Land Use/Cover Change Detection in the Twin Cities Metro Area

Land Use/Cover Change Detection in the Twin Cities Metro Area

Paul Walters and Katie Blake

FR 3262

Remote Sensing Project Report

Objective

Remote sensing can be used to monitor change in land cover over time. Urbanization and conversion of vegetated land to impervious surfaces can dramatically influence the environment. Water quality, city planning, and information related to infrastructure can all benefit from this type of data. For our project we evaluated Landsat images from 1984 and 2005 using ERDAS Imagine. (to determine the change of vegetation to impervious surfaces in the twin cities.)

Process

Using the Landsat images that were provided from 1984 and 2005, we evaluated the sevencounty Metro Area. This includes Anoka, Carver, Dakota ,Hennepin, Ramsey, Scott, and Washington county. We then obtained the DNR Data Deli Minnesota Counties shapefile and used ArcMap to clip a shapefile of these seven counties.We imported this shapefile into ERDAS as a vector layer to clip the Landsat images so we were only analyzing the area of interest. ERDASwas then used to re-project the Landsat images to UTM Zone 15 N to match the DNR county shapefile.

Next a Supervised classification of both of the imageswas performed. We identified 4 classes: urban, agricultural, water, and vegetation. We selected 20 training sites for each classin an AOI layer and added them to the signature editor. The sites chosen were in a well distributed fashion, and made sure they were representative. We chose them to the best of our ability using what we knew about the area. We then merged them into their respective classes, and ran the classification. We changed the colors for each class to be identifiable and logical and created a legend. The water class was blue, urban was red, agriculture was yellow and vegetation was green.

We then performed an Image Difference and used a threshold of 10 and 20 percent. The 10 percent threshold was more sensitive and therefore more change was detected that would indicate vegetation changes. The 20 percent was less sensitive and therefore only showed the significant changes.

Next we performed a thematic change detection and acquired the tables from the Matrix Union Summary to evaluate “from-to” changes.We were unable to perform any quantitative accuracy assessment because we had no reference photo of the area in question.

Results

Below are the Matrix Union Summary Tables for each class. Essentially there were some inaccurate and confusing results which we will discuss in a later section. 50,693.8 hectares of vegetation was converted to agriculture which was 38.46 percent. 47,944 hectares of vegetation was converted to urban which was 27.18% percent. 14,076.2 hectares of agriculture was converted to urban which was 7.98 percent. This finding is consistent with the knowledge of urbanization in which vegetation is reduced and urban area increased.

Discussion

We had some confusing results which can be explained by a few things. There was cloud cover in the 2005 Landsat image that was mistakenly classified as Urban. This influenced the thematic change detection and threw off the numbers. Our supervised classification was not entirely accurate but when we performed an unsupervised classification the results were even worse. In order to get better results we would need more knowledge of the area to select better training sites. In conclusion we learned of the potential of remote sensing in the hands of more experienced individuals.

Below are our resulting images:

Temporal Image 1: Metro Area 1984

Temporal Image 2: Metro Area 2005

Supervised Classification Image 1: Metro Area 1984

Supervised Classification Image 2: Metro Area 2005

Image 1: 20% Highlight Change

Image 2: 10% Highlight Change