Presenter Information

Alexa Junker, Colby CollegeFollow

Location

Parker-Reed, SSWAC

Start Date

1-5-2014 9:00 AM

End Date

1-5-2014 10:00 AM

Project Type

Poster

Description

This study attempts to recreate existing datasets of land-use classifications for Brevard County in east-central Florida from Landsat 5 satellite images using two different classification tools provided by the GIS program ArcMap. After acquiring the Landsat images from the USGS for the years 1995, 2000, 2004, and 2009, the unsupervised classification tool was used in order to identify 6 different land-use classes: urban, agriculture, non-forested land, forest, water, and wetland. Then, 6 sample polygons for each class were used as a basis for the software to recognize different land-use types, and the classification was repeated. In order to measure the effectiveness of one tool versus the other, this study compared the classifications from both tools to the existing datasets for land-use in Brevard County. Because the existing datasets used several hundred distinct land-use classes, they had to be condensed into the same 6 classes as above. After converting the polygons into a raster layer (a continuous image consisting of pixels rather than lines and polygons), a binary (match or no match) analysis was used to assess the values of the pixels in the land-use datasets versus the unsupervised and the trained classifications. The analysis shows that the trained classifications consistently display a higher percent match; with differences as high as 17% between 50% and 67%, and as low as 1% between 53% and 54% for the unsupervised and trained classifications, respectively.

Faculty Sponsor

Manny Gimond

Sponsoring Department

Colby College. Environmental Studies Program

CLAS Field of Study

Interdisciplinary Studies

Event Website

http://www.colby.edu/clas

ID

817

Share

COinS
 
May 1st, 9:00 AM May 1st, 10:00 AM

Recreating Ground-Truth Land-Use Classifications from Landsat Images

Parker-Reed, SSWAC

This study attempts to recreate existing datasets of land-use classifications for Brevard County in east-central Florida from Landsat 5 satellite images using two different classification tools provided by the GIS program ArcMap. After acquiring the Landsat images from the USGS for the years 1995, 2000, 2004, and 2009, the unsupervised classification tool was used in order to identify 6 different land-use classes: urban, agriculture, non-forested land, forest, water, and wetland. Then, 6 sample polygons for each class were used as a basis for the software to recognize different land-use types, and the classification was repeated. In order to measure the effectiveness of one tool versus the other, this study compared the classifications from both tools to the existing datasets for land-use in Brevard County. Because the existing datasets used several hundred distinct land-use classes, they had to be condensed into the same 6 classes as above. After converting the polygons into a raster layer (a continuous image consisting of pixels rather than lines and polygons), a binary (match or no match) analysis was used to assess the values of the pixels in the land-use datasets versus the unsupervised and the trained classifications. The analysis shows that the trained classifications consistently display a higher percent match; with differences as high as 17% between 50% and 67%, and as low as 1% between 53% and 54% for the unsupervised and trained classifications, respectively.

http://digitalcommons.colby.edu/clas/2014/program/4