4th NOAA Workshop on Leveraging AI in Environmental Sciences

All information about workshop logistics can be found in the handbook for participants.

Slack is the official space for discussion during the workshop. To facilitate continuous discussion during and after the workshop, Zoom chat will be disabled during the session and all participants should join the workshop Slack workspace. You can continue the conversation during the session in Slack. When replying to other participants, please use “Reply in thread” to help track the conversation.

• If you are within NOAA, you can directly sign up using your noaa.gov email address.
• If you are outside of NOAA, please use this invitation link to join the Slack workspace. [Link updated on September 6, 2022]
• If you joined the workshop space in 2021, you don’t need to rejoin since we are using the same workspace.
• Slack only allows 100 people to use the invitation link. If you are having trouble joining the space please email ai.workshop@noaa.gov.

During this opening session, plenary speakers will discuss the big picture in which emerging technologies including Digital Twins for the Earth System align with select institutions’ visions. This session touches on all three focused themes of the workshop.

Session agenda:
09:30: Welcome! - Yuhan “Douglas” Rao, Chris Slocum, Heather McCullough, Rob Redmon
09:35: Dr. Sarah Kapnick (NOAA, Chief Scientist)
09:50: Prof. Stephen Belcher (UK Met Office, Chief Scientist)
10:05: Dr. Nadine Alameh (Open Geospatial Consortium, CEO)
10:20: Panel Q&A with Speakers and Additional Panelists - moderated by Eric Kihn
Dr. Kapnick, Dr. Alameh
Dr. Jacqueline Le Moigne (NASA, PM Advanced Information Systems Technology)
Dr. Peter Dueben (ECMWF, Head of Earth System Modeling)
11:00: Session Close

Notes:
Times above are in Mountain Daylight Time (UTC - 6). We will use Slido during the Q&A. We'll use Slack to build community throughout the workshop.

Slido for Q&A: sli.do/2022AI/

In this session, speakers representing different institutions present their short and long term plans in DT for the Earth System (DT4ES). What are the ultimate goals?

Session agenda:
11:30: Welcome / Housekeeping
11:35: Sid Boukabara (NOAA): Global Observation and Prediction of the Earth System Environment Using AI-Powered Digital Twin Technology: NOAA’s Activities
11:47: Thomas Geenen (ECMWF): Destination Earth
11:59: David M. Hall (NVIDIA): NVIDIA - Interoperable Digital Twins of the Earth
12:11: Kirstine Dale (UKMO): Environmental Digital Twins – vision and plans at the UK Met Office
12:23: Jacqueline Le Moigne (NASA): NASA Advanced Information Systems Technology (AIST) - Building Earth System Digital Twins
12:35: Q&A
13:00: Session Close

Notes:
Times above are in Mountain Daylight Time (UTC - 6). Lightning talks are roughly 10-12 minutes each followed by a Q&A with the panel. We will use Slido during the Q&A. We'll use Slack to build community throughout the workshop.

Slido for Q&A: sli.do/DIGITALTWIN/

Google Documents

About Fire Weather and Impact Theme: This year's AI Workshop is piloting a collaboration development activitu modeled after the hackweek which brings participants with different backgrounds together to form collaboration and develop various use cases that are related to research and applications for fire weather and impacts. The core objective of FireWxHack are two-fold:

1. Foster collaboration building in applying AI/ML for fire weather research: Provide opportunities for fire weather researchers and data scientists to develop project ideas on how to use AI/ML and open environmental data for fire weather research & application development.
2. Collect user feedback on open environmental data for AI/ML in fire weather: Collect user feedback on how to improve open environmental data that can accelerate the adoption of AI/ML in fire weather research.
This event is NOT a data science competition and there is no fixed target for each individual team.

Session information:
This session will use interactive collaboration tools, breakout rooms, and facilitated discussion to help fire weather theme participants to form teams that are interested in similar use cases to develop use case ideas and outline the project development plans.

Relevant materials:
1. About FireWxHack: Document summarizing the goal of fire weather theme activities.
2. Introduction slides: Slides used during meet and greet to describe the timeline of fire weather theme.
3. Suggested use cases: These are three suggested use cases with relevant environmental data for fire weather research.
4. JupyterHub Environment: https://oasis.ncics.org (require GitHub user name to sign in).

During this session, we will hear from a selction of lightning talks from different organizations to understand the current landscape of challenges and solutions of using AI in ocean conservation and sciences.

