Scaling AI (GeoAI) for National Mapping at Country Level

Welcome to part 5 of my 7-part blog series which is continued from Part 4 The AI workplace and ArcGIS Deep Learning Workflow. When I started writing this section my goal was to keep the article short. Unfortunately, when I started putting the content together, I found the topic of scaling AI to national geographic regions to be very complex. So, to make sure this article was more valuable I made the executive decision to split Part 5 into part 5A and 5B. Part 5A will focus on Providing Artificial Intelligence GeoAI solutions at a National Scale with the Geospatial Data Scientist and GIS professional. Part 5B will focus on leveraging cloud capabilities like Kubernetes and ArcGIS Enterprise. To start Part 5A’s blog purpose is to help department and program managers with a GeoAI based approach to scaling current and future geospatial workflows with Artificial Intelligence (AI) and ArcGIS. National Mapping Managers daily roles are typically focused on running programs of record for national aviation charting and obstruction information, topographic geospatial vector and baseline imagery production, and maritime hydrographic navigation and bathymetric information management. Technologies such as ArcGIS are used to scale a nations national information production and analysis in the air, land, and sea domains.

Figure 1 Scaling National Mapping Data Collection and Production

National Mapping Department and Program managers have the demanding task of maintaining the currency of a nation’s foundation baseline imagery, elevation, and geospatial vector information over large geographic regions often with the following challenges of incorporating Artificial Intelligence into their workforce, workflows, and developing human and machine teams by:

• Providing adequate Physical compute infrastructure to collect, train, inference AI models
• Leveraging a Spatial Data Infrastructure (SDI) for Deep Learning Packages (DLPK) models and workflow development
• Training and education of leadership and staff on GeoAI
• Overcoming existing organizational silo’s by promoting information sharing between teams
• Reducing duplication by using/modifying prebuilt models and using many times to optimize the investment in Artificial Intelligence
• Setting expectations, GeoAI change is a process not a single implementation, “How do you eat a whale? One Bite at a time”
• Creating a GeoAI strategy for the organization at the executive, project, and team levels

   National Mapping Production Managers all have the responsibilities of running organizational operations with limited resources while learning new technologies like AI and 3D. These daily demands are not easy to manage and may often seem overwhelming. One question faced is, as a leader, how can I prepare my organization for the inevitability of Artificial Intelligence being part of my team’s operations?

         One area to start is the organizations current production systems such as ArcGIS Production Mapping and ArcGIS Defense Mapping where they will be used to accelerate data production using GeoAI with data from multiple remote sensing collection platforms. It is expected that in upcoming releases that prebuilt Artificial Intelligence workflows will be included. Some of the new features added would include enhanced Labelling, pretrained starter AI models, workflows for detecting objects, and advanced Quality Assurance (QA) capabilities where attributes will feed national mapping data models. Additionally, prebuilt National Mapping scripts can then run for inferencing providing information on confidence levels when processing the massive amounts of national data.

Figure 2 Accelerating National Mapping Production with GeoAI

With ArcGIS Production Mapping and ArcGIS Defense Mapping there would exist a seamless workflow for GIS professionals and Data Scientists. The process starts initial data collection and surveying of Imagery, Sonar, and LiDAR. The ArcGIS professional will focus on GeoAI labeling and data preparation. When starting the Training Loop (TL) the GeoAI data scientist is then ready to begin the training process where trained models are developed, confidence levels are examined, and models are examined for acceptable results. At this point the Data Scientist can determine if the AI model is good or more data is needed to retrain. The workflow is then handed over to the ArcGIS Professional to accelerate the production of information for the organization’s enterprise.

Figure 3 National Mapping Team Roles and Relationships

Realizing Value with ArcGIS Production Mapping and ArcGIS Defense Mapping

It’s important to realize that when extracting large amounts of data organizations will need to focus not just on automation but effectively implementing the human and machine team. The human machine team will be where real savings will be made in creating information that supports an authoritative decision-making process. To empower this human and machine team, Artificial Intelligence models need to be able to be made within the organization that are specific to an organizations mission. This is important because object identification and extraction for buildings, landcover, transportation networks, and shoreline often require specific imagery and are expensive when working with outside contractors. Many internal policies also must be considered because often public and hybrid clouds are not an option due to the sensitivities of the data. By ensuring the organization maintains expertise inside the organization significant savings can be realized by using their own production shops that are implementing GeoAI.

Providing National Scalability – Modernizing National Mapping capabilities 

We’ve mentioned earlier in this blog that to achieve real ROI on AI driven projects organizations need to move beyond desktop AI applications and onto AI applications that use the resources of a larger enterprise. On the below diagram on the left is standard enterprise architecture that can be stood up on premise or cloud. The standard enterprise whether it’s on premise, off premise, or a hybrid consists of the desktops, web applications, and APIS that utilize web services. This is illustrated inside the red box that cloud hosted services content and local data from users.

