Next Article in Journal
Monitoring Policy-Driven Urban Restructuring and Logistics Agglomeration in Zhengzhou Through Multi-Source Remote Sensing: An NTL-POI Integrated Spatiotemporal Analysis
Previous Article in Journal
Impact of Assimilating FY-4B/GIIRS Radiances on Typhoon “Doksuri” and Typhoon “Gaemi” Forecasts
Previous Article in Special Issue
Spatiotemporal Dynamics of Total Suspended Solids in the Yellow River Estuary Under New Water-Sediment Regulation: Insights from Sentinel-3 OLCI
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Remote Sensing
Volume 17
Issue 17
10.3390/rs17173106
remotesensing-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Phenological Characteristics of the Yellow Sea Spring Bloom: A Comparative Evaluation of Multiple Diagnostic Methods

by
Kangjie Jin
1,
Chen Dong
1,
Xihan Liu
2,
Yan Sun
1,
Jibo Liu
1 and
Lei Lin
1,3,*
1
College of Ocean Science and Engineering, Shandong University of Science and Technology, Qingdao 266590, China
2
Hebei Technology Innovation Center for Geographic Information Application, Institute of Geographical Sciences, Hebei Academy of Sciences, Shijiazhuang 050011, China
3
State Key Laboratory of Estuarine and Coastal Research, East China Normal University, Shanghai 200241, China
*
Author to whom correspondence should be addressed.
Remote Sens. 2025, 17(17), 3106; https://doi.org/10.3390/rs17173106
Submission received: 1 July 2025 / Revised: 4 September 2025 / Accepted: 5 September 2025 / Published: 6 September 2025
(This article belongs to the Special Issue Satellite Remote Sensing for Ocean and Coastal Environment Monitoring)
Browse Figures
Versions Notes

Abstract

The phenological characteristics of the spring phytoplankton bloom in the mid- and high-latitude oceans, including its initiation, duration, and intensity, can be assessed using various diagnostic methods. However, there is currently a lack of systematic comparisons among these different methods. To elucidate the differences in spring bloom characteristics derived from different approaches and to identify suitable methods for shelf seas, this study comprehensively compares and evaluates the multiple methods for characterizing the spring bloom in the central Yellow Sea, based on satellite-derived chlorophyll-a (Chl-a) data from 2003 to 2020. The methods examined include concentration threshold (CT), cumulative concentration threshold (CCT), rate of change (RoC), and curve-fitting methods for determining bloom initiation; threshold and symmetric methods for estimating duration; and peak, mean, integral, and relative intensity index methods for assessing intensity. The results show that the bloom initiation determined by the CT method occurs earliest (average: Day of Year (DOY) 64), whereas the RoC method identifies a notably later initiation (average: DOY 100), approximately 40 days later. The CCT method yields an intermediate bloom initiation (average: DOY 70), with minimal interannual variability. Notably, curve-fitting methods often produce outliers (e.g., DOY 1) due to the fluctuations in Chl-a time series during winter. The threshold method yields a shorter bloom duration (average: 70 days), while the symmetric method results in a duration of more than 10 days longer. The four intensity assessment methods indicate that bloom intensity initially increased and subsequently decreased from 2003 to 2020, but the peak year varies depending on the method used. Overall, the CCT, symmetric, and relative index methods are more suitable for the Yellow Sea, as their computational results exhibit fewer outliers and relatively low standard deviations. The interannual variations in spring bloom characteristics assessed by different methods display distinct patterns and weak correlations, indicating that methodological choices can lead to divergent interpretations of spring bloom dynamics. Therefore, it is essential to carefully select methods based on research objectives and dataset characteristics.

