Subtask2_12_ArtificialIntelligenceMachineLearningTools

Subtask 2.12: AI and ML Tools to
Support the Development of
Blueprint
North Carolina Flood Resiliency Blueprint
Prepared for the North Carolina Department of Environmental Quality by AECOM
October 2023

North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | i
Table of Contents
Definitions ...................................................................................................................... iii
Acronyms ........................................................................................................................ iv
1 Introduction ............................................................................................................... 1
1.1 Connections to Other NC Flood Resiliency Blueprint Tasks ...................................................... 1
1.2 Branches of AI Applicable to Blueprint ....................................................................................... 1
1.3 Data Governance ......................................................................................................................... 3
2 Data Enhancement & Cost Savings .............................................................................. 4
2.1 Standard 2D Model Layers .......................................................................................................... 4
2.1.1 Infiltration Layer .......................................................................................................... 4
2.1.2 Building Footprints ...................................................................................................... 5
2.1.3 Stormwater Networks ................................................................................................. 5
2.1.4 Automatic Mesh Creation Tool .................................................................................... 5
2.2 Probabilistic Modeling ................................................................................................................ 5
3 Improved Data Sharing & Interagency Collaboration ................................................... 7
4 Improved Engagement & Decision-Making Processes ................................................... 8
4.1 Engagement at Varied Scales of Complexity .............................................................................. 8
4.2 Engagement with Public Platforms ............................................................................................ 8
4.3 Funding Solutions ....................................................................................................................... 9
4.4 Reporting ................................................................................................................................... 10
5 Digital Twin .............................................................................................................. 11
5.1 Agile System of Continuous, Automatic Updates .................................................................... 11
5.2 Phasability, Synergy, and Packability ...................................................................................... 11
5.3 Mimic the Reality, Better Decisions .......................................................................................... 12
6 Conclusion ................................................................................................................ 13
7 Appendix .................................................................................................................. 14
7.1 Existing tools ............................................................................................................................. 14
7.1.1 Flood Recognition ..................................................................................................... 14
7.1.2 Land Use Recognition ................................................................................................ 14
7.1.3 2D Floodway .............................................................................................................. 15
7.1.4 Microsoft Planetary Computer .................................................................................. 15
7.1.5 Google Flood Hub ...................................................................................................... 16
North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | ii
7.1.6 Fathom Flood ............................................................................................................. 17
7.1.7 StreamFlow Prediction .............................................................................................. 18
7.1.8 FloodMapp ................................................................................................................. 19
7.1.9 Premonition ............................................................................................................... 20
7.1.10 IBM's Water Management Platform .......................................................................... 21
7.1.11 Aquantix ..................................................................................................................... 22
7.1.12 Flood.io ...................................................................................................................... 22
7.1.13 Digital Twin and AI ..................................................................................................... 23

Figures
Figure 1-1: Artificial Intelligence Technology Branches ............................................................................. 2
Figure 1-2: Data Strategy Operation Model ................................................................................................. 3
Figure 7-1: Generating Metadata via Local Satellite Imagery ................................................................... 14
Figure 7-2: Integration of Orthoimagery and LiDAR to generate Land Use Data ..................................... 15
Figure 7-3: Microsoft Planetary Computer's Water Occurrence Change Intensity Layer
(Goldsboro, NC) ...................................................................................................................... 16
Figure 7-4: Google Flood Hub Flood Alerts & River Forecasts (Kinston, NC) ............................................ 17
Figure 7-5: Fathom Flood's Relative Risk Product .................................................................................... 18
Figure 7-6: Prediction of Hurricane Ida at Little Tennessee at Needmore ............................................... 19
Figure 7-7: FloodMapp PostCast Flood Inundation Impact Product........................................................ 20
Figure 7-8: Microsoft's Premonition Platform ........................................................................................... 21
Figure 7-9: AECOM's Digital Twin's Australian Community Model ........................................................... 23

North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | iii
Definitions
A comprehensive list of definitions applicable to multiple Flood Resiliency Blueprint documents is
provided in a separate document.

North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | iv
Acronyms
AI – Artificial Intelligence
AEP – Annual Exceedance Probability
BCA – Benefit Cost Analysis
FEMA – Federal Emergency Management
Agency
GIS – Geographic Information System
GPS – Global Positioning System
HEC-RAS – Hydrologic Engineering Center's
River Analysis System
IoT – Internet of Things
IP – Internet Protocol
LiDAR – Light Detection and Ranging
LiMWA – Limit of Moderate Wave Action
ML – Machine Learning
NC – North Carolina
RARR – Risk Assessment and Risk Reduction
ROI – Return on Investment
SIFT – Sediment Fingerprinting Tool
USGS – United States Geological Survey
WSEL – Water Surface Elevation
2D – Two-Dimensional

