Waldo Canyon Fire 2012
Discussion of basic remote sensing techniques to detect landcover change caused by wildfires.
CONTENTS :
Introduction to the Waldo Canyon Fire, and Principles of Remote Sensing
True and False Color Composites
Image Processing in Google Earth Engine
Vegetation regeneration
Difficulties of Remote Sensing: Burn Severity and Mountain Shadow Discrepancy
Cartographic Posters
Conclusion
Colorado Springs, Colorado
Waldo Canyon
Waldo Canyon was a recreation destination with a 6.5 mile trail and elevated views of the city.
Waldo Canyon Fire
In June of 2012, a fire broke out and burned over 15,000 acres and nearly 350 homes, closing the area for restoration.
Data For the Fire Investigation
Our investigation uses data gathered from Landsat5 & 8 imagery
Landsat 5 Satellite
Launched in March of 1984 from Vandenberg Air Force Base. The rocket dropped its payload at roughly 500miles above the earths surface.
Fire Investigation usingLandsat 5 Imagery from 2011
This investigations first multispectral image was taken in June of 2011 by Landsat 5.
Landsat 8 Satellite
Launched in February 2013 from Vandenberg Air Force Base. Its first images were relayed from orbit in March of 2013.
Landsat 8 imagery from 2013
This investigations second multispectral image was taken in September of 2013 by Landsat 8.
USGS Eros Data Center
This investigations data came from the USGS Eros data center in Sioux Falls, South Dakota, where Landsat Imagery is processed and archived to be found and downloaded from Earth Explorer, GloVis, and other online viewers.
Image processing in Google Earth Engine
Google Earth Engine is a powerful web based analytics tool. Users code in Java and program google earths engine to filter imagery databases for specific areas, times and cloud cover to begin analysis.
Enhanced Vegetation Index processed in Google Earth Engine.
The enhanced vegetation index (EVI) came about through the evolution of the vegetation index. This index uses vegetation reflectance properties in the red and near-infrared wavelengths to saturate vegetation in an image.
Histogram of 2011 EVI image pixel values
Together the image and associated histogram show a snap shot of vegetation health in the Waldo Canyon area prior to the Waldo Canyon Fire in June of 2012.
The full extent of the fire is seen by a dramatic change in EVI from the previous 2011 image.
Histogram of 2013 EVI image pixel values
Normalized burn ratios processed in Google Earth Engine.
The normalized burn ratio uses the near infrared and shortwave infrared wavelengths to emphasize charred areas in an image. The image to the left is from 2011, before the Waldo Canyon fire happened.
This image from 2013 was taken about a year after the Waldo Canyon fire.
The low reflectance in the near infrared and high reflectance in the shortwave infrared wavelengths of charred areas is the opposite reflectance pattern of vegetation, and is drawn to attention by the blue color.
The delta normalized burned ratio compares the normalized burn ratios of before and after the fire and is used to analyze the severity of the burn. This image has had the color palette changed for visual purposes.
Histogram of the delta normalized burn ratio between images from 2011 and 2013.
Vegetation Regeneration
Remote Sensing Challenges: Interpreting shadows in the Waldo Canyon images.
First, observe both images by moving the slider to one side or the other.
The left image is a typicalities classification with vegetated and burned areas in mountain shadows assigned to different categories. The image on the right is the delta normalized burn ratio image with classes of burn severity assigned as different categories.
Comparative slider of land cover classification and burn severity classification.
These two images were compared using cross tabulation inTerrSet, which compares categories designated to a location in one image to the categories designated in the other. The result of the cross tabulation showed high burn severity in the delta normalized burn ratio image coincided with burned areas in mountain shadows.
Cross tabulation of land cover classification and burn severity classification.
This discrepancy comes about because of the way the index normalizes information from an image. Heres more on why.
Reflectance values for shadowed and non-shadowed areas in the pre-fire 2011 image.
The shadowed areas, represented by the blue dotted lines, had lower reflectance values than the non-shadowed areas in both the 2011 and 2013 images; which is an important note when normalizing. Larger numbers, when normalized, become smaller decimals and the opposite is true for smaller numbers. Take for example [3-2]/[3+2] =0.2 while a smaller normalized ratio like, [2-1]/[2+1]= 0.3.
Normalized Burn Ratio of shadowed and non-shadowed areas.
The result of normalizing the reflectance values in both the near infrared and shortwave infrared using the normalized burn ratio ([NIR-SWIR2]/[NIR+SWIR2]) of the shadowed and non-shadowed areas returned values in the shadowed areas to be be higher than the non shadowed areas.
When the pre-fire image was subtracted from the post-fire image to calculate the change (delta, dNBR) in the Normalized Burn Ratios (NBRpre-fire - NBRpost-fire) the difference of the apparent higher valued shadowed area remained higher than the non-shadowed areas and gave a false impression that the burn severity was different. In fact it is unclear what the difference in the burn severity is.