Lab 6: Digital Change Detection
Goals and Background:
The goal of this lab was to practice various functions to detect change in images over time. Quick qualitative change detection was first practiced with the Write Function Memory Insertion method, which is the assigning of different monitor color guns to bands of multi-date images. This practice only gives a quick visual overview of the Then, the changes in total land cover of each specific land use/land cover (LULC) class in a pair of images (one from 2001 and another from 2011) were calculated along with the percent change. Finally, amounts of LULC change of for specific pairs of LULC classes that a government agency would be interested in were calculated. These included Agriculture to Urban/Built-up and Wetland to Agriculture, among others. This was performed via the Wilson-Lula algorithm. All functions were performed in ERDAS Imagine (including the use of the Model Builder) and Microsoft Excel.
Methods:
Part 1: Change detection using Write Function Memory Insertion
A first qualitative method that was used to detect change was Write Function Memory Insertion. This method consisted of simply assigning the different color guns of the monitor (red, green, and blue) to bands of a pair of images of central western Wisconsin, one image being from 1991, and the other being from 2011. This was easily accomplished via the Set Layer Combinations window available from the Bands pane of the Multispectral tab in ERDAS Imagine. This resulted in an image that showed a pinkish color for areas that had changed. The result of this method can be seen below in the results section.
Part 2 Section 1: Post-classification quantitative changes
For this process, two classified images (2001 and 2011) were used of the Milwaukee Metropolitan Statistical Area which used the same LULC classification system and spatial resolution. Data of the totals for each class in each image were copied to an Excel spreadsheet from the raster attribute table summary data shown in ERDAS Imagine. The data was then converted to area in hectares, from pixels. This required the spatial resolution (30m x 30m), and the conversion from square meters to hectares (1 square meter = 0.0001 hectares). After getting these totals, the difference was found, and then the percent change was found.
Part 2 Section 2: Development of From-to change detection data
This premise of this process was that a government agency wanted to know not just how much change in specific circumstances happened (Agriculture to Urban/built-up, Wetlands to Urban/built-up, Forest to Urban/built-up, Wetland to Agriculture, Agriculture to Bare Soil), but where these specific changes occurred. This process required the use of the Wilson-Lula algorithm. This algorithm takes the two images as inputs to all combinations the user wants to find. It then finds each type within each combination by employing an Either If Or function to assign a 1 value where the input raster has a pixel of the desired type or a 0 if not (Function 1). The resulting rasters are stored in temporary rasters, and then are input to an & bitwise function to find the areas containing the LULC classes desired in both the 2001 and 2011 images.
Examples of the functions used are: EITHER 1 IF ($n1_milwauke_2001==7) OR 0 OTHERWISE, and $n1_memory & $n2_memory.
The model results in rasters of each combination of LULC classes.
Results:
Part 2 Section 1 resulted in a table showing the classes in the area of interest and their percent change. Open Space and Bare Soil had the largest apparent growth, while Shrubland and Agriculture had the largest decrease.
The goal of this lab was to practice various functions to detect change in images over time. Quick qualitative change detection was first practiced with the Write Function Memory Insertion method, which is the assigning of different monitor color guns to bands of multi-date images. This practice only gives a quick visual overview of the Then, the changes in total land cover of each specific land use/land cover (LULC) class in a pair of images (one from 2001 and another from 2011) were calculated along with the percent change. Finally, amounts of LULC change of for specific pairs of LULC classes that a government agency would be interested in were calculated. These included Agriculture to Urban/Built-up and Wetland to Agriculture, among others. This was performed via the Wilson-Lula algorithm. All functions were performed in ERDAS Imagine (including the use of the Model Builder) and Microsoft Excel.
Methods:
Part 1: Change detection using Write Function Memory Insertion
A first qualitative method that was used to detect change was Write Function Memory Insertion. This method consisted of simply assigning the different color guns of the monitor (red, green, and blue) to bands of a pair of images of central western Wisconsin, one image being from 1991, and the other being from 2011. This was easily accomplished via the Set Layer Combinations window available from the Bands pane of the Multispectral tab in ERDAS Imagine. This resulted in an image that showed a pinkish color for areas that had changed. The result of this method can be seen below in the results section.
Part 2 Section 1: Post-classification quantitative changes
For this process, two classified images (2001 and 2011) were used of the Milwaukee Metropolitan Statistical Area which used the same LULC classification system and spatial resolution. Data of the totals for each class in each image were copied to an Excel spreadsheet from the raster attribute table summary data shown in ERDAS Imagine. The data was then converted to area in hectares, from pixels. This required the spatial resolution (30m x 30m), and the conversion from square meters to hectares (1 square meter = 0.0001 hectares). After getting these totals, the difference was found, and then the percent change was found.
Part 2 Section 2: Development of From-to change detection data
This premise of this process was that a government agency wanted to know not just how much change in specific circumstances happened (Agriculture to Urban/built-up, Wetlands to Urban/built-up, Forest to Urban/built-up, Wetland to Agriculture, Agriculture to Bare Soil), but where these specific changes occurred. This process required the use of the Wilson-Lula algorithm. This algorithm takes the two images as inputs to all combinations the user wants to find. It then finds each type within each combination by employing an Either If Or function to assign a 1 value where the input raster has a pixel of the desired type or a 0 if not (Function 1). The resulting rasters are stored in temporary rasters, and then are input to an & bitwise function to find the areas containing the LULC classes desired in both the 2001 and 2011 images.
Examples of the functions used are: EITHER 1 IF ($n1_milwauke_2001==7) OR 0 OTHERWISE, and $n1_memory & $n2_memory.
The model results in rasters of each combination of LULC classes.
 |
| Figure 1: Wilson Algorithm Model |
Part 1: Change detection using Write Function Memory Insertion
One feature that stood out as having changed is the new highway 53 on the east side of Eau Claire. This highway supplied a reference color for which was sought after in the rest of the image. Many roadways, agricultural lands, and areas around rivers seemed to have changed.
 |
| Figure 2: Part 1 Result |
Part 2 Section 1: Post-classification quantitative changes
 |
| Figure 3: Land Use Change by Class |
Part 2 Section 2: Development of From-to change detection data
This part resulted in data which was subsequently mapped. The following figure was made including both the per class percent changes and the new mapped data from this part.
 |
| Figure 4: Prominent Land Cover Changes |
The figure above show substantially more change in Waukesha and Washington Counties, this reflecting the more recent development of these lands away from the shore and away from Milwaukee where development occurred long before this change detection period.
Comments
Post a Comment