North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 1
1 Introduction
Purpose: Subtask 2.12 – Identify Artificial Intelligence/Machine Learning (AI/ML) tools that can be used
or developed to improve the North Carolina (NC) Flood Resiliency Blueprint (Blueprint) (e.g., reduce
costs of updates, increase the accuracy of web-based decision tools, improve engagement and
decision making).
AI technologies have recently seen a surge in momentum and will become ubiquitous and integrated
into floodplain management. It is not a question of when this technology will become available
because much of it is already open-sourced with accompanied guidance. Leveraging existing open-
source AI can have several benefits for the Blueprint, including cost savings, faster production time,
flexibility and scalability of AI solutions that accommodates changing project conditions. A data
governance strategy, in conjunction with a data driven decision making process will be adopted in the
operation process to increase the transparency of decision makings and foster trust between users
and administrators. It is essential to identify and understand the best use case and applicability to the
functionality of the online data decision support tool as part of the development of the Blueprint.
The general benefits of AI include reduced information production time, rapid model adjustments,
enforced data governance plan, most advanced technology adoption and cost savings. General
limitations include the authority, accuracy, integrity, accessibility, and usability of data, technical
expertise to develop the AI tool, domain expertise to interpret the tool and public perception of AI.
1.1 Connections to Other NC Flood Resiliency Blueprint Tasks
There are various interconnections between this Identification of AI/ML Tools for Blueprint (Task 2.12)
and other tasks in the Blueprint, listed below.
• Task 3.11 - AI/ML Utilization Recommendations

1.2 Branches of AI Applicable to Blueprint
AI is a continuously developing field that incorporates mathematics, statistics, sciences, data
processing, and engineering to develop algorithms and apply rules that reach an objective. Two
primary schools of thought define AI as 1) anything that replaces human intelligence is AI and 2)
anything that can make predictions on unknown information based on existing and available data is
AI1. While what is technically defined as AI remains a highly debated topic, artificial intelligence
technology can essentially be broken into “branches,” separated by application, including machine
learning, computer vision, language processing, and robotics (Figure 1-1). It is important to note that
these branches are not strictly independent of one another and often rely on or include aspects of
each other. The following three branches of AI can be used or developed to improve Blueprint and are
defined as…

1 https://medium.com/@bharathkumar.h/artificial-intelligence-101-tech-terms-explained-for-non-tech-folks-
3190d7e711cd
North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 2

Figure 1-1: Artificial Intelligence Technology Branches
Machine Learning: “computational algorithms that are designed to emulate human intelligence by
learning from the surrounding environment. They are considered the working horse in the new era of
the so-called big data2.”
Computer Vision: “artificial intelligence (AI) that enables computers and systems to derive meaningful
information from digital images, videos, and other visual inputs — and take actions or make
recommendations based on that information. If AI enables computers to think, computer vision
enables them to see, observe and understand.3”
Language Processing: “combines computational linguistics—rule-based modeling of human
language—with statistical, machine learning, and deep learning models. Together, these technologies
enable computers to process human language in the form of text or voice data and to ‘understand’ its
full meaning, complete with the speaker or writer’s intent and sentiment.4”
In this report we apply these three AI branches to four specific Blueprint use cases including improved
data enhancement & cost savings, data sharing & interagency collaboration, improved engagement &
decision making, and digital twin.
A digital twin acts as a sandbox replica of real-world systems that allows users to develop, run, and
test virtually modeled scenarios. While a digital twin is not AI technology, AI can be integrated to
increase the predictive power of the generated models through accessing existing input data,
learning, and ultimately increasing the complexity of input data by identifying and filling gaps as well
as extending data’s spatial and/or temporal scope.

2 https://link.springer.com/chapter/10.1007/978-3-319-18305-3_1
3 https://www.ibm.com/topics/computer-vision
4 https://www.ibm.com/topics/natural-language-processing
North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 3
1.3 Data Governance
To ensure the performance of AI/ML, Data Governance is the first step to ensure the data foundation is
mature enough to feed into the AI/ML process. Data Governance can ensure accessibility, usability,
relevance, accuracy, timeliness, integration of our data to support our decision-making with a reliable
data pipeline.
Figure 1-2 shows the overall design of the data governance operation model. The data governance
plan can span over multiple years, considered a long-term goal. Along the long-term plan
implementation, with pain points identified, AI pilots can be implemented and applied as short-term
initiatives. AI solutions can also be applied to those areas where the data foundation achieves
maturity.

Figure 1-2: Data Strategy Operation Model

North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 4
2 Data Enhancement & Cost Savings
2.1 Standard 2D Model Layers
The 2D (two-dimensional) modeling industry is not standardized, allowing a range of techniques and
parameters to be acceptable. This leads to discrepancies, confusion, and distrust. Rework and
duplicate work occur due to incorrect or unreasonable ranges of freedom allowed on current models.
Developing statewide standards and tools will reduce the expense and redundancy of reworks.
AI technology can be used to standardize the production of infiltration layer, building footprints and
stormwater networks for the following reasons:
1. AI will have a scientific process that can be reproduced as opposed to tasking a team, which would
have differing engineering judgments and bias.
2. AI could continuously update these layers and verify the data.
3. AI can apply a localized and specific approach as opposed to a national approach that may relate
poorly to the area.
4. AI can help close the resource gap of underserved communities by increasing the accessibility and
usability of resources otherwise unavailable to them.