Session agenda:

11:30 - 11:35: Welcome (Ben Richards, NOAA Fisheries/PIFSC)
11:35 - 11:45: TagLab, an AI-assisted labelling tool for reefscape analysis (Gaia Pavoni, Italian National Research Council)
11:45 - 11:55: Evaluating AI Solutions for Reducing Benthic Habitat Annotation Bottlenecks (Bryan Costa, NOAA/NOS/NCCOS)
11:55 - 12:05: Tator, an open source, web-based collaborative analysis platform for video and imagery (Benjamin Woodward, CVision AI)
12:05 - 12:15: Mapping Seafloor Habitats with AI (Bill Iannacci, Jeff Clark, JAXON AI)
12:15 - 12:25: Flukebook as a collaboration tool (Tanya Stere, Wild Me)
12:25 - 12:35: Optimize and efficient monitoring operation using repeatable computer vision solution (Monica McEwe, Rafi Lav, Deloitte)
12:35 - 12:45: Machine Learning for real-time maritime transparency at Skylight (Patrick Beukema, Mike Gartner, AI2)
12:45 - 12:55: AI applications in satellite oceanography (Guangming Zheng, NOAA/NESDIS/STAR)

Notes:
Times above are in Mountain Daylight Time (UTC - 6). Lightning talks are roughly 10-12 minutes each followed by a Q&A with the panel. We will use Slido during the Q&A. We'll use Slack to build community throughout the workshop.

Slido for Q&A: sli.do/OCEAN/

Form a community vision for an interoperable DT4ES ecosystem by discussing how the previous session's and other varying visions match, overlap, are compatible or not. Discuss how these visions could evolve to ensure compatibility, interoperability, standard interfaces, etc.

Session agenda:
13:30: Guidance
13:40: Breakout Discussion (10-15 persons/room, randomly assigned)
14:05: Main room discussion at halfway point (Slido)
14:15: Breakout Discussion
14:45: Main room discussion
15:00: Session Close

Notes:
Times above are in Mountain Daylight Time (UTC - 6). Lightning talks are roughly 10-12 minutes each followed by a Q&A with the panel. We will use Slido during the Q&A. We'll use Slack to build community throughout the workshop.

Slido for Q&A: sli.do/DIGITALTWIN/

Google Documents for Breakout Rooms

About Fire Weather and Impact Theme: This year's AI Workshop is piloting a collaboration development activitu modeled after the hackweek which brings participants with different backgrounds together to form collaboration and develop various use cases that are related to research and applications for fire weather and impacts. The core objective of FireWxHack are two-fold:

1. Foster collaboration building in applying AI/ML for fire weather research: Provide opportunities for fire weather researchers and data scientists to develop project ideas on how to use AI/ML and open environmental data for fire weather research & application development.
2. Collect user feedback on open environmental data for AI/ML in fire weather: Collect user feedback on how to improve open environmental data that can accelerate the adoption of AI/ML in fire weather research.
This event is NOT a data science competition and there is no fixed target for each individual team.

Session information:
During this session, each use case team will share the initial project ideas and receive feedback from other workshop participants during the first 45 minutes. Participants will have the opportunity to switch use cases that are better fit for their personal interest. Use case teams will reconvene in the breakout room to continute the discussion.

Relevant materials:
1. About FireWxHack: Document summarizing the goal of fire weather theme activities.
2. Introduction slides: Slides used during meet and greet to describe the timeline of fire weather theme.
3. Suggested use cases: These are three suggested use cases with relevant environmental data for fire weather research.
4. JupyterHub Environment: https://oasis.ncics.org (require GitHub user name to sign in).

During this session, we invite participants to join the breakout rooms that are focusing on different topics to identify key challenges and needs to accelerate the adoption of AI for ocean sciences and applications.

Session agenda:

13:30 - 13:35: Welcome and logistics
13:35 - 14:05: Breakout room discussion
14:05 - 14:20: Main room sharing and report out
14:20 - 14:50: Breakout room discussion
14:50 - 15:00: Look ahead and wrap up

Breakout rooms are separated by topics:
1. Passive Acoustics & Telemetry (Facilitator: Bryan Costa)
2. Underwater Imagery (Facilitator: Hanu Singh)
3. Ocean Data & Physical Modeling (Facilitator: Megan Cromwell)
4. Airborne/Space Remote Sensing (Facilitator: Brendan Turley)
5. Environmental, Species & Stock Assessment Modeling (Facilitator: Ben Richards)

Participants can self select which topic room to join.