1. Introduction

Phytoplankton, as primary producers in marine ecosystems, account for approximately half of global primary productivity. Variations in their biomass significantly influence global marine biogeochemical cycles [1,2,3,4,5]. In temperate and subpolar seas, seasonal changes in biomass are primarily driven by fluctuations in light availability and nutrient supply, resulting in seasonal phytoplankton blooms. Among these phenomena, the spring phytoplankton bloom (hereafter referred to as the spring bloom), characterized by a substantial increase in phytoplankton populations, was one of the earliest recognized marine phenomena [6] and continues to be a prominent feature of the annual phytoplankton biomass cycle [7,8,9]. The bloom has a significant impact on ecosystem energy flow and further influences the growth and reproduction of fish and other higher-trophic-level organisms [10,11,12]. Therefore, investigating and elucidating the phenological characteristics and variations in the spring bloom is essential for understanding marine ecosystem dynamics [8,13].
Recent advancements in satellite remote sensing have greatly expanded the spatial coverage and temporal range of marine data, significantly enhancing research on spring blooms [14,15,16,17,18]. Analyzing continuous satellite-derived sea surface chlorophyll-a (Chl-a) data is a key pathway for diagnosing spring bloom characteristics. However, different studies employ varying methods and criteria when quantifying seasonal bloom features [19,20]. For instance, Siegel et al. [21] defined the initiation of the North Atlantic bloom as the point at which Chl-a concentrations exceeded by 5% of the annual median. In contrast, Niu et al. [17] established a threshold of the annual median plus 15% to determine the initiation of the spring bloom in the South Yellow Sea. Ueyama et al. [22], in their study of seasonal blooms in the North Atlantic, utilized a sigmoidal curve fit to cumulative Chl-a variance time series to identify bloom initiation and termination. Meanwhile, Zhai et al. [23], investigating phytoplankton bloom phenology in the Northwest Atlantic, employed Gauss curve fitting to analyze Chl-a concentrations for assessing bloom characteristics and interannual variations. Brody et al. [24], using satellite Chl-a data, compared three methods—rate of change, concentration threshold, and cumulative concentration threshold—for determining the initiation of the spring bloom in the North Atlantic and reported significant discrepancies (up to 20 days) among the results. This highlights the potential for different diagnostic methods to yield varying conclusions in studies of seasonal blooms. Key characteristics of the spring bloom include initiation, duration, and intensity [13,17,25,26], each of which can be evaluated using multiple calculation methods. However, there is a notable lack of comprehensive summarizations and comparisons of these diagnostic methods. Thus, the differences and limitations among these methods have not been fully understood yet.
This study aims to elucidate the differences in the characteristics of spring bloom obtained by various methods and to identify a suitable method for a typical shelf sea—the Yellow Sea (Figure 1). First, we systematically reviewed the primary methods currently employed for diagnosing spring bloom characteristics. Subsequently, we applied these multiple methods to calculate bloom initiation, duration, and intensity in the central Yellow Sea using satellite Chl-a data in order to assess the differences and consist in spring bloom characteristics. Finally, we conducted a comparative analysis of the results obtained from different methods, highlighting their respective strengths and limitations. Our findings may serve as a valuable reference for future research on spring bloom phenology.

2. Study Area

The Yellow Sea, a typical temperate shelf sea located in the western Pacific Ocean, spans an area of approximately 380,000 km2 with an average depth of 44 m (Figure 1). The spring bloom is one of the most significant phenomena in the annual cycle of phytoplankton in the Yellow Sea, accounting for approximately 40% of the Yellow Sea’s annual primary productivity [27] and exerting a considerable impact on its fishery resources. Existing studies suggest [28,29,30] that the spring bloom in the Yellow Sea typically commences in March and concludes in early May, with a peak occurring in April. However, variations in physical conditions and nutrient availability lead to interannual variability in the characteristics of the bloom [29]. Several methods—including the concentration threshold method, cumulative concentration threshold method, Gauss curve fitting, and sigmoidal curve fitting—have been employed to determine bloom initiation in the Yellow Sea [15,25,29,31]. For example, Shi et al. [25] utilized the cumulative concentration threshold method to identify the initiation of the spring bloom in the central South Yellow Sea, while also employing the Peak Method and symmetric method to estimate bloom intensity and duration. Liu et al. [15] determined bloom initiation and termination by fitting a sigmoidal curve to Chl-a time series and assessed bloom intensity using a relative intensity index. Meanwhile, Kim et al. [31] applied Gauss curve fitting in their investigation of phytoplankton blooms in the South Yellow Sea to define bloom initiation. These studies illustrate that various methods have been utilized to characterize the spring bloom in the Yellow Sea; however, their consistency and discrepancies have yet to be thoroughly examined.

3. Materials and Methods

This section summarizes the primary computational methods for diagnosing bloom characteristics, with a focus on three key characteristics: bloom initiation, duration, and intensity.

3.1. Bloom Initiation

Common methods for determining bloom initiation using Chl-a time series mainly include the concentration threshold method, cumulative concentration threshold method, curve-fitting method, and rate of change method (Table 1).

