Mississippi River Flood Analysis

David Baird
Marc Courville
Claire Pennington
Remote Sensing/GE 324
Final Project
December 7, 2001

Plan of Action

The first step in our project was to develop a plan of action that would serve as a model for our project. After studying the materials distributed in class, we determined that it would be important to proceed with the following steps:

1. Development of image processing techniques
2. Classification of the image
3. Assessment of accuracy
4. Detection of change
5. Calculations of the change
6. Interpretation of the analysis

Executive Summary

The primary objective of this project was to assess the amount of change that occurred between the years of 1988 to 1993. The region under analysis is a portion of the Mississippi River watershed. In 1993 a severe flood burdened much of the area. By comparing the satellite images of the same location in these two different time periods, an assessment of flood damage can be interpreted.

Image Processing Techniques

Classification Process and Results

Five classes were classified in the miss88.img: Vegetation, Urban Land Use, Residential, Water, and Barren Land. In the signature editor, the values for each class in the miss88.img file was set in order from 1-5, and for the class values for the miss93.img file, the values were to be multiplied by ten so the values ranged from 10-50.

Accuracy Assessment

Six samples were selected from each class, for example six areas of vegetation and six areas of residential. In an accuracy assessment of the miss88 image, the results were more accurate than the flooded image of miss93. From the five classes the number correct to the total number is as follows:

1988 image
Total 28/30 = 93%

1993 image
Total 27/30 = 90%

Change Detection Modeling and Analysis

A model was constructed from the model maker on the ERDAS Imagine model tab. From the model tools, a raster image tool was selected to represent the raster layers and included the supervised miss88 image and the supervised miss93 image as separate layers. The formula created was:

PV x 625 = # of acres changed
4046.86

Through careful calculation, the total amount of acreage that changed from land to water was determined to be 12,447.92.

Mississippi River Flood Analysis

Introduction

Study Area Characterization

The region that was analyzed, a portion of St. Charles, Missouri, is a low-lying river valley. Some of the area is urbanized, however, the largest amount is agricultural land.
This is most likely an area which has flooded many times in past. It is a possibility that the river was at one time on a different course. The evidence of this feature can be seen in the meander scar, which appears directly north of the river in the 1988 image. When the flooding occurred in 1993, it extended to the northern edge of the meander scar. There is not a substantial amount of human settlement or infrastructure in this region. The southern bank of the river has more man made structures, but they are still limited in number. This would also reflect that the citizens expected this area to be prone to flooding.

Image Processing Techniques

In order to support the classification of the image, three separate image-processing techniques were performed. These techniques were the NDVI, band ratioing, and band combinations.
NDVI
The NDVI index is very effective in distinguishing land use and land cover in both large and small land area. Specifically for the purposes of this project, it has been used to determine the health of vegetated areas. Also, the brighter an area is, the healthier the vegetation. In the 1993 image, there are many brighter areas that were less visible in the earlier image. This shows that the health of the vegetation has improved.
Band Ratios
Band ratios allow the user to detect and determine specific characteristics and features of an image. By enhancing subtle changes in color and spectral reflectance, different band combinations are useful for different types of analysis. After considering a variety of different ratios, it was decided that the 3/7 and 4/3 ratios were the best choices for our purposes.

· 3/7 ratio- This is effectively used for determining water turbidity. This ratio was
applied because one can easily delineate land from water pixels (Avery 442). In examining the images, it was then extremely apparent as to where the boundaries of the water and the land lie. This is the primary function of change detection, meaning that this ratio can be used as a reference. Since the latter band is higher in value than the first in the original image, which ranges from 0 to 255, it cannot be divided unless a adequate model was developed. By creating a model that changed the pixel values to 1-256, a 3/7 ratio could then be applied.
· 4/3 ratio- This ratio is appropriate for defining the distribution of vegetation
(Avery 442). The 1988 image showed a large amount of vegetation, however, in the 1993 image, much of this area had been flooded. Within the land that was flooded, some pockets of the vegetation were still evident. With this ratio the lighter tones distinguish the larger density of vegetation. Residential areas also are clearly defined in contrast to the agricultural land.

Band Combinations
The 3,2,1 and 5,4,1 combinations were helpful in the analysis of the miss88 and miss93 images. Several situations arose when a comparison was needed in order to determine certain features more accurately. The 3,2,1 combination is true color and the 5,4,1 combination clearly shows healthy vegetation.

Classification Process and Results
Five classes were classified in the miss88.img:
1. Vegetation
2. Urban Land Use
3. Residential
4. Water
5. Barren Land

A supervised classification was performed on these five classes. These classifications were given a name for organization (ex; 88veg1, 88veg2, and etc.). Histograms were viewed on these samples to determine the
quality and if another sample should be chosen. There were six samples taken for each class, and once there were classifications from all six samples, they were merged in the
Signature Editor of ERDAS. The merged set was then given a name, for example; 88veg_merge.sig, which contained all six samples of each classification.