Notes:
Times above are in Mountain Daylight Time (UTC - 6). Lightning talks are roughly 10-12 minutes each followed by a Q&A with the panel. We will use Slido during the Q&A. We'll use Slack to build community throughout the workshop.

Slido for Q&A: sli.do/OCEAN/

Please join the workshop participants at the Wonder space to share your experiences during the day and discuss any additional ideas/topics that you want to share.

This plenary session provides a high-level overview of the benefits, current status, and needs of AI for ocean sciences including perspectives from NOAA, academia, and industries.

Session agenda:

09:30 - 09:35: Welcome (Douglas Rao, Benjamin Richards)
09:35 - 09:45: Mark Osler (Senior Advisor for Coastal Inundation and Resilience at NOAA's National Ocean Service)
09:45 - 09:55: Evan Howell (Director of the Office of Science and Technology at NOAA Fisheries)
10:00 - 10:10: Margrit Betke (Professor, Boston University)
10:15 - 10:25: Vinay Viswambharan (Principal Product Manager at Esri)
10:30 - 10:40: Anthony Hoogs (Vice President of AI at Kitware)
10:40 - 11:00: Plenary Panel Discussion

Notes:
Times above are in Mountain Daylight Time (UTC - 6). Lightning talks are roughly 10-12 minutes each followed by a Q&A with the panel. We will use Slido during the Q&A. We'll use Slack to build community throughout the workshop.

Slido for Q&A: sli.do/2022AI/

About Fire Weather and Impact Theme: This year's AI Workshop is piloting a collaboration development activitu modeled after the hackweek which brings participants with different backgrounds together to form collaboration and develop various use cases that are related to research and applications for fire weather and impacts. The core objective of FireWxHack are two-fold:

1. Foster collaboration building in applying AI/ML for fire weather research: Provide opportunities for fire weather researchers and data scientists to develop project ideas on how to use AI/ML and open environmental data for fire weather research & application development.
2. Collect user feedback on open environmental data for AI/ML in fire weather: Collect user feedback on how to improve open environmental data that can accelerate the adoption of AI/ML in fire weather research.
This event is NOT a data science competition and there is no fixed target for each individual team.

Session information:
Lightning talks:
11:35: Christina Kumler (NOAA/GSL/CIRES) – Using Machine Learning as a Tool for Improving Wildfire Intensity Forecasting
11:50: Kyle Hillburn (CSU/CIRA) – The Value of Low Latency Satellite Data for Initializing Coupled Fire Models
12:05: Michael Seablim (NASA) – NASA Wildland Fire Program and FireSense
12:20: Use case daily check in and feedback

Each use case team have five minutes to present use case ideas and get feedback from lightning talk presenters and other workshop participants.

Relevant materials:
1. About FireWxHack: Document summarizing the goal of fire weather theme activities.
2. Introduction slides: Slides used during meet and greet to describe the timeline of fire weather theme.
3. Suggested use cases: These are three suggested use cases with relevant environmental data for fire weather research.
4. JupyterHub Environment: https://oasis.ncics.org (require GitHub user name to sign in).

This session gets to the practical level to talk about how DT4ESs and key components from major institutions and community collaboration could dovetail with each other, leverage each other, be compatible with each other, etc.

Session agenda:
11:30: Welcome / Housekeeping
11:35: Sue Haupt (NCAR): Using AI for Physics Parameterizations and Downscaling to High Resolution
11:50: Karthik Kashinath (NVIDIA): Key components of NVIDIA's Earth-2 Initiative: Progress and Challenges towards building Digital Twin Earths
12:05: Ann Bostrom (University of Washington, AI2ES): Risk communication design principles for responsible and trustworthy Digital Twin Earth systems
12:20: Tiago Quintino (ECMWF): Interoperability Destination Earth
12:35: Q&A
13:00: Session Close

Notes:
Times above are in Mountain Daylight Time (UTC - 6). Lightning talks are roughly 10-12 minutes each followed by a Q&A with the panel. We will use Slido during the Q&A. We'll use Slack to build community throughout the workshop.

Slido for Q&A: sli.do/DIGITALTWIN/

During this session, we will feature a selection of lightning talks to discuss the data needs and chalenges for applying AI in ocean science and conservation including ongoing effort from the community to advance AI-ready data. These lightning talks will set the stage for the follow-up breakout room discussion.

