Decline Trends of Chlorophyll-a in the Yellow and Bohai Seas over 2005–2024 from Remote Sensing Reconstruction

Abstract

Chlorophyll-a (Chl-a) concentration is a key indicator of coastal ecosystem health, reflecting both primary productivity and the ecosystem’s response to climate change and human activities. This study quantifies long-term Chl-a trends in the Yellow and Bohai Seas using a multi-source remote sensing reconstruction dataset generated with deep learning algorithms. Quantile regression was applied to assess changes across the 75th, 50th, and 25th percentiles, and environmental drivers—including sea surface temperature, mixed layer depth, wind speed, and sea surface height anomalies—were evaluated in representative regions such as estuaries, aquaculture zones, and offshore waters. From 2005 to 2024, Chl-a concentrations declined across the 75th, 50th, and 25th percentiles, with rates of −4.82 × 10⁻³, −4.50 × 10⁻³, and −4.09 × 10⁻³ mg·m⁻³·a⁻¹, respectively (where “a” denotes year). The decline also showed strong seasonal differences, with summer decreases (−0.0638 mg·m⁻³·a⁻¹) substantially greater than winter (−0.04 mg·m⁻³·a⁻¹). Spatially, the decline was more pronounced in high-concentration nearshore waters, with rates of −0.0283 mg·m⁻³·a⁻¹ in the Qinhuangdao region, compared to −0.0137 mg·m⁻³·a⁻¹ in deeper offshore waters. Mixed-layer depth and wind speed emerged as the primary physical controls, with nearshore declines driven by enhanced vertical mixing and offshore changes dominated by mesoscale oceanic processes. These findings provide new insights for modeling and managing coastal ecosystems under combined climate and anthropogenic pressures.

1. Introduction

Chlorophyll-a (Chl-a), as a core indicator of phytoplankton biomass, serves not only as a key parameter for assessing water eutrophication levels but also influences marine food web structures and biogeochemical cycling processes by regulating primary productivity [1,2]. By analyzing the concentration levels of Chl-a within marine ecosystems, we can investigate the biogeochemistry of these ecosystems [3]. Examining Chl-a variations and their relationship with environmental factors enables us to respond promptly to environmental changes caused by human development and other factors [4].

The Yellow and Bohai Seas, as a significant semi-enclosed continental shelf region in northern China, fulfill multiple functions including supplying fishery resources, providing ecological barrier protection, and supporting coastal economies. The stability of their ecosystems directly impacts regional sustainable development and ecological security. The Yellow and Bohai Sea region lies within the East Asian monsoon zone, characterized by a complex natural environment: prevailing southeasterly and southerly winds dominate in summer, while northwesterly winds prevail in winter. The surface current field exhibits wind-driven characteristics, with surface currents primarily influenced by wind forces. The region primarily develops coastal water masses, a central cold-water mass, and a high-salinity warm water mass in the southern Yellow Sea [5]. The semi-enclosed topography and unique hydrodynamic conditions render this region particularly sensitive to climate change and anthropogenic disturbances. In recent decades, driven by dual pressures—climate stressors such as sea surface warming and abnormal wind speeds, alongside human activities including nutrient inputs from land-based sources, fishing, and coastal aquaculture—the issue of marine eutrophication has become increasingly pronounce [6,7]. Consequently, Chl-a concentrations in the Yellow and Bohai Seas exhibit complex spatiotemporal dynamics.

Research into Chl-a concentrations in the Yellow and Bohai Seas commenced in the 1960 [8]. Research methodologies have progressively evolved from traditional in situ observations [8,9] to remote sensing monitoring [10], with monitoring products becoming increasingly diverse and their accuracy steadily improving [11]. In trend studies, scholars have produced numerous findings based on diverse satellite data sources and analytical methodologies [11,12,13,14,15]: Kiyomoto et al. utilized CZCS (Coastal Zone Colour Scanner) and OCTS (Ocean Colour and Temperature Scanner) data to reveal springtime algal blooms in the East China Sea [16]; Qian Li et al. employed MODIS Level 3 inversion data to identify a gradual increase in Chl-a concentrations in the Bohai Sea between 2002 and 2009 [17]. Meng Qinghui et al. employed the OC3 algorithm to invert observed data, proposing a V-shaped variation pattern for Chl-a concentrations in the Bohai Sea [18]. Zhai et al. utilized MODIS-Aqua data to indicate a declining trend in Chl-a concentrations within the Bohai Sea and northern Yellow Sea from 2012 to 2018, hypothesizing a correlation with climate change [19]. In spatial and driving mechanism research, Ma Aohui, Tang, and Tian Hongzhen et al. employed multi-source remote sensing data alongside regional analysis and modal decomposition techniques to elucidate the spatiotemporal variations in Chl-a concentrations across different bathymetric zones of the Yellow and Bohai Sea [20]. They further examined the influence of environmental factors on these concentration [21,22]. Mamum et al. [23] and Abbas et al. [24] found that monsoons dominate algal growth in the basin. Even minor variations in water temperature and wind can exert considerable environmental impacts on the ecosystem [25].

