Skip navigation
All Places > GIS > Applications > ArcGIS Pro > Blog
1 2 3 Previous Next

ArcGIS Pro

33 posts

Welcome to the third act of “This Week’s Picks” - ArcGIS Pro, Product Advocacy’s unofficial recurring GeoNet blog.  If you haven’t seen the first two posts, the aim of this series is to highlight some of my favorite ArcGIS Pro content.


This week we’re going to look at some quick mapping tips and tricks for anyone looking to make compelling maps in ArcGIS Pro.  For some, every day is Dia de Cartografo (cartography day) but these resources can be used in many mapping scenarios whether you are a daily mapmaker or find yourself creating maps only occasionally. Now on to the picks!

 This Week's Picks-ArcGIS Pro #3


One Minute Map Hacks Series

I love John Nelson’s “One Minute Map Hacks” series. These short (1 min or less) clips cover a lot of area, too numerous to discuss individually but suffice it to say there is something in the series for everyone designing maps whether you’re a map nerd or not! Want to tweak a projection? Check. Give your map a tattered paper effect or create a vignette? Check and check. Oh, derive hillshade that doesn’t wash out your map? He’s got you covered there too.


John has posted two map hack blogs with 5 hacks each. The first, covering 1-5 can be seen here while the second, covering 6-10 can be seen here. If you want to see the full playlist that can be found here. This contains additional map hacks not covered in the aforementioned blogs.


Symbolizing 2D data in 3D with preset layers

Feature attributes in your 2D data could be used to create 3D symbology, making for a more impactful presentation and overall map. Using preset layers to make energetic symbology is also another potential way to understand the data. This video from Esri Canada shows how this can be done with simple hurricane point data and preset layers to create a dramatic 3D cylinder effect to emphasize the hurricanes intensity. Check it out here.


If you want to experiment with this there are hurricane datasets available on ArcGIS Online that you can export for use with preset layers.  In about 5 minutes with a 2017 Atlantic hurricane season feature layer, I was able to use a preset to show the intensity difference between Harvey and Irma as they made landfall:

Hurricane gif


Additional resources:

Preset Layers


Mercator, it’s not hip to be square

When creating maps, the coordinate system plays a part in how the data will be interpreted. If you’re using Web Mercator for your thematic web maps could the results be misleading? Perhaps there is a better option for your web mapping needs? This blog  discusses the history of Mercator, what's best for, and proposes some alternative workflows for when you are dealing with smaller scales and thematic content where visual comparison is essential for interpreting the data. Topics also include vector and raster tiles and the when to choose an equal area projection.


Additional resources:

Projection Wizard


I hope you enjoyed this week’s picks, mapping tips and tricks (or hacks). Thanks for reading and stay tuned for the next round and if you are interested, you can also check out “This Weeks Picks” for ArcGIS Online and ArcGIS Enterprise!



Welcome to round two of “This Week’s Picks” for ArcGIS Pro, Product Advocacy’s unofficial recurring GeoNet blog, and Happy Halloween!


To recap, the aim of this series is to highlight noteworthy ArcGIS Pro content, and with the first installment I shared some of my recent favorites in the world of Data Management in ArcGIS Pro with an emphasis on editing.


This week’s topics largely fall into imagery capabilities in ArcGIS Pro 2.4.x with an emphasis on multidimensional datasets, deep learning, imagery and Remote Sensing and The ArcGIS Living Atlas of the World.


The planet, its systems and how they change over time will always demand attention. With scientific data and imagery support being foundational to a complete GIS, I wanted to showcase resources that tie-in to this theme. As such, the picks show how these capabilities can be used to map or visualize temporal and land cover changes, automate damage assessment after a natural disaster, or analyze and manage raster data (whether multidimensional or representing a singular phenomenon). Some of the tools require certain licensing or Image Analyst so that will be noted. With that, on to the picks!


Ever wanted to clip an image service from ArcGIS Living Atlas in ArcGIS Pro? How about taking a custom AOI from an image service or raster and then performing some analysis, such as calculating change over time?


This first blog shows just that! It highlights the changing landscape in the Las Vegas area and more importantly, the shrinking Lake Mead to the East that supplies its water. Taking it a step further, and to complete this story of loss, the author uses an artistic approach to symbolize this change. This can be adapted to your own use cases though it may require tweaking to the specifics of your area. Note: The NLCD layer requires an ArcGIS Online subscription.


