in NASA GSFC September 2001.
 Y. Asaoka and Y. Kominami, “Incorporation of satellite-derived snow-cover area in spatial snowmelt modeling for a large area: determination of gridded degree-day factor,” in Annals of Glaciology vol. 54(62), 2013, 205-213.
 H.S. Negi et al., “Estimation of snow cover distribution in Beas basin, Indian Himalaya using satellite data and ground measurements,” in Journal of Earth System Science vol. 118(5), 2009, 525-538.
Tyler Lynn is a recent graduate from Auburn University working with DEVELOP at the Mobile County Health Department node as an independent research consultant on the Southeastern Arizona Water Resources II project.
Elaina Gonsoroski is a recent graduate from the Ohio State University working with DEVELOP at the Mobile County Health Department node as an independent research consultant on the Southeastern Arizona Water Resources II project.
Charles Barrow III is a recent graduate from the University of South Alabama working with DEVELOP at the Mobile County Health Department node as an independent research consultant on the Southeastern Arizona Water Resources II project.
Saranee Dutta is a recent graduate from Mississippi State University working with DEVELOP at the Mobile County Health Department node as an independent research consultant on the Southeastern Arizona Water Resources II project.
Farnaz Bayat is a recent graduate from the University of South Alabama working with DEVELOP at the Mobile County Health Department node as an independent research consultant and worked on the Southeastern Arizona Water Resources project.
Katie Harville is a graduate from the University of South Alabama. Since working with DEVELOP at the Mobile County Health Department on the Southeastern Arizona Water Resources project, Katie has returned to her work with FEMA as an Environmental and Historic Preservation Specialist.