1
Landisch and Zahler
Quantifying Land Cover Change
And Urban Development
In Rochester, MN
FR 3262
Patrick Landisch and Stephanie Zahler
5/10/13
Introduction
Humans have had an incredible impact on the environment over the course of our existence. Over recent years, our population has continued to grow exponentially and this has caused major changes in land use and land cover. With more than 80% of the world’s population living in urban areas, studying the growth and development of cities is critical in assessing land cover change (Veldkamp and Verburg 2004). It is essential to study the effects of urbanization and understand the various implications a changing land cover can have on the environment such as the loss of forest and agricultural fields, increased runoff due to impervious surfaces, CO2 emissions from urban areas etc. There are many more environmental concerns out there and awareness and action are necessary, but so is urban development. People need a place to live, but we need to consider Earth, our true home, as we change it in developing more living space.
For our project we decided to monitor urban development in Olmsted County. The city of Rochester is located in this county and is the third largest city in Minnesota. From 1980 to 2000 the city saw a rise in population from 57,890 to 70,745 (U.S. Census Bureau). In order to accurately study the land cover change in this area, we used data downloaded from the internet and the ERDAS Imagine software program to assess urban development from 1985 to 2001.
Methodology
-Defining the study area-
Olmsted County has a total of 654.50 square miles of land and water. In order to do any analysis of Rochester, we first needed to define an exact location as our area of interest. After examining the Public Land Survey (PLS) township boundaries, we concluded that a total of four townships will be sufficient enough for the scope of this project: Cascade, Haverhill, Marion, and Rochester. This amounted to a total of 106.7 square miles. These four counties encompass the city of Rochester as well as the surrounding area that may or not be subjected to urban development and land use change.
-Data acquisition-
To obtain the necessary data for this project, we used two sites to gather the preliminary resources. First, we accessed the MN Data Deli and downloaded the PLS township boundary shapefile. Second, through the United States Geologic Survey Glovis website, we were able to select Landsat 5 satellite images of our study area. We selected images from the spring of 1985 and 2001. However, we were not able to find images from the same month so we settled for April and May. To eliminate cloud cover, we used a 10% cloud cover filter on both of the images in hopes that it would negate any obstruction that clouds may present. Both images were downloaded and imported to ERDAS Imagine for further image processing
-Image Preprocessing-
In order to successfully import Landsat images, it is necessary to use the Layer Stack tool in ERDAS. This step requires the user to stack the .tif files on top of each other, creating a final output image. Once this was successfully done the next step was to georeference each image so that they were both projected in the same reference system. We used the UTM NAD83 Zone 15N projection and datum for our images.
Meanwhile, we imported the PLS township boundary data into ArcMap to define our study area. By opening the attribute table and selecting the appropriate townships, we highlighted our study area. In order to use this file in ERDAS, we exported the selected features to a USB flash drive as a shapefile and were then able to import it to ERDAS.
Within the same 2D Viewer of ERDAS Image, we opened both our Landsat satellite raster layer and our four township vector layer. To cut out our study area from the satellite images, we needed to perform a simple clip operation. The first step in doing this was to open a new Area of Interest file. By clicking on each township and copying it, we were able to paste it to our Area of Interest file. The clip tool allowed us to essentially cut out our study area from the Landsat image. After doing this for both of our satellite images, we were ready for the classification step.
-Image Classification-
We chose to perform a supervised classification on both of our images. It was agreed upon that through a supervised classification, we would be able to more accurately select our training areas. We decided that a total of 5 classes would be used for this classification: Urban, Water, Bare Soil/Vegetation, Green Vegetation, and Forest. The choice to separate bare soil/vegetation and green vegetation was in hopes that it would make the identification, classification and quantification easier. We selected 10 to 20 training areas for each class. After each training area was created, it would be added to the Signature Editor tool. Finally, with all the training areas added we then merged them for each defined class. After which we were able to run the classification. The distinction between agriculture and forest was fairly easy to make in that forested areas had a courser texture, irregular shape, and a darker shade of red than the brighter color and uniform shape of agricultural fields. Urban areas were easy to detect, however some areas appeared to have mixed pixels, which we believed were suburban areas.