Link: Mapping Loss: The Faces of Landscape Change in Las Vegas


Additional resources:

Frequency statistic Tool

Clip Raster Tool

Raster to Polygon

Use Living Atlas content in ArcGIS Pro


Have you ever asked: “what are multidimensional data and how can they be used in ArcGIS Pro?” This could go either way, in short you can work with data that has a multidimensional characteristic and 2.4 has some new functionality.


Whether you frequently work with scientific data or are simply curious about the multidimensional functionality available in ArcGIS Pro this resource has you covered! Note: Some of the tools require either ArcGIS Pro Standard or Advanced licensing and/or the ArcGIS Image Analyst or Spatial Analyst extension.


Link: Let’s do data science! Multidimensional analysis in ArcGIS Pro


Additional Resources:

Image Analyst Extension Introduction

An Overview of multidimensional raster data

Raster Functions

Raster functions general tab


There has been a lot of buzz around machine learning or “deep learning.” It is no surprise that these methods are seeing their day in the context of natural disasters and disaster response efforts. With ArcGIS Pro, you can use machine learning classification methods to automatically classify remote sensing imagery to identify areas of high risk and then use this data for recovery efforts.


This type of automation was shown during this year’s user conference plenary session by USAA in a collaborative effort with Esri. In the aftermath of the devastating Woolsey Fire in California last year, they demonstrated how deep learning could be utilized for disaster response. They were able to train a model to help classify buildings in the fire perimeter and perform automated damage assessment on affected structures. This helped them identify which of their members were impacted. Be sure to watch the video from the plenary below if you missed it. Note: Deep learning is available with an ArcGIS Image Analyst license.


Link: Damage assessment using deep learning in ArcGIS


Additional Resources:

VIDEO: Remote Sensing for Catastrophe Response

Deep Learning in ArcGIS Pro


I hope you enjoyed this week’s picks. Thanks for reading and stay tuned for the next round and if you are interested, you can also check out “This Weeks Picks” for ArcGIS Online and ArcGIS Enterprise!



The Product Advocacy team at Esri is excited to introduce an unofficial recurring GeoNet series: “This Week’s Picks.” As Product Advocacy Leads, we are always on the lookout for content that is compelling and insightful but also relevant. "This Week's Picks" aims to share some of our favorites with you.


For This Week's Picks – ArcGIS Pro, the theme is editing within the broader context of Data Management. We all know that data management is a key aspect of the ArcGIS Platform. Data prep, compilation and management are essential to get to the eventual end goal. That goal may also require creating, modifying, querying or deleting features. As such, I will be highlighting 3 blogs this week that touch on some of these concepts.


           1. For users newer to ArcGIS Pro editing, an excellent blog from this summer explaining the editing experience:

            Don't Fret It, Just Edit! (Demystifying how editing works in ArcGIS Pro)


            Why I like it: A common ArcGIS Pro editing support topic involves how it is “different” than ArcMap. For users             new to ArcGIS Pro editing, this blog breaks down ArcGIS Pro editing and explains in a clear way how it differs             from ArcMap. It also clarifies some of the misconceptions about editing in Pro.


Further readingEditing in ArcGIS Pro


2. The next blog is geared toward user’s who have an editing workflow involving ArcGIS Enterprise alongside ArcGIS Pro and want to increase editing and querying performance:

            Performance Enhancements: Using sourceSpatialReference with feature services in ArcGIS Pro


Why I like it: This blog discusses how to leverage sourceSpatialReference with feature services in ArcGIS Pro  (new at 2.4) in your editing and querying workflows. This enhancement allows server to do less work on the back end by avoiding projecting the data on the fly. This is particularly good for large datasets and complex  geometries. (NOTE: Requires ArcGIS Server 10.7.1 and later)


3. The last blog showcases some of the expanded offline editing capabilities in ArcGIS Pro, specifically with sync and mobile geodatabases:

           Go off the grid with ArcGIS Pro: Offline editing with sync enabled feature services


Why I like it: Details a lightweight workflow for extending feature service capabilities to use mobile geodatabases and sync capabilities so that you can bring data offline for editing and other analysis and still sync back with enterprise environments.


Supplemental Video: Offline Editing in ArcGIS Pro


Stay tuned for more picks on a variety of ArcGIS Pro related topics and if you are interested, you can also check out “This Weeks Picks” for ArcGIS Online and ArcGIS Enterprise!

If you use Concurrent Use licensing for ArcGIS Pro or ArcMap, you will need to upgrade your ArcGIS License Manager to 2019.0 before upgrading to ArcGIS Pro 2.4 or ArcMap 10.7.1.


If you are using ArcGIS Enterprise to manage your ArcGIS Pro Named User license, you will need to upgrade your ArcGIS License Manager to 2019.0 before upgrading to ArcGIS Pro 2.4.


