Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Artificial intelligence and downscaling global climate model future projections

Rasmus E. Benestad

TL;DR

Problem: AI/ML methods for downscaling GCM future projections risk non-stationarity and non-representative training data. Approach: the author reviews historical AI/ML efforts, defines downscaling vs interpolation/bias-adjustment, and contrasts with empirical-statistical downscaling (ESD), stressing proper benchmarking and transferability. Findings: AI/ML can emulate RCMs and extract meaningful features, but performance is strongest when trained on future-period simulations and evaluated across GCMs; ESD often offers greater robustness and efficiency. Significance: the paper argues for rigorous, physics-informed, and performance-based evaluation to avoid hype and to determine the practical value of AI/ML in climate downscaling.

Abstract

A critical review of artificial intelligence and deep machine learning (AI/ML) applied to downscaling of global climate model simulations provides some words of caution, based on past experiences and well-established principles. Recent papers tend to ignore more subtle successes with statistics and mathematical based downscaling, and there are examples of inappropriate evaluation strategies and incomplete accounts of the scientific progress when it comes to climate downscaling. An incomplete description state-of-the-art and a dogmatic approach to evaluation may give a deceiving impression that AI/ML is superior to more statistics and mathematics based methods.

Artificial intelligence and downscaling global climate model future projections

TL;DR

Problem: AI/ML methods for downscaling GCM future projections risk non-stationarity and non-representative training data. Approach: the author reviews historical AI/ML efforts, defines downscaling vs interpolation/bias-adjustment, and contrasts with empirical-statistical downscaling (ESD), stressing proper benchmarking and transferability. Findings: AI/ML can emulate RCMs and extract meaningful features, but performance is strongest when trained on future-period simulations and evaluated across GCMs; ESD often offers greater robustness and efficiency. Significance: the paper argues for rigorous, physics-informed, and performance-based evaluation to avoid hype and to determine the practical value of AI/ML in climate downscaling.

Abstract

A critical review of artificial intelligence and deep machine learning (AI/ML) applied to downscaling of global climate model simulations provides some words of caution, based on past experiences and well-established principles. Recent papers tend to ignore more subtle successes with statistics and mathematical based downscaling, and there are examples of inappropriate evaluation strategies and incomplete accounts of the scientific progress when it comes to climate downscaling. An incomplete description state-of-the-art and a dogmatic approach to evaluation may give a deceiving impression that AI/ML is superior to more statistics and mathematics based methods.
Paper Structure (8 sections, 1 figure)

This paper contains 8 sections, 1 figure.

Table of Contents

  1. Introduction
  2. Truly representative data
  3. Downscaling climate change
  4. Proper testing and benchmarking
  5. Data and mathematics
  6. Good use of AI/ML for climate downscaling
  7. Discussion
  8. Conclusion

Figures (1)

  • Figure 1: Clouds have many different scales, and different cloud types and structures illustrate different situations where links between large and small scales vary. They can also provide a metaphor for Earth's climate going through transitions from a known state to new unknown one with different internal structures. An AI/ML trained to predict small-scale features from the large scales in the clouds of left panel may not work well for those in the right hand panel. Often AI/ML involves cloud computing, pun intended.