In summary, AI can identify patterns and trends to leverage large volumes of historical, real-time, and
predicted future data to ultimately support the creation of state-wide standardized datasets. AI
integration would create more accurate forecasts, optimize resource allocation, and reduce
operational costs.
Some statewide layers, such as terrain, structures, and survey, do not require AI. The statewide layers
that could be enhanced with AI are discussed in the following sections:
2.1.1 Infiltration Layer
The infiltration layer will contribute to the completeness of hydrology and hydraulics routing during
the modeling process. The advantage of using an infiltration layer is that each cell has its own
calculation instead of a single averaged calculation for the entire watershed. For example, with an
infiltration layer the rain that falls on a street mostly becomes runoff, while rain that falls on sand is
absorbed into the ground. Unfortunately, almost all the models for North Carolina currently average
the street and the sand to be considered the same. Even more challenging is that each model uses
different values for the same thing, often leading to results that do not match reality (e.g., sand
specified as street). A standard layer allows all users to use the same input parameters and closes
loopholes, allowing communities to use extreme parameters to achieve goals. This would provide
cost savings as each study would not need to put in the effort to create its own layer while also
correcting flood damage estimates from previous flawed studies.
AECOM has already created an AI–image processing tool that uses satellite imagery to determine land
use at a much greater accuracy. This tool could be further expanded with LiDAR (Light Detection and
Ranging) point cloud data to consider vegetation density within the land cover. It would use machine
learning to produce a scientific process to calculate important model parameters such as a curve
number, Manning’s n value roughness coefficient, etc. The tool could also be fully automated to
continuously scan new satellite imagery to detect recent changes and adjust with the seasons. ML
North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 5
could be further used to calibrate the standard infiltration layer based on data such as radar rainfall,
antecedent moisture conditions, gage flows and stages, etc. (see Appendix: Existing Tools - Land Use
Recognition).
2.1.2 Building Footprints
While multiple building footprint sources already exist, the challenge is keeping these sources
accurate and up to date with new construction and demolition activities. AI – image processing can
scan new satellite imagery and LiDAR to detect these changes and update the layer, as previously
discussed, with the infiltration layer. The reason to standardize this as a statewide layer is that the
models should account for these changes, but currently do not most of the time. Additionally, having
a standard of hydraulically accounting for these buildings would be beneficial.
2.1.3 Stormwater Networks
Large-scale modeling typically ignores that flooding, especially in urban areas, is often diverted and
conveyed through stormwater drainage networks. It is important to capture these systems as they
significantly affect model results as most mitigation projects are in areas with stormwater networks.
The initial task would be to compile a GIS (geographic information system) database of all municipal
and state operated stormwater systems as a unified, statewide layer. However, there are many private
systems and many municipalities that may not have readily available data. These missing segments
could be generated with AI tools. Culverts, outfalls, storm drains, and other structures could be
identified and estimated with aerial imagery through image processing. Many of these would be
simple culverts under driveways, roads, railroads, etc. These strategies would save costs as they
would reduce the need to manually modify and rework layers to account for missing segments.
2.1.4 Automatic Mesh Creation Tool
A large expense is required to refine a 2D model’s mesh, which few, if any, of North Carolina’s models
are at the needed stage of advancement. There are no practical layers to assist this process. For
example, the North Carolina Department of Transportation’s transportation lines layer often follows
aerial imagery and not the terrain. Models will need to be refined and would benefit from an AI tool
assisting with this task instead of manually creating them for the entire state. A refined mesh is
particularly important, especially in local applications such as a mitigation project’s return on
investment (ROI). Refined meshes include aligning cell faces with the direction of flow, aligning cell
faces to extraneous features in the terrain (often via breaklines), variation in cell size (often via
regions) and shape, and cell placement within the appropriate averaged calculation.
Image processing approach could be used to determine the breakline locations instead of drawing
them manually. It can save a lot of labor hours. Leveraging the breakline auto creation, 2D mesh can
be created or enhanced from there. The task of 2D modeling is costly and presents a significant hurdle
for future flood management. A high quality, refined mesh is one solution that would increase cost
savings, efficiency, and overall data accuracy.
2.2 Probabilistic Modeling
The existing standard practice of deterministic modeling only provides the “worst-case scenario”
simulation. The modeling approach is conservative. In contrast, the probabilistic approach uses a
Monte Carlo simulation, generating a range of simulated events. The probabilistic approach applies
North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 6
the proper weight to each simulated event and can return much more realistic results than the
deterministic approach. The probabilistic approach was applied to FEMA’s Flood Risk 2.0 program.
With completed simulation results, ML can learn the data trend and predict the simulation results
from targeted events, thus, this process can significantly reduce the operation cost of model
simulations.