Session agenda:

11:30 - 11:35: Welcome (Stacie Robinson, Erin Moreland)
11:35 - 11:45: An ML-ish perspective on "getting to know your AI for conservation project" (Dan Morris, Google)
11:40 - 11:50: TBA (Eric Egland, Microsoft)
11:50 - 12:00: Detecting anomalous ocean observations from the ocean’s surface to the seafloor through machine learning (Paige Lavin, UMD/CISESS)
12:00 - 12:10: Underwater column sonar data (Carrie Wall, NOAA/NESDIS/NCEI/CIRES)
12:10 - 12:20: Scientific data integration from uncrewed systems ((Josh) Sage Osborne, UW/CICOES)
12:20 - 12:30: AI and Cloud Developments to Meet the Mission Needs of the NOAA Office of Ocean Exploration (Mashkoor Malik, NOAA/Ocean Exploration)
12:30 - 12:40: Seize the data! Effectively leveraging data annotations to build an accurate machine learning model (Ann Allen, NOAA/PIFSC)
12:40 - 12:50: Working toward a standard for AI-ready data (Stacie Robinson, NOAA/PIFSC)
12:50 - 13:00: Overflow Q&A

Notes:
Times above are in Mountain Daylight Time (UTC - 6). Lightning talks are roughly 10-12 minutes each followed by a Q&A with the panel. We will use Slido during the Q&A. We'll use Slack to build community throughout the workshop.

Slido for Q&A: sli.do/OCEAN/

Join the session in Wonder space here (access code is NOAAAI2022).

To best facilitate the networking and community building during the virtual workshop, we invite organizations, groups, and individuals to propose topics for a 60-minutes virtual exhibition and networking.

- For organizations, you can propose demonstrations of existing products, data, and services that people might be interested in learning about.
- For individuals, you can propose research, application, or discussion topics that you want to connect with other workshop participants.

The virtual booth will be in different virtual circles in the workshop Wonder space.

To propose a topic, fill out this short google form.

Current proposed topics:
1. Ryan Berkheimer (NOAA/NCEI) – A Bidirectional Information Exchange in Support of Standardizing the Digital Twin Earth Framework Specification (DTE-FS)
"The Digital Twin Earth Framework Specification (DTE-FS) is a first-of-its-kind specification that derives a set of axioms and describes formal axiom-based requirements for construction, use, and management of, and participation in, a fully democratic digital twin earth that is grounded by the Group on Earth Observations (GEO) defined principles of syntactic, schematic, semantic, and legal interoperability, and that also fully satisfies the functional requirements of a Digital Twin Earth.
The DTE-FS provides a multi-viewpoint, Web 3.0 (semantic web) treatment of the concept of Digital Twins, and introduces many attractive and novel concepts in support of that treatment. These concepts include a distributed and open-architecture process framework that is fully API exposed for layered construction and user-specific implementation; a 'fuzzy semantics' based knowledge graph for evolutionary governance; fully-declarative two-layer data models for long-term sustainability; and process 'self-ownership' for ensuring security and consent-driven control.
The DTE-FS was organized at NOAA, which maintains organizing authorship, but has been widely circulated for academic and industry review at venues such as the first ever Earth Science Information Partners (ESIP) session on digital twin earth and the recent NSF/Harvard/GMU/UW/NASA/OGC led Spatiotemporal Computing (STC) I/UCRC phase 3 steering workshop that sets the long-term academic priority for national STC research. This virtual booth will provide additional opportunities for discussion and debate around the DTE-FS, as well as an opportunity for potential review and authorship toward a standards publication."

2. Ugur Cayoglu (Simulation and Data Life Cycle Lab Earth System Sciences) – NetCDF visualization in the browser with WASM
I would like to demo our web application for visualizing NetCDF data. It is written in Javascript and Rust. It runs natively on the client side using WASM and has a minimal memory footprint. It supports NetCDF2 and 3. A demo of the application is at https://alpha.datasets.earth/. It would be great to get feedback about the tool and possible collaborators for the development.

3. Chintam Venkateswarlu (Andhra University) – AI for Oceans and related topics

4. Hasan Nawaz (University of the Punjab Pak) – Soumi VIIRS Fire Products
I will demonstrate the importance of fore monitoring in Pakistan and adjoining regions of India during the fire season.

