You can take
1) a baseline climate run (eg 1xCO2, or 1% increase a year, whatever)
2) then add some extra CO2 on top (say constant 2xCO2) and let your AI
stick a geoengineering layer on top of that, to try and dial it back down
to the baseline
The detection AI doesn't know which is the baseline -
The first two don’t require ANN.
I don’t get what your detection network is supposed to do. We have simulations
of the baseline climate and we know what it is. If we generate simulations of
a climate that is usefully different, that’ll be pretty clear. If the
differences are subtle, then it
I can think of a couple of counterpoints
1) a cheap climate model may allow a neural network to learn some basic
strategies pretty quickly - like mixing CCT with SAI to control precip
2) training AI on weird climates (8x CO2, anoxic oceans) might prepare us
for weird stuff we'd otherwise miss
3)
Thanks!
I was actually thinking of reinforcement learning in my previous response…
agree that GAN is not at all applicable here.
I think a stronger statement is in order… this is NOT a problem where any
ANN-based algorithm is likely to be of much value in the next 10+ years, and
maybe never.
Hi everyone, I asked a data scientist who works on deep learning -- Alex
Orona (https://www.linkedin.com/in/alexorona/) -- about this. Here is his
email:
"Can't comment on the climatological aspects of the question (it's well
outside my domain), but here is my feedback on the machine
Title: Climate Engineering Newsletter
Climate Engineering Newsletter
for Week 05 of 2019
(new) 6.02.2019, Colloquium: Daniel L. Sanchez – Near-term opportunities for carbon dioxide removal from bioenergy, Berkeley CA / USA
(new) 30.01.2019, Meetup: CARBON RUSH --
https://www.su.se/english/about/working-at-su/jobs?rmpage=job=7945=UK
Postdoctoral Fellow in International Relations, with an emphasis on
Environmental Policy Ref. No. SU FV-0079-19
Share
Share
at the Department of Economic History and International Relations. Closing
date: 15 March 2019.
The
Andrew
I did something similar to optimize the control of a wave energy
generator in a 1/100 scale test tank. We want to control forces on the
model in three degrees of freedom, pitch heave and surge during the
passage of a 51.2 second repeatable 'random' sea with 16 adjustments per
second.
