Text S1.
The spatial SOMs of the 1994–2018 time series of sea-surface velocity (SSV) were computed from the HYCOM + NCODA GOFS 3.1 reanalysis and analysis (described in the main document). The resulting spatial patterns from this HYCOM instantaneous (HI) SSV time series were compared to the spatial patterns from the HYCOM 1994–2018 climatology (HC) by correlating both the SSV angle and the SSV magnitude at each grid cell. The magnitude correlation was always above 0.7 (Figure S3); thus, the angle correlation between SSV vectors was used to identify the spatial pattern of the climatology that corresponded most closely to the instantaneous time series SOMs (Figure S4). The climatological pattern number 5 has a null frequency, i.e., it rarely happens; thus, it was ignored. Except for HC 5, each climatological spatial pattern has one clear maximum correlation with one of the instantaneous spatial patterns. The correspondence between HC and HI patterns according to the highest angle correlation is shown in Table S1. Using the correspondence in Table S1 to rename the HI spatial patterns so that the names of the two patterns that correlate the highest match, we can compare the temporal evolution of matching spatial patterns according to BMUs (Figure S5). Each year between 1994 and 2018, the period over which we have a temporal evolution of instantaneous spatial patterns, the HI BMUs are similar to the temporal evolution of the climatological spatial patterns, i.e., the HC BMUs. Starting in 2016, and especially in 2017, HI and HC BMUs differ more than in most previous years. It is unclear if the ocean dynamics were different these years, or if the different model setup may be partly responsible; data from 1994 to 2015 is from the HYCOM GOFS 3.1 reanalysis, while the velocity starting in 2016 comes from HYCOM GOFS 3.1 analysis.