However, satellite remote sensing data are prone to significant gaps due to cloud interference. Although studies have employed multivariate interpolation methods such as DINEOF for data reconstruction [26,27,28,29], existing research exhibits notable limitations: firstly, reliance on traditional methods like linear regression can only reveal overall long-term trends in Chl-a [30], struggling to characterize differentiated variations across different concentration levels; Secondly, insufficient attention is paid to the coupled effects of environmental factors and the heterogeneity of driving mechanisms along the “nearshore-offshore” gradient. Furthermore, some studies employ relatively short time series, making it difficult to reflect long-term dynamic patterns.

To address the aforementioned research gap, this study utilized Chl-a concentrations from the Yellow and Bohai Seas (117–127° E, 31–41° N) as its subject matter. By employing deep learning algorithms, it constructs a multi-source remote sensing reconstruction dataset for Chl-a concentration. For the first time, it introduces quantile regression methods to systematically analyses long-term trends in Chl-a at the 75th, 50th, and 25th quantile levels, along with their seasonal and spatial heterogeneity. Concurrently, utilizing univariate linear regression models, this study quantitatively assessed the independent contributions and interactive effects of sea surface temperature (SST), mixed layer depth (MLD), wind speed, and sea level anomaly (SLA). To elucidate regional variations in driving mechanisms, this study selected three representative study areas: a portion of the northern Yellow Sea (characterizing nearshore aquaculture zones), waters adjacent to Qinhuangdao (a nutrient input zone), and the deepwater southern Yellow Sea (an offshore zone) (Figure 1): By contrasting nearshore regions, the study discerns variations in human activity impacts; through comparisons between nearshore and deep-water zones, it reveals differences in the regulation of nearshore processes versus open-ocean dynamics. These findings provide scientific support for eutrophication management and ecosystem restoration in the Yellow and Bohai Seas, while also offering methodological references for multi-scale dynamic studies of nearshore ecosystems under climate change.

Figure 1. Geographical Location of the Yellow and Bohai Seas and Schematic Diagram of the Study Area.

Compared with previous regional studies, which primarily focused on mean or linear trends in Chl-a, the present study extends the physical interpretation of long-term Chl-a variability in the Yellow and Bohai Seas in two key aspects. First, the application of quantile regression allows for the identification of heterogeneous trend behaviors across low-, median-, and high-Chl-a regimes, providing insights that cannot be captured by conventional mean-based approaches. Second, the use of a deep learning–based reconstructed Chl-a dataset reduces data gaps and enhances the robustness of trend detection, particularly in regions affected by cloud contamination and coastal complexity. Together, these advances enable a more comprehensive understanding of the physical and environmental mechanisms governing Chl-a trends in both nearshore and offshore environments.

2. Data and Methods

2.1. Data

2.1.1. Construction of the Chl-a Reconstruction Dataset

Chl-a data is derived from the multi-source remote sensing dataset integrated by the European Space Agency’s GlobColour project (https://hermes.acri.fr/index.php?class=archive, accessed on 8 July 2025). This dataset combines satellite remote sensing data from MODIS, MERIS, VIIRS, SeaWiFS and other sources. However, due to factors such as cloud cover, the raw data exhibits an average pixel missing rate as high as 84%, rendering it unsuitable for long-term continuous spatiotemporal dynamic analysis. Therefore, to fill these data gaps, this study employs interpolation reconstruction based on convolutional autoencoders.

The DINCAE (Data-Interpolating Convolutional Auto-Encoder) algorithm is employed for data reconstruction. This algorithm, proposed by Barth et al. [31], integrates the feature extraction capabilities of convolutional neural networks with the unsupervised learning advantages of autoencoders. When processing image-type data, it preserves more small-scale feature information [31] and overcomes the limitations of traditional DINEOF methods in capturing spatiotemporal nonlinear relationships [30,31,32,33,34]. It has been proven to be an effective tool for reconstructing remote sensing data such as Chl-a.

The core workflow of DINCAE comprises three stages: preprocessing, training, and reconstruction (Figure 2). The preprocessing stage follows the standard procedure outlined by Barth et al. [31], which shall not be elaborated upon herein. The training stage employs a symmetric architecture comprising five encoding layers and five decoding layers: encoding layers alternate between convolutional and pooling layers, with the final convolutional layer followed by two fully connected layers to achieve a nonlinear combination of extracted features; the decoder comprises convolutional layers and nearest-neighbor interpolation layers, restoring spatial resolution through upsampling. Jump connections are established between the interpolation layers and the pooling layer outputs from the encoder to capture small-scale features lost during encoding.

Figure 2. DINCAE Flowchart.

The reconstructed Chl-a dataset was evaluated using the coefficient of determination (R²), root mean squared error (RMSE), and mean bias (Bias) (Figure 3). The reconstructed data exhibit a high level of agreement with the original valid observations, with an R² value of 0.9764, indicating that the reconstructed dataset captures most of the variability in the observed Chl-a fields. The slope of the fitted regression line is close to 1, suggesting that the reconstructed values generally follow the observations, although they are not identical. The RMSE (0.088 mg·m⁻³) and bias (−0.012 mg·m⁻³) further quantify the reconstruction errors, both of which are relatively small, indicating limited deviations between reconstructed and observed Chl-a values. These differences are mainly observed in high-concentration regions, such as nearshore estuaries.