5. Humberto García Montano (UNAN-Managua) – Temperature Anomaly Studies and Lightning Studies

6. Shekhar Guddimath (Aventurine Technologies Inc) – Realtime species detection using Edge AI Cameras
How to use AI/ML on Edge device to detect, annotate and label the species underwater using Edge AI Camera. This will save time on manual annotation.

7. Daria Micovic – AI in Ocean Robotics
Could AI be useful in Ocean Robots? Where could it fit in?

8. Richie Singh (Graphcore) – How Graphcore IPUs can accelerate weather and climate discoveries.
Come and listen to how Graphcore IPUs (Intelligence Processor Units) are being utilized to achieve the next round of machine learning breakthroughs

We will use this session to come up with community ideas for standardization, and gaps in interoperability.

Session agenda:
14:00: Guidance
14:05: Breakout rooms by Component (attendee self selection)
14:40: Main room discussion, sharing from each group
15:00: Session Close

Notes:
Times above are in Mountain Daylight Time (UTC - 6). Lightning talks are roughly 10-12 minutes each followed by a Q&A with the panel. We will use Slido during the Q&A. We'll use Slack to build community throughout the workshop.

Slido for Q&A: sli.do/DIGITALTWIN/
Google Docs for Breakout Rooms

About Fire Weather and Impact Theme: This year's AI Workshop is piloting a collaboration development activitu modeled after the hackweek which brings participants with different backgrounds together to form collaboration and develop various use cases that are related to research and applications for fire weather and impacts. The core objective of FireWxHack are two-fold:

1. Foster collaboration building in applying AI/ML for fire weather research: Provide opportunities for fire weather researchers and data scientists to develop project ideas on how to use AI/ML and open environmental data for fire weather research & application development.
2. Collect user feedback on open environmental data for AI/ML in fire weather: Collect user feedback on how to improve open environmental data that can accelerate the adoption of AI/ML in fire weather research.
This event is NOT a data science competition and there is no fixed target for each individual team.

Session information:
Use case teams will continue the development during this 60 minutes session.

Relevant materials:
1. About FireWxHack: Document summarizing the goal of fire weather theme activities.
2. Introduction slides: Slides used during meet and greet to describe the timeline of fire weather theme.
3. Suggested use cases: These are three suggested use cases with relevant environmental data for fire weather research.
4. JupyterHub Environment: https://oasis.ncics.org (require GitHub user name to sign in).

During this session, attendees will choose a breakout room focusing on specific topics to discuss the requirements of ocean science data to advance the adoption of AI in ocean conservation and sciences.

Session agenda:

14:00 - 14:05: Welcome and logistics
14:05 - 14:45: Breakout room discussion
14:45 - 15:00: Breakout room sharing & wrap up

Breakout rooms are separated by topics:
1. Passive Acoustics & Telemetry (Facilitator: Kathy Broughton)
2. Underwater Imagery (Facilitator: Kathryn Lohr)
3. Ocean Data & Physical Modeling (Facilitator: Jebb Q. Stewart)
4. Airborne/Space Remote Sensing (Facilitator: Erin Moreland)
5. Environmental, Species & Stock Assessment Modeling (Facilitator: Stacie Robinson)

Participants can self select which topic room to join.

Notes:
Times above are in Mountain Daylight Time (UTC - 6). Lightning talks are roughly 10-12 minutes each followed by a Q&A with the panel. We will use Slido during the Q&A. We'll use Slack to build community throughout the workshop.

Slido for Q&A: sli.do/OCEAN/

Please join the workshop participants at the Wonder space to share your experiences during the day and discuss any additional ideas/topics that you want to share.

This plenary session focus on high-level vision on status, challenges, and future-oriented visions for leveraging AI/ML for fire weather and impacts research.

Session agenda:
09:30: Welcome & announcement (Douglas Rao, Jebb Stewart, Imme Ebert-Uphoff
09:35: Ko Barrett (NOAA Senior Advisor for Climate, Recorded address on Climate Ready Nation)
09:45: Robyn Heffernan (NOAA/National Weather Service)
10:05: Everett Hinkley (US Forest Service)
10:25: Q&A

Slido link for live Q&A: https://sli.do/2022AI

Develop a list of potential roles for AI in DT4ES component evolution.