ArcGIS License Manager 2019.0 is backwards compatible with earlier versions of ArcGIS Pro and ArcMap.


Concurrent Use licensing enables multiple users to share access to ArcGIS Desktop applications (ArcGIS Pro and ArcMap) on a network. ArcGIS License Manager, installed on the network, manages access and use of the Concurrent Use licenses for the users of the network.


Learn how to upgrade ArcGIS License Manager.

On January 14, 2020, Microsoft is ending support for their Windows 7, Windows Server 2008, and Windows Server 2008 R2 operating systems.


As are we.


After January 14, 2020, we will no longer support Windows 7, Windows Server 2008, and Windows Server 2008 R2 for Esri software. If you are still using Esri software on these operating systems, we highly recommend that you upgrade to Windows 10 or newer version of Windows Server, such as Windows Server 2016 or Windows Server 2019, before January 2020.


If you stay on these older operating systems after January 14, 2020, you can continue to use Esri software, but it will not be supported. Esri will not be able to address defects related to Microsoft operating systems no longer supported.


These changes are published in the Deprecated Features – Year-End 2018 document on the Esri Support site.

Take away message first: If you know anybody using ArcGIS Desktop 10.6.1, share this with them and make sure that they install the Buffering Degenerated Polygon patch.  Don't be fooled by the name.  Whether working with buffers or not, everybody running 10.6.1 needs to install this patch as the crash is often caused by editing.


Backstory: Back in November last year we made an Announcement out on the ArcGIS Blog, providing some information about why it is important to provide at least a valid email address when submitting error reports.  While we wanted to provide some education around how the error reporter works and what we do when we receive these reports, that announcement used the Buffering Degenerated Polygon patch as a specific example where we were seeing a high number of crashes reported with no email address, meaning no way for us to contact those users.


In January this year we posted another blog about improvements in the error reporter called ArcGIS Desktop Error Reporter Learns Its Manners that also explains why it is important to include a valid email address if you ever experience a crash.


We're still seeing a very high number of users crashing in ArcMap 10.6.1 and submitting the error report with no email address.  The crash that they are seeing is fixed in 10.7 and with a 10.6.1 patch if they are not able to upgrade.  But we can't tell them that, because we have no contact details.


Here is where I'm personally asking for your help.  If you have friends, friends of friends, or heck, even enemies, who are using ArcGIS Desktop 10.6.1, please let them know that they need to install the Buffering Degenerated Polygon patch.  


Please share far and wide.  Email this blog link, Tweet it, Facebook it, pull your phone out and show it to your co-workers.  I want to see the numbers from this crash go down. 


I appreciate your help!

Esri Educational Services has published an excellent 14 page document that can help you get started with ArcGIS Pro. It discusses essential tasks for migrating your organization migrate from ArcMap to ArcGIS Pro. Great job by Olivia Iannone on the ArcGIS Pro team!


High level tasks:

  1. Get to know ArcGIS Pro
  2. Set up ArcGIS Online
  3. Assign ArcGIS Pro
  4. Download and install
  5. Move content to ArcGIS Pro
  6. Explore with hands-on learning
  7. Learn more


Download the ArcGIS Pro Migration Guide:



Hope this helps,

We’ve had many users ask about plans to support the ArcMap geodatabase replication workflows in ArcGIS Pro and we want to clarify a few things regarding replication workflows moving forward.  The need to access authoritative GIS data from anywhere at any time is more important than ever. This is causing a shift in how we access and interact with data. Web GIS patterns provide the means to share, access, and work with data in a variety of ways extending the ArcGIS Platform.  Because of this shift our general direction has been moving from the client/server model (directly accessing the geodatabase via a database connection) to a web GIS services model. We believe that there are inherent advantages in a services architecture.


In ArcGIS Pro and ArcGIS Enterprise, we have already been actively developing functionality that supports the new feature service sync technology. With ArcGIS Pro 2.1 we introduced offline editing workflows to allow maps to be taken offline when disconnected from the network. This takes the feature service datasets offline to a local geodatabase. Users can perform edits locally, and then sync those edits with the server. The bi-directional sync process allows the offline geodatabase to share changes made and receive updates others have made to the web feature layer. See the ArcGIS Pro documentation for more information:


We have a mid-term project planned to further incorporate geodatabase replication workflows into ArcGIS Pro. We are still early in the planning phase of this project, but one aspect of the project involves leveraging the feature service sync technology. We want to leverage sync as it is available across the platform in both our online and enterprise products. The existing geodatabase replication tools will continue to work with data that is compatible with ArcMap and we encourage you to continue to use them until an ArcGIS Pro solution is available.