Figure 3. Scatter plot density diagram of raw data versus DINCAE reconstructed data. Color scale denotes local density (%). The black dashed line represents the 1:1 line. The red solid line indicates the best-fit line.

As shown in Figure 4, the reconstruction was applied to daily Chl-a data from 2005 to 2024. The daily average missing rate was 26%, while the overall average missing rate across the entire study area reached 84%, primarily due to extensive spatial data gaps in offshore regions. To improve the reliability of the reconstructed dataset, days with missing rates greater than 98% were excluded from the analysis [31]. The final product comprises a Chl-a dataset for the Yellow and Bohai Seas, covering the temporal range from January 2005 to December 2024 and the spatial extent of 117–127° E, 31–41° N, with a uniform spatial resolution of 4 km × 4 km. The robustness of DINCAE in reconstructing Chl-a fields in both coastal and offshore environments has been extensively validated in previous studies [32,33].

Figure 4. Average Missing Rate in the Study Area.

2.1.2. Environment Variable Data

The environmental drivers selected for this study comprise Sea Surface Temperature (SST), Sea Level Anomaly (SLA), Sea Surface Wind (SSW), and Mixed Layer Depth (MLD). The data sources and fundamental parameters for each are as follows:

SST, SLA, SSW, and SH data were derived from satellite remote sensing products provided by the Copernicus Marine Environment Monitoring Service (CMEMS), covering the period from January 2005 to December 2024. The spatial resolution for SST is 0.05° × 0.05°, while SLA, SSW, and SH data feature a spatial resolution of 0.25° × 0.25°.

MLD data originate from the Hycom reanalysis dataset (https://orca.science.oregonstate.edu/1080.by.2160.monthly.hdf.mld030.hycom.php, accessed on 7 October 2024), featuring a spatial resolution of 1/12° and a temporal resolution of 8 days.

All environmental variable data were resampled to 4 km × 4 km using bilinear interpolation to align with the spatiotemporal resolution of the Chl-a dataset, ensuring the validity of subsequent correlation analyses.

2.2. Methods

QR

To refine the characterization of long-term trends in Chl-a across different concentration levels, this study employs quantile regression (QR) for trend analysis. This method, proposed by Bassett and Koenker in 1978 [35], constitutes an extension of classical least squares regression (LSM). Subsequently introduced into ecological research by Cade et al. [36], it has been demonstrated to effectively capture the heterogeneous effects of environmental factors on Chl-a, revealing detailed interactions between the two [37].

LSM focuses on fitting the mean trend of the dependent variable, constructing a linear model by minimizing the sum of squared errors between predicted and observed values:

$${\mathit{y}}_{\mathit{i}} = \mathit{k}{\mathit{t}}_{\mathit{i}}+\mathit{b}+{ℇ}_{\mathit{i}},$$

(1)

Here, $\mathit{k}$ denotes the regression coefficient, $\mathit{b}$ represents the intercept, and ${ℇ}_{\mathit{i}}$ signifies the error term. The dependent variable represented by ${\mathit{y}}_{\mathit{i}}$ exhibits a linear relationship with time ${\mathit{t}}_{\mathit{i}}$.

Unlike LSM, quantile regression simultaneously quantifies the trend of the dependent variable at different quantiles (τ) by minimizing the weighted absolute deviation and estimating regression coefficients, thereby fully capturing the spatiotemporal dynamics of the data distribution. For a given quantile τ, its objective function can be expressed as:

$$\mathit{M}\mathit{i}\mathit{n}\mathit{i}\mathit{m}\mathit{i}\mathit{z}\mathit{e}{\sum }_{\mathit{i}=1}^{\mathit{n}}{\mathit{\rho }}_{\mathit{\tau }}({\mathit{y}}_{\mathit{i}}-{\mathit{x}}_{\mathit{i}}\mathit{\beta }(\mathit{\tau }\left)\right),$$

(2)

where ${\mathit{\rho }}_{\mathit{\tau }}\left(\mathit{u}\right)$ is the check (pinball) loss function, defined as

$${\mathit{\rho }}_{\mathit{\tau }}\left(\mathit{u}\right)=\mathit{u}\left(\mathit{\tau }\right)-\mathit{\Pi } (\mathit{u}<0),$$

and $\mathit{\Pi } (\mathit{u}<0)$ is ts the indicator function, which equals 1 when $\mathit{u}<0$ is 1, and 0 otherwise.

Given the susceptibility of extreme quantiles to outlier interference, this study employs the 0.05–0.95 quantile range for analysis, focusing on the variation characteristics of the 75th, 50th, and 25th quantiles (corresponding to high, medium, and low concentration levels, respectively). The 75th quantile is defined as the high-concentration threshold for Chl-a, serving to distinguish response differences across varying concentration gradients.

This study employs the conventional meteorological seasonal classification for this sea area: March to May constitutes spring, June to August summer, September to November autumn, and December to February winter.

