Coastal Phytoplankton Bloom Trends and Climate Change

To assess the impacts of climate change on long-term trends in coastal phytoplankton blooms, we correlated the annual mean bloom frequency and the associated SST and SST gradient in various coastal current systems for grid cells with significant changes in bloom frequency (Fig. 3c). The SST and SST gradient were averaged over the growth window within a year, assuming that the changes within the growth window, either in water temperatures or ocean circulations, play more important roles in the bloom trends compared to other seasons^{32 (link)}.
We determined the growth window of phytoplankton blooms for each 1° × 1° grid cell (Extended Data Fig. 9a) using the following method: first, we estimated the proportion of cumulative bloom-affected pixels within the grid cells for a year. Second, a generalized additive model⁷² was used to determine the shape of the phenological curves (Extended Data Fig. 9b), where a log link function and a cubic cyclic regression spline smoother were applied^{73 ,74 (link)}. Third, the timing of maximum bloom-affected areas (TMBAA) was then determined by identifying the inflection point on the bloom growth curve (Extended Data Fig. 9c). To facilitate comparisons across Northern and Southern Hemispheres, the year in the Southern Hemisphere was shifted forward by 183 days (Extended Data Fig. 9c). We characterized the similarity of the bloom growth curve between different grid cells and grouped them into three distinct clusters using a fuzzy c-means cluster analysis method^{75 ,76 (link)}. We found uniform distributions of the clusters over large geographic areas. Cluster I is mainly distributed in mid-low latitudes (<45° N and <30° S), where the maximum bloom-affected areas were expected in the early period of the year. Cluster II was mostly found in higher latitudes, with bloom developments (quasi-) synchronized with increases in SST. Cluster III was detected along the coastlines, where the bloom-affected areas increase throughout the entire year. In practice, the growth window for clusters I and III was set as the entire year, and that for cluster II was set from day 150 to day 270 within the year. We further found that the TMBAA for cluster II showed small changes over the entire period (Extended Data Fig. 9d), indicating relatively stable phenological cycles for those phytoplankton blooms^{32 (link),77 (link)}.