The Pagami Creek Wildfire Adam Burger and John Habib Remote Sensing, FNRM 3262 December 8, 2014
Stats Large August - September 2011 Unprecedented speed and coverage due to unique conditions Took place within the Boundary Waters Canoe Area Wilderness John
The BWCAW Popular wilderness area Forest fires have ecological benefits Pagami Creek Fire historically large 100,000 acre fire every half-century Adam bwca important bc it is one of the most popular wilderness areas in the country. Picture is of Pagami Creek fire and some of the loss of vegetation that occured. This pic was taken this year although some might be sad about destruction, forest fires natural ecological process Many tree species need forest fires to reproduce and fires are also important for nutrient cycling Fires are actually fairly common in the boundary waters although not necessarily fires as large as the pagami creek fire historical data from tree cores in the boundary waters indicated that 100,000 acre forest fire occurs every half-century, 90,000 acres
Pagami Creek Fire Map on the left shows the BW and the fire’s location within it. You can see the size and shape of the fire in the second map. Also how fast it spread. See smoke from Chicago
Goals Goal 1: To determine the extent of forest cover change caused by the fire Goal 2: To assess the forest recovery of the area affected by the fire John
Imagery Before: May 2011, Landsat 4 After: October 2011, Landsat 4 Present: October 2014, Landsat 8 Adam 3 images came from USGS glovis website Minimal amount of cloud cover and from the same sensor. we chose Landsat Before image is from May 2011, After from October 2011 Before and after images were used to assess the total extent of the fire the present day image from Oct 2014. After and present images used together to assess the recovery of the forest We wanted the after and present images to be from same time of year since we were trying to analyze change in vegetation If images were from different times of year, we may get skewed results based on temporal variation Landsat 4, Landsat 8 Prior to stacking the Landsat 8 layers, we eliminated 4 of Landsat 8’s bands that least matched with wavelengths of Landsat 4
Unclassified Images Pre-fire Post-fire May 15th 2011 October 6th 2011 In the post-fire image, you can clearly see where the fire is May 15th 2011 October 6th 2011
Supervised Classification 4 classes Fire Forest Water Other 10 training sites per class Minimum-distance algorithm Minimum distance algorithm to assign pixels to classes. Originally we used maximum-likelihood, but found that it result in much more classification error
Supervised Classification As I previously mentioned, we used 4 classes but decided to make the other class the same color as the forest class because we only really cared about the burned forest area The burned forest area is fairly apparent in the After image and not present in the Before If you look closely you can see some small dots of red in the before image where classification error occured Before After
Summary by Zone Post-classification change detection: from-to change of pixels Produced a summary report Count, %, and total hectares of pixels changed from the zone name class in the ‘before’ image’ to the ‘class name’ classes in the after image For example if we look at the fire class
Summary by Zone For example if we look at the fire class, 720.81 hectares were classified as fire in the before image and also classified as fire in the after image These represent error because there should be no fire area in the before image 23,613 pixels falsely classified as fire Next step is to calculate the total area that was burned
Summary by Zone To calculate total area of the fire, took total number hectares that changed to the fire class with exception of the wrongly classified fire pixels Added hectares that changed from forest to fire...
Goal 1 Results Total area 33,453 hectares Reported area by USFS around 37,000 This is our final image created showing the burned area An image like this can potentially be used by foresters to create a map for visitors or the public who are interest in knowing the extent or areas affected by the fire A lot of people may want to avoid burn areas when visiting the BWCA and it can also be used for post-fire management so the ability to produce this kind of information is pretty important
NDVI Processing “Before” Image NDVI Image Difference Subset Statistical Analysis “After” Image NDVI Image Difference Subset Statistical Analysis “Present” Image NDVI John
NDVI Processing John
NDVI Processing John
Goal 2 Results 4% John
Goal 2 Discussion % of 2010 NDVI values BUT... John Years since 2010
Errors Classification Seasonal Spectral Image Extent Clouds Lakes The list goes on... Also we could talk about mitigating these errors with subset, a Lakes shapefile, etc, and/or ways to improve the project.
Questions?
References http://www.queticosuperior.org/blog/wp-content/uploads/PagamiCreekFireMap9192011.jpg http://boundarywaters.com/wp-content/uploads/2012/03/pagami-map.jpg http://www.fs.usda.gov/detail/superior/home/?cid=stelprdb5341928 http://www.mprnews.org/story/2012/09/12/environment/pagami-creek-fire-one-year http://www.paddlinglight.com/articles/fire-management-in-the-boundary-waters-canoe-area-wilderness-the-pagami-creek-fire/