Fig. 1: Study domain and the eleven North American Large Marine Ecosystems (LMEs), with significantly higher prediction skills from the decadal prediction system against reconstruction persistence in predicting habitat viability in the upper-600 m ocean.

The prediction skills are assessed as Anomaly Correlation Coefficient (ACC). The decadal prediction system refers to the Community Earth System Model - Decadal Prediction Large Ensemble (CESM-DPLE)²² system. The eleven LMEs are: 01-Eastern Bering Sea (EBS; 261 grid points), 02-Gulf of Alaska (GoA; 316 grid points), 03-California Current (CC; 378 grid points), 04-Gulf of California (GoC; 26 grid points), 05-Gulf of Mexico (GoM; 270 grid points), 06-Southeast U.S. Shelf (SEUS; 61 grid points), 07-Northeast U.S. Shelf (NEUS; 62 grid points), 08-Scotian Shelf (SS; 68 grid points), 09-Labrador-Newfoundland (LN; 263 grid points), 10-Insular Pacific Hawaiian (IPH; 163 grid points), and 11-Aleutian Islands (AI; 35 grid points). They are grouped into the northwest, northeast, southwest, and southeast coast LMEs, based on their geographical locations. The red contours on the map represent the 200-m isobath, based on bathymetry data from ETOPO1⁵⁴. The boundaries of the LMEs are plotted as thick black lines on the map. For the panel of each LME, y-axis represents the prediction skill in ACC, and x-axis represents the lead year (LY) ranging from 1 to 10. The ACCs significantly nonzero at the 95% confidence level are marked with solid dots. Only ACCs of the medium-\({E}_{{{{{{\rm{o}}}}}}}\) trait (temperature sensitivity of hypoxia vulnerability; 0.4 eV) species are presented. Data to reproduce the figures are shared on Figshare⁵³.