Create a hosted imagery layer

To identify infrastructure vulnerable to natural hazards such as fires, floods, and landslides, you must know the infrastructure locations as well as the areas where those hazards are most likely to occur. You’ll use deep learning analysis in ArcGIS Image for ArcGIS Online to automatically extract building footprints from aerial imagery and perform analysis with raster functions to classify areas in the landscape according to their susceptibility to landslides. ArcGIS Image for ArcGIS Online is a complete software as a service (SaaS) offering for hosting, analyzing, and streaming imagery and raster collections. With the Creator user type and an add-on license for ArcGIS Image for ArcGIS Online, you can manage imagery collections, stream tiled and dynamic image services for analysis, and perform advanced analysis, such as deep learning, in the cloud.

Download data and create a tiled imagery layer

In this section, you’ll create a hosted imagery layer for a portion of Grenada. To do this, you’ll use a guided workflow in your web browser to upload the imagery to the Esri cloud and create the layer in your online organization.

Before creating the hosted imagery layer, you will download the imagery files required for the lesson.

  1. Download the file Grenada TIFF files.
  2. Locate the .zip file on your computer and extract it to a folder that you will use for this lesson.

    Now that you have the data, you will sign in to ArcGIS Online and create the hosted imagery layer.

  3. Sign in to your ArcGIS organizational account.
    Note:

    If you don't have an organizational account, you can sign up for an ArcGIS free trial.

  4. On the ribbon, click the Content tab.

    Content tab on ribbon

  5. Ensure that the My Content tab is selected.

    My Content tab

  6. On the My Content tab, click the New Item button.

    New item button

  7. In the New item window, click Imagery layer.

    Imagery layer option

    The Create imagery layer window appears. First, you will choose the type of imagery layer to create, either a tiled imagery layer or a dynamic imagery layer.

  8. In Step 1 – Get started, choose Tiled Imagery Layer and click Next.

    Tiled Imagery Layer option

    When you create a tiled imagery layer, ArcGIS Image for ArcGIS Online converts your data to cloud raster format (CRF). The CRF format is optimized for reading and writing large files in a distributed processing and storage environment, which makes tiled imagery layers ideal as inputs to raster analysis tools that generate persistent output results. A tiled imagery layer is the recommended option for this workflow, since you’ll run a raster analysis tool to extract building footprints in the next topic. Dynamic imagery layers can also be used in raster analysis, but they are not recommended for this workflow because they do not support distributed processing, unless your data is already in a tiled cloud-optimized format such as CRF. Also, the ongoing costs to host dynamic imagery are a little higher than for tiled imagery layers. The real power of dynamic imagery layers is that they support server-side on-the-fly processing and analysis, which is not covered in this lesson.

  9. Click Next.
  10. In Step 2 – Configure layer, click One Mosaicked Image.

    One Mosaicked Image option

    Note:

    Since you’ll be uploading a collection of 64 individual images, the two valid choices for this step are One Mosaicked Image and Multiple Imagery Layers. For this workflow, One Mosaicked Image is the correct choice, since you’ll want the input images to be mosaicked into a single imagery layer.

  11. Click Next.
  12. In Step 3 – Define imagery, click Browse. Browse to the 3-band TIFF JPEG 90 file that you extracted and open it. Select all the image files (there are 64) by pressing Ctrl+A or by clicking the first file and pressing Shift while clicking the last file.

    Selected imagery files

  13. Click Open.

    The image files begin to upload, and the progress bars shows the upload status.

    Upload progress

    You can advance to the next screen while the imagery is uploading to complete the workflow.

  14. Click Next.
  15. In Step 4 – Set item details, enter the following information:
    • For Title, type Grenada_aerial_imagery.
    • For Tags, type Grenada.
    • For Summary, type Aerial imagery of the island of Grenada.

    Title, Tags, and Summary

    Note:

    If you or someone in your organization has completed this workflow, you may get an error message that says the layer name already exists. If you receive this message, provide a unique name for your tiled imagery layer, such as by appending your initials to the end

  16. Accept the default folder in which to store the imagery layer and click Create.

    When the upload is complete, ArcGIS Image for ArcGIS Online begins the process to create your hosted imagery layer. After the image processes for a few minutes, a message appears stating that you can now safely leave this page. When the layer is created, you will be automatically directed to its item details page. You have created a mosaicked imagery layer in ArcGIS Online that contains multiple TIFF images for Grenada. Now you’ll open your hosted imagery layer in a web map and visually inspect it for buildings.

  17. On the Item Details page of your imagery layer, click Open in Map Viewer Classic.

    Open in Map Viewer Classic

    Note:

    If your button reads Open in Map Viewer, click the drop-down arrow, and choose Open in Map Viewer Classic.

