Display the lake

To calculate the lake's change in area over time, you'll compare imagery of the lake taken by Landsat satellites between 1984 and 2014. The Landsat satellite program has been in operation for over 40 years, making its imagery vital for monitoring major planetary changes. First, you’ll set up the project and compare the imagery visually.

Open the project

You'll download the project and open it in ArcGIS Pro.

  1. Download the Lake Poyang Project package file. Locate the downloaded file on your computer and move it to a location where you can easily find it, such as your Documents folder.
    Note:

    Depending on your web browser, you may be prompted to choose the file's location before you begin the download. Most browsers download to your computer's Downloads folder by default.

  2. If you have ArcGIS Pro, double-click the Lake_Poyang_Project.ppkx project package to open it. If prompted, sign in using your licensed ArcGIS account or ArcGIS Enterprise account.
    Note:

    If you don't have access to ArcGIS Pro or an ArcGIS organizational account, see options for software access.

    Default project

    The project opens to central-eastern China. To get you oriented, three layers (and the basemap) are turned on: the Three Gorges Dam, the Yangtze River, and a simple representation of Lake Poyang. Lake Poyang is several hundred miles downstream of the Three Gorges Dam.

Visually compare Lake Poyang over time

The project also contains three imagery layers, which are currently turned off. These layers show Lake Poyang at the peak of its rainy season during three different years: 1984, 2001, and 2014. You'll compare the imagery visually to get a sense of how the shape of the lake has evolved over time.

  1. In the Contents pane, right-click the Lake Poyang layer and choose Zoom To Layer.

    Lake Poyang

    Lake Poyang is mostly long and narrow, extending south from the Yangtze River. Its narrow shape means that even small shrinkages in lake surface area can lead to fragmentation of aquatic habitat. Additionally, several cities around the lake depend on the fishing and transport trades provided by the lake. On a national level, the lake is China's largest source of freshwater. The shrinking of the lake can have devastating ecological and economic ramifications.

  2. In the Contents pane, uncheck the box next to the Three Gorges Dam, Yangtze River, and Lake Poyang layers to turn them off.

    Layers turned off

  3. In the Contents pane, check the box next to the June 1984.tif layer to turn it on.
  4. On the map, pan and zoom with the mouse so that the entire image is visible.

    Lake Poyang in 1984

    This image shows the lake in June 1984. The image was acquired by the Landsat 5 satellite.

    Note:

    The Landsat satellite program is a joint initiative between two American government agencies: the United States Geological Survey (USGS) and the National Aeronautics and Space Administration (NASA). You can learn more about the Landsat program on the USGS Landsat Missions page.

    The imagery shows a bright and clear distinction between the blue lake and the green vegetation nearby. Although these colors may seem natural, they are actually a combination of colors on the electromagnetic spectrum that are normally invisible to the human eye.

  5. In the Contents pane, click the arrow next to the June 1984.tif layer.

    Lake Poyang band combinations

    Landsat imagery comes with different bands of colors on the electromagnetic spectrum. There are as many as 7 to 11 bands per image, based on the Landsat type, as you can see in the table below.

    Band Landsat 5 Landsat 7 Landsat 8

    1

    Blue

    Blue

    Coastal aerosol

    2

    Green

    Green

    Blue

    3

    Red

    Red

    Green

    4

    Near Infrared 1

    Near Infrared 1

    Red

    5

    Near Infrared 2

    Far Infrared

    Near Infrared

    6

    Thermal

    Thermal

    Far Infrared 1

    7

    Mid Infrared

    Far Infrared 2

    Far Infrared 2

    8

    Panchromatic

    Panchromatic

    9

    Cirrus

    10

    Thermal Infrared 1

    11

    Thermal Infrared 2

    Because all the bands cannot be depicted at the same time, you usually pick a combination of three bands that you display through the color channels red, green, and blue, which can be seen by the human eye. As you can see in the legend, the 1984 image uses Near Infrared 2 for the red color channel, Near Infrared 1 for the green color channel, and Red for the blue color channel.

    All three images in your project use band combinations that emphasize vegetation, making the boundaries between the lake and the surrounding landscape clearer and more distinct. Next, you'll compare the 1984 imagery to the later imagery to see how the lake has changed.

  6. Check the box next to the June 2001.tif layer to turn it on.

    Lake Poyang in 2001

    This image was taken by Landsat 7 instead of Landsat 5, so its colors are different. Without a side-by-side comparison, it's difficult to see exactly what has changed. You'll use the Swipe tool to compare the images side by side.

