Artificial intelligence could improve climate patterns – how good is that?

Article by Eric Worrell

If you have a possible missing variable problem, the solution is to add more arbitrary adjustments to your model?

AI reveals accelerating climate change: 3°C temperature rise is imminent

Published by IOP December 28, 2024

Artificial intelligence-enhanced research shows that regional warming will exceed critical thresholds sooner than expected, with most regions on track to exceed 1.5°C by 2040. Adapt to action.

Three leading climate scientists used artificial intelligence (AI) to analyze data from 10 global climate models to improve accuracy. Their findings suggest that regional warming thresholds may be reached earlier than previously estimated.

…

Elizabeth Barnes said: “Our study highlights the importance of incorporating innovative artificial intelligence techniques such as transfer learning into climate modeling to potentially improve and constrain regional forecasts and provide insights for policymakers, scientists and researchers around the world. and communities provide actionable insights.

…

Learn more: https://scitechdaily.com/ai-exposes-accelerated-climate-change-3c-Temperature-rise-imminent/

reference research;

Combining climate models and observations to predict remaining time to regional warming thresholds

Elizabeth Barnes^*Noah S Diffenbaugh and Sonia I Seneviratne

Published December 10, 2024 • © 2024 The Authors. Published by IOP Publishing Ltd.
Environmental Research Letters, Volume 20, Number 1 citation Elizabeth Barnes et al. 2025 environment. resource. Wright. 20 014008DOI 10.1088/1748-9326/ad91ca

abstract

The importance of climate change in causing adverse climate impacts has prompted significant efforts to understand the rate and extent of regional climate change in different parts of the world. However, despite decades of research, there remains considerable uncertainty about the time remaining until a specific regional temperature threshold is reached, with climate models often disagreeing about the warming that has occurred so far and how much warming the region is likely to experience. . Here, we employ recent machine learning methods to train convolutional neural networks to predict the timing of reaching different regional warming thresholds (and their uncertainties) based on the current state of the climate system. In addition to predicting regional rather than global warming thresholds, we include a transfer learning step in which a climate model-trained network is fine-tuned with limited observations, which further improves real-world predictions. Using observed 2023 temperature anomalies to define the current climate state, our approach yields central estimates of the 1.5 °C threshold in 2040 or earlier, and 2.0 in 2040 or earlier, for all regions where transfer learning is possible The centers of the °C thresholds estimate the °C thresholds for 31 of the 34 regions. For 3.0 °C, the threshold is projected to be reached by 2060 at 26 °C in 34 regions. Produces finite predictions based on future temperatures.

Learn more: https://iopscience.iop.org/article/10.1088/1748-9326/ad91ca

If I understand correctly, they are essentially using artificial intelligence as a complex black box polynomial correction to their rather imprecise climate models to try to squeeze out better answers. Polynomials are trained by comparing observed temperature data to model output, and then extrapolated to a final merger of climate models and artificial intelligence polynomial corrections to try to predict future events.

The problem with this approach is It creates the illusion of accuracy without actually knowing whether greater accuracy has been achieved. AI used in this way applies complex arbitrary “corrections” to input data to produce a near-perfect match to whatever data was used to train the AI. But artificial intelligence knows nothing about the underlying physical phenomena. If AI has enough data, it might be able to infer physical phenomena—or Artificial intelligence can make stuff upespecially if the data set used to train the artificial intelligence is missing unknown key input data.

Artificial intelligence does play a role in scientific analysis. In areas such as drug discovery and complex optimization problems, artificial intelligence can produce outstanding results.

But as we all know, artificial intelligence also has a tendency to go off track, that is, to “hallucinate” wrong results.

AI glitches are not a problem if you can immediately test the quality of the AI ​​results. But using AI to try to figure out how to correct climate models where no one will know if the AI ​​is correct for years or decades, and then using those AI corrections to predict future events, seems like a dubious use of AI.

Relevant