Differences in bloom phenology and seasonal chlorophyll-a variability in the Fram Strait's hydrographic subregions resolved by a machine learning approach

Highlights: • Machine learning separates Fram Strait into 3 hydrographical sectors • Blooms (Chl-a > 1 mg m-3) occur earlier in Arctic and later in Atlantic sector • Normalised Chl-a variability in 2018 is higher in the Arctic sector • Phenological differences between sectors are robust during years 2016–2021 Abstract The Fram Strait (FS) exhibits pronounced variability in chlorophyll-a (Chl-a) concentration on intra- and inter-annual scales, which is linked to the complex hydrography of the region, marked by two opposing currents separated by a frontal system. Here, we propose a machine learning approach that partitions the FS into three hydrographical subdomains (sectors) based on remotely sensed sea surface temperature, the main property for water mass identification besides salinity. The sectors resolve the cold Arctic, warm Atlantic, and intermediate front-influenced regions. This sectorisation allows for a nuanced analysis of variability, accounting for the dynamic behaviour of the currents and frontal system. Differences revealed in bloom phenology are reflected in remote-sensing Chl-a concentrations within these hydrographical sectors. Blooms (Chl-a > 1 mg m-3) tend to occur in May in the Arctic-influenced sector, which is subject to variable sea ice cover. They appear later, in July, in the Atlantic-influenced sector. The absolute magnitude of Chl-a variability increases with its average concentration. However, concentration-normalised, the variability is notably higher in the Arctic-influenced sector, emphasising the impact of environmental differences within the hydrographic regimes on phytoplankton dynamics. Between 2016 and 2021, we could not detect clear trends in the sectors’ average Chl-a concentration or major changes in bloom timing. This study contributes to a broader comprehension of the expected variability ranges of Chl-a in the Fram Strait. Furthermore, it provides a valuable tool for analysing large geospatial data sets, especially when in-situ measurements of the physical seawater properties are limited in spatial and temporal coverage.