  7. In the Contents pane, click the June 2001.tif layer to select it.
  8. On the ribbon at the top of the application, click the Raster Layer tab. In the Compare group, click Swipe.

    Swipe tool

    When you move the pointer over the map, it changes to an arrow.

  9. Drag the map in the direction the arrow is pointing.

    Swipe on map

    The selected layer is hidden where you drag the pointer. You can now compare the two images. As you drag the Swipe tool back and forth (or up and down), you'll see that most of the change happens on the southern and eastern ends of the lake. Areas where the lake has receded are generally a dull orange, because there is no vegetation there. Overall, in 2001, the lake had visibly less surface area than in 1984. Both of these images were taken before the Three Gorges Dam's opening in 2003, so the cause of the lake's decline during this period may be due to drought or other meteorological trends.

  10. In the Contents pane, turn on the May 2014.tif layer.

    Lake Poyang in 2014

    This image was taken by Landsat 8. The orange areas that indicated bare earth from the receding lake in the 2001 image now show up as bright green due to vegetation growth, indicating long-term water level change.

  11. In the Contents pane, click the May 2014.tif layer to select it.
  12. Use the Swipe tool to compare the 2014 image to the 2001 image.

    The lake appears to have undergone additional surface area loss, mostly in its southern and western parts. Visually, it's clear that the lake has diminished between 1984 and 2014 (at least during the rainy season, when all three of these images were taken), although the exact amount is unknown.

  13. On the ribbon, on the Map tab, in the Navigate group, click the Explore button to exit the swipe mode.

    You can navigate the map normally again.

The next step is to quantify the area loss more precisely.


Classify land cover to identify the lake

To quantify the change in lake surface area from 1984 to 2014, you'll classify the land cover in both images, identifying the areas covered with water and distinguishing them from other land cover, such as vegetation or urban areas. In multispectral imagery, such as Landsat, every individual pixel (or cell) in the image has a value for every spectral band. As you can see from the vibrant imagery of Lake Poyang, there are many possible color values for all varieties of shades and hues. However, all the pixels representing the same land cover tend to have somewhat similar spectral values. By classifying the image, you'll identify the pixels that are similar in value and group them together to represent a small number of classes, such as water, vegetation, or urban areas.

You'll use a particular type of classification technique, known as an unsupervised classification, in which the software uses statistical analysis to decide which values are similar enough to each other to be grouped together into classes. You'll only need to specify how many classes you want to obtain, and the tool will produce that exact number of classes. The tool you'll use to do that is the Iso Cluster Unsupervised Classification.

Classify land cover in 1984

First, you'll classify the 1984 image.

  1. On the ribbon, click the Analysis tab. In the Geoprocessing group, click Tools.

    Tools button

    The Geoprocessing pane appears.

  2. In the Geoprocessing pane, type Iso Cluster Unsupervised Classification in the search box. Click the result with the same name.

    Iso Cluster tool search results

    The Iso Cluster Unsupervised Classification tool opens. This tool runs an unsupervised classification on the imagery layer, or raster, of your choosing. It uses the Iso Cluster algorithm to determine the characteristics of natural groupings of cells and creates an output layer based on the number of classes you want. You'll run the tool on the 1984 image layer.

    You'll classify the pixels of the 1984.tif imagery into four classes only, because you primarily want to see the lake, so there is no need for a large number of classes that would distinguish many types of land cover.

  3. In the Iso Cluster Unsupervised Classification tool parameters, choose the following values:
    • For Input raster bands, choose June 1984.tif.
    • For Number of classes, type 4.
    • For Output classified raster, type Iso_1984 at the end of the poyang.gdb location.
    • Leave the other parameters unchanged.

    Parameters of Iso Cluster tool

  4. Click Run.

    After the tool finishes, the output layer is added to the map. The colors on your map might differ from the ones in the example images.

    Classified 1984 image

    The new layer resembles the original June 1984 imagery, but there are now only four colors that represent each of the four classes that were generated by the classification tool. All image layers comprise grids of pixels, also known as cells, but in the original imagery the pixels had thousands of different colors. The Iso Cluster Unsupervised Classification tool took all the pixels in the original image and sorted them into four value classes, based on their spectral similarity. It then chose four colors at random to symbolize each class. It looks like all water cells were classified into one class (Value 1), while vegetation, cloud cover, and other land cover types are captured in the three other classes.