3. Results and Discussion

3.1. Long-Term Spatiotemporal Variation Characteristics of Chl-a Concentration

The annual average Chl-a concentration in the Yellow and Bohai Seas is 1.23 mg/m³, exhibiting significant spatial and temporal heterogeneity. Seasonally, Chl-a concentrations gradually increase from autumn onwards, peaking in winter (1.31 mg/m³). Summer Chl-a levels (1.07 mg/m³) are markedly lower than those in spring (1.22 mg/m³) and autumn (1.23 mg/m³) (Figure 5). Spatially, Chl-a concentrations in the Bohai Sea were generally higher than in the Yellow Sea, with high-value zones concentrated along the coast of Qinhuangdao (corresponding to the QHD characteristic zone) (1.69 mg/m³), southern Laizhou Bay, northern Liaodong Bay, and the northern Yellow Sea coast (corresponding to the characteristic zone NYS) (1.53 mg/m³). The average concentration in the Yellow Sea was only 1.09 mg/m³, remaining at a relatively low level overall.

Figure 5. Seasonal distribution of Chl-a concentration in the study region. Boxes indicate the interquartile range (25th–75th percentiles), the central line denotes the median, whiskers extend to 1.5 times the interquartile range, and circles represent outliers.

3.1.1. Long-Term Trend

Quantitative analysis of long-term trends in Chl-a concentrations was conducted using least squares method (LSM) and quantile regression (QR). Results indicate a significant decline in c Chl-a concentrations across the entire sea area, with the LSM-fitted annual decline rate at −0.0182 mg·m⁻³·a⁻¹ (Figure 6).

Figure 6. Spatial distributions of mean Chlorophyll-a (Chl-a) concentrations in the Yellow and Bohai Seas for the annual (a), spring (b), summer (c), autumn (d), and winter (e) periods during 2005–2024. Chl-a concentrations are expressed in mg·m⁻³.

The quantile regression results further reveal distinct decline characteristics across different concentration levels: the annual decline rates corresponding to the 75th, 50th, and 25th quantiles were −4.82 × 10⁻³ mg·m⁻³·a⁻¹, −4.50 × 10⁻³ mg·m⁻³·a⁻¹, and −4.09 × 10⁻³ mg·m⁻³·a⁻¹ respectively. The decline trend at the 75th percentile (high concentration level) closely approximates the LSM linear trend (−4.50 × 10⁻³ mg·m⁻³·a⁻¹). Declines were more pronounced in high-percentile sea areas (e.g., Bohai Strait, NYS, and QHD), with spatial heterogeneity in trends being particularly evident in high-latitude regions (e.g., Bohai Sea and NYS) (Figure 7).

Figure 7. Annual trends in Chl-a concentration for the linear trend (a), 75th percentile (b), 50th percentile (c), and 25th percentile (d) using the classical least squares method (LSM).

Comparative analysis of Chl-a concentration changes across four major representative regions—the Bohai Sea and Yellow Sea (BSYS), QHD, NYS, and YSDA—reveals markedly divergent decline trends between areas. The rate of Chl-a decrease intensifies with increasing percentile rank [38], and therefore implies that regions or periods characterized by higher productivity are more sensitive to long-term environmental changes. The annual decline rates in LSM were most pronounced in QHD and NYS, reaching −0.0268 mg·m⁻³·a⁻¹ and −0.0315 mg·m⁻³·a⁻¹ respectively. This indicates a gradual shift in the characteristics of coastal waters influenced by human activities, which historically exhibited phytoplankton proliferation typical of eutrophicated bodies. Within the QHD region, a pronounced overall decline was observed, with a faster rate than other areas. This acceleration was particularly evident at high percentiles (>75%), where the annual decline exceeded −0.03 mg·m⁻³·a⁻¹ when the percentile surpassed 80% (Figure 8). This phenomenon may be associated with regulatory measures implemented in Qinhuangdao, a major coastal city and tourist destination. Strict land-based pollution control policies introduced in recent years have effectively reduced nutrient inputs into the sea, thereby curbing phytoplankton growth at its source [39,40].

Figure 8. Trends in Chl-a concentrations corresponding to the 5th to 95th percentiles throughout the year in the Bohai and Yellow Sea (BSYS), Qinhuangdao waters (QHD), northern Yellow Sea (NYS), and Yellow Sea deep area (YSDA). Trend values are expressed in mg·m⁻³·a⁻¹. Vertical bars indicate the standard error of each regression.

3.1.2. Seasonal-Scale Variation Characteristics

To investigate the significant differences in Chl-a concentrations between winter (December to February) and summer (June to August), quantile regression analysis was conducted to further clarify variations within the seasonal scale:

The results indicate that the LSM fitting result for winter was −0.04 mg·m⁻³·a⁻¹. The trends across all quantiles were generally consistent but varied in magnitude, with the decline rates corresponding to the 75th, 50th, and 25th quantiles being −0.0475 mg·m⁻³·a⁻¹, −0.0303 mg·m⁻³·a⁻¹, and −0.026 mg·m⁻³·a⁻¹, respectively (Figure 9). The decline trend intensified with increasing percentile, where higher percentiles correspond to higher Chl-a concentration levels, indicating more pronounced reductions in winter Chl-a within high-concentration zones (Figure 10). Notably, the high-percentile Chl-a decline trend was particularly pronounced in waters near the Gulf of Korea, suggesting this region is significantly impacted by dual influences of anthropogenic disturbances (aquaculture, industrial discharges, agricultural runoff, and other human activities) and climate change [40].