-Change Detection-
After our classifications were complete, it was necessary to quantify the amount of change that happened from 1985 to 2001. In order to observe this change quantitatively, we performed a thematic change detection on both the images. With the thematic change we were able to also create a Summary Report. This gave us the amount of land changed from each zone throughout the 16 year period and to what class it was converted to. This is the advantage of a thematic change detection, it gives the user “from-to” information about the two images. With this information we were able to quantify how much land from water, vegetation, and forest was converted to urban land over the span of 16 years.
-Accuracy Assessment-
We performed an accuracy assessment on the 2001 image. To do so, we downloaded NAIP imagery of the state of Minnesota from the USDA NRCS website. We applied the same image processing steps to slim down the image to just our study area. After doing so, this image was used as our reference data for the accuracy assessment. However, it is important to note that the earliest land cover image provided from the NAIP imagery set was 2006. This might cause some error in accuracy, but it was the best reference data we were able to find. A stratified random sample method was used with 105 total points created and a minimum of 15 for each class. The error matrix shows our 2001 image had an accuracy of 71% and a Kappa statistic of 0.64. These numbers are fairly close, so we feel confident in our assessment. We believe the majority of error is due to our attempt to distinguish between green vegetation and bare soil.
Results (all information can be found in the Appendix)
After the completion of all these steps it is clear that there has been an increase in urban development. With the increase in urban space comes a decrease in land cover and land use for forests and agriculture. Almost 24% of the classified vegetation was converted to urban land. This amounted to a total of 3,685.58 acres converted from agricultural to urban development. As for forestland converted to urban, 15% had been converted to urban which amounted to 4,829.1 acres of land. As one can see from the images in our Appendix, the urban growth is most prominent in the northwest and south central region of images. The central area of Rochester also has a clear rise in urban development, most likely due to the continual development of urban infrastructure such as commercial buildings, new roads, housing etc.
Discussion
It is comes to no surprise that the city of Rochester has experienced an increase in urban development. This may be directly correlated to the rise in population from 1980 to 2000. The population of 70,745 in 2000 in comparison to 57,890 in 1980 explains the significant increase in urban land cover and the need to convert forest and agricultural land to urban areas. One of the drawbacks to our project was the fact that we did not use Landsat 7 information from 2012. Due to the scan lines in the Landsat 7 imagery, it would have been a challenging task to accurately and efficiently classify and detect land cover change. However, this project would have been more relevant. Another thing we would have done differently would be to classify agriculture as one class instead of bare soil/vegetation and green vegetation. Another setback to our project was the absence of a reference map for the 1985 classification image. This prevented us from successfully performing an accuracy assessment on the 1985 image.
Land cover change poses a variety of environmental issues that require attention and mitigation plans. An increase in impervious surfaces due to urban development may lead to an increase in soil erosion, water pollution, C02 emissions etc. Although our objective was not to study the environmental impacts of urban development, studying and knowing which areas were converted to urban surfaces is important in identifying starting points for mitigation plans.
Overall, we feel confident in the work we accomplished for this project. In retrospect, there are a couple of things we would have differently. However, for the first time using ERDAS Imagine to complete a project with satellite imagery, we believe that we have provided a detailed procedure for studying land cover and land use change. This was a learning experience working with ERDAS, and upon completion we feel that we have stronger and deeper understanding of the process of image classification and applying satellite imagery to real life situations
Appendix
Supervised Classification of 1985 Image
Supervised Classification of 2001 Image
2006 NAIP Reference Image of MN Land Cover
Percent Change Chart
Percent Change to Urban Area 1985-2001Percent Change / Acres
Bare Agriculture to Urban / 0.1367 / 3119.48
Water to Urban / 0.1998 / 37.42
Green Vegetation to Urban / 0.1085 / 566.1
Forest to Urban / 0.1513 / 4829.1
Urban to Urban (Acres in 1985) / 15503
Total / 0.5963 / 24055.1
Accuracy Assessment for 2001 Supervised Classification
References
Veldkamp, A., and P. H. Verburg. "Modelling land use change and environmental impact." Journal of Environmental Management 72.1 (2004): 1-3.
http://www.census.gov/
FR 3262 Lab Manual 7 Image Operations and Clip
FR 3262 Lab Manual 11 & 11a Supervised Classification and Accuracy Assessment
FR 3262 Lab Manual 12 Change Detection