We are interested in hearing your workflows with distributed data. Please feel free to comment with your business requirements and how you are currently working with distributed data.

This blog post shows how to create a Map Tile Package from a Image Service and include it in a Mobile Map Package for offline use.  But what about the Esri World Imagery Basemap?  Well, ArcGIS Pro 2.2 and later provides a convenient way to clip out a portion of the Esri World Imagery basemap for Inclusion in a Mobile map.


Start ArcGIS Pro and Open a Project.

On the Insert Tab click the New Map button.

On the Map tab click the Basemap Gallery button and choose the Imagery basemap.


Zoom to the extent of the imagery that you want to include in your Mobile Map.


On the Map tab click the Download Map button.   Check include basemap & tile layers and click the Download button.


The Map Tile Package layer will be added to the map when the export completes. (remove the space in the layer name.. see notes below)


A few things of interest:

It is best to set the desired max scale

For the Extent Pictured the default scale was 1:142.  This created a Map Tile Package file (.tpk) that is 845 mb in size.

Choosing a scale of 1:565 creates a .tpk that is 84 mb in size.


So, consider the scale that your map users need if file size is a concern.


Some extra notes about using Map Tile packages in Mobile Map packages.

Do not use spaces in the name of the Tile Package layer in the Contents pane.  It is OK to rename the layer, but do not use spaces in the name. Tile Package are not written into the package correctly when spaces are used in the name, this issue will be addressed at a future release.


Make sure the coordinate system of the Map Tile Package (.tpk) matches the coordinate system of the Map. Custom applications built with the runtime and Esri Apps like Explorer for ArcGIS will not display Tile Package layers if the coordinate system is not the same as the map.


ArcGIS Pro allows you to apply symbology to Tile Package layers.  Symbology parameters are not supported by the Esri Runtime. You will just see the TPK as it was originally created in the runtime applications.  You can think of the symbology settings in ArcGIS Pro as over-rides for the symbology defined in the TPK.


If you create a Basemap that has a Map Tile Package layer and / or Vector Tile Package layers , do not include a Feature Class (point, poly or line features) in the basemap.  When a feature class layer is present in the basemap with a tile package layer the Mobile Map package is not written correctly, and some of the basemap layers will not display in runtime applications.  This issue will be addressed at a future release.



I just mentioned Vector Tile Package.  We are often asked how to get Esri Vector Tile Basemaps into a mobile map package, that capability will be possible in ArcGIS Pro 2.3 (it didn't make it in) its in ArcGIS Pro 2.4.  If you have a vector tile package you can add it to the map like any other layer and create a mobile map package. As described in  Use ArcGIS Pro to make an offline map - part 1 


Until next time.


... continued from Using ArcGIS Pro 2.2 Sharing and the Publisher Extension to make a public offline map with hillshading


Create a Map Tile Package

Click the burger menu on the Geoprocessing pane and Click Create Map Tile Package.

Fill in the required arguments for the tool.

Select Map for Input Map.

Uncheck the Package for ArcGIS Online | Bing Maps | Google Maps check box.

Enter an Output File name, GTNP_hillshade.tpk, for the Map Tile Package.

Select JPEG for the Tiling Format.

For the hillshade effect, 14 levels of detail will look pretty good, so type in ‘13’ for Level of Detail.

For Service, browse for and select the Tiling Scheme that was created in the previous step. GCS_NAD_1983.xml

On the Extent drop down choose Current Display Extent



Click Run to create the Map Tile Package.


Add the Tile Package to the Current Map. Click the Map tab, press add data button, and add the Map Tile Package (.tpk).


On the Catalog Pane – Project – Right click on Map and Convert it to a Basemap.

Remove the Terrain_Layer from the Basemap.



Copy the GrandTetonBM Basemap layer from the Operational map and Paste it in the Basemap.

Right click Map_BM and Paste the GrandTetonBM layer into the basemap.


On the Appearance tab adjust the Transparency for the GrandTetonBM layer to your preferred display.



Use the Basemap with the Operational map.   Click the Operational map tab to activate it. On the Map tab click the Basemap gallery and click the Map_BM basemap.


You have just added Hillshading to the Grand Teton National Park map.


Now, Let use ArcGIS Pro 2.2 to and the Publisher Extension to share the map with everyone and make the map useable by anyone who has Explorer for ArcGIS.


Check to see if you have the Publisher Extension. Click the Project tab.


Check the Licensing Status for the Publisher Extension.