Figure 9. Winter Chl-a concentration trends: (a) Linear trend using the Least Squares Method (LSM), (b) 75th percentile trend, (c) 50th percentile trend, (d) 25th percentile trend.

Figure 10. Trends in Chl-a concentrations corresponding to the 5th to 95th percentiles in the Bohai Sea and Yellow Sea (BSYS), Qinhuangdao waters (QHD), northern Yellow Sea (NYS), and Yellow Sea deep area (YSDA) during winter. Trend values are expressed in mg·m⁻³·a⁻¹.Vertical bars indicate the standard error of each regression.

The annual decline rate of summer LSM was −0.0638 mg·m⁻³·a⁻¹, significantly higher than the annual average and winter values. The 75th, 50th, and 25th percentile trends were −0.0690 mg·m⁻³·a⁻¹, −0.0456 mg·m⁻³·a⁻¹, and −0.0375 mg·m⁻³·a⁻¹, respectively (Figure 11). As the percentile increases, the decreasing trend becomes more pronounced, and this pattern also holds in summer (Figure 12). The most pronounced decline was observed near the Yangtze River estuary. Key contributing factors include recent upstream Three Gorges Dam flood storage and dry season discharge operations, which have resulted in lower water temperatures and reduced water transparency during the dry season, thereby inhibiting phytoplankton growth. Concurrently, flood-season flow regulation has diminished nutrient inputs. These dual effects collectively contribute to the interannual decline in Chl-a concentrations within this region [41].

Figure 11. Trends in summer Chl-a concentrations: (a) Linear trend using the least squares method (LSM), (b) 75th percentile trend, (c) 50th percentile trend, (d) 25th percentile trend.

Figure 12. Trends in Chl-a concentrations corresponding to the 5th to 95th percentiles in the Bohai Sea and Yellow Sea (BSYS), Qinhuangdao waters (QHD), northern Yellow Sea (NYS), and Yellow Sea deep area (YSDA) during summer. Trend values are expressed in mg·m⁻³·a⁻¹.Vertical bars indicate the standard error of each regression.

3.1.3. Comparative Analysis of Representative Sea Areas

Comparative analysis of Chl-a concentration changes across four representative regions—the Bohai Sea and Yellow Sea as a whole (BSYS), QHD, NYS, and YS-DA—reveals a general pattern in which higher quantiles, corresponding to higher Chl-a concentration levels, tend to exhibit stronger declining trends, although the magnitude and underlying mechanisms of this relationship vary among regions. Moreover, the annual decline rates corresponding to each quantile did not exceed −0.05 mg·m⁻³·a⁻¹ [37], confirming the core characteristic of more pronounced declines in high-value Chl-a concentration zones within the Yellow and Bohai Seas. Concurrently, heterogeneous differences exist in the decline trends across regions: the annual LSM decline rates in the nearshore waters of QHD and NYS were most pronounced, reaching −0.0268 mg·m⁻³·a⁻¹ and −0.0315 mg·m⁻³·a⁻¹ respectively. Within the QHD region, a pronounced overall decline was observed, with a faster rate than other areas. This acceleration was particularly evident at higher percentiles (>75%), where the annual decline exceeded −0.03 mg·m⁻³·a⁻¹ when the percentile surpassed 80%. This phenomenon is closely linked to the control measures implemented in Qinhuangdao, a significant coastal city and tourist destination. The stringent land-based pollution control policies enforced in recent years have effectively reduced nutrient inputs into the sea, thereby curbing phytoplankton growth at its source. It should be noted that the influence of nutrient input reduction policies is inferred based on the observed Chl-a trends and regional management measures; however, this effect has not been quantitatively assessed and should be interpreted with appropriate caution.

3.2. Analysis of the Driving Mechanisms of Chl-a Concentration Variations by Relevant Environmental Factors

The Yellow and Bohai Seas, as semi-enclosed shelf seas, experience strong land–sea interactions and abundant nutrient supply due to the combined influence of the East Asian monsoon, multiple river inputs, and oceanic processes such as the Yellow Sea Warm Current and cold water masse [8,9]. Previous studies have shown that the interannual variability of Chl-a concentration reflects the combined influence of multiple environmental factors, which together explain only part of its variability, rather than being driven by a single dominant factor [42]. Building upon the established long-term decline in Chl-a concentrations and its pronounced spatiotemporal heterogeneity, this chapter focuses on four key environmental factors: sea surface temperature (SST) (Figure 13 and Figure 14), mixed layer depth (MLD) (Figure 15), wind speed (Figure 16), and sea level anomaly (SLA) (Figure 17). Through Pearson correlation analysis and regional comparisons, we systematically elucidate the regulatory roles of these factors on Chl-a variations across the entire Yellow and Bohai Seas and their typical subregions (Northern Yellow Sea NYS, Qinhuangdao coastal QHD, Southern Yellow Sea deep area YSDA), elucidating the regional differentiation of driving mechanisms (Table 1).

