Methods
Methods
Embedded Files
Choosing the Data:
Changes in climate typically take place over large time scales, whereas changes in weather can occur year to year (Blunden, 2020). Therefore, when choosing data, it was important to compare imagery that was spaced at least a decade apart in order to attribute any changes in snow extent to climate change. Additionally, it was important to find imagery that was taken as close to anniversary dates as possible for comparison. Comparing an image from June of 2013 to an image of September of 2024 would not provide accurate results, because those dates are from significantly different times of year which impacts the amount of snow cover (Paul, 2013).
For this study, I chose a Landsat 8 image from September 25, 2013, and a Landsat 9 image from September 15, 2024. These images are only 10 days apart in terms of their anniversary date, which makes them ideal for comparison. Additionally, these images are from the end of the summer season, meaning that most of the excess snow has melted, exposing just the snow and glacial ice that is able to last year round. Images from this time of year are the most useful for identifying changes in glacier size (Lindsey, 2024).
Downloading Data
The Landsat data for this project was downloaded from the Earth Explorer website run by the United States Geological Survey (USGS).
Landsat 8 image from September 25, 2013 with band combination 4, 3, 2 (truecolor). Landsat 8 satellites have a sensor called the the Operational Land Imager (OLI). This sensor provides images with a multi-spectral spatial resolution of 30 meters for visible, NIR, and SWIR bands. The radiometric resolution of Landsat 8 is 12-bits. Source: NASA. https://landsat.gsfc.nasa.gov/satellites/landsat-8/
Landsat 9 image from September 15, 2024 with band combination 4, 3, 2 (truecolor). Landsat 9 satellites have a sensor called the Operational Land Imager 2 (OLI-2). This sensor provides images with a multi-spectral spatial resolution of 30 meters for visible, NIR, and SWIR bands. The radiometric resolution of Landsat 9 is 14-bits, which is an increase from Landsat 8. Source: U.S. Geological Survey (USGS). https://www.usgs.gov/landsat-missions/landsat-9
Classification Methods
One method for detecting change in snow and ice extent was to create land cover classifications for the imagery. These classification processes were performed in Google Earth Engine (GEE) due to the large processing power needed, and which is more efficient in the cloud based solution. Unsupervised and supervised classification methods were used. Unsupervised classification methods such as the K-Means, X-Means, and LVQ method were tested (Carlton University GIS tutorials). These methods provided fair results, but there was considerable misclassification between areas of snow versus areas without snow.
Supervised Classification:
A supervised classification was performed on the imagery which resulted much more accurate classification of snow covered areas vs. areas without snow. Performing the supervised classification involved creating training polygons of the different land cover types for the Classification and Regression Trees (CART) decision tree algorithm to identify (Bittencourt, 2003). This method had minimal classification errors when compared to the unsupervised methods. The only errors in this classification were in shaded areas on the snowpack that were misclassified as water or other landcover, especially in the 2nd image (2024). However, the areas of shadow were fairly uniform between the two images I used, even if the misclassifications are different classes between the two images. Therefore, I believe these misclassifications to be noteworthy, but not significant enough to disqualify the results.
Supervised classification of Landsat 8 image from September 25, 2013.
Supervised classification of Landsat 9 image from September 15, 2024. Note: It was not necessary for land cover classification surrounding the snow extent to be accurate, because the goal of the study is simply to detect changes in the snow, and not surrounding land cover types.
Change Detection
To visualize the change occuring between the two images, I created binary images based on the above classifications. These new images have only two classes--areas of snow, and areas without snow. Below is an animation that shows these two binary images. The animation allows for comparison of the extent of the snowpack between the two years. Next, I created a change detection map from the two binary images by subtracting the 2024 image from the 2013 image.
Images showing the snow extent in the Saas Fee region in 2013 and 2024. Grey areas are all landcover classes that are not snow. White areas include land covered by snow and ice.
Change detection map showing areas that saw a change in snowpack extent between the two years. Red areas represent areas that changed, and yellow represents no change.
Spectral Indices
To further accentuate the spectral properties of the snow and ice cover in the imagery, I used the Normalized Difference Snow Index (NDSI). This spectral index helps to display snow and ice in sharp contrast with the surrounding land cover, which is very useful for understanding the changes occuring in the region. To create the new NDSI images, I used the following NDSI formula using multi-band images created in QGIS. The SWIR1 band on Landsat 8 and 9 is band 6, and the green band is band 3.
Source: NV5geospatialsoftware
NDSI image created from Landsat 8 image from September 25, 2013.
NDSI image created from Landsat 9 image from September 15, 2024.
Differenced Normalized Difference Snow Index (dNDSI)
Subtracting the 2024 NDSI image from the 2013 NDSI image produces a difference image which helps to visualize what type of change occurred. In this image, areas that are orange/red lost snow cover from 2013 to 2024. Areas in blue/green gained snowcover. This image shows that areas in blue/green are largely closer to the center/higher areas of the mountain range. Areas of red/orange are largely in areas of lower elevation around the edge of the snow extent.
Page updated
Report abuse