If you are licensed for the Publisher Extension you can create maps that can be used by anyone.  If you are not licensed for the Publisher extension, users of the Mobile Map (.mmpk) that you create will need to be signed into ArcGIS Online or Enterprise organization to use it.


Share a Mobile Map - On the Share tab click the Mobile Map button to open the Package Mobile Map pane.


Fill out the inputs on the Package Mobile Map pane. If you have the Publisher Extension you can Enable the map for anonymous use.


Click the Package button to upload the Mobile Map to your Organization and share it with everyone.

Note: You can also use the Create Mobile Map Package geoprocessing tools to create a mobile map and share it.  You can use python to automate the mobile map creation process.

If the map was created for anonymous use any Explorer user can use Explorer without signing in.

And Search For “Grand Teton National Park with Hillshade” to download the map and use it.


Woot!  That’s it for now…


In this post we will continue forward from what was covered in the blog post Use ArcGIS Pro 2.1 to make an offline map.

In this exercise we will modify the package created in that post and re-share it.


So, let’s get to it.


1. Start ArcGIS Pro 2.2 and open a new blank project


In the Contents pane click All Portal and Search for “grand_teton_national_park owner:mark_nitro” and Right click on it to open it.

We will enhance the map by adding a hillshade layer to it. Specifically, we will create a tile package of hillshade imagery to add to our map. The Esri Living Atlas has an Imagery Service that we can use to do this.  ArcGIS Pro makes it easy to use Living Atlas data.  


Click the View tab and open the Catalog Pane.

Click Portal, the Living Atlas button, Search for ‘World Hillshade’ and Add ‘Terrain: Hillshade’ to a New Map.

Remove the “World Basemap” layers from the map.

On the View Tab click Link Views Center and Scale.

Click the Operational Map tab to activate the map and zoom to the extent of the basemap.

Click the Map map tab to activate it, and see that it is zoomed to the desired extent.

This is the extent of the data that we want to include in a Map Tile Package.  The tile package needs to be in the same coordinate system as our Operational map, which Is GCS North American 1983. Esri Apps like Explorer for ArcGIS and applications developed with the Esri runtime cannot project Map Tile Packages (.tpk) on the fly. If the tile packages coordinate system is different from the map’s coordinate system, it will not display.  Many of the Living Atlas services are in the WGS 84 Web Mercator Auxillary Sphere coordinate system, so creating a Map Tile Package (.tpk) from the map as it is now, will create a tile package in that Coordinate system .  For image services there is an easy way to get the data into the coordinate system we want, using the Make Image Service Layer geoprocessing tool.


Right Click on the Terrain:Hillshade layer in the table of contents and Click Properties. Click Source and copy the Location URL for the Terrain: Hillshade layer.


Click the Analysis tab, click Tools and search for Make Image Server Layer in the Geoprocessing pane.  Click Make Image Server Layer to open the tool.

For Input paste in the ImagerServer URL

For Processing Template choose Grayscale_Hillshade


On the Environments tab set the Output Coordinate system to GCS_North_America_1983

Click Run

The layer will be added to the map.

Remove the Terrain: Hillshade layer from the map.

Right Click the map in the Contents Pane, click Properties and Set the Coordinate System of the map to GCS North American 1983.

Since we have the Map Properties window open, click Metadata and enter a Description for the map. A map description is required when creating a Map Tile Package

We are going to use this layer to provide shading for the Grand Teton National Park map.  Let’s lighten up the shading a little bit.


Right Click on the Terrain_Layer and click Symbology to open the Symbology Pane.  On the Symbology Pane click the Color Scheme and Click ‘Format color scheme…’

Lighten up the color scheme by adjusting the color of the first stop.  Click the first stop and change its color to a lighter shade of gray and click OK

In the Geoprocessing Pane Search for Create Map Tile Package and Open the tool.

The Create Map Tile Package tool is set up to create content in WGS 1984 Web Mercator Auxillary Sphere coordinate system, this won’t work for our example map, so un-check the box.

Notice that the Create Map Tile Package tool requires an existing service or .xml tiling scheme file to create the tile package in a coordinate system that is not WGS84 Web Mercator Auxillary Sphere.

We will use the Generate Tile Cache Tiling Scheme geoprocessing tool to create a tiling scheme.


Click the burger menu on the geoprocessing pane and click Open Another Tool.


Search for Generate Tile Cache Tiling Scheme and open the tool.


Generate Tile Cache Tiling Scheme requires and input data source. Ideally, we could use the Terrain_Layer we created with the Create Image Server Layer tool, but this won’t work, if you use it, a tile scheme for WGS84 Web Mercator Auxillary Sphere will be created.  To create an input data layer that we can use, do the following:


Click the Terrain_Layer in the content pane, Click the Imagery tab and click the Process Button to Create a temporary raster clipped to the current display extent.