Figure 13. 75th percentile SST trends for the whole year (a), winter (December to February) (b), and summer (June to August) (c).

Figure 14. Trends in the 5th–95th percentiles of annual, winter, and summer SST in the Bohai and Yellow Seas. Vertical bars indicate the standard error of each regression.

Figure 15. 75th percentile MLD trends throughout the year (a), winter (December to February) (b), and summer (June to August) (c).

Figure 16. 75th percentile wind speed trends throughout the year (a), winter (December to February) (b), and summer (June to August) (c).

Figure 17. 75th percentile SLA trends throughout the year (a), winter (December to February) (b), and summer (June to August) (c).

Table 1. Correlation coefficients between Chl-a and other environmental factors in the Bohai Sea and Yellow Sea (BSYS), Qinhuangdao Sea Area (QHD), Northern Yellow Sea (NYH), and the Yellow Sea Deep Area (YSDA) during summer.

	SST	Windspeed	MLD	SLA
BSYS	−0.2211	−0.1667	0.5226	0.1877
NYS	−0.0875	−0.3190	−0.4006	−0.0651
QHD	−0.1987	−0.5166	−0.9831	0.8955
YSDA	−0.0446	−0.1231	−0.0264	0.2553

3.2.1. SST

Against the backdrop of global warming, the study area exhibited a pronounced overall upward trend in SST from 2005 to 2024, with marked seasonal variations: the average warming rate at the 75th percentile during summer (0.3314 °C/a) exceeded that of winter (0.2709 °C/a, correlating with the more pronounced summer decline in Chl-a (Figure 6).

Across the entire marine domain, SST exhibits a weak negative correlation with Chl-a (r = −0.2211), indicating that ocean warming generally exerts a certain inhibitory effect on phytoplankton growth. Previous studies have shown that elevated temperatures can reduce the solubility and availability of key nutrients (e.g., dissolved oxygen, silicate, and phosphate) in the water column, and may also exceed the optimal thermal range for phytoplankton growth, thereby suppressing photosynthesis and biomass accumulation [43].

Regarding regional differentiation characteristics, SST exhibited extremely weak correlations with Chl-a variations in the Northern Yellow Sea (NYS) and the deep southern Yellow Sea (YSDA) (r = 0.0875, r = −0.0446), indicating limited direct regulation of Chl-a by SST in these regions. Concentration variations are primarily governed by other factors such as monsoon-driven water mixing and terrestrial nutrient inputs. SST may indirectly influence phytoplankton growth through coupled interactions with factors like mixed layer depth (MLD) and wind speed [43]. In contrast, the negative correlation between SST and QHD (r = −0.1987) is more pronounced, reflecting greater phytoplankton sensitivity to temperature fluctuations in this area. Warming-induced reductions in nutrient supply and deteriorating growth conditions further exacerbate the decline in Chl-a concentrations [44].

3.2.2. MLD

Across the entire marine area, MLD exhibits the strongest correlation with Chl-a (r = −0.5226), making it the key physical factor influencing the spatiotemporal variation in Chl-a. The core of this regulatory mechanism lies in the fact that mixed layer depth directly determines phytoplankton’s light exposure conditions and nutrient availability: excessive mixed layer depth removes phytoplankton from the photic zone, reducing average light exposure levels; simultaneously, enhanced vertical mixing dilutes surface nutrient concentrations. These dual effects collectively inhibit phytoplankton growth, particularly in certain groups such as picoplankton, which are more sensitive to light limitation and mixing intensity [45].

In distinct characteristic regions, the differential effects of MLD manifest as follows: MLD exhibits a significant negative correlation with Chl-a in the Northern Yellow Sea (NYS) (r = −0.4006). Considering the monsoon characteristics of this region, the weakening of winter wind speeds results in a shallower mixed layer. While this reduces vertical diffusion of phytoplankton, it also constrains the upwelling supply of nutrients from the bottom layer, ultimately manifesting as an inhibitory effect on Chl-a. In the Qinhuangdao coastal area (QHD), an extremely strong negative correlation (r = −0.9831) highlights MLD as the primary regulatory factor governing Chl-a variations in this region. Given the shallow water depth in this area, minor variations in mixed layer depth significantly alter surface water stability. Compounded by high nutrient backgrounds from land-based inputs, deepening MLD disrupts stratification and dilutes phytoplankton concentrations, becoming a key driver of declining Chl-a concentrations. In the southern deep waters of the Yellow Sea (YSDA), the correlation between MLD and Chl-a is weaker, indicating that this region is more significantly influenced by oceanic dynamic processes (such as cold-water mass fluctuations). The regulatory role of mixed layer depth is mitigated by other factors, with Chl-a being indirectly affected primarily through interactions with water mass exchange.