    The image layer appears in the map.

    Imagery layer displayed in the map

  18. Zoom in and pan around the map to inspect the building locations more closely.

    Zoomed in to buildings

    There are hundreds of buildings over this part of Grenada. You could manually create each building and store the footprints as features in a hosted feature layer, but this would be tedious and time-consuming. You will use artificial intelligence and deep learning to extract information from imagery such as building footprints, land use, and land-cover types, and more. The resulting layer will store the building footprints but will take much less time to complete.

You’ve uploaded a collection of images to ArcGIS Online and created a tiled imagery layer, which is optimized for distributed processing and analysis in the cloud.


Extract features from imagery

Building and training your own deep learning models or fine-tuning existing trained models is an advanced task. The most difficult aspect of using deep learning is to create a series of training samples to teach a model to recognize the specific type of information or objects that you are interested in. To save time in the lesson, you’ll use an existing trained deep learning model, rather than building your own. ArcGIS Image for ArcGIS Online provides a growing library of trained deep learning models. By leveraging these models, you can get started right away with using artificial intelligence to extract information and gain insights from your imagery. Next, you will use a trained model in ArcGIS Living Atlas of the World to detect building footprints from your hosted imagery layer. With these building footprints and the results of a landslide susceptibility analysis that you will perform later, you can identify buildings on Grenada that are at risk of landslides.

Extract building footprints using deep learning

Next, you will use artificial intelligence with a deep learning model to extract building footprints from the imagery and create a feature layer of buildings that you will use for further analysis.

  1. On the ribbon, click Analysis. In the Perform Analysis pane, click Raster Analysis.

    Raster Analysis option

    The Raster Analysis tools pane appears.

    Raster Analysis tools

    All tools in this pane are only for raster analysis.

  2. On the Raster Analysis pane, click Deep Learning to expand it, and click Detect Objects Using Deep Learning.

    Detect Objects Using Deep Learning tool

    The Detect Objects Using Deep Learning tool pane appears.

  3. For Choose image used to detect objects, ensure that Grenada_aerial_imagery is selected.
  4. For Choose deep learning model used to detect objects, click the drop-down arrow and choose Choose Deep Learning Model.

    Choose Deep Learning Model

    The Choose Deep Learning Model window appears.

    Choose Deep Learning Model window

  5. Click My Content and choose Living Atlas.

    Choose Living Atlas

    A list of deep learning packages with trained deep learning models appears.

    Available deep learning packages

    There are many trained models managed by Esri in ArcGIS Living Atlas.

  6. For Building Footprint Extraction – USA, click Select.

    Select button

    The Choose deep learning model used to detect objects parameter updates to the package that you selected.

    Building Footprint Extraction - USA package added

    Once you select the deep learning package, the next parameter, Specify deep learning model arguments, initially displays the text Querying model arguments.

    Querying model arguments in parameter

    When the model arguments are finished being queried, the arguments and values appear.

    Model argument names and values

  7. For threshold, type 0.6

    Threshold value updated

    The threshold argument controls the sensitivity of the analysis, which determines how many buildings are detected and how many of those are false positives. The optimal value for any given analysis may vary depending on your tolerance for false positives and false negatives, and how closely your imagery matches the imagery used to train the model. Testing has shown that a value of 0.6 produces good results with this imagery of Grenada.

  8. For Result layer name, type Grenada_buildings.

    Result layer name parameter

    The tool is ready to run. Anytime you perform analysis in ArcGIS Online, there is a cost in credits for Esri to use its resources to process your data. You can determine how many credits will be consumed by a tool before you run it. Once you run the analysis, the credits will be deducted from your organization's available credits.

  9. Uncheck Use current map extent and click Show credits.

    Show credits button

    A pop-up appears, showing how much it will cost to run the tool.

  10. Close the Credit Usage Report window.
    Note:

    The cost of running a tool in ArcGIS Image for ArcGIS Online is based on the complexity of the analysis and the number of pixels to be processed. You can reduce the cost by leaving Use current map extent checked and zooming in to a smaller area for the analysis.

  11. At the bottom of the tool pane, click the Run Analysis button.

    The tool may take 10 to 15 minutes to run. While the tool runs, the tool pane closes, and the Content pane appears. The output layer is unavailable initially, with a wait indicator to show that the tool is still running.

    When tool the tool is complete, the results layer name becomes available and displays in the Contents pane.

    Extracted building footprints in the map

    The extracted building footprints appear in the map with your imagery layer.

    Grenada buildings layer

  12. Zoom in to the map and inspect the results of the analysis.

    Zoomed in to buildings

  13. In the Contents pane, change the Grenada buildings layer visibility to see the buildings in the imagery and the buildings that were detected using deep learning.

