Use Change Detection Workflow to Monitor Flooding

来源: https://vis-webcontent.s3.amazonaws.com/quickguides/Spectral+Analysis/Change+Detection/index.html#/

Lesson content

Lesson 1 of 1

Use the Change Detection Workflow to Monitor Flooding

In this quick guide, you will:

•

Open and display two Sentinel-2 images before and after Hurricane Delta struck Lake Charles, Louisiana. * •

Use the Change Detection Workflow to assess where open water increased after the hurricane.

Sample Data

Download sample data below. Then extract the contents of the .zip file to a local directory.

[Sentinel2_ChangeDetection.zip

137.8 MB

DownloadArrow down with horizontal line beneath it](assets/Sentinel2_ChangeDetection.zip)

About Change Detection

When aerial or satellite sensors acquire data from the same region multiple times, it provides an opportunity to study how the landscape changes over time.

Change detection is the process of identifying, describing, and quantifying differences between two images of the same geographic area. The two images should both reflect similar observation conditions – close in the same year, or in the same season in different years. Ideally, the weather and solar illumination angles should be similar between the two images. Finally, they must be co-registered with one another. This ensures that the perceived changes come from actual surface processes and not by differences in geographic positioning (displacement), angle of incidence or from calibration issues.

Open and Display Sentinel-2 Images

1

Select File > Open from the Menu bar. An Open dialog appears. 2. 2

Go to the location where you downloaded the sample data, and select the files HurricaneDeltaAfter_2020-10-10.dat and HurricaneDeltaBefore_2020-09-30.dat. Click Open. The images are added to the Layer Manager and displayed in the Image window. They cover a large area around the city of Lake Charles. 3. 3

Click the Zoom to Full Extent button in the Toolbar. 4. 4

In the Layer Manager, uncheck the HurricaneDeltaAfter_2020-10-10.dat layer to hide it and to view the "before" image.

The "before" image, acquired on 30 September 2020.

The "after" image, acquired on 10 October 2020.

The "after" image shows increased sediment in the Calcasieu River and flooded fields to the east.

Run the Change Detection Workflow

1

In the Toolbox, expand the Workflows folder and double-click Change Detection Workflow. The workflow begins with the Select Data panel. 2. 2

Click the Browse button next to the Input Raster 1 field. This is where you specify the "before" image. The Data Selection dialog appears. 3. 3

Select HurricaneDeltaBefore_2020-09--30.dat and click OK. 4. 4

Click the Browse button next to the Input Raster 2 field. This is where you specify the "after" image. The Data Selection dialog appears. 5. 5

Select HurricaneDeltaAfter_2020-10-10.dat and click OK.

6

Click the Next button to proceed to the Image Registration panel. 2. 7

The images are already well aligned with one another. No image-to-image registration is necessary. Keep the default selection of Skip this step and click the Next button to proceed to the Calculate Change panel.

Select a Change Detection Method

When we talk about “change” in images over time, what exactly does this mean? Various techniques have been developed to interpret change. The following options are provided in the Calculate Change panel:

•

Band Difference: This is the default method. It subtracts the pixel values in one band (Time 1) from those in another image of the same band at a different time (Time 2). * •

Vegetation Index Difference: If change analysis is focused on specific features—such as vegetation, water, or built-up areas—then this is a good option for assessing changes to those features. You will select this option. * •

SAM Image Difference: This option performs Spectral Angle Mapper (SAM) classification on the input rasters. The spectral angle for each pixel is measured, then similarity between the rasters is determined by calculating the angle between the spectra and treating them as vectors in space, with dimensionality equal to the number of bands. * •

ICA/MNF/PCA Difference: Rather than subtracting brightness values from one image to another, another technique is to compute an image transform such as Independent Component Analysis (ICA), Minimum Noise Transform (MNF), or Principal Component Analysis (PCA) on the input spectral bands. The data is transformed to a different space in which redundant spectral information is reduced and unique spectral characteristics are highlighted. When applied to change detection, each component (band) highlights subtle changes that occurred over time in different materials.

1

Click the Method drop-down list and select Vegetation Index Difference. 2. 2

Click the Vegetation Index drop-down list and select Modified Normalized Difference Water Index. The MNDWI index(opens in a new tab) enhances open water features while suppressing noise from built-up land, vegetation, and soil. 3. 3

Enable the Preview option. A preview of the MNDWI difference image is displayed. Bright pixels indicate an increase in open water extent between September 30 and October 10.

In the steps that follow, you will set a threshold on this image to specify "how much" change occurs over time.

4

Click the Next button to proceed to the Threshold panel.

Specify a Change Threshold

The Threshold panel displays a histogram of the MNDWI difference image. It determines a suitable threshold and displays the result in the Image window. Blue pixels indicate a "positive" change, i.e., increased open water extent. The few red pixels you see indicate a "negative" change, or decreased open water extent.

1

Click and drag the right slider back and forth to adjust the "positive change" threshold. Moving the slider to the right results in fewer blue pixels. Moving to the left results in more blue pixels. 2. 2

In the Max Value field, enter a value of 0.63 and press the Enter key. This was the original, default value. 3. 3

Click and drag the left slider all the way to the left. This removes all "negative change" (red) pixels. We are only interested in the positive change, i.e., where water increased over time.

4

Click the Next button to proceed to the Vectorize Changes panel.

Convert "Change" Pixels to Vectors

At this step, the workflow will convert the blue pixels to vector polygons. The Minimum PIxels and Smooth Kernel Size parameters control the amount of detail the vectors will show.

1

Click the Zoom drop-down list in the Toolbar and select 200% (2:1). 2. 2

Enable the Preview option. 3. 3

In the Layer Manager, uncheck all layers except for Preview. A preview of the current vector polygons is displayed in the Image window.

4

To further reduce the number of small polygons, increase the Minimum PIxels value to 21. Polygons with fewer than 21 pixels are removed. 2. 5

Keep the default value of 3 for Smooth Kernel Size.

6

Click the Next button to proceed to the Export Results panel. The final vector polygons are displayed in the Image window.

Tip:If the result still contains too many polygons or too much detail for your liking, you can optionally click the Back button to return to the Threshold panel. Then increase the Max Value field and continue with the rest of the workflow.

Create Output Products

1

Enable the Export Classification Raster option, and enter a file name of LakeCharles_Flooding_Class.dat. 2. 2

Enable the Export Shapefile option, and enter an output file name of LakeCharles_Flooding.shp. 3. 3

Click the Finish button. The output classification image and shapefile are added to the Layer Manager and displayed in the Image window. The shapefile has blue outlines. The classification image has two classes: Unclassified (black) and Change (green).

4

This concludes the exercise.

In summary, the Change Detection Workflow helps you create output products that show the spatial distribution of changes between two dates. "Change" can be based on:

•

Differences in brightness values between two dates for a given band * •

Differences in feature indices such as vegetation extent, water, or built-up areas * •

Differences in spectral angles * •

Differences in image transforms (ICA, MNF, PCA)

Since water extent was the focus of this use case, you used a Modified Normalized Difference Water Index (MNDWI) as the basis for assessing change in the Change Detection Workflow.

Your input is important to us, please take a few moments to fill out ourQuick Guide Feedback(opens in a new tab)form.