The miss93.img image was to be manipulated the same as the miss88.img image; therefore, the same operations were performed with both images. In the signature editor the values for each class in the miss88.img file was set in order from 1-5 (vegetation=1, residential=2, urban=3, water=4, & barren land=5), and for the class values for the miss93.img file, the values were to be multiplied by ten so the values equaled 10, 20, 30, 40, & 50. This must be completed in order to manipulate and compare the miss88 image and the miss93 image with the modeler.

Accuracy Assessment

An accuracy assessment was performed on the supervised image. A topographic map was downloaded from topozone.com and was used as a reference for the areas selected on the image. Six samples were selected from each class, for example six areas of vegetation and six areas of residential. In an accuracy assessment of the miss88 image the results were more accurate than the flooded image of miss93. In comparison to the topographic map of the Saint Charles quad and the supervised image, 27 of the total 30 areas selected matched the primary selection, therefore, the total percentage correct for accuracy was 90%. From the total the five classes and number correct (out of six areas) was broken down as follows for the miss88 image:

· Vegetation 6/6
· Residential 5/6
· Urban 5/6
· Water 6/6
· Barren Land 6/6
· Total 28/30 = 93%
The total for the five classes in the miss93 image are as follows:
· Vegetation 5/6
· Residential 6/6
· Urban 5/6
· Water 6/6
· Barren Land 5/6
· Total 27/30 = 90%

Change Detection Modeling and Analysis

To detect change from one image to another in ERDAS, organization is very important. A model is constructed from the model maker on the ERDAS Imagine model tab. From the model tools, a raster image tool was selected to represent the raster layers and included the supervised miss88 image and the supervised miss93 image as separate layers. A circle in the model tools represents the function operator to perform on the images to add them for comparison. The function definition table and calculator from the model function were then used to add the two images, and then an output is produced in the final raster layer in the interface. The purpose in combining these two images was one of the first steps taken to determine change between the images from 1988 to 1993. The Run button (lightening bolt) is then used to perform the functions to produce the output layer. Once the output layer is produced, it is then opened after it is displayed as pseudo color and "fit to frame", and once this is completed the raster attributes can then be manipulated to visually display changes in the image using colors of choice.

The five different classes that were chosen in this project and were given a significant color. The supervised miss93 image's values were multiplied by ten and the supervised miss88 image's values were only 1-5 and this equaled a grand total of 55 different values to be manipulated. The values were (supervised miss93):
· Vegetation = 10
· Residential = 20
· Urban = 30
· Water = 40
· Barren Land = 50
And in the supervised miss88 image the values were:
· Vegetation = 1
· Residential = 2
· Urban = 3
· Water = 4
· Barren Land = 5
When these values were determined, the first steps were then taken to determine change. For example, to determine how much vegetation changed to water from the miss88 image to the miss93 image the number to search would be 41. 40 would be the number that the areas have changed to, and 1 would represent what the landscape has changed from. The next step was to open the Raster Attribute table in the Viewer. Then a consistent color was assigned to the values, for example; numbers 11-15 in the raster attribute table would represent vegetation and assigned red, 21-25 would be residential and given green, 31-35 would signify urban with the color yellow, 41-45 would embody water with blue as the assigned color, and the numbers 51-55 would represent barren land and comprise the color orange. Each number in the ranges 1-55 would have a significant pixel value and these values were then plugged into a formula to determine the area of change. The formula was:

PV x 625 = # of acres changed
4046.86
PV = pixel value

· Example if the pixel value for #41 in the raster attribute table was 2432;
2432 x 625 = 1,520,000 ---------> 1,520,000/4046.86 = 375.6 acres
Then it was determined that 375.6 acres of vegetation changed to water from the miss88 image to the miss93 image. The same steps were taken for all five classes.
The actual results were pretty astonishing because some of the numbers brought certain things to attention. Each class showed a significant amount of change, but some of the results were questionable. 69.5 acres of residential land changed to barren land this number is not 100% percent accurate because the reflection may have dominated certain pixels but not fill the pixel and the pixel then represented 25m2 regardless if it filled the pixel. To explain the reason from residential changing to barren land could be from the destruction of old homes, house fires, and etc. Each class was placed into a table that showed the change from the other classes (Table 1). As seen from Table 1 a total of 12,447.92 acres changed from the other classes to water, which is not surprising because of the flood. One area of the change results became suspect when the numbers did not seem credible. The change from urban to residential was shown as 4656.39 acres and to correct this, more fieldwork and analysis must be done. To view a complete analysis of the change, refer to Table 1.

Conclusion

In the analysis of a portion of the Mississippi River Watershed that runs through St. Charles, Missouri, there were several different ways in which the project could have been approached. After determining a plan of action, a distinct path could be followed. With the use of ERDAS, we were able to assess the amount of land the flood encompassed. Through careful calculation, the total amount of acreage that changed from land to water was determined to be 12,447.92. This was of course the most important aspect to the study; however, there were many different categories of analysis such as vegetation to residential and barren land to urban.
In order to complete the study, several procedures had to be performed. These ranged from simple changes in band combinations to the complex task of supervised classification. During the project many obstacles were encountered, but through repeated efforts a final analysis was created.

Home