  5. In the Contents pane, for the Iso_1984 layer, right-click Value 1 and change the color to Yogo Blue.

    Yogo Blue color

  6. Change the other values (2, 3, and 4) to No color.

    Only the water value remains visible. You'll compare it to the original June 1984 image to make sure the classification is correct.

  7. In the Contents pane, turn off the the May 2014.tif and June 2001.tif layers.

    Classification and imagery on map

  8. Click the Iso_1984 layer to select it. On the ribbon, on the Raster Layer tab, turn on the Swipe tool to compare the two 1984 layers.

    Although the lake boundaries mostly match up, the classified value also includes smaller bodies of water around the lake. You'll remove some of these smaller water bodies in the next section. There is also a portion of the lake that was not classified into the same value as the rest of the lake due to cloud cover.

    Clouds in image

    Clouds often obscure ground features in satellite imagery. The cloud cover in this image is relatively minor, so it won't have a significant impact on the analysis, but the analysis could be improved by using imagery with even less cloud cover.

  9. On the ribbon, on the Map tab, reactivate the Explore tool.
  10. On the Contents pane, turn off the Iso_1984 and June 1984.tif layers.

Classify land cover in 2014

Next, you'll classify the 2014 image to see how the lake has changed over time. The Geoprocessing pane is already open to the Iso Cluster Unsupervised Classification tool, so you'll only change the parameters before running the tool again.

  1. In the Contents pane, turn on the May 2014.tif layer.
  2. In the Geoprocessing pane, change the following parameters:
    • For Input raster bands, change June 1984.tif to May 2014.tif.
    • For Output classified raster, type Iso_2014 at the end of the poyang.gdb location.
    • Leave the other parameters unchanged.

    Parameters for Iso Cluster tool

  3. Click Run.

    Classified 2014 image

    As before, the water cells were classified in Value 1 of the new layer.

  4. In the Contents pane, for the Iso_2014 layer, change the color of Value 1 to Light Apple.

    Light Apple color

  5. Change the other values (2, 3, and 4) to No color.
  6. On the ribbon, on the Raster Layer tab, turn on the Swipe tool to compare the two 2014 layers.

    Here again, the water classification seems quite accurate.

  7. On the Contents pane, turn off the May 2014.tif layer and turn on the Iso_1984 layer.

    Result of the two classification processes

    The visible blue areas indicate areas that were covered with water in 1984 but not in 2014, more clearly showing the lake's reduction between the two points in time.

  8. On the Quick Access Toolbar, click the Save button.

    Save button

    Note:

    If you receive a message telling you this project was created with an earlier version of the software, click Yes to proceed.

You classified the 1984 and 2014 images to show land cover, creating a single value for water from the many water values in the original image. Next, you'll smooth out the lake boundaries and remove many of the smaller water features that were classified alongside Lake Poyang's.


Clean up the classification

You'll now clean up your classified images with generalization analysis tools to remove small errors or minor water bodies around the lake. You'll also smooth the lake's boundaries.

Filter individual pixels

First, you'll clean up the small, isolated pixels that were classified as water but are not part of Lake Poyang. Some of these pixels belong to tiny ponds or water bodies, while others were classified incorrectly. Either way, they should not be counted when calculating Lake Poyang's area, so you'll run a geoprocessing tool to eliminate as many of them as possible.

  1. If necessary, open the Poyang Land Cover project in ArcGIS Pro.
  2. On the ribbon, click the Analysis tab and click Tools to open the Geoprocessing pane.
  3. In the Geoprocessing pane, click the Back button to return to the search box.

    Back button

  4. In the search box, type Majority Filter. Click the Majority Filter tool.

    Majority Filter search results

    The Majority Filter tool is a data generalization tool. It replaces cells in an image or raster layer based on the value of the majority of the neighboring cells. If a cell has been classified as Class 1 (water), but three of its four neighboring cells have been classified as Class 2, the tool will change the cell value to fit the surrounding values, in other words, to a Class 2. You'll run the tool two times, once for each classified image.

  5. In the Majority Filter tool, for Input raster, choose Iso_1984.
  6. Change the Output raster name to Filter_1984 at the end of the poyang.gdb location.

    The other parameters allow you to choose how many neighboring cells the tool will use and whether a majority of contiguous cells must be the same value or if only half must be. To generalize the maximum amount of individual pixels and create a greater smoothing effect, you'll use half.