North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 7
3 Improved Data Sharing & Interagency Collaboration
Flood resiliency management is a complex and multi-faceted operation that requires coordinated
efforts across various government agencies and related Non-Governmental Organizations (NGOs), to
prepare, prevent, and recover from flood events. The rapid advancement of the AI sector via a
comprehensive suite of machine learning applications often unsettles traditional project
management practices. There is an imperative need for regulators to efficiently streamline the data-
sharing process and stay updated to continuously develop resiliency strategies to be more
incrementally optimized in cost as the latest technology allows.
Advantages of committing to well-sought interagency collaboration for improved data sharing include
comprehensive data collection for more holistic risk assessments that go beyond flood management
for better prevention and recovery. Furthermore, there is the ability to strategize a timely, real-time
emergency response and long-term resiliency planning via effective communication/quality control of
resource allocations among different programs. By leveraging the power of language processing,
flood resiliency planning can integrate resources such as text analysis for large complex datasets,
ensuring scalability, which humans relatively lack the ability to adequately process in time. It can also
offer sentiment analysis (e.g., citizen science on Twitter) to tailor specific community needs locally as
a swift “bottom-up” response approach, as opposed to the “top-down” method of tediously running
through the entire chain of commands of distinct entities to enact.
To streamline the multi-agency decision-making process, language processing can be applied to
automate the above various assessments and ultimately provide an AI-powered chatbot to serve as
the definitive source of information. The chatbot could automatically facilitate valuable, unified
insights in a short-term and long-term timeframe with 24/7 support, which otherwise may have
required stringent permitting via multiple meetings, detailed manual analyses of numerous
standards/guidelines, and much larger funding requirements merely to come up with an agreement.
An available service like Kapa.ai already exists today, which the teams at OpenAI and others are
actively using to create a project tailored “ChatGPT” to help reduce the need for internal support or
save labor costs. The product not only reduces human reliance, but also identifies gaps in the existing
documentation and products by analyzing the generated answers. Similarly, the chatbot can
automate the data sharing and cleaning from all associated stakeholders’ inputs, which may be
extracted from multiple databases, in a consistent manner to improve the decision-making process
for better accountability and fair distribution of available resources.

North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 8
4 Improved Engagement & Decision-Making Processes
As complexity in floodplain management increases, so will the need for a simple product. Mutual
understanding needs to be established between the public, government, and practitioners. Programs
are trending towards an authoritative and “black box” approach. The “black box” approach works as
a device or system which will return final products to the end users without enough understanding of
the internal working logic. This is a concern for Floodplain management advises, which can be
extremely complicated, especially without transparent and clear communication of how and what
decisions been made. By setting the level of understanding and engagement at a public level, other
levels such as federal or private will also be able to understand and govern well. To reach this level of
understanding AI will play a key role in simplifying the communication and representation of
stakeholders by deploying a strategic data governance plan, which can be easily shared and
interpreted by decision-makers. AI also enables the Blueprint platform to offer 24/7 communication to
users. It can effectively convey flood information based on extensive data-driven methods, in more
straightforward or complex terms to be understandable by users with a wide range of expertise,
depending on the details of the requested prompt.
The Blueprint online decision support tool, supported by AI technology, can allow users to make more
informed and equitable floodplain management choices by providing clear visuals and reporting of
how North Carolina is managing flooding. A common challenge communities face is the ability to
make flood mitigation decisions that fully account for ongoing and/or existing projects in their
applicable hydrologic service areas. Progress of flood management can be accounted for with factors
such as, but not limited to, risks of current property (including recently developed), cost of mitigation
efforts minus the benefits (cost expenses avoided), upward mobility, etc. This would be a human-
intensive effort subject to bias and external influences. An AI system would be best to account for
these complex factors being data-driven and able to recognize patterns not seen by a human analyst.
AI would allow continual, transparent feedback to ensure responsible governance on how funds are
distributed and how well funds are used relative to project goals. An AI system could account for a
holistic overview and allow Blueprint to set goals and see progress in reaching those goals.
4.1 Engagement at Varied Scales of Complexity
A virtual assistant powered by AI can clarify the concept of a 1% AEP (annual exceedance probability)
storm event as the terminology can be confusing even for professionals in the field, tailored to the
specific user’s needs. For example, the virtual assistant might define the 1% AEP event by presenting
historical photos from a comparable flooding incident, using the user’s IP (internet protocol) address
for geospatial information. It could then, through a series of automated data networking and
calculations, inform the user that their area of interest faces a 30% chance of flooding over the course
of a 30 years.
4.2 Engagement with Public Platforms
Most residents of North Carolina have a camera phone or the ability to document and communicate
flooding and many freely share this information through social media. There are also many cameras
for security, monitoring, or other purposes that can capture flooding. In addition, there are historical
newspapers and other records documenting flooding and help with the human perception of
flooding. Overall, there is a vast wealth and dense network of existing and developing documented
North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 9
flood information that is rarely used that would further densify the network and may be a useful
source to document flooding.
People often use social media platforms during an event to post about their experiences and seek
help. Image and video processing and ML algorithms can be used to analyze these posts and identify
the areas where people most need help. For example, suppose people are posting about being
stranded on the roof of their house or about being trapped in a flooded area. In that case, AI
algorithms can identify these posts and alert emergency responders. This information can be used to
prioritize rescue efforts and allocate resources more effectively. This is wider than just an event. AI can
continually look at historical social media posts to mine data and improve floodplain management.
A common weakness in floodplain management is the lack of validating sources to establish ground
truth. The high costs of acquiring and operating many gauges and sensors prohibit the establishment
of a dense network. The typical solutions to high costs are to obtain more funding and/or find cheaper
alternative sensors (also often increasing maintenance costs). However, there may be an out-of-the-
box solution to the traditional sensors using photographs (in combination with other sources).
AI tools (such as AECOM’s tool) that translate the photos of flooding to locations and inundation
extents can be expanded to determine depth, time, and other variables. This AI tool would enable the
vast wealth of photos to become low-cost sensors with a dense urban network (See Appendix: Existing
Tools – Flood Recognition for more detail).
4.3 Funding Solutions
AI, specifically machine learning and language processing technology, can efficiently identify,
continuously learn about, and update project funding opportunities for flood resiliency projects. By
populating project or program profiles specific to North Carolina, these solutions can provide
matching and scoring for alignment of existing funding, perform a gap analysis on funding, and
identify where more funding is needed. Additionally, these solutions provide users with the expert
guidance and program insights required to navigate the rigorous requirements of the Infrastructure
Investment and Jobs Act and the federal commitments of the Justice40 Initiative.
Most projects need to be justified, often accomplished by seeing if the benefits outweigh the costs.
Having the ability to provide a rough estimate for a ROI analysis will allow projects to progress to
further stages. Otherwise, expenses must be paid upfront that may be wasted if the ROI is too low. AI
could produce ballpark ROI or identify areas that would have large benefits. For example, single-
access neighborhoods that become flooded or washed out, trap people inside and make it difficult for
outside help to access the area. Mitigation is one relatively low-cost solution where the benefits
typically outweigh the costs (for example, upsizing a culvert). Identifying areas in the model that
experience flooding, single access neighborhoods, and underserved communities could lead to
mitigation efforts in previously overlooked communities.
AI could then identify and prioritize potential mitigation sites, having calculated the cost to replace a
road and the associated loss of service (flooding often causes a loss of service to transportation,
utilities, employment, schooling, emergency services, etc.). The expense of the loss of services is
usually not considered for large-scale planning purposes. Even without mitigation, identification of
the risk alone would prepare damage cost at the time of a major event and assist stakeholders in
making decisions with an optimal ROI.
North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 10
4.4 Reporting
It is expected that various agencies and users would want reports covering project accounting
material. An applicable example of this strategy in action is the City of Charlotte’s Flood RARR (Risk
Assessment and Risk Reduction) tool, which calculates the cost saved from mitigation such as home
elevation when real storms occur. Charlotte’s RARR tool also displays a graph of flood risk decreasing
over the years. AI powered chatbot can be used to return answers to users’ questions and provide
automatic reports based on users’ needs. That information can be used for inter-agencies
communications and public education. Thus, this will improve the transparency of the decision-
making process.