    Comparison of buildings in imagery and detected buildings

    The deep learning model did a good job of detecting buildings in the imagery and created polygons from them. Now that you have the buildings as a feature layer, you can use it for many types of operations, including spatial analysis that involves other layers to determine landslide risk.

    Note:

    While the deep learning package extracted buildings from the imagery, you may still need to postprocess the layer to make it more accurate. For example, there may be duplicate buildings that you can remove, or you can regularize the footprints.

You’ve used deep learning with a trained model from ArcGIS Living Atlas to extract building footprints from imagery and store the results in a hosted feature layer. As you examined the results in the map, you saw that the model was successful in detecting nearly all the buildings. Now that you have the building locations, you next task is to determine whether any of them are in areas susceptible to landslides.


Perform landslide susceptibility analysis

Now that you know the building locations, your next task is to perform raster analysis to identify areas on the island of Grenada that are susceptible to landslides. For this analysis, you’ll use imagery layers and a raster function template that are already hosted in ArcGIS Online. You’ll use land use, distance to rivers, and soil types as the main factors that contribute to an area being at a high risk for landslides. With the result of this analysis, you’ll be able to identify the buildings that are in high-risk areas.

Open web map and explore analysis layers

To analyze landslide susceptibility on Grenada, you’ll open a shared web map and explore the imagery layers that you’ll use to perform the analysis.

  1. Open the Grenada landslide analysis web map from ArcGIS Online.

    If you are not signed in with your ArcGIS Online account, sign in now.

    Note:

    This web map was created using data provided by the Government of Grenada, supplied by the United Kingdom. This is a secondary product and has not been verified and is not sanctioned by the Government of Grenada or the United Kingdom.

  2. On the Item Details page, click Open in Map Viewer Classic. If you only see the Open in Map Viewer option, click the drop-down arrow and choose Open in Map Viewer Classic.

    Open in Map Viewer Classic

    The web map appears and shows the island of Grenada. No layers other than the topographic basemap display because they are turned off. Before you view the layers, you will save the web map to your account, as it is currently hosted in another ArcGIS Online organization.

  3. From the ribbon, click Save and choose Save As.

    Save As option

  4. For Title, remove Copy from the end so it is named Grenada landslide analysis. Accept the remaining defaults and click Save Map.

    Now you have a copy of the web map saved in your ArcGIS Online account.

  5. Click the Content button. In the Contents pane, for each layer, check its check box to display it in the map.

    The analysis layers appear in the map, but most of the layers are covered by the one that is drawing on top of it.

    Analysis layers

    Note:

    Grenada DEM 5m and Distance to rivers store continuous data, while Soil types and Sentinel 2 land use are categorical. If you examine the data while zoomed in to a small area, you will see that they have different pixel sizes and are not pixel-aligned with one another. These observations are important to consider when you choose a resampling method.

  6. Explore each of the layers and their legends by panning and zooming, changing visibility, and clicking the Legend button to see what the symbols represent.

    Legend of analysis layers

The layers that you will use in the analysis include land use, soil types, elevation, and distance to roads. All these factors contribute to an area being more susceptible to landslides. For example, the more clay in the soil, the more prone the area is to landslides. Next, you will use the layers you added to the map to perform landslide susceptibility analysis using an online raster function template.

Create a suitability layer using a raster function

Next, you will perform analysis using a raster function and the analysis layers you added to the map.

  1. On the ribbon, click Analysis. In the Perform Analysis pane, click Raster Analysis.

    Raster Analysis option

    The Raster Analysis tools pane appears.

  2. In the Raster Analysis pane, click the Browse Raster Function Templates button.

    Browse Raster Function Templates button

    The Custom Analysis Tools and Raster Functions pane appears.

    Custom Analysis Tools and Raster Functions pane

    You’ll access a raster function template that has been shared to ArcGIS Online. First, you will set the search filter to search only in ArcGIS Online.

  3. Under Custom Analysis Tools and Raster Functions, click System and choose ArcGIS Online.

    Set filter to ArcGIS Online.

    Now the search will only look for raster functions shared to ArcGIS Online.

    Raster functions are operations that allow you to preview the results before generating a new imagery layer and can be chained together for complex workflows as a raster function template. With ArcGIS Image for ArcGIS Online, you can also choose how existing raster function templates are shared in ArcGIS Online. You’ll use an existing template to analyze landslide susceptibility in Grenada.

  4. In the search box, type Grenada landslide.

    The raster function template Landslide Susceptibility Grenada appears.

    Landslide Susceptibility Grenada raster function template

    The Landslide Susceptibility Grenada raster function chain in this template normalizes the input layers to a common range based on landslide susceptibility, (for example, areas with steeper slopes or closer to rivers are more susceptible to landslides). Once the layers are normalized to a common range, the function performs a weighted summation of the normalized values and classifies and symbolizes the summed values into five classes of landslide susceptibility.

