Cooperative Institute for Research in the Atmosphere

AI Institute

Overview of AI Institute (aka AI2ES)

NSF AI Institute for Research on Trustworthy AI in Weather, Climate, and Coastal Oceanography (AI2ES).

• AI2ES was funded by the NSF in 2020 with a $20M grant over 5 years.
• AI2ES is headquartered at the University of Oklahoma, and CSU is one of its founding members.
• The graphic below gives a basic overview of the primary research areas AI2ES seeks to address and how they are connected.  For a more detailed explanation of AI2ES activities and philosophy, see https://www.ai2es.org/research/

AI2ES activities at CSU

Within CSU the core departments involved in AI2ES are

• CIRA – led by Imme Ebert-Uphoff (also serves as lead of the entire AI2ES-CSU team)
• Atmospheric Science – led by Elizabeth Barnes
• Computer science – led by Chuck Anderson

AI2ES activities at CIRA

Several of our CIRA members are funded by this NSF grant and are working on topics that are of interest to both CIRA and AI2ES.

CIRA activities related to this AI institute include:

• Tropical cyclones:
  • Develop ML methods that can simulate passive microwave imagery from geostationary satellite imagery.
  • Goal:  Generate imagery that reveal internal structure of tropical cyclones at high temporal resolution.
• XAI methods for weather and climate:
  • Select and adjust methods for eXplainable Artificial Intelligence (XAI) specifically for use in environmental applications.
• Incorporating forecasters’ feedback: 
  • Work with an interdisciplinary team of risk communication experts, atmospheric scientists and AI experts on soliciting feedback from forecasters on which kinds of AI tools, visualization and explanations are actually useful for them.
  • Take that feedback into account to develop AI tools for forecasters that they actually find useful.
• Ethical consideration for development of AI tools for environmental science 
  • Many people think that because AI is a mathematical tool it is objective in all aspects.  That’s not true.  There are many ways in which bias can sneak into the algorithms, such as
    • Bias in data (e.g., unevenly distributed sensor network, measurements only possible under certain conditions, etc.)
    • Bias can also arise from algorithm choices, e.g., spatial resolution of predictions greatly influences whether underrepresented groups are “averaged out”

Primary contact

Imme Ebert-Uphoff