Deep embedded clustering of chlorophyll-a reveals distinct seasonal regimes in the Beibu Gulf, China

The concentration of chlorophyll-a (Chl-a) in nearshore waters is a pivotal indicator of marine ecosystem health and productivity. However, the spatial heterogeneity of Chl-a across coastal–offshore gradients and the dominant driving mechanisms at sub-regional scales remain poorly quantified, partly due to the reliance on predefined geomorphological boundaries. This study aims to objectively partition the Beibu Gulf (Guangxi) into functionally homogeneous sub-regions and to clarify the spatiotemporal variability and driving factors of Chl-a within each zone. MODIS-derived Chl-a data, combined with in situ observations, were used to conduct regionalization based on Deep Embedded Clustering (DEC), enabling data-driven delineation of On-shore, Mid-shore, and Off-shore areas. The spatiotemporal patterns and seasonal dynamics of Chl-a in each partition were analyzed using the Mann–Kendall trend test and global spatial autocorrelation (Moran’s I), while a geographic detector model and a self-organizing map (SOM) neural network were applied to identify dominant environmental drivers and their nonlinear influences. The results demonstrate that DEC-derived sub-regions exhibit higher internal similarity and lower dispersion of Chl-a variability in nearshore waters compared with conventional contour-line and buffer-zone divisions, supporting the ecological rationality of the proposed functional zonation. Distinct sub-regional phenological patterns were observed, with stronger seasonal and interannual variability nearshore and more stable but differentiated regimes offshore. The DEC-derived On-shore region shows a much lower Chl-a coefficient of variation (CV = 0.292) than contour line (CV = 0.881) and buffer zone (CV = 0.668) divisions, indicating higher internal homogeneity. Driver analysis reveals nearshore Chl-a variability is closely linked to terrestrial factors (wetlands, urban areas), while offshore dynamics are primarily driven by climatic and oceanographic processes, with sea surface salinity (q = 0.641) and temperature (q = 0.587) as dominant drivers, where q quantifies each factor’s contribution to Chl-a variance (0-1 scale). Overall, the results demonstrate that Chl-a spatial heterogeneity in the Beibu Gulf arises from the coupled effects of terrestrial inputs and climate-driven processes, and that DEC-based functional regionalization provides a more robust framework for interpreting Chl-a dynamics, offering a scientific basis for zoned marine management and climate-change adaptation.