    Note:

    ArcGIS Image for ArcGIS Online includes a raster function template editor that you can use to create your own templates or open existing templates like this one to see how they are built.

  5. In the Landslide Susceptibility Grenada raster function template, click the Select button.

    Select button in raster function template

    The Landslide Susceptibility Grenada raster function template appears in the tool pane.

    Landslide Susceptibility Grenada raster function tool

    The input imagery layers in the first parameter, Select the input data and set parameters, are already defined for this tool. Before you run the tool, you will adjust some other parameters.

  6. For Result layer name, type Landslide Susceptibility Grenada.
  7. For Save result as, verify that it is set to Tiled imagery layer.

    Result layer name

  8. At the top of the tool pane, next to Landslide Susceptibility Grenada, click the options button to open the Analysis Environments window.

    Open Analysis Environments window.

    The Analysis Environments window appears.

  9. In the Analysis Environments window, under Raster Storage, for Resampling method, confirm that Nearest neighbor (for discrete data) is selected.

    Nearest neighbor resampling method selected

    Resampling ensures that all pixels used in the analysis have the same size and are aligned while preserving the precision of the input datasets. Pixels in the source data are resampled before they are passed into the analysis. The nearest neighbor method performs a nearest neighbor assignment and is appropriate for analysis that includes discrete or categorical data. For analysis that includes only continuous data, such as elevation or precipitation, the Bilinear Interpolation and Cubic Convolution methods are more appropriate choices. However, these methods will result in some smoothing of the input values.

  10. In the Analysis Environments window, click Apply.
  11. On the raster function template tool pane, under all the parameters, click the Show Preview button.

    Show Preview button

    After a few seconds, a preview of the analysis results is displayed for the visible map extent.

    Preview of analysis results

    The preview displays more quickly if you zoom in to a small area first.

  12. Zoom in to a smaller area of the island.

    Zoomed-in analysis results preview

  13. Click Show Preview again to turn it off.
  14. On the raster function template tool, click the Run Analysis button.

    Run Analysis button

    After you run the analysis, the tool pane closes automatically, and the Details pane reopens. The output layer for the analysis appears in the map table of contents and is initially unavailable, with a wait indicator to show that the tool is still running. The analysis will complete in about 2 minutes if you run it for the full extent of the data.

Add building footprints and compare layers

You have run the raster function to create a landslide susceptibility imagery layer. Next, you will add the building footprints that were detected using deep learning to see which structures are in high-risk areas.

  1. On the ribbon, click Add and choose Search for Layers.

    Search for Layers option

    The default location should be set to My Content, which is where you stored the buildings layer.

    My Content

    You should see your Grenada_buildings layer in the list. If not, you can type its name in the Search for layers box and search for it.

  2. In the list of results, for the Grenada_buildings layer, click the Plus button to add it to the map.

    Add buildings layer to map

  3. Next to My Content, click the Back button to return the Content view.

    Back button

  4. In the Content view, turn off all layers except Grenada buildings and Landslide Susceptibility Grenada.

    Visible layers in Contents

  5. In the Contents pane, point to the Grenada buildings layer and click Show Legend.

    Show Legend button

    The legend for the buildings appears.

    Grenada buildings layer

  6. In the Contents pane, point to Grenada buildings, click the More Options button, and choose Zoom to.

    Zoom to Grenada buildings.

    The map zooms to the buildings that were detected using the deep learning model.

    Buildings and Landslide Susceptibility Grenada layer

  7. In the Contents pane, for the Landslide Susceptibility Grenada layer, show its legend.

    Landslide Susceptibility Grenada legend

  8. Pan and zoom through the map to identify relationships between the buildings and the five classes of landslide susceptibility.

    It appears that most buildings in Grenada are in lower-risk areas based on distance to rivers, elevation, land use, and soil types.

In this lesson, you created several layers that you used in a landslide susceptibility analysis for Grenada. You created tiled imagery layers in your ArcGIS Online organization using a simple guided workflow in the web browser. Once you created the tiled imagery layers, you accessed a trained deep learning model in ArcGIS Living Atlas that used artificial intelligence to automate the extraction of buildings from an imagery layer. You used an existing raster function template shared to ArcGIS Online to perform raster analysis to classify the landscape according to landslide susceptibility. All your analysis results are available for use in further mapping and analysis workflows by you or by other members in your organization. For example, view the analysis results in a 3D web scene. In the scene, you will find the same layers that you created and worked with in the lesson. The buildings display as 3D objects because the flat building footprints have a 20-meter extrusion applied, so they appear taller and as 3D shapes.