Overview



During the course of this study, each group was assigned a year; 2003, 2005, 2006, 2008, 2009 or 2011 for which they were provided with LANDSAT 5 imagery to determine impervious surface percentages for each watershed, which includes readings from sensors beyond the abilities of human perception, to interpret different factors such as land cover, land use change, vegetation, and soil cover.

The motivation behind conducting this remote sensing was to detect change in impervious surfaces and find any possible relation to poor water quality.


Methodology


The satellite images used in this study were taken by Landsat 5 and downloaded from glovis.usgs.gov. Once loaded into ERDAS Imagine, subsets were taken of each image to narrow the view to the Sandy Springs area, using GA 400, Interstate 75, and Interstate 285. Next, unsupervised classifications were created. We preset the number of classes at 18 and then 20 once it was apparent they needed more discrimination. Identifying each individual grouping of pixels took considerable time, and was susceptible to subjective decisions as each group had their own year. Eight classes were decided upon: Water, Grass, Sparse Forest, Medium Forest, Dense Forest, Impervious - Light, Impervious - Medium, and Impervious - Heavy. Each class was identified using specific clusters around the municipality of Sandy Springs. Grasses were identified using golf courses; Sparse Forest was identified using the Memorial Park on Mount Vernon Highway; Dense Forest was identified using a thickly forested area just south of the Sope Creek Parking Area; Medium Forest consisted of every other forested area; Light impervious areas were identified with suburbs all over the image, as there is no shortage of single residence homes; Medium impervious was found using apartment/townhouse homes that surround major highways; Heavy impervious was effortless, as Cumberland Mall is a giant cluster of commercial buildings. A recode is performed to aggregate each of the 20 classes into a single value per category.

Finally, an area column is added in acres to calculate the percent area of each class. For example, 20% of a given area may be forested, or 45% business/commercial areas. These percentages are multiplied by the impervious quotient as defined by the Atlanta Regional Commission and in use by the Georgia Adopt-A-Stream. Different land cover receives different impervious quotients. Forest is generally 5% impervious while business/commercial areas have a much higher impervious quotient at 80%. The resulting numbers are the percentages for a given land cover's impervious surface in proportion to its percent of land cover within the watershed. So if 20% of a watershed is forest: 20% X .005 = .2% impervious surfaces contributed by forests. Once these final percentages are calculated for all classes of land use, they are summed and this is your total impervious surface for each watershed.


Remote Sensing

Marsh Creek
2003
2005
2006
2008
2009
2011

Long Island Creek
2003
2005
2006
2008
2009
2011


Conclusions

Marsh Creek
Analyzing the above maps, from 2003 to 2011, one result is absolutely present:a definite increase in impervious surfaces and a definite decrease in vegetation. In 2003 and 2005, vegetation is significantly thicker at the Western section and just Southeast of the center of the watershed. The most change is observed
by 2011. Where the Western section remains somewhat undisturbed while the vegetative section to the Southeast of Center has seen a significant increase in
impervious surfaces, mainly light and medium. At the Southeastern border of the watershed and beyond, there is a disturbing increase of heavy impervious surfaces around Georgia 400.

RS_2003_vegetation_focus.jpg
2003 - Focus
RS_MC_2011_focus.jpg
2011 - Focus

Long Island Creek
At the headwaters of the Long Island Creek Watershed, lying in the heart of Sandy Springs, the landscape has known significant urban growth. In 2003, south of Interstate - 285, a large swathe of forest lies on the outskirts of downtown sandy springs. The road "Lake Forest Drive" is aptly named, because it runs directly through this area of dense forest. As the years progress, this area, specifically, has lost most of this forested area. The circled area on the maps below shows this significant change.The majority of this watershed appears to have remained mostly unchanged. This may be due to the fact that most of this area is heavily suburban. Many of these homes have been here for the length of our study and do not allow for medium or heavy impervious growth.


RS_2003_focus.jpg
2003 - Focus
RS_2011_focus.jpg
2011 - Focus




Drawbacks

Despite the extremely applicable nature of remote sensing there are certain drawbacks, especially in a group project setting. The process in which the satellite images are prepared can be extremely subjective, requiring one to make judgement calls. In an unsupervised classification, the software groups pixels into similar classes and the user goes through labels these classifications based on knowledge of the local area and up-to-date maps. Many times certain classes will contain pixels covering areas with very different landforms and the user must decide. In this way, analysis on maps created by different users may vary significantly.

In two cases, the images themselves were drawbacks. These images had spots of clouds which has two effects. The first being the actual cloud interfering with the landscape, the second being the shadow of the cloud, which the unsupervised classification identifies as water.

RS_2011_cloud_focus.jpg
Left Focus - Shadow | Right Focus - Cloud
Another drawback lies with the software itself. There comes a point in time where accuracy must be sacrificed due to lack of time. A person could theoretically classify each individual pixel within a satellite image and the end result would be incredibly accurate, but this process would take countless hours for each image. Therefore, certain landforms, such as roads and commercial areas may be grouped together by the software, making it impossible to differentiate without using the surroundings to identify.

Introduction | Stakeholders | Land Use Maps | Methods | Study Sites | Results | Conclusions | Restoration