Can Transfer Learning Be Used to Identify Tropical State-Dependent Bias Relevant to Midlatitude Subseasonal Predictability?

Abstract Previous research has demonstrated that specific states of the climate system can lead to enhanced subseasonal predictability (i.e., state-dependent predictability). However, biases in Earth system models can affect the representation of these states and their subsequent evolution. Here, we present a machine learning framework to identify state-dependent biases in Earth system models. In particular, we investigate the utility of transfer learning with explainable neural networks to identify state-dependent biases in tropical precipitation important for subseasonal predictability of upper-level circulation over the North Pacific using the historical simulations of the Energy Exascale Earth System Model, version 2 (E3SMv2). Using a perfect model framework, we find that transfer learning may require substantially more data than provided by present-day reanalysis datasets to update neural network weights, imparting a cautionary tale for future transfer learning approaches focused on subseasonal modes of variability. Significance Statement In this study, we explore how two machine learning methodologies (transfer learning and explainable artificial intelligence) can be used to identify biases in Earth system models. We are particularly interested in identifying biases associated with states in our climate that can lead to enhanced predictability on 2-week to 2-month time scales. Identifying and then subsequently correcting these biases can lead to improved prediction skill on these time scales. However, we find that our machine learning framework for identifying biases requires more data than our observational datasets provide.