Methods

Region of Interest

We used data from Landsat 5, 7, and 8, acquired from the USGS at http://glovis.usgs.gov/, to do the change detection for the local area. To minimize the error between the seasonal growth of vegetation and fluctuations of weather we selected data sets from peak summer months for the Alaskan region, May-September, focusing on data-sets from July. Our time range for the study was from 1999-2016, using images from mostly Landsat 7 and 8 to minimize problems that could arise from differences in spatial resolution, temporal resolution, and spectral resolution. There were issues with cloud coverage for available images, so datasets were chosen for specific months based on the amount of cloud coverage, we aimed for a coverage of less than 10%, but some images were up to 40-50%, and this was taken into account when processing.

Normalized Difference Vegetation Index (NDVI)

We used the NDVI band ratio to enhance the visualization of vegetation in the 1999 and 2016 scenes.

The darker the color, the less vegetated the pixel. While, the brighter the shade of red, the more vegetated the pixel.

As you can see, in 1999, nearshore and the alluvial fan pixels were much darker than in 2016.

Inland pixels were closer in value but 1999 inland pixels are still slightly less brilliant than 2016.

Temperature Maps & Raster Slices

Left: A temperature output image of the 2016 scene in rainbow color.

Right: A raster color slice of the temperature map.

Raster Color slice legend in degrees Kelvin.

The raster color slice allowed us to more precisely classify pixel temperature values than the computer classified rainbow color image.

This is easily visible in the raster color slice as temperature resolution is increased with more classes specified.

Modified Normalized Difference Water Index (MNDWI)

We used the MNDWI to delineate water bodies in our scene. It allowed for better classification and noise reduction than standard NDWI.

We applied further band math to the MNDWI output file to map water bodies as pixel value 1 and everything else as 0. This enabled us to create a shape file for the water bodies and ultimately apply this mask to the temperature scene.

The purpose for this was to avoid false readings due to emissivity, which greatly affects surface temperature readings. We decided to take advantage of the plethora of water bodies in our scene as surface temperature readings from water is far more reliable.

Left: An example of ENVI creating the water body Shape file.

Right: Temperature map only of water bodies. Purple values are equal to zero as those regions were not classified as water.

Shortwave Infrared Composite

Preliminary method for viewing vegetation in both years without much processing beyond radiometric correction.

False color composite displayed in Red: Short wave infrared band 1, Green: Near infrared, Blue: Green.

This allowed for us to easily visualize the stressed vegetation by taking advantage of vegetation spectral reflectance in these bands.

Spectral Profiles

Viewing the short wave infrared composite and the spectral profiles for both years supports that vegetation in 2016 does have a healthier vegetation profile.

Further averaging may be done to improve this inference with the creation of ROI's and performing a spectral mean analysis for those regions.

The spectral profile in 1999 shows a snapshot of vegetated pixel with a spectral reflectance peak in the short wave infrared which is consistent with stressed vegetation whereas the 2016 spectral profile for the same pixel shows the spectral peak in the near infrared band.

Healthy vegetation's spectral profile peaks highest in NIR and this is the reason NDVI band ratio uses the NIR band in its algorithm.

Change Detection Difference Map

Initial year: 1999
Final year: 2016

Shown is a difference map of the NDVI images from 1999 and 2016.

Red indicates positive change in vegetation while blue indicates negative change in vegetation.

Software

Google Sites

Report abuse