3.2.3. Wind

Wind speed serves as a primary dynamic factor regulating vertical mixing in the upper ocean and sea surface temperature. The mean wind speed in the Yellow and Bohai Seas during 2005–2024 was 4.4625 m/s. Previous studies have reported that Chl-a concentrations tend to remain relatively high under low wind conditions [46], while higher wind speeds (exceeding approximately 2 m/s) are generally associated with reduced Chl-a levels due to enhanced vertical mixing [47]. At the pan-oceanic scale, both wind speed and wind speed-derived Chl-a exhibit an inhibitory effect on Chl-a (r = −0.1667). This is primarily due to enhanced wind speed promoting vertical mixing and surface evaporation cooling. This dual action reduces Chl-a concentrations by lowering SST and displacing phytoplankton from the photic zone.

In the nearshore waters of the northern Yellow Sea (NYS), wind speed exhibits a negative correlation with Chl-a (r = −0.319), with pronounced seasonal variation. Winter wind speed reduction leads to shallower mixed layers and insufficient nutrient upwelling, inhibiting phytoplankton growth. In summer, even a slight increase in wind speed may induce re-mixing processes, partially mitigating the decline in Chl-a. Conversely, in the Qinhuangdao coastal area (QHD), the correlation between the two is exceptionally strong (r = −0.5166), with the inhibitory effect of wind speed being particularly pronounced. This region’s shallow waters and significant terrestrial inputs mean stronger winds readily disrupt local water stability, diluting or vertically dispersing phytoplankton and markedly reducing Chl-a concentrations. This aligns with the strong negative correlation between MLD and Chl-a observed here. Compared to the aforementioned zones, wind-driven mixing in the deep southern Yellow Sea (YSDA) exerts limited direct regulation on Chl-a (r = −0.1231), more likely influencing phytoplankton growth conditions indirectly through coupled interactions with cold water mass activity and MLD variations.

3.2.4. SLA

From 2005 to 2024, the study area exhibited an overall upward trend in Sea Level Anomaly (SLA). Spatially, the Yellow Sea showed a higher rate of increase (0.0048 m/a) than the Bohai Sea (0.0042 m/a), while the central sea area (0.006 m/a) increased faster than the coastal sea area (0.0024 m/a). Seasonally, the rise in sea surface height (SSH) was greater during summer than winter (Figure 10). The spatiotemporal variations in SLA fundamentally reflect the combined effects of regional water mass movements, sea level changes, and oceanic dynamic processes.

Across the entire marine domain, SLA exhibits a positive correlation with Chl-a (r = 0.1877). The core mechanism involves sea-level change indirectly regulating phytoplankton growth by influencing local water transport, nutrient redistribution, and upwelling intensity. However, regional differentiation is markedly pronounced: SLA is the most strongly correlated factor in the Qinhuangdao coastal area (QHD) (r = 0.8955), where sea-level rise enhances water exchange capacity and nutrient redistribution efficiency, thereby creating favorable conditions for phytoplankton growth. This is consistent with previous findings that Chl-a variability in QHD is jointly regulated by terrestrial nutrient inputs from river discharge and coastal runoff and by local hydrodynamic processes, which can be modulated by sea-level-related changes in water exchange and nutrient redistribution [39,48]. The stimulatory effect of rising SLA on Chl-a in the southern deep waters of the Yellow Sea (YSDA) is relatively weak (r = 0.2553), with its influence primarily mediated indirectly through interactions with water mass exchange and MLD variations, reflecting the complexity of open-ocean dynamics. Sea-level rise exerts an extremely weak influence on the northern Yellow Sea nearshore region (NYS). Previous studies suggest that phytoplankton dynamics in this area are primarily controlled by coastal upwelling intensity and nutrient availability; thus, weakened upwelling and insufficient nutrient supply may exert a slight inhibitory effect on Chl-a, while the direct regulatory role of SLA remains limited [49,50].

Overall, the long-term decline in Chl-a concentrations across the Yellow and Bohai Seas represents the synergistic effect of four primary factors: MLD, wind, SST, and SLA. Moreover, the driving mechanisms exhibit significant regional heterogeneity:

• Across the entire marine domain, MLD exhibits the strongest association with Chl-a variability (r = −0.5226), suggesting that vertical mixing processes play an important role in regulating phytoplankton biomass. Wind speed (r = −0.1667) and SST (r = −0.2211) are also negatively correlated with Chl-a, potentially reflecting enhanced vertical diffusion and increased thermal stress combined with nutrient limitation, respectively. In contrast, SLA shows a weak positive correlation with Chl-a (r = 0.1877), implying that oceanic dynamic processes associated with sea-level variability—such as water transport and nutrient redistribution—may exert a modest stimulatory influence on phytoplankton growth.
• In the Qinhuangdao coastal area (QHD), a typical region characterized by strong anthropogenic–natural coupling, Chl-a variability shows particularly strong correlations with local physical conditions. Specifically, Chl-a is extremely negatively correlated with MLD (r = −0.9831) and significantly negatively correlated with wind speed (r = −0.5166), suggesting a strong sensitivity to changes in water column stability and vertical mixing. Given the shallow water depth in this region, even relatively small variations in physical forcing may substantially influence Chl-a concentrations by modifying water stability and phytoplankton dilution processes. Meanwhile, the strong positive correlation with SLA (r = 0.8955) further suggests that sea-level variability may play an important role in regulating water exchange and nutrient supply in such complex nearshore ecosystems.
• In the offshore deep-sea area (YSDA), correlations between Chl-a and individual environmental factors are generally weak. This suggests that Chl-a variability in this region is more likely influenced by large-scale oceanic dynamic processes, such as cold water masses and water mass exchange, as well as the coupled effects of multiple factors, rather than by any single local driver, consistent with previous studies [51].
• The North Yellow Sea (NYS), as a transitional zone, exhibits driving characteristics intermediate between coastal and offshore regions. Chl-a variability in this region is more strongly associated with monsoon-regulated wind speed (r = −0.319) and MLD (r = −0.4006). Reduced winter monsoon intensity may lead to a shallower mixed layer and weakened nutrient entrainment, which together could limit phytoplankton growth in this region.