North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 11
5 Digital Twin
A digital twin is the “holy grail” of floodplain management, a sandbox replica of the real world. All the
AI/ML tools discussed within this report, as well as the sources of Blueprint could power the digital
twin. A major advantage of the digital twin is that the range of what it can be widely varies, allowing it
to be phased and designed as needed. While the initial cost to develop a digital twin is high, the
benefits far outweigh the costs as a long-term investment.
5.1 Agile System of Continuous, Automatic Updates
The main advantage of a digital twin is that it is dynamic; automatically interpolating a myriad of
information. For example, a digital twin continuously scans satellite imagery to detect new land
development, new building footprints, forest fires, risk of debris jams, or other major changes to flood
risk. This allows for good housekeeping as data is quickly updated and can be linked to automatically
update flood risk (with the assistance of human supervision). The functionality of real-time flood risk
updates starkly contrasts the traditional study based on course, outdated national datasets and
engineering judgement. There are a plethora of examples of potential automatic updates and data
inputs. Suffice it to say that a digital twin is agile and able to continuously progress with future
demands.
5.2 Phasability, Synergy, and Packability
A digital twin can be created in phases allowing flexibility in costs and schedule. A digital twin can
start small and straightforward, including no more capability than what a normal study would entail.
As resources become available, more can be built upon the previous work. For example, a FEMA
(Federal Emergency Management Agency) study of a 2D (Two-Dimensional) model (Specifically for NC
the 2D Advisory Flood Hydrologic and Hydraulic Modeling) could be uploaded as the foundation of
digital twin. Data such as terrain, flood hazard data layers, property value data, property damage lost
data coupled with flood insurance quotes, flood control facility dataset can be all uploaded to build
the digital twin to show the flood risk related information to the agencies and public. When funds
become available the digital twin could be taught to establish machine learned parameters. These
learned parameters could then allow for quick flood risk assessments and bridge the gaps between
flood events. In a later phase, a city could create a digital database of their stormwater networks and
use the digital twin to create a 2D model of the networks, improving the planning ability of urban
areas (or as the technology improves such as the next version of HEC-RAS).
The ability to phase the digital twin enables these phases to be packaged and shared with others.
More importantly it allows each community to decide what is important for them. For example, a city
struggling to control development with discontent and frustrated residents. This city could invest in
the digital twin to display flood risk through an immersive virtual reality/augmented reality teaching
experience that would better communicate the communities flood risk. A further example, where a
city lacks flood depth gages and decides to instead invest in using historic and public photographs to
determine urban flood depths. Each of these developments could be packaged and applied to other
cities at a fraction of the cost because most of the work has been done by an interested party. In many
instances shared interest can lead to synergy in developing these packages. Third party packages also
can be incorporated from private companies or other states and nations. To ensure the data quality
North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 12
uploaded to the digital twin, a well-established data governance plan needs to be in place to ensure
the data quality, integrity and timeliness.
5.3 Mimic the Reality, Better Decisions
The concept of digital twin is to mimic the reality in a digital world. With good data governance plan in
place, using proper AI tools and sensors to collect real data, the digital twin can have a much better
representative of the reality. Many digital twin applications have simulation data as well; thus, the
simulated data and the monitored data can supplement each other, meanwhile cross validate each
other. The rich and near-real digital twin can be the trust data source for floodplain management
decision making process. With the transparency of the data displayed and accessible to multiple
agencies and the public, it can improve the communication on disaster mitigation and disaster
responses.
The availability of big data, advanced computing power, and modern technology has removed
barriers that previously limited AI/ML. Data processed in a digital twin can be looped back into the
digital twin to improve itself. For example, the digital twin could take rain depths and hurricane paths
and compare them to forecasts from the European and American Agencies and determine that the
European Agency provides better rain forecasts while the American Agency provides better path
forecasts. The next time a hurricane is forecasted, the digital twin would have learned how to better
interpret and apply this data. A digital twin can derive patterns from data to gain understanding
beyond human capacity. This will have an initial cost, but eventually it will operate automatically in
the background for minimal cost.
North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 13
6 Conclusion
The purpose of this document is to identify AI/ML tools that can be used or developed to improve the
Blueprint by reducing costs of updates, increasing the accuracy of web-based decision tools, and
improving engagement and decision making. Using the AI/ML tools identified in this report, Subtask
3.11 - Recommendations for the Utilization of Artificial Intelligence and Machine Learning to Inform
Blueprint – builds on this previous work to provide recommendations on the utilization of AI/ML to
inform the development and maintenance of the Blueprint.