Session agenda:
11:30     Welcome / Housekeeping
11:35     Dev Niyogi (UT Austin): Urban Digital Twin
11:50     Jonathan Kyle (AWS): Putting Context to Digital Twin Use Cases - Introducing a Leveling Framework
12:05     Dale Durran (UW): Deep Learning Weather Prediction and Earth System Modeling
12:20     Q&A - post your questions to Slack anytime (#04_digital_twin_for_earth_systems)
12:30     Breakout rooms by Component (attendee self selection)
12:55     Main room general discussion
13:00     Session Close

Notes:
Times above are in Mountain Daylight Time (UTC - 6). Lightning talks are roughly 10-15 minutes each followed by a Q&A with the speakers. We'll use Slack to pose questions to the speakers and to build community throughout the workshop.

About Fire Weather and Impact Theme: This year's AI Workshop is piloting a collaboration development activitu modeled after the hackweek which brings participants with different backgrounds together to form collaboration and develop various use cases that are related to research and applications for fire weather and impacts. The core objective of FireWxHack are two-fold:

1. Foster collaboration building in applying AI/ML for fire weather research: Provide opportunities for fire weather researchers and data scientists to develop project ideas on how to use AI/ML and open environmental data for fire weather research & application development.
2. Collect user feedback on open environmental data for AI/ML in fire weather: Collect user feedback on how to improve open environmental data that can accelerate the adoption of AI/ML in fire weather research.
This event is NOT a data science competition and there is no fixed target for each individual team.

Session information:
This session will use interactive collaboration tools, breakout rooms, and facilitated discussion to help fire weather theme participants to continue the use case development.

Relevant materials:
1. About FireWxHack: Document summarizing the goal of fire weather theme activities.
2. Introduction slides: Slides used during meet and greet to describe the timeline of fire weather theme.
3. Suggested use cases: These are three suggested use cases with relevant environmental data for fire weather research.
4. JupyterHub Environment: https://oasis.ncics.org (require GitHub user name to sign in).

T3. Google Cloud: R Workloads in Jupyter Notebooks – Ed Rodgers

A demonstration of running R in Jupyter Notebooks using both user-managed and Google-managed deployments within the Vertex AI service framework.

In this demonstration, I will provide a brief overview of the R kernel within a JupyterLabs server, deploy both user-managed and Google-managed servers in a sandbox environment, and compare them.

Link to the presentation notes: https://docs.google.com/presentation/d/1Y155mytKNgTmn1R-YfO8UbiO4AomLuSmorMxaGn7vpk

T2. TagLab - the learning pipeline – Massimiliano Corsini & Gaia Pavoni (Italian National Research Council)

Over the past decade, underwater photogrammetry has become one of the most popular tools for studying coral reefs. Coral demographic studies require the generation of highly accurate semantic segmentations on large area orthoimages (ten to hundreds of square meters). This heavily time-consuming task forces scientists to adopt AI-based labelling solutions. TagLab is an open-source labelling tool designed for ecological applications TagLab enables researchers to annotate efficiently large orthoimages and facilitates the analysis of changes in coral reef ecosystems. In this tutorial we will focus on how to use the assisted annotation tools in TagLab to speed up the semantic segmentation of seafloor and, in particular, on how to create automatic custom classifiers for the data under investigation.

T1. VIAME 101 (Basics) – Matt Dawkins, Bryon Lewis (Kitware)

VIAME is an open-source computer vision software platform that was developed by Kitware in cooperation with NOAA’s Automated Image Analysis Strategic Initiative. It contains web, desktop, and command line interfaces covering a range of tasks including video/image annotation, object detection, object tracking, video search and rapid model generation, full frame classification, image registration and mosaicing, camera calibration, stereo measurement, and 3D scene modelling. This tutorial will cover the basics of VIAME including a short overview presentation, followed by an interactive tutorial on its public web (or desktop, for those who prefer it) client covering a typical object detection workflow such as annotation, model training, model inference, and model re-training.

T4. VIAME-Web in the NOAA Fisheries Google Cloud – Sam Woodman

**NOTE: Users interested in following along that do not already have a GCP project should request access to Qwiklabs using this Google Form. Qwiklabs is a hands-on learning platform that provides temporary credentials to a Google Cloud Platform project, in this case allowing you to practice deploying VIAME-Web in GCP during and after this training. The Qwiklabs environment will be active until Saturday, September 10, 2022 9:00AM EDT.**

VIAME is an open-source computer vision software platform that was developed by Kitware in cooperation with NOAA’s Automated Image Analysis Strategic Initiative. Kitware hosts a public deployment of VIAME-Web, but this deployment has limited computing resources shared across all users and cannot connect to NOAA Google Cloud Storage (GCS) buckets because of NOAA Fisheries Cloud security requirements. The solution to this problem is for users to deploy their own instance of VIAME-Web in their own GCP project. This training will walk users through the GitHub repo viame-web-noaa-gcp, which includes docs with NOAA-specific instructions on deploying, provisioning, and accessing GCP resources, as well as general instructions and guidelines on managing the VIAME-Web deployment. This training is tailored for NOAA users, but much of it will also be applicable to any users looking to deploy their own instance of VIAME-Web in GCP.