This will add a new layer, Clip_Terrain_layer, to the Content Pane.  We will use this layer as Input to the Generate Tile Cache Tiling Scheme tool.


Fill out the input for the Generate Tile Cache Tiling Scheme tool.

For input data source choose the “Clip_Terrain_Layer”

Enter a name “GCS_NAD_1983” and location for the output tiling scheme XML file.

Set number of Scales to 20, the scales will fill in automatically.

In Advanced Options, set the Tile Format to JPEG.


Click Run to create the Tiling Scheme. You can open the XML file in a webbrowser or text editor to confirm that the coordinate system is correct.



On the Content pane Right Click “Clip_Terrain_Layer” and remove it from the map.


Now we will create a Map tile package, the tile package will be created based on the current display of the map, that is why we adjusted the cartography to our liking, before creating the tile package.  


Continued - Using ArcGIS Pro 2.2 Sharing and the Publisher Extension to make a public offline map with hillshading - Part 2

ArcGIS Pro 2.2, Esri’s flagship 64-bit desktop GIS, has been released and is available. Now is the perfect time to migrate to ArcGIS Pro


ArcGIS Pro 2.2 is the largest update to ArcGIS Pro yet and brings a slew of new features and functionality. It adds and improves your highly requested workflows, features new innovations that take advantage of ArcGIS Pro’s unique 3D and 64-bit environment, and connects your desktop more tightly with the rest of the ArcGIS platform.


Slice Tool

Explore content hidden behind or within other content with the new interactive 3D exploration tool, Slice. You can slice through content in your scenes using planes or volumetric shapes. Slice is included among the other Interactive Analysis tools in the 3D Exploratory Analysis tools introduced in ArcGIS Pro 2.1.


Full Motion Video (FMV)


Play and analyze full-motion video (FMV) data that is geospatially enabled with the ArcGIS Image Analyst extension. Enable the projection and display of the video frame footprint and sensor position on the map while the video plays. You can also collect features in the video player and visualize them on the map, or collect features in the map and see them displayed in the video player.


New Styles

Inferno, Magma, Plasma, and Viridis scientific color schemes are now included in the ArcGIS Colors system style. These color schemes are particularly useful with imagery, LAS symbology, unclassed, and graduated colors symbology. They are also effective for grey scale environments and color-blind users.


Additional Innovations and Updates

ArcGIS Pro 2.2 is a big release. Here are some more new features:

  • Support for reading Autodesk® Revit™ files enabling access to architectural model data inside ArcGIS.
  • Stream layers: a new layer type that displays real-time streaming data.
  • Apply photographic textures when interactively editing 3D objects.
  • Pause drawing of a map or scene and still interact with it. While paused, you can navigate, add layers, or change the symbology; the state of the map will not refresh until paused drawing is turned off.
  • Clip a measured grid to only show coordinates within its UTM zone boundary. This is especially useful when mapping areas that cross UTM boundaries.
  • 50 new geoprocessing tools and batch geoprocessing to automate the running of a tool multiple times using many input datasets or different parameter settings.


Get the full details and watch video from the ArcGIS Pro developers on what’s new in ArcGIS Pro 2.2.

ArcGIS Image Analyst for ArcGIS Pro is now available for ArcGIS Personal Use and ArcGIS for Student Use


ArcGIS for Personal Use and ArcGIS for Student Use subscriptions have just gotten more powerful with the addition of ArcGIS Image Analyst at no additional cost. The Image Analyst extension provides powerful image visualization, interpretation, and classification tools to efficiently unlock the potential of every pixel.


ArcGIS Image Analyst extends ArcGIS Pro making it an image analysis workstation. Based on years of cross-domain experience in remote sensing and GIS, ArcGIS Image Analyst is designed for analysts, scientists, and photogrammetrists. It gives an advantage when working with image processing, interpretation, exploitation, analysis, and the creation of information products from remotely sensed data.


ArcGIS Image Analyst provides intuitive image visualization and tools for advanced interpretation of imagery and other raster data. There’s no need to switch between remote sensing and GIS software. Gain access to stereo and image space visualization, powerful image processing, advanced mensuration and 3D feature compilation tools, plus machine learning classification tools creating an environment to an interpret, analyze, and exploit imagery with simplicity and speed. 