North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 14
7 Appendix
7.1 Existing tools
The following tools are included as examples of how AI/ML can be used in flood resiliency related
online tools. Each tool touches on at least one concept presented in the main body of the report while
providing ground truthed methods that Blueprint can consider integrating during future phases of
work.
7.1.1 Flood Recognition
AI flood recognition can locate photos from metadata, text extraction, reverse image search, user
camera clustering, etc. (Figure 7-1). It can also approximate inundation extent by projecting images
(focal length, GPS [Global Positioning System]) location, altitude, roll, tile, direction, and SIFT
(Sediment Fingerprinting Tool)), then identify flooded areas (image thresholding and Kmeans
Clustering) and convert it to a polygon. This is a proprietary solution which uses open-source
packages.

Figure 7-1: Generating Metadata via Local Satellite Imagery
• Sources:
o https://www.thetimes.co.uk/article/ai-used-to-process-a-flood-of-information-lkdzf9xts
o https://www.nature.com/articles/s41612-023-00388-1#Sec12
7.1.2 Land Use Recognition
AI land use recognition can automatically develop roughness and imperviousness data used in flood
modelling by data-driven methods using imagery and LiDAR (Figure 7-2). Traditionally, practitioners
have relied on a national dataset or manually digitized land use features with low detail and abundant
inconsistency. AI is used to delineate features such as building extents, roads and vegetation with high
precision and minimal bias at a massive scale. In 2021, the Moreton Bay Regional Council in
partnership with AECOM developed an innovative approach that uses x-ray vision and point cloud
data science to calculate vegetation density below the canopy. This innovative approach was
successfully ground truthed and significantly improves the understanding of hydraulic roughness.
This is a proprietary solution which uses open-source packages.
North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 15

Figure 7-2: Integration of Orthoimagery and LiDAR to generate Land Use Data
• Source: https://www.surfacewater.biz/wp-
content/uploads/2021/11/20210427_FMA21_ArtofRoughness_FINALPaper.pdf
7.1.3 2D Floodway
ML techniques are applied to limit the number of iterations to determine the floodway (most
dangerous part of the river where development is very restricted) in 2D models. Being able to quickly
determine where the floodway will be in 2D models will be important to restrict development as time
and funding may take longer than the development.
AECOM developed a proof of concept in applying ML process to optimize the 2-Dimension Floodway
Development. The ML process leveraged depth and velocity raster data to predict a better start point
for engineers to begin floodway model iterations. The ML model was trained using a completed
floodway simulation dataset, cross-validated internally with an accuracy of 99%. Furthermore, by
applying the ML model to a different project, which is the external test, showed a successful reduction
of model iteration time by 90%. This is a proprietary solution which uses open-source packages.
7.1.4 Microsoft Planetary Computer
The Microsoft Planetary Computer is an initiative by Microsoft to create a cloud-based platform that
combines various forms of data from Earth's systems, such as the atmosphere, oceans, and land, to
provide researchers, scientists, and policymakers with the tools and data they need to better
understand and address global environmental challenges.
The platform aims to aggregate and analyze petabytes of data from sources such as satellites,
scientific instruments, and social media, using advanced machine learning algorithms and artificial
intelligence tools to generate insights and predictions about changes in the Earth's environment.
These insights can be used to inform decision-making around issues such as climate change, natural
resource management, disaster response, and urban planning.
North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 16
The Microsoft Planetary Computer is built on top of the Azure cloud platform and makes use of
Microsoft's extensive global network of datacenters to process and store vast amounts of data. It also
incorporates a range of open-source technologies, including Jupyter notebooks, Apache Spark, and
TensorFlow, making it easy for researchers to develop and share their own models and analyses.
Overall, the Microsoft Planetary Computer represents an important step forward in the use of
technology to help address some of the most pressing environmental challenges facing the planet
today. This is a proprietary solution which uses open-source packages.