4. Conclusions

This study reconstructed Chl-a concentration datasets for the Yellow and Bohai Seas from 2005 to 2024 using the DINCAE algorithm. By combining quantile regression with linear regression, it systematically examined the spatiotemporal distribution, long-term trends, and key environmental drivers of Chl-a variability in the region. The main conclusions are as follows:

Chl-a concentrations exhibit a long-term decline with clear concentration dependence and pronounced spatiotemporal heterogeneity. From 2005 to 2024, Chl-a levels showed a marked overall decrease, with an average annual rate of −0.0182 mg·m⁻³·a⁻¹. Seasonally, the decline was strongest in summer (−0.0638 mg·m⁻³·a⁻¹) and more moderate in winter (−0.04 mg·m⁻³·a⁻¹). Quantile regression further revealed that high-concentration waters (75th percentile and above) experienced substantially greater reductions (−4.82 × 10⁻³ mg·m⁻³·a⁻¹) than low-concentration waters (25th percentile: −4.09 × 10⁻³ mg·m⁻³·a⁻¹), highlighting the concentration-dependent nature of the decline. Spatial patterns showed a clear gradient, with stronger declines nearshore and weaker declines offshore. High-concentration zones along the northern Yellow Sea coast and around the Gulf of Korea exhibited especially pronounced reductions at higher quantiles. Nearshore Qinhuangdao, where local hydrodynamic disturbances and land-derived inputs play a major role, recorded a decline rate (−0.0268 mg·m⁻³·a⁻¹) exceeding the regional average. In contrast, the deep-water southern Yellow Sea showed only slight decreases, largely governed by basin-scale oceanic processes rather than local forcings.

The influence of environmental factors on Chl-a exhibits clear regional differentiation, with mixed layer depth (MLD) emerging as one of the dominant drivers. At the basin scale, MLD shows the strongest correlation with Chl-a (r = −0.5226) and thus acts as the dominant physical factor regulating its variability. By altering vertical mixing, MLD affects both light availability and nutrient supply, ultimately contributing to the observed decline in Chl-a concentrations. Sea surface wind speed also shows a significant negative correlation with Chl-a (r = −0.1667). Strong winds enhance vertical mixing and reduce light penetration, thereby inhibiting phytoplankton growth. Sea surface temperature (SST) displays a long-term warming trend, with a faster increase in summer (0.3314 °C/a) than in winter (0.2709 °C/a). SST is weakly negatively correlated with Chl-a overall (r = −0.2211), with an even stronger negative relationship in nearshore regions such as Qinhuangdao (r = −0.1987). Rising temperatures can reduce nutrient solubility and push conditions beyond the optimal thermal range for phytoplankton growth, suppressing biological productivity. In contrast, sea level anomaly (SLA) shows a positive correlation with Chl-a (r = 0.1877), which is more pronounced in the Qinhuangdao coastal area (r = 0.8955). SLA affects Chl-a by modulating local water transport and nutrient redistribution. However, in the deep waters of the southern Yellow Sea, Chl-a dynamics are governed primarily by mesoscale processes—including eddies and water mass exchange—rather than by local physical forcing.

Regarding the mechanisms driving nearshore–offshore differences, the study reveals clear spatial contrasts. In nearshore areas such as Qinhuangdao, strong wind–wave disturbance and the deepening of the mixed layer enhance vertical mixing, leading to light limitation and sediment resuspension. These processes tend to suppress phytoplankton growth and may accelerate the decline in Chl-a. In contrast, phytoplankton dynamics in open waters are regulated mainly by mesoscale processes—such as fronts and eddies—while the direct effects of local physical forcing are comparatively weak.

This study effectively addressed gaps in satellite remote-sensing data using the DINCAE algorithm and demonstrated the value of quantile regression for analyzing Chl-a trends across different concentration levels. Together, these methods offer a robust framework for multidimensional analyses of coastal ecosystem dynamics. The identified “nearshore–offshore” gradient in Chl-a decline, along with the region-specific driving mechanisms, provides actionable insights for ecological management in the Yellow and Bohai Seas. Nearshore zones require continued strengthening of land-based pollution control and more effective regulation of aquaculture activities. In contrast, offshore regions demand close monitoring of changes in oceanic dynamic processes linked to climate variability. Furthermore, the more pronounced decline in high-concentration areas suggests that recent eutrophication control measures may have contributed to the observed changes, offering scientific support for ongoing ecosystem restoration and sustainable development in the region.