Learn more about ArcGIS Image Analyst

Guest Post by Dmitry Kudinov, Esri


Calculating travel times is a foundational piece in transportation logistics, urban design, asset management, retail, etc. At Esri, we just completed a research project where we used artificial intelligence (AI) and machine learning to train an artificial neural network to predict travel times for transportation networks with a large number of complex, hard-to-model, and hidden variables. For this project, we partnered with NVIDIA, who provided us with GP100 and GV100 cards, which made this experiment feasible form the computation standpoint.


In this blog post, we will briefly discuss the details of this project, including the neural network architecture, training data format, efficient ways to evaluate training quality, and overall results which allow for a flexibility modelling intricate transportation aspects, and a significant throughput of the trained network.



Building a route from A to B these days is trivial: numerous services and applications can do this for you quickly and for free. But what if you need to build a route that’s a little more complex? One which starts at your home, then goes to 3 different friends in various parts of town, then to a local produce store where you need to pick up an order you placed yesterday? But wait, the grocery store expects you at about 5pm and gets closed at 5:30pm, and it is actually located near one of the friends along the way whom you initially planned to visit first in the morning… and there is also that pesky road traffic which always gets in the way and ruins the plans.


Things become suddenly quite trickier when you want to find the best visiting sequence which also has expected arrival times.


Challenge 1: Computational complexity

Logistics companies work with even more challenging requirements, scheduling not just multiple stops, but for multiple vehicles simultaneously. Large companies, while doing next day planning, schedule thousands of stops with hundreds of vehicles per day as a single optimization problem. Now you can start getting a sense of the computational complexity and resources involved in such operations.


Challenge 2: Model complexity

Another challenging area is hard-to-model aspects of transportation:

  • Changes in road speeds caused by seasonality and periodic weather patterns,
  • User preferred routes,
  • Individual driving habits and/or vehicle features affecting performance,
  • Individual commute preferences (especially important in urban areas and multimodal transportation, e.g. predict how long will it take a person to get to a chosen store to pick up her online-placed order), etc.


Although some of these aspects are even hard to formalize and even harder to represent with traditional algorithms, these are the integral properties of modern transportation and are already captured but buried deep inside individual GPS tracks.


The experiment

While the former large-scale logistics challenge asks for high throughput computations, the latter scenarios demand greater degree of flexibility without increasing complexity of the model.


Here at Esri, we decided to see if both requirements can be met with the help of machine learning. We used a simulated set of 300 million “journeys” (GPS tracks represented only by two locations - departure and destination, departure time, and how many minutes it took to travel, Figure 1) covering the region of California and Nevada roads to train an artificial neural network to predict travel times on the transportation graph.

Figure 1. "Journeys" used to train the neural network. X1, Y1 – coordinates of the departure location; X2, Y2 – destination location; START_TIME – departure time in UTC milliseconds since Jan 1st, 1970; NA_COST – time it took to travel in minutes.

Figure 1. "Journeys" used to train the neural network. X1, Y1 – coordinates of the departure location; X2, Y2 – destination location; START_TIME – departure time in UTC milliseconds since Jan 1st, 1970; NA_COST – time it took to travel in minutes.


Despite the simplicity of the input data, the neural network, after being trained, was able accurately predict travel times between any two locations in California and Nevada taking departure time into account, effectively embedding the road congestion factor into its function.


Once trained, the neural network can produce predictions with enormous throughput: a single desktop machine with a NVIDIA GV100 card can calculate over 300,000 ETAs per second, which is two-to-three orders of magnitude faster than common traditional deterministic algorithms. Of course, a prediction produced by a neural network is an approximation, but with a controllable accuracy - we will talk more about it below. For now, it is important to mention, that such a throughout may address the first challenge: logistics companies use various algorithms to solve multivehicle scheduling problems, and at the core of most of them lies the so-called Origin-Destination Cost matrix which needs to be calculated first, filled with ETAs for any possible combination of two stops, i.e. if we need to visit 1,000 stops, the OD Cost Matrix will have 1,000,000 ETAs. Our neural network can completely populate this matrix in only three seconds!


The second challenge, flexibility, has a promising future too: while being trained with just simple two-point GPS tracks, the neural network successfully figured out accurate representation of road congestion patterns, which makes it flexible enough for further finetuning with user preferred routes, or adapt to individual driving habits, or commuter preferences.


The details

For this experiment we partnered up with NVIDIA team, who provided us with multiple GP100 and GV100 cards. The strong GPUs gave us the ability to train neural networks of realistic size and the various experiment times were shortened by twenty to fifty-plus times (thanks to massive parallelization of matrix operations needed for training). This made the search for optimal neural network architecture and numerous hyperparameter values feasible and effective. A simple example: we spent about eight months of running one of the GP100 cards 24-7 in a search for an efficient architecture, spatial and statistical distributions of the training set, good values for multiple hyperparameters. The machine had 4 (8 hyperthreaded) Xeon(R) CPU E5-1620 v2 @ 3.70GHz CPU cores. After we compared single epoch training time between the GP100 and of the same machine CPU – the difference was over fifty times! This translates the above eight months of GPU time into over 30 years of CPU!


