How to Use ArcGIS Pro for Supervised Classification

Jeff Oppong

Updated:

One of the emerging technologies that steers that frontier of urban development and land use planning is geospatial tools and methodologies.

Remote sensing provides a proactive way of addressing issues by implementing its data collection, processing, and map presentation techniques in solving a wide range of problems that affects the environment, health, safety, transportation, land use demarcation, among several others. Satellite imagery has been used substantially in the land use management and planning arena for the management and monitoring of land systems.

This tutorial contributes to the several land use assessment and management techniques in the geospatial industry by presenting a step-by-step approach of applying Remote Sensing and GIS techniques to perform supervised classification on Landsat 8 imagery. Even though several classification algorithms have evolved over the years, this tutorial utilizes the Maximum Likelihood algorithm.  

Figure 1 shows a summary of the methodological flow employed in performing supervised classification in ArcGIS Pro.

A flow chart showing how to use a GIS software to perform supervised classification.
Figure 1: Summary of flow of method used in performing supervised classification.

Landsat Imagery and Creating Composite Bands

For the purposes of this tutorial, Landsat 8 imagery is accessed from the USGS Earth Explorer’s website, after a free sign-up process.



Free weekly newsletter

Fill out your e-mail address to receive our newsletter!
Email:  

By entering your email address you agree to receive our newsletter and agree with our privacy policy.
You may unsubscribe at any time.



Cloud cover for the Landsat 8 imagery should be set to “less than 10 percent.” To correct for atmospheric errors, the following video is helpful: Satellite Imagery Cloud Removal and Correction In ArcGIS Pro.

The following band combinations may be used for band compositing processes; natural color (bands 4,3,2), false color (bands 7,6,4), color infrared (bands 5,4,3), among many others. These band combinations make it easier to detect specific land use classes during sample training for classification. For this tutorial, (bands 5,4,3) are used.

To perform band compositing in ArcGIS Pro, the following steps are adhered to:

  1. Open ArcGIS Pro and create a new project. 
  2. From the Map tab, select “Add data,” and navigate to the location of the Landsat 8 imagery.
Screenshots showing how to add data in ArcGIS Pro.
Figure 2: Illustration of “add data” in ArcGIS Pro.
  1. Select Bands 5,4,3 and click “Ok.”
  2. From the search bar of the geo-processing toolbox, type and search “Composite bands.”
  3. Set “Input Raster” to bands 5,4,3.
  4. Set “Output Raster” to the desired output name and location.
  5. Click “Run.”
Screenshot showing the ArcGIS pro composite tool.
Figure 3: Illustration of “Composite Bands” in ArcGIS Pro.

Clipping, Training Sample Manager, and Classify

To streamline the Landsat imagery to a desired area of interest, the following steps are followed;

  1. Type and search “Extract by Mask” from the search bar of the geo-processing toolbox.
  2. Set “Input Raster” to the stacked image (results obtained from the composite band process).
  3. Set “feature mask” to the desired area of interest.
  4. Set “Output Raster” to the desired output name and location. 
  5. Click “Run.”
Screenshot showing the extract by mask tool in ArcGIS Pro.
Figure 4: “Extract by Mask” in ArcGIS Pro.

To train sample data for classification in ArcGIS Pro, the following steps are followed;

  1. From the drop-down menu of “Classification Tools” on the Imagery tab, select “Training Samples Manager.”
  2. From the “new schema” tab, select “Edit Properties.”
  3. Set “name” to desired schema output name, and click “Save.”
Screenshot of the schema creation tool in ArcGIS pro.
Figure 5: Schema creation in ArcGIS Pro.

New classes can be added to the schema, and training data can be selected by;

  1. Select “Add New Class” as shown in figure 5 above. 
  2. Set “Class Name” and “Value” to the intended class to be trained.
  3. Click “Ok.”

For this tutorial, the following classes are used; dense forest, open forest, water, and built-up areas. To select training samples for water; 

  • Click water from the schema panel and select a drawing tool.
  • Hover the mouse pointer over the stacked imagery while sketching over water bodies in the area of study.
  • Repeat steps for other land use classes; open forest, dense forest and built-up areas.
  • To save the training samples, click “save as” as shown in figure 6.
Screenshot of the ArcGIS pro training manager tool.
Figure 6: “Training Sample Manager” in ArcGIS Pro.

To classify the Landsat imagery using the collected and saved training sample data, the following steps are used;

  • From the drop-down menu of “Classification Tools” on the Imagery tab, select “Classify.”
  • From the “Classify” interface, set classifier to maximum likelihood.
  • Set “training sample” to results of the “training sample manager.”
  • Click “Run.”
Screenshot of the image classification tool in ArcGIS Pro.l
Figure 7: “Classify” in ArcGIS Pro.

The end result is this classified layer that you can then stylize:

Screenshot of a land use map for Ghana in ArcGIS Pro with shades of green for the categories.
Figure 8: Classified map of Wassa East District in Ghana.

Accuracy Assessment of Classified Image in ArcGIS Pro

To validate the classification results, the following steps are followed;

  1. Type and search “Create Accuracy Assessment Points” from the search bar of the geo-processing toolbox.
  2. Set “Input Raster” to the classified image.
  3. Set “Output Accuracy Assessment Points” to the desired name and location of the automatically generated accuracy assessment points.
  4. Set “Number of Points” to 500 and Sampling strategy to “Stratified Random.”
  5. Click “Run.”
Screenshot showing the Accuracy Assessment Points tool in ArcGIS Pro.
Figure 9: Create Accuracy Assessment Points in ArcGIS Pro.
  1. Open the attribute table of the generated accuracy assessment points.

The default ground truth values for the data points as shown in the attribute table are -1. 

  1. Click on the “Edit” tab from the menu bar.
  2. Set the ground truth values for each of the accuracy assessment points (500) by matching the points to the land use class of the base map.
  3. Turn the column which represents the values of the various land use classes of the “Classified image” upon which these accuracy assessment points were generated off while undertaking step 8.
  4. Save changes. 
Screenshot showing editing the attribute table for groundtruthing data in ArcGIS pro.
Figure 10: Editing Ground truth column in ArcGIS Pro.

To generate a confusion matrix from the accuracy assessment points, the following steps are followed;

  1. Type and search “Compute Confusion Matrix” from the search bar of the geo-processing toolbox. 
  2. Set “input accuracy assessment points” to the generated accuracy assessment points and set “output confusion matrix” to the desired name and location of the matrix.
  3. Click “Run.”

Table 1: Confusion matrix of classification

ClassValueWaterOpen forestDense forestBuilt up areasTotalU_AccuracyKappa
Water100001010
Open forest5194402030.955665020
Dense forest1723712460.963414630
Built up areas02146490.938775510
Total162032424750800
P_Accuracy0.6250.95560.979330.9787200.9586614170
Kappa0000000.93111

More ArcGIS Pro Tutorials

Photo of author
About the author
Jeff Oppong
Jeff Oppong holds a BSc in Geomatic Engineering from KNUST in Ghana. Jeff is a prolific researcher and a GIS/Remote sensing expert who aspires to be a change-agent and a renowned Engineer. Jeff Oppong can be contacted via email