Figure 4 Providing National GeoAI Scalability - Enterprise Architecture

In the below diagram on the red box on the right is Esri’s image server which provides the ability to perform inferencing using various machine learning scripts. Using ArcGIS Image Server provides the ability to scale inferencing over large geographic regions

Image server tiles up regions into image chips or can automatically read an imagery service to run inferencing against. This inferencing then can be distributed out across imagery services.

Figure 5 Providing National GeoAI Scalability - Scaling Imagery & Inferencing

Esri Provides national scalability to any National Mapping Organizations modernization efforts. One advancement that can be implemented is the option to be native cloud ready for your organization. This includes scaling such things as Binary Large Objects (BLOB) which have several options such as Azure or S-3 using an Amazon Web Service (AWS) or a locally shared file. These storage types contain the raw input data that you GeoAI will use for inferencing or object detection.

Figure 6 Leveraging ArcGIS Notebook Server and Notebooks

Leveraging ArcGIS Notebooks 

ArcGIS notebooks provide a Jupyter experience with visualizations, maps, and data science tools. ArcGIS notebooks includes an editor interface where you can write and execute code and document workflows in one convenient place. The three main ways that Notebooks are used within ArcGIS are for web GIS administration, content management, and analysis and data science. There are several advantages to working with ArcGIS Notebooks within Geospatial organizations.  Firstly, they are directly integrated within the ArcGIS user interface, which allows users easy access an organization’s content and automates insertion of Python code snippets. A great advantage to using ArcGIS Notebooks is that Notebooks enables an organizations ability to share and integrate their internal knowledge and helps provide standardized environments that potentially could contain hundreds or thousands of open-source Python libraries.

ArcGIS Notebooks can run on a workstation using ArcGIS Pro, in the cloud using ArcGIS Online, or in the Enterprise using ArcGIS Notebook Server. With ArcGIS Notebook Server, users can deploy Notebooks within their own enterprise infrastructure. This allows them to easily connect to their organization’s maps and data, and gives them the ability to document, share, and automate workflows. Notebook Server also enables the construction of GeoAI and Machine learning models using Esri’s spatial tools and hundreds of open source Python libraries.

Scaling GeoAI at Kuwait Public Authority for Civil Information Authority (PACI)

Figure 7 Deep Learning Helps Kuwait Automate Mapping for PACI Kuwait

The Kuwait Public Authority for Civil Information (PACI) mission is to capture and maintain the countries geospatial information. Currently Kuwait’s development plans include increased spending on infrastructure for the county by 11 percent in order to modernize Kuwait into a world class financial and commercial center that is meant to foster competitiveness and increase productivity.

The challenge the country faced was to quickly capture and deliver content through their GIS App called the Kuwait finder which is used by over 650,000 citizens every day.

In order to streamline the capture of geospatial features PACI turned to ArcGIS and GeoAI. The GeoAI and Machine learning used satellite imagery, built models that helped capture and maintain their data bases through ArcGIS automation which reduced costs and saved time. “Before, this would have taken us 119 days, so we are comparing a week to 119 days,” said Maher Abdel Karim, GIS project manager at PACI. You can read the full article Deep Learning Helps Kuwait Automate Map Updates to Better Serve Citizens.

PACI did a great job modernizing workflows and delivery results. What made this possible was the use of GeoAI automation to become more responsive, faster delivery of products, and utilizing automation to scale their internal capacities, workforce, and realize the benefits of GeoAI across the government. “You can imagine the productivity increase by applying machine learning and deep learning models to automate the GIS workflow for base map updates” said Karim. PACI took a process which used to take 5 persons workload 74 days to complete to just 1.5 hours processing time with 2 days of QA/QC using ArcGIS’s GeoAI capabilities.

Realizing the Opportunity of GeoAI within National Mapping

National Mapping Department and Program managers are in an exciting transition time for the implementation and scaling of GeoAI workflows within their organization. By leveraging GeoAI Geospatial organization can quickly update the currency of a nation’s foundation baseline imagery, elevation, and geospatial vector information over large geographic regions and deliver the ability of providing decision making information faster and more accurately. GeoAI with ArcGIS Production Mapping and ArcGIS Defense Mapping are the future of accelerated information production for National Mapping. 

An important part of the evolution is moving beyond the desktop into enterprise environment that fosters the success of human and machine teams. These enterprise systems can leverage technology like ArcGIS Notebook Server to ensure security and proper access to shareable AI models across the organization. Also, organizations like PACI are now successfully implementing GeoAI to quickly capture and deliver content.

The next section 5B will be going into more details on ArcGIS Enterprise, Kubernetes, GeoAI, becoming Cloud Native, and National Spatial Data Infrastructures to support scaling GeoAI for National Mapping. If you’d like to learn more about this series, please feel free to read the previous 4 parts of this 7-part series.

part 1 of the series - Future Impacts on Mapping and Modernization by GeoAI

Part 2 of the series - GIS, Artificial Intelligence, and Automation in the Workplace

Part 3 of the series - Teaming with the Machine – AI in the Workplace

Part 4 of the series - The AI workplace and ArcGIS Deep Learning Workflow

Part 5A of the series - Scaling AI (GeoAI) for National Mapping at Country Level