Users that already have a GCP project should ideally install and configure the Google Cloud CLI on their local machine before the training. Interested parties without a GCP project can pre-register for a Qwiklabs environment via this form to follow along with the training, but are also welcome to simply watch the training to gain familiarity with the viame-web-noaa-gcp repo and docs, managing resources with Terraform, and VIAME-Web admin options."

Presntation link

Explore potential roles for AI in enabling DT4ES interoperability. Explore communicating benefits and limitations of DT4ES ecosystems.

Session agenda:
13:30: Main room discussion, sharing from each group from morning session
13:50: Breakout rooms by Component (attendee self selection)
14:30: Main room discussion, sharing from each group
15:00: Session Close

Google Docs for Breakout Rooms

Notes:
Times above are in Mountain Daylight Time (UTC - 6). We'll use Slack to build community throughout the workshop.

About Fire Weather and Impact Theme: This year's AI Workshop is piloting a collaboration development activitu modeled after the hackweek which brings participants with different backgrounds together to form collaboration and develop various use cases that are related to research and applications for fire weather and impacts. The core objective of FireWxHack are two-fold:

1. Foster collaboration building in applying AI/ML for fire weather research: Provide opportunities for fire weather researchers and data scientists to develop project ideas on how to use AI/ML and open environmental data for fire weather research & application development.
2. Collect user feedback on open environmental data for AI/ML in fire weather: Collect user feedback on how to improve open environmental data that can accelerate the adoption of AI/ML in fire weather research.
This event is NOT a data science competition and there is no fixed target for each individual team.

Session information:
This session will use interactive collaboration tools, breakout rooms, and facilitated discussion to help fire weather theme participants to continue the use case development.

Relevant materials:
1. About FireWxHack: Document summarizing the goal of fire weather theme activities.
2. Introduction slides: Slides used during meet and greet to describe the timeline of fire weather theme.
3. Suggested use cases: These are three suggested use cases with relevant environmental data for fire weather research.
4. JupyterHub Environment: https://oasis.ncics.org (require GitHub user name to sign in).

T6A. Make your data "AI-ready": A hands-on data assessment activity (Acoustics data) – Tyler Christensen, Dan Woodrich
T6B. Make your data "AI-ready": A hands-on data assessment activity (Imagery data) –  Stacie Robinson, Erin Moreland

Data wrangling can be one of the most time-consuming components at the outset of a new AI-project. Having data that is high-quality, well-documented, easily accessible, and annotated can streamline application in AI/ML models and make data more open and valuable to collaborators. In this tutorial we will familiarize participants with characteristics of AI-ready data using this AI-Readiness Checklist. We will evaluate datasets to identify information gaps in data documentation and challenges in data preparation. We will also ask participants to provide feedback on the checklist. Participants are encouraged to come with their own datasets (only metadata is necessary) or download metadata for publically available datasets such as NOAA's Climate Data Records.

Use this link to make a copy of the AI-readiness assessment worksheet. Once copied, you can rename the worksheet with your name or dataset name.

T7. Tator - AI Assisted Video Analysis – Benjamin Woodward (CVision AI)

This tutorial will cover the ability of Tator to enable rich analysis of video using custom metadata and platform sensor data, as well as integration of algorithms for tasks such as detection, tracking, and classification. In addition, we will cover integration with third-party algorithm APIs, as well as examples of simple visual trackers, color correction algorithms, and QA/QC workflows.

T5. VIAME 201 (Advanced) – Matt Dawkins, Bryon Lewis (Kitware)

This tutorial will briefly cover some of the more advanced features of the VIAME toolkit such as AI-assisted annotation, stereo measurement, video search and rapid model generation, model parameter tuning, image enhancement, detector ensembles, and the different forms of object trackers in the system.

