Lucas’s Presentation — Tackling Climate Change with Machine Learning

  • The length of the hydrocarbon determines whether it is methane, kerosene, or gas
    • C6 - C10 hydrocarbons, become crude oils
    • A single barrel of oil 42 gallons turn into:
      • 19 gallons of gasoline
      • 11 gallons of diesel
      • 7 gallons of other products (ink, crayons, dishwashing liquids)
      • 4 gallons of highly refined gasoline
  • Svante Arrhenius was the first to posit the greenhouse gas effect (as the quantity of carbonic acid increases, temperature will increase linearly)
  • Earth re-reradiates heat as blackbody radiation, GHG particles catch that heat
  • Carbon dioxide PPM is in cycles (peaks and valley in a year), peak is winter in northern hemisphere, dead trees are leaking our carbon, valley is in the middle of summer in the northern hemisphere

Positive and negative feedback loops:

  • Negative feedback:
    • Ocean absorbs CO2
    • More plan growth in some areas
    • Increase in dust, evaporation, volcanic activity
      • Increase in volcanic activity is because as the earth warms, glaciers melt, causing change in geologic pressures
  • Positive feedback loops:
    • Poleward shift of forests
    • Drying of peatlands and methanols bubbling in permafrost
    • Decrease in biodiversity (biodiversity fixes carbon)
    • Increase in forest temperatures

Main effects of climate change (affecting humans)

  • Food instability, drought, hunger
  • Increase in conflict following natural disasters
    • Syrian war is a result of drought, something to follow up on
  • Climate refugees
  • Rightward shift in many governments of the world
  • More extreme weather events
  • Greater spread of disease vectors

Enter Machine Learning

Summarizing the paper

Tackling Climate Change with Machine Learning Useful Machine Learning Techniques

  1. Generative modeling: Statistical models that create simulated observations” of real world phenomena
    • Applications:
      • Generate structural model of buildings with less carbon intense material
      • Generate energy signatures of people to help model in data poor environments
      • Dynamic price generation of grid prices to help optimize for lowered GHG emissions
  2. Personalization:
    • ML grid price signals
    • NLP has been used to extract plane ticket info and shopping receipts from email to quantify a person’s carbon impact
    • Counterfactual AI has been used to create what-if scenarios
    • Psychological research - Distance from climate change psychologically is a big determinant on someone’s climate change policies
  3. Digital Twin models
    • Make a good representation of machinery in a computer base, and use that representation to do (and model) things
  4. Image ML
    • Precision agriculture
      • Scan for disease, yield, identify spots for fertilizer
      • There is a digital revolution going on in agriculture going on right now. (Farmers tend to like flying drones)
    • Identify and count species
    • Scan satellite photos and identify good spots for solar panels or predict good wire siting (work in India used minimum spanning trees)
    • Agriculture is fundamental to climate change: Ensuring that we have high yields is essential for making the most of the environmental consequences that are occurring
    • Can also visualize the effects of flooding on homes, can change your perception of climate change
  5. Natural language processing
    • Venugopalan (2015) applied this to analyzing solar patent applications to build a general model of solar innovation
    • Provide personalized recommendations for people who want to reduce their carbon footprints
    • Analyze social media to understand discourse around climate change
    • Automated identification and scoring of climate risks in company’s public disclosures

[went to make dinner, so don’t have much information from 4-2 unfortunately]

  1. Reinforcement Learning and Optimization
    • Control the charging of EVs to help stabilize price grid
  2. Combinatorial ML for material discovery
    • Basically search ML, search the space of materials experiments faster

uuid: 202005072038 date: May 7, 2020 tags: #raise #presentation


Date
February 22, 2023