  7. For Replacement threshold, choose Half.

    Majority Filter tool

  8. Click Run.

    Filtered 1984 image

    The generalization removed many of the individual pixels, but many still remain. Additional generalization may be warranted, but generalization also runs the risk of removing data that you do want; in other words, you may lose cells that represent Lake Poyang. You'll fix some of the remaining issues when you smooth out the boundaries later, but for now, you'll run the tool on the other image layer.

  9. In the Geoprocessing pane, change Input raster to Iso_2014 and the Output raster name to Filter_2014.
  10. Click Run.

    The generalized 2014 image is added to the map.

    Now that you have generalized versions of the two classified images, you no longer need the original classified images on the map, so you'll remove them.

  11. In the Contents pane, right-click the Iso_2014 layer and choose Remove.

    Remove option

  12. Similarly, remove the Iso_1984 layer.

    If you need to access these layers again, you can find them (as well as all the other layers you create in this project) in the poyang.gdb database in the Catalog pane.

Clean image boundaries

You've removed some of the individual pixels in each image. Next, you'll clean the boundaries between values in each image to remove the pixelated, granular edges.

  1. In the Geoprocessing pane, click the Back button.
  2. Search for and open the Boundary Clean tool.

    The Boundary Clean tool smooths boundaries between classes by expanding the boundaries and then shrinking them back to their original size. Doing so generally removes individual pixels and replaces them with the value of the pixels around them. Its results achieve a similar effect as the Majority Filter tool, but with a focus on class boundaries.

  3. In the Boundary Clean parameters, choose the following values:
    • For Input raster, choose Filter_1984.
    • Change the Output raster name to Clean_1984.

    Boundary Clean tool

    The Sort type parameter determines whether values with larger or smaller areas are prioritized during expansion, and the check box determines the number of times the process is run. You'll accept the default options for these parameters.

  4. Click Run.

    Cleaned 1984 image

    The new 1984 layer is added to the map. The differences are small, but the boundaries between values are smoothed out. Additionally, more of the small, individual pixels scattered throughout the image are removed. While a few remain, the generalization tools have cleaned up the image substantially. If you want to see the differences for yourself, try using the Swipe tool and zooming in close to the image to compare. Next, you'll run the Boundary Clean tool on the other image.

  5. Run the Boundary Clean tool on the Filter_2014 raster. Change the Output raster name to Clean_2014.

    The new 2014 layer is added to the map. You no longer need the images created by the Majority Filter tool, so you'll remove them.

  6. In the Contents pane, remove the Filter_2014 and Filter_1984 layers.
  7. Save the project.

You've cleaned up your original classified image to remove a lot of the small errors or stray pixel groupings. Next, you'll calculate the area of the lake for 1984 and 2014 and determine how much lake surface was lost between these two dates.


Calculate area over time

Now, you'll calculate the area of Lake Poyang in hectares for 1984 and 2014. First, you'll decide on the appropriate formula.

Develop area computation formula

You'll investigate to decide how to compute the lake area.

  1. In the Contents pane, right-click the Clean_1984 layer and choose Attribute Table.

    The table appears.

    Attribute table

    Each of the layer's four values (for the four classes) has a pixel count. Value 1, which corresponds with water, has approximately 3 million pixels. That's a lot of pixels, but how big is a pixel in real-world terms? You'll find out by checking the image's resolution, which measures how many real-world units correspond to a single pixel.

    Note:

    Deriving the area of a feature directly based on the number of pixels listed in the attribute table, as you are about to do, is only valid if you use a projected coordinate system that does not distort areas too much. As you will see later, the coordinate system used here is WGS 1984 UTM Zone 50N, which is acceptable. Other systems, such as Web Mercator, often used for web maps, would not be acceptable.

  2. Keep the attribute table open. In the Contents pane, right-click the Clean_1984 layer and choose Properties.

    The Layer Properties window appears.

  3. On the left side of the Layer Properties window, click Source.

    Source tab

    The Source tab contains information about the layer's data type and location on your computer, the extent of the data, and how the data is being projected on the map.

  4. Click Raster Information.

    Raster Information section

    The Cell Size X and Cell Size Y parameters refer to the length (X) and height (Y) of each cell or pixel. In this case, each pixel on the map corresponds to a real-world area of 30 units by 30 units. However, you still don't know the unit of measurement. Is it 30 inches? 30 kilometers? You want to calculate hectares, so knowing the unit of measurement is important.

