Evaluating the transferability potential of deep learning models for climate downscaling

TL;DR

This paper assesses how well deep learning models for climate downscaling can transfer knowledge across different datasets, architectures, and tasks to improve their robustness and applicability in diverse climate scenarios.

Contribution

It systematically evaluates the transferability of CNNs, FNOs, and ViTs for climate downscaling across multiple datasets and variables, highlighting their generalization capabilities.

Findings

CNNs, FNOs, and ViTs show varying transferability performance.

Models trained on diverse datasets improve robustness.

Architectures differ in their ability to generalize across tasks.

Abstract

Climate downscaling, the process of generating high-resolution climate data from low-resolution simulations, is essential for understanding and adapting to climate change at regional and local scales. Deep learning approaches have proven useful in tackling this problem. However, existing studies usually focus on training models for one specific task, location and variable, which are therefore limited in their generalizability and transferability. In this paper, we evaluate the efficacy of training deep learning downscaling models on multiple diverse climate datasets to learn more robust and transferable representations. We evaluate the effectiveness of architectures zero-shot transferability using CNNs, Fourier Neural Operators (FNOs), and vision Transformers (ViTs). We assess the spatial, variable, and product transferability of downscaling models experimentally, to understand the generalizability of these different architecture types.