Infrared Reflectance of Glacier National Park Lakes using Landsat 5 Thematic Mapper, 1984-2010

Introduction

        Mountain glaciers worldwide have been retreating since the end of the Little Ice Age in the mid-nineteenth century. These glaciers, as opposed to the enormous ice sheets of Greenland and Antarctica, are passive reactors to global climate and not drivers of it. They are very sensitive to changes in temperature and precipitation over the course of years and manifest these changes in a very visible way. They are therefore a target of scientific interest as a near real time record of climate change.

        Against this backdrop of glacial retreat, there are associated effects which merit investigation. Glacier National Park in Montana, USA is home to 132 named lakes ranging in size from 27.8 km² to .001 km². Many of these ponds and lakes are fed directly from glacial meltwater. Glacial runoff is characterized by its bright blue coloring and its milky appearance. The blue color is caused by sediments in the water, specifically "rock flour". Rock flour is produced by the scouring of surface rock by the glacier as it moves, and it is very fine. Particles on the order of 2 μm preferentially scatter light in the blue-green part of the visible spectrum, giving glacial lakes their color. Larger particles such as sand at around 100 μm scatter visible light non-preferentially, giving these lakes their milky appearance.

        The hypothesis is that the known decrease in glacier size in Glacier National Park will correlate with a downward trend in Landsat Thematic Mapper band 4 near infrared reflectance from lakes in the Park, because smaller glaciers will produce less sediment than larger glaciers.

data

Landsat Thematic Mapper Band 4

        To capture the spectral signature of the rock flour suspended in the lakes, Landsat TM band 4 was used because of the linear relationship of fine sediment levels to reflectance in this range of wavelengths (.76-.90 μm). The comparison above shows an RGB composite image on the left, and band 4 on the right. While still very dark compared to ice and vegetation, the lakes on the right do exhibit infrared reflectance levels 50% or more higher than lakes without significant sedimentation.

        Cloud-free, atmospherically corrected imagery was obtained from 1984-2010 during the month of July in order to capture the lakes during a time of rapid ice melting and high levels of sediment.

analysis

image classification

        The first step in the analysis was to separate the liquid surface of the lakes from adjacent shoreline, ice, vegetation, etc. Using vector data from the USGS National Hydrography Dataset (NHD) as well as raster quality assurance files included with the reflectance data, a decision tree (above) was created in the image processing software to isolate just the liquid surface of the lakes.

        Sample results of the image classification are shown to the right. Following the coloring in the decision tree above, the classifications on the right show pixels outside the NHD vector boundaries as black; pixels classified as ground by Landsat QA data as red; pixels classified as ice by Landsat QA files as cyan; pixels with a user-defined reflectance cutoff level to eliminate false positives in the water layer as white; and finally surface water remains as blue. The RGB composite image is shown for comparison.

results

        The requirement that the satellite images be free of clouds and taken in the month of July dramatically reduced the available scenes that could be used for park-wide analysis. However, scenes with partial cloud cover were useful for analysis of individual lakes which increased the sample size. For example 14 scenes were captured of Grinnell Lake and its source the Grinnell Glacier, a fairly large glacier-lake system with a well documented history of retreat. The figure below shows mean, minimum, and maximum reflectance values for Grinnell Lake, as well as a linear regression trend line. Contrary to the hypothesis, there is no statistically significant change in the reflectance values over the course of 25 years of documented retreat of Grinnell Glacier.

        Further refinement is possible in the selection of appropriate Landsat scenes. The sample size could be increased by expanding the acceptable time frame of image capture, e.g. from June-September. Also, the cloud threshold could be raised thereby increasing the number of scenes in which at least some of the lakes are cloud free. Furthermore, true glacier-fed lakes such as Grinnell could be isolated from other lakes in the park which lie farther from glacial water sources (such as Lake McDonald) which allows the sediment to settle and reduces infrared reflectance. Finally, ground truth data of in situ sediment levels would be beneficial in order to establish a relationship between infrared reflectance and actual sedimentation.