  5. Click Raster Information again to close it. Click Spatial Reference.

    Spatial Reference section

    The Linear Unit parameter refers to the unit of measurement that all spatial calculations involving the layer use by default. In this instance, the unit is meter, which means that each pixel represents a 30 meters by 30 meters (or 900 square meters) area in the real world.

    Under Spatial Reference, you can also see that the Projected Coordinate System used is WGS 1984 UTM Zone 50N, as discussed earlier.

  6. Close the Layer Properties window.

    To find the area of each value in the image, you'll multiply the pixel count by 900 to convert it to square meters. Then, you'll divide the result by 10,000, which is the number of square meters in a hectare. The overall formula is as follows:

    Hectares = (Count × 900) / 10,000

    Now that you have developed the formula, you'll apply it to compute the lake area.

Compute the lake area lost in hectares

You'll now compute the lake area in hectares for 1984 and 2014. Then, you'll find the number of hectares lost between the two dates.

  1. In the Clean_1984 attribute table, click the Add Field button.

    Add Field button

    The Fields view opens, allowing you to manage the fields in the attribute table. A new field is added to the end of the list.

  2. For the new field, change the Field Name value to Hectares. Change Data Type to Float.

    Hectares field

    Float is a data type that allows numbers with decimals.

  3. On the ribbon, on the Fields tab, in the Changes group, click Save.

    Save button

  4. Close the Fields: Clean_1984 view to return to the attribute table.

    The table now contains the Hectares field, but it has no values. Next, you'll calculate the hectares for each class value using the conversion equation discussed earlier.

  5. Right-click the heading of the Hectares field and choose Calculate Field.

    Calculate Field option

    The Calculate Field tool appears.

  6. In the Calculate Field tool, confirm the following.
    • For Input Table, the value is Clean_1984.
    • For Field Name (Existing or New), the value is Hectares.
    • For Expression Type, the value is Python.
  7. Under Hectares =, create the expression (!Count! * 900) / 10000.
    Tip:

    You can add the Count field to the expression by double-clicking Count in the Fields box.

    Expression to calculate hectares

  8. Click OK.

    Hectares field values

    The Hectares field in the attribute table is populated with the area in hectares of each value in the image. Value 1, which shows water, is approximately 270,000 hectares—the area of the lake in 1984.

  9. Close the attribute table.

    Similarly, you'll now calculate the area of the lake in 2014 in hectares. Because the spatial resolution and other characteristics of the two images are the same, you'll use the same formula as before.

  10. In the Contents pane, right-click the Clean_2014 layer and choose Attribute Table. In the attribute table, click the Add Field button.
  11. For the new field, change Field Name to Hectares and Data type to Float.
  12. On the ribbon, on the Fields tab, in the Changes group, click Save. Close the Fields: Clean_2014 view to return to the attribute table.
  13. Right-click the heading of the Hectares field and choose Calculate Field. Confirm that Input Table is Clean_2014, Field Name (Existing or New) is Hectares, and Expression Type is Python.
  14. Under Hectares =, create the expression (!Count! * 900) / 10000. Click OK.

    Hectares for the 2014 map

    The Hectares field in the attribute table is populated, and Value 1, which shows water, is approximately 200,000 hectares. That's the area of the lake in 2014.

  15. Close the attribute table.

    The area of the lake in 1984 is approximately 270,000 hectares, while the area in 2014 is approximately 200,000 hectares. Between 1984 and 2014, nearly 70,000 hectares of Lake Poyang were lost: close to 2,300 hectares a year on average.

    Note:

    There are other methods to compute the area after the classification is completed. For instance, you could first use the Raster to Polygon tool to generate a polygon feature of the lake, which would include an area in square meters. Then you would convert the area to hectares by dividing it by 10,000.

  16. Save the project.

In this tutorial, you compared visually and classified Landsat imagery of Lake Poyang to understand how much the lake's area has changed over time. Your findings indicate a severe problem: the lake has lost thousands of hectares in only 30 years. Your calculations don't reveal the causes of Lake Poyang's reduction, but they do provide factual evidence of a serious problem and provide a starting point for environmental scientists and others to conduct further research.

Note:

If you want to know more about the pace of Lake Poyang's change within that 30-year period, you can perform the same analysis on imagery from intermediary dates, such as the June 2001 image. This way, you can plot the change over time and find out whether the pace is stable or perhaps accelerating.

You can find more lessons such as this on the Introduction to Imagery & Remote Sensing page.

You can find more tutorials in the tutorial gallery.