Figure 7-3: Microsoft Planetary Computer's Water Occurrence
Change Intensity Layer (Goldsboro, NC)
• Source: https://planetarycomputer.microsoft.com/
7.1.5 Google Flood Hub
The Google Flood Hub is an online platform developed by Google that provides information and tools
related to flooding and flood management. The platform was launched in 2021 as part of Google's
broader efforts to support climate action and help communities prepare for and respond to the
impacts of climate change.
The Google Flood Hub offers a range of resources for individuals, communities, and organizations,
including:
• Flood forecasts and alerts: The platform includes real-time flood forecasts and alerts for more
than twenty countries, providing users with up-to-date information on potential flood risks and
impacts.
• Flood mapping: Users can access detailed flood maps that show areas of potential flooding, as
well as information on flood depths and flow velocities.
• Preparedness and response resources: The platform provides guidance and resources on how to
prepare for floods, including information on emergency kits, evacuation plans, and flood
insurance. It also offers resources for responding to floods, such as information on sandbagging
and flood cleanup.
North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 17
• Data and analytics tools: The platform include a range of data and analytics tools that can be used
to analyze flood risks and impacts, including satellite imagery, weather data, and flood modeling
tools.
This is proprietary solution which uses open-source packages.

Figure 7-4: Google Flood Hub Flood Alerts & River Forecasts (Kinston, NC)
• Source: https://sites.research.google/floods/l/0/0/3
Overall, the Google Flood Hub is designed to help individuals, communities, and organizations better
understand and manage the risks associated with flooding, and to support efforts to build more
resilient communities in the face of climate change.
7.1.6 Fathom Flood
Fathom Flood is a web-based platform developed by Fathom, a data visualization and analytics
company, which provides flood risk assessments for properties in the United Kingdom. The platform
was launched in 2018 and is designed to help property owners and insurance companies better
understand and manage flood risk.
Fathom Flood uses a combination of data sources, including satellite imagery, topographic data, and
flood models, to create detailed flood risk assessments for individual properties. The platform
considers a range of factors, such as the likelihood of flooding, the potential depth and extent of
flooding, and the impact of flooding on the property and surrounding area.
Users of the platform can enter an address or postcode to generate a flood risk report for a specific
property that includes information on flood risk levels, flood history, and potential impacts on the
property. The report also includes recommendations for flood prevention measures that can be taken
to reduce the risk of flooding and minimize potential damage.
Fathom Flood is designed to be user-friendly and accessible to a broad range of users, including
property owners, insurers, and local authorities. The platform is intended to help users make more
informed decisions about flood risk management and prevention, and to support efforts to build more
resilient communities in the face of climate change. This is a proprietary solution which uses open-
source packages.
North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 18

Figure 7-5: Fathom Flood's Relative Risk Product
• Source: https://www.fathom.global/
7.1.7 Streamflow Prediction
Hydroforecast is a company using science-guided AI solution to do streamflow prediction. It learns to
represent hydrologic processes by identifying relationships between satellite observations, basin
characteristics, meteorological forecasts, and streamflow measurements. The product can be used for
hydropower operations, energy trading, dam safety, emergency preparedness, water supply
management, restoration, and conservation. This is a proprietary solution which uses open-source
packages.
North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 19

Figure 7-6: Prediction of Hurricane Ida at Little Tennessee at Needmore
7.1.8 FloodMapp
FloodMapp is a software platform that uses real-time data and machine learning algorithms to
provide flood forecasts and alerts to help people and organizations better prepare for floods. The
platform, founded in 2017, is based in Australia.
FloodMapp uses a variety of data sources, such as rainfall data, river levels, and topographic data, to
generate flood forecasts and alerts. The platform uses machine learning algorithms to analyze this
data and provide highly accurate and up-to-date flood risk assessments for specific locations. This
allows people and organizations to take appropriate action to prepare for floods and minimize the
risk of damage and loss of life. This is a proprietary solution which uses open-source packages.
North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 20

Figure 7-7: FloodMapp PostCast Flood Inundation Impact Product
The FloodMapp platform includes a range of features, such as real-time flood warnings, flood depth
and velocity modeling, and impact assessment tools. The platform is designed to be user-friendly and
accessible to a broad range of users, including emergency services, local authorities, and businesses.
FloodMapp is designed to help individuals, communities, and organizations better understand and
manage the risks associated with flooding, and to support efforts to build more resilient communities
in the face of climate change. The platform is currently being used in Australia and is expanding to
other parts of the world.
• Source: https://www.floodmapp.com/
7.1.9 Premonition
Premonition is a company that uses data analytics and machine learning algorithms to help identify
potential disease outbreaks, including those caused by mosquito-borne viruses like Zika and Dengue
fever. The company was founded in 2013 and is based in Miami, Florida.
Premonition collects data from a variety of sources, including social media, news articles, and
government data, and uses machine learning algorithms to analyze the data for patterns and trends
that may indicate the emergence of a disease outbreak. The company's algorithms can detect
outbreaks earlier than traditional surveillance methods, allowing public health officials to respond
more quickly and effectively to prevent the spread of disease.
In addition to disease surveillance, Premonition also offers a range of other data analytics services,
such as predictive modeling and risk assessment, to help clients make more informed decisions and
reduce risk.
Premonition's technology has been used to identify potential disease outbreaks in several countries
around the world, including Brazil, India, and the United States. The company's work has been
North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 21
recognized by the World Health Organization and other international organizations as a valuable tool
in the fight against infectious diseases. This is a proprietary solution which uses open-source
packages.

