top of page
Image by NASA

FARR mini research opportunities for  advancing AI-readiness in earth sciences

The FARR (FAIR in ML, AI-readiness and Reproducibility) Research Coordination Network is supporting efforts to increase AI-readiness in earth sciences communities (farr-rcn.org). As one aspect of its activities, FARR is inviting proposals for focused, limited-term efforts to advance some aspect of AI-readiness in an earth sciences domain. Examples include, but are not limited to:

  • Developing a training dataset for future machine learning work,

  • Improving the quality and documentation of an existing dataset,

  • Developing training materials, or offering training on specific AI/ML skills,

  • Organizing a community hackathon event,

  • Creating recommended practices or other guidance documents,

  • Working within a community to increase use of FAIR principles to advance AI-readiness, including developing/refining domain-specific aspects of the FAIR principles.


Proposals should be for 1 year or less and must clearly describe expected outcomes (e.g., written document, conference presentation, community event, etc.) and how success will be measured that will be produced within one year. Extension to a second year possible after initial progress is demonstrated.

Successful applicants will receive support from FARR in the form of a $2,500 honorarium directly to the lead (not through an institutional subcontract). Projects co-led by an established scientist and a student or early career scientist (within 10 years of final degree) can receive two $2,500 honoraria, if roles are justified for each. FARR project team will also support selected proposals to request cloud computing credits or other computing resources to support technical development as needed. For selected projects, a limited amount of travel support will be made available for a small in-person meeting held in conjunction with the
FARR Workshop (scheduled for October 9-10, 2024 in Washington DC).

To apply, please submit this application form, including a description of the task with specific deliverables/outcomes and their schedule, the expected impact on AI-readiness of a community, and the skills/expertise you bring to the work. (500 words maximum); and whether you are potentially interested in travel support for an in-person meeting and/or compute resources.

Applicant Requirements:

  • Must be over 18 years of age.

  • If you are not a U.S. citizen, then you must have a U.S. Taxpayer Identification Number (ITN) and have the appropriate visa type (J-1, H-1B, F-1OPT).

  • Confirm your eligibility for honorarium payments with your home institution. There are a few restrictions, for example, NSF employees are ineligible and University of California awardees must be faculty members.


Applications are closed.  Please contact community@farr-rcn.org with any questions.

 

FARR Selects Six Mini Research Opportunities in AI-readiness

 

The FARR (FAIR in ML, AI Readiness and Reproducibility) Research Coordination Network is pleased to announce its first five mini research projects to increase AI-readiness in earth sciences communities. Projects include: (1) developing open educational content for the application of generative AI tools in Earth science education, (2) applying FAIR principles to design and create a training dataset for ML models to estimate forest characteristics, (3) constructing a robust library of geologic map symbols that will provide a solid data foundation for intelligent understanding of future geologic maps, (4) creating an AI/ML model of daily detection with explanations and a catalog of daily greenhouse gas emission events,  (5) creating an open-access, high-quality, multi-imagery training dataset for rooftop solar panel detection through a community mapathon, and (6) creating a comprehensive time series dataset for tundra lake segmentation in Western Siberia using satellite optical and synthetic aperture radar (SAR) images, as well as historical cartographical maps. Recipients listed below.


Advancing Earth Science Education Through Generative Artificial Intelligence
Sanjib Sharma, Howard University
    The primary objective of this project is to develop open educational content for the application of generative AI tools in Earth science education. This project will (i) train undergraduate and graduate students on how to use generative AI tools, (ii) prepare open-source educational materials, and (iii) disseminate this learning experience to the wider Earth science community through GitHub, social media and newsletter.

Preparing a time series database for applications in forest ecology and wildfire resilience in the Western United States
Jazlynn Hall and Winslow D. Hansen, Cary Institute of Ecosystem Studies
    By combining field-based empirical work with remote sensing, process-based simulation, and machine learning (ML),  the Western Fire and Forest Resilience Collaborative (WFFRC) will 1) evaluate historical changes to forest and fire regimes in the Western US, 2) project future regimes under various management and climate scenarios, and 3) assess and project consequences of changing forest and fire regimes on ecosystem services.

GeoSymbolNet: Leveraging Data Augmentation to Decipher Geological Map Symbols
Wenjia LI and Xiaogang Ma, University of Idaho
    This project is dedicated to the use of open source scanned material - geologic maps sourced from the Idaho Geological Survey website - combined with data augmentation methods, with the aim of constructing a robust library of geologic map symbols that will provide a solid data foundation for intelligent understanding of future geologic maps. By applying advanced data processing and augmentation techniques to existing scanned geologic maps, including but not limited to rotation, scaling, and contrast adjustment, we aim to address the challenges faced due to variable resolution and inconsistent symbols in historical archived images, and thus achieve automated and accurate geologic map recognition and analysis.

Monitoring Greenhouse Gas Emitters at Night with Machine Learning Insights on NASA’s Black Marble
Srija Chakraborty, USRA
    In this work, we propose to improve an Earth Science training dataset by exploring detection explanations using SHapley Additive exPlanations (SHAP) and ancillary (land cover; land water mask) layers and to cluster the detected pixels with GHG events into subclasses based on most relevant bands and context and enhance the dataset to create a repository of VIIRS-based GHG event detection for six classes: fire, flare(land), flare(offshore), ships, volcanoes, background at a pixel level (along with detection confidence) from daily top-of atmosphere Black Marble data (Task 1). This dataset 'Nighttime-Emission' will be generated using geospatial standards in GeoTIFF or zarr format over five Black marble tiles from 2018-2021 (Task 2) and serve as training dataset for the greenhouse gas monitoring community. We will share our models, code using FAIR principles in github (Task 3) and create and share (dataset and workflow documentation over NASA Black Marble website) for transparency (Task 4).

Mapping Rooftop Solar across New England
Denys Godwin and Hamed Alemohammad, Clark University
    In this research, we propose to create an open-access, high-quality, multi-imagery training dataset for rooftop solar panel detection through a community mapathon at Clark University. This dataset will enhance AI readiness in earth sciences by publishing a new open-access benchmark, training students to label aerial imagery and introducing them to the world of geospatial AI.

Multisource Time Series Dataset for Tundra Lake Segmentation in Western Siberia
Ivan Sudakow, The SETI Institute and Jian Gong, University of Wyoming
    This project aims to create a comprehensive time series dataset for tundra lake segmentation in Western Siberia spanning a century (1922-2022) using satellite optical and synthetic aperture radar (SAR) images, as well as historical cartographical maps. The dataset will facilitate the training of machine learning architectures tailored for tundra lake recognition.

bottom of page