Representative Bias of Marine Heatwaves in the North Pacific Ocean Simulated by CMIP6 Models: A Clustering-based Approach

Abstract

Climate models are the primary tool for projecting marine heatwaves (MHWs). Assessing the simulation performance of the climate model is crucial before projecting future changes in MHW, as it helps comprehend how model bias can influence the interpretation of future changes. Previous research has mainly focused on assessing the common bias produced by multi-model ensembles of climate models. However, in the evaluation of individual models, certain groups exhibit comparable bias patterns, and clustering them can provide a variety of representative biases. In this study, we aimed to evaluate the MHW characteristics from 30 Coupled Model Project Intercomparison Phase 6 (CMIP6) models in the North Pacific Ocean compared to OISST reanalysis data for 33 years (1982–2014), and to cluster CMIP6 models with similar bias patterns in order to understand the representative bias. Based on similar patterns of MHW bias, hierarchical clustering was used to classify the 30 CMIP6 models into five groups. In terms of total MHW days, the first group showed a spatial distribution of overestimation similar to the Pacific decadal oscillation pattern. In addition, other groups showed a concentrated overestimation in the Kuroshio Extension, the southwestern North Pacific and the Bering Sea for their respective spatial distributions. Regarding the mean MHW intensity bias, more than 80% of the CMIP6 models showed a dipole pattern in the Kuroshio Extension. Thus, the performance of the ocean current simulation was assessed using the same model for each of the clusters. The clusters that show a dipole pattern in mean intensity indicate a simulation of the Kuroshio Current that extends further north than observed. By presenting representative biases that differ from the common biases but are shared by multiple models, this study suggests that it is necessary to identify representative biases to interpret the future MHWs simulated by specific climate models