Please join the workshop participants at the Wonder space to share your experiences during the day and discuss any additional ideas/topics that you want to share.

This plenary session provides a high-level overview of the needs, current status, and challenges of developing ethical and responsible AI applications for environmental sciences.

Session agenda:

09:30 - 09:35: Welcome (Douglas Rao, Amy McGovern)
09:35 - 10:00: David Danks (UC San Diego)
10:00 - 10:25: Christine Custis (Partnership on AI)
10:25 - 10:45: Plenary Panel Discussion

Notes:
Times above are in Mountain Daylight Time (UTC - 6). Lightning talks are roughly 10-12 minutes each followed by a Q&A with the panel. We will use Slido during the Q&A. We'll use Slack to build community throughout the workshop.

Slido for Q&A: sli.do/2022AI/

We'll explore overarching questions and summarize the discussions we’ve had throughout the week to form a community vision for an interoperable DT4ES ecosystem. How can our vision evolve to ensure compatibility, interoperability, standard interfaces, etc?

Session agenda:
11:30: Guidance
11:40: Breakout Discussion (10-15 persons/room, randomly assigned)
12:20: Main room discussion, sharing from each group over entire week
13:00: Session and DT4ES Workshop Theme Close

Google Doc for Breakout Rooms

Notes:
Times above are in Mountain Daylight Time (UTC - 6). We'll use Slack to build community throughout the workshop.

About Fire Weather and Impact Theme: This year's AI Workshop is piloting a collaboration development activitu modeled after the hackweek which brings participants with different backgrounds together to form collaboration and develop various use cases that are related to research and applications for fire weather and impacts. The core objective of FireWxHack are two-fold:

1. Foster collaboration building in applying AI/ML for fire weather research: Provide opportunities for fire weather researchers and data scientists to develop project ideas on how to use AI/ML and open environmental data for fire weather research & application development.
2. Collect user feedback on open environmental data for AI/ML in fire weather: Collect user feedback on how to improve open environmental data that can accelerate the adoption of AI/ML in fire weather research.
This event is NOT a data science competition and there is no fixed target for each individual team.

Session information:
This session will give participating teams the opportunity to summarize the development experience during the workshop and share feedback about the development process. All participants will also discuss the next steps for the development.

Relevant materials:
1. About FireWxHack: Document summarizing the goal of fire weather theme activities.
2. Introduction slides: Slides used during meet and greet to describe the timeline of fire weather theme.
3. Suggested use cases: These are three suggested use cases with relevant environmental data for fire weather research.
4. JupyterHub Environment: https://oasis.ncics.org (require GitHub user name to sign in).

This final session for the ocean theme will have a mix of lightning talks and breakout discussions to form a community roadmap on how to advance the adoption of AI in research and operations for ocean conservation and sciences. Participants will be invited to contribute to a community whitepaper to summarize the discussion.

Session Agenda:

11:30 - 11:35: Welcome
11:35 - 11:40: An Unconstrained Look to the Future: Using Simulation for Data Collection (Casey Sapp, Blue Ring Imaging)
11:45 - 11:55: How AI Enables a National Operational Coastal Webcam Network (Greg Dusek, NOAA/NOS/CO-OPS)
12:00 - 12:10: TBA (Grace Young, X, the moonshot factory)
12:15 - 12:25: Migrating AI Workflows from On-Premises Processing to a Cloud Computing Platform: Things to Consider (Michael Schwartz, Brandon Krumwiede)
12:30 - 12:40: R2X and AI2ES: Discussion of Paths to Trustworthy AI Operational Models (Philppe Tissot, John Williams)
12:45 - 12:55: Maxar imagery for AI applications (Manny Gonzalez-Rivero, Francois Smith)
12:55 - 13:25: Breakout room discussion on next steps

Breakout rooms are separated by topics:
1. Passive Acoustics & Telemetry
2. Underwater Imagery
3. Ocean Data & Physical Modeling
4. Airborne/Space Remote Sensing
5. Environmental, Species & Stock Assessment Modeling

Participants can self select which topic room to join.

Notes:
Times above are in Mountain Daylight Time (UTC - 6). Lightning talks are roughly 10-12 minutes each followed by a Q&A with the panel. We will use Slido during the Q&A. We'll use Slack to build community throughout the workshop.

Slido for Q&A: sli.do/OCEAN/

Please join the workshop participants at the Wonder space to share your experiences during the day and discuss any additional ideas/topics that you want to share.