OK, let’s get back to the details. We used TensorFlow + Keras libraries to build a dense fully connected neural network (multilayer perceptron (MLP)) with sixteen hidden layers and ten million trainable parameters total. To reduce the overfitting, we added a Dropout node right before the output layer. The input was represented by normalized pairs of coordinates for departure and destination locations, and departure time; the output – single value showing the number of minutes it took to travel from A to B at given time.


We used Mean Squared Error (MSE) as the loss function, and Adamax optimizer with initial learning rate of 1e-3.


Training was performed for 4,000 epochs total on consecutive subsets of 20 million journeys, simulating “online” training. By the end of training, the MSE value on validation set was at about ~13.5.


But how good is MSE of ~13.5? Can the neural network be usable at this point? Well, MSE of 13.5 translates into 3.7 minutes of standard deviation of predicted values being off from the ground truth… but the routes in California-Nevada region may differ significantly in size: 3.7 minutes difference may be OK for an hour-long route, but for a route which is under 10 minutes - that’s a big difference. So, a chart showing how prediction accuracy varies depending on route length can tell a better story – Figure 2.


Figure 2. Variation of prediction accuracy as a function of route length.

Figure 2. Variation of prediction accuracy as a function of route length.


Another great tool for evaluating prediction accuracy which we built here in Esri, is a WMS REST service endpoint wrapping our trained neural network. The service returns a geographically bound PNG containing travel time surface, where every pixel is colored proportionally to the time it takes to reach it from the central pixel. Once constrained by a maximum travel time value, such surface looks like a isochron polygon. Figure 3 shows isochrones being built around San Francisco:

Figure 3. Isochron polygon constrained by 20-minute travel time.

Figure 3. Isochron polygon constrained by 20-minute travel time.


If such isochrones remind you of Network Analyst Service Area polygons, you are not mistaken: ultimately, in such form, both represent “reachability” zones and, if our neural network was trained well, these two should match closely. Figure 4 shows how neural network produced isochron (blue) matches Service Area polygon (red) built for the same departure location and time of day. Note how closely the boundaries of two polygons match in places where they both intersect with the streets. It is also important to note, that San Francisco area is particularly challenging due to an intricate coastal line and uneven distribution of transportation graph elements, and nevertheless, our neural network gets a good grasp on this complexity producing very compelling predictions.

Figure 4. Neural network isochrons (blue) and matching Service Area polygons (red).

Figure 4. Neural network isochrons (blue) and matching Service Area polygons (red).


So, what about the road congestions which we mentioned before? Here is the last animation for today, Figure 5, showing how a 25-minute isochron changes over a 24 hours period. You can see how the isochron shrinks during the business hours, and how it expands back during the night.

Figure 5. Road congestion patterns captured by the neural network during training. There corlor rings were added for visualization purposes and are similar to isolines of a continuous 3d surface, where the 3rd dimension is time. That animation has 12 rings colored (or left transparent) for ranges of travel times falling into 125 second buckets – 12 total, summing up to 25 minutes.

Figure 5. Road congestion patterns captured by the neural network during training. There corlor rings were added for visualization purposes and are similar to isolines of a continuous 3d surface, where the 3rd dimension is time. That animation has 12 rings colored (or left transparent) for ranges of travel times falling into 125 second buckets – 12 total, summing up to 25 minutes.


The road ahead

Although we have achieved here some impressive results, there is room for improvement. One particular path which we want to explore further down the road, is to check the applicability of one-dimensional convolutional network instead of MLP. The reason for this is simple: there is a strong correlation between the coordinates, and multiple repeating patterns in the training data – this makes our scenario a good candidate for a convolutional architecture which will scale better for larger geographical areas.


Another area of improvement can be illustrated with the Figure 2 above: we want smaller standard deviation values for shorter routes, and this can be achieved by more accurate selection of the training data, giving shorter routes a bigger share in the training set.


And, of course, the final step – using the trained neural network in transportation analysis and planning.


We will keep you updated on the progress.

FYI, this is a great 3 min video that explains how ArcGIS Pro licenses are assigned in ArcGIS Online and ArcGIS Enterprise, by the organization's administrator. A nice overview.


Named User Licensing In ArcGIS Pro - YouTube