Figure 7-8: Microsoft's Premonition Platform
• Source: https://www.microsoft.com/en-us/research/product/microsoft-premonition/
7.1.10 IBM's Water Management Platform
IBM's Water Management Platform is a cloud-based platform that uses Internet of Things (IoT)
technology, data analytics, and machine learning algorithms to help organizations manage water
resources more effectively. The platform was launched in 2018 and is designed to help organizations
better understand water usage, identify areas of inefficiency, and optimize water management
strategies.
The Water Management Platform collects data from a variety of sources, including sensors, weather
data, and satellite imagery, and uses machine learning algorithms to analyze the data for patterns and
trends that may indicate water management challenges. The platform can also help organizations
monitor water quality, detect leaks, and manage irrigation systems.
In addition to its core water management capabilities, the platform also includes features for
predictive maintenance, asset management, and environmental monitoring. The platform is designed
to be scalable and customizable, making it suitable for a wide range of organizations, from small
businesses to large municipalities.
IBM's Water Management Platform has been used in a variety of applications around the world,
including in agriculture, industry, and urban water management. The platform is intended to help
organizations reduce water waste, improve operational efficiency, and build more sustainable water
management systems. This is a proprietary solution which uses open-source packages.

• Sources:
North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 22
o https://www.ibm.com/blogs/think/2018/11/49707/
o https://portal.mwater.co/#/resource_center
7.1.11 Aquantix
Aquantix is a software and data analytics company that provides solutions for water utilities and
organizations that manage water resources. The company, founded in 2017, is based in Melbourne,
Australia.
Aquantix offers a range of software platforms and services to help water utilities improve their
operations and reduce costs. The company's platforms use advanced data analytics and machine
learning algorithms to optimize water usage, detect leaks, and monitor water quality in real-time.
Aquantix also provides consulting services to help utilities design and implement more efficient water
management strategies.
One of Aquantix's flagship products is WaterWorX, a cloud-based software platform that uses machine
learning algorithms to optimize water treatment processes and reduce energy consumption.
WaterWorX provides utilities with real-time data on water quality, flow rates, and chemical dosing,
allowing them to make more informed decisions about their operations.
Aquantix also offers a range of other software platforms and services, such as leak detection, asset
management, and billing management solutions. The company's solutions are designed to help water
utilities and organizations improve their efficiency, reduce costs, and ensure a reliable supply of safe
and clean water for their customers. This is a proprietary solution which uses open-source packages.
• Source: https://www.aquantix.ai/
7.1.12 Flood.io
Flood.io is a cloud-based load testing platform that helps developers and businesses assess their
websites, applications, and APIs (application programming interface) for performance under heavy
traffic and high load conditions. Flood.io was founded in 2014 and is based in Australia.
The platform allows users to simulate real-world traffic scenarios by generating massive user loads
from around the world. Users can create and run load tests with thousands or even millions of virtual
users, and measure how their systems perform under different conditions. The platform also provides
real-time monitoring and analysis of performance metrics, allowing users to quickly identify and
troubleshoot issues.
Flood.io's load testing platform is designed to be flexible and customizable, allowing users to
configure and fine-tune their load tests to match their specific needs. The platform supports a wide
range of protocols and technologies, including HTTP, WebSockets, and JMeter.
Flood.io's load testing platform has been used by businesses of all sizes, from startups to Fortune 500
companies. The platform is intended to help businesses ensure their websites, applications, and APIs
can manage high traffic loads and provide a smooth user experience, even under peak load
conditions. This is a proprietary solution which uses open-source packages.
• Source: https://www.flood.io/

North Carolina Flood Resiliency Blueprint
Subtask 2.12: AI and ML Tools to Support the Development of Blueprint | 23
7.1.13 Digital Twin and AI
A recent project led by AECOM engineers and planners developed digital twins with integrated AI
solutions to enhance stakeholder engagement efforts in an Australian community experiencing
chronic flooding. The project team developed the digital twin with hydrologic variables and AI
generated underlying data, including buildings and vegetation, to depict 2-4m2 (5-10km2), 3D models
of the community environment. During the engagement process, participants navigated a model of a
simulated flood path from a recent storm event where they validated the model’s accuracy and
identified aspects the digital twin may have missed like surcharging sewer maintenance holes. The
engagement effort further used the 3D modeled flood event simulation to walk through potential
flood risk management strategies such as stream restoration and open space
establishment/preservation with participants. This is a proprietary solution which uses open-source
packages.

Figure 7-9: AECOM's Digital Twin's Australian Community Model
• Source: https://digital.aecom.com/article/using-artificial-intelligence-and-digital-twins-to-
inform-our-most-vulnerable