3.1.1. Concentration Threshold Method (CT)

The concentration threshold method (hereafter referred to as the CT Method) identifies the initiation of blooms by establishing a specific threshold concentration of Chl-a. The bloom initiation is designated when Chl-a levels surpass this threshold [24]. However, the selection of concentration thresholds varies among studies. A prevalent method is to define the threshold as a percentage of the annual median Chl-a concentration [21,32,33,34,35]. For instance, Siegel et al. [21] established the threshold at 5% above the annual median Chl-a concentration (i.e., threshold = annual median × 105%) in their investigation of the North Atlantic spring bloom using satellite data (hereafter referred to as the CT-S Method). They tested thresholds ranging from 1% to 30% above the median and found minimal differences in the resulting outcomes. This method is regarded as a reliable indicator for bloom initiation, as it effectively captures the initial growth phase of blooms rather than relying solely on peak concentrations [21]. Henson et al. [32,33] enhanced this method by requiring that Chl-a concentrations remain above the threshold for three consecutive days to mitigate the effects of short-term extreme events (hereafter referred to as the CT-H Method). Brody et al. [24] utilized a similar method but with a modification: they began from the annual maximum Chl-a concentration and searched backward to identify the first instance where Chl-a exceeded the threshold, provided it was preceded by two consecutive days of sub-threshold concentrations (hereafter referred to as the CT-B Method). Some studies have employed fixed empirical Chl-a concentration values as thresholds. For example, Fleming et al. [36] defined the initiation of the spring bloom in the Baltic Sea as occurring when Chl-a concentrations first exceeded 5 μg L−1. However, due to significant regional variations in baseline Chl-a levels, this fixed-threshold method lacks universality [32].
Table 1. An overview of methods for calculating bloom characteristics (more details in Section 3). Threshold = 105% of the median annual Chl-a concentration.
Table 1. An overview of methods for calculating bloom characteristics (more details in Section 3). Threshold = 105% of the median annual Chl-a concentration.
Bloom CharacteristicsMethodPrincipleReference
InitiationCT-SWhen the concentration of Chl-a first exceeds the threshold.[21]
CT-HWhen the concentration of Chl-a exceeds the threshold for three consecutive days, the first day of these three days is selected.[32,33]
CT-BWhen Chl-a first exceeds the threshold following two consecutive days of values below this threshold (searching backward from the peak).[24]
CCT-XWhen the daily integrated Chl-a concentration reaches 25% of the total integration from January to August.[25,37,38]
CCT-Y Consistent with CCT-X, but the starting point for the integration is the date of minimum Chl-a concentration rather than DOY 1.[24]
RoCThe point at which the rate of change in Chl-a (dChl-a/dt) reaches its maximum.[24,39]
Gauss curve-fittingBased on the time series of Chl-a fitted using a Gauss model, initiation was defined as t_Peak − 2σ (the peak time and standard deviation of the Gauss fitting curve, respectively).[40]
Sigmoidal curve-fittingThe point at which the slope of the sigmoidal fitting curve first equals 20% of the maximum slope.[22]
DurationThresholdThe bloom ends when the Chl-a concentration falls below the threshold for the first time after reaching its peak. The duration is the period from its initiation to its termination.[17,19]
SymmetricTwice the interval between bloom initiation and peak.[25]
IntensityPeakPeak concentration of Chl-a during the bloom.[13,28,29]
IntegralThe integral value of Chl-a concentration during the bloom duration.[41]
MeanThe average Chl-a concentration during the bloom duration.[41]
IndexThe ratio of the mean Chl-a concentration during the bloom period to the non-bloom period.[22]

3.1.2. Cumulative Concentration Threshold Method (CCT)

The cumulative concentration threshold method (hereafter referred to as the CCT Method) involves integrating the Chl-a concentration time series, with bloom initiation defined as the point at which the integrated Chl-a value exceeds a specified threshold (CCT-X Method) [24,37,38]. This threshold is typically established as a percentage of the total cumulative Chl-a biomass. For example, Shi et al. [25] aggregated daily Chl-a concentration data from January to August and proposed that the initiation of spring blooms occurs when the cumulative value exceeds 30% of the total Chl-a biomass. Brody et al. [24] modified the preprocessing step of this method by utilizing the minimum value in the Chl-a time series as the starting point (before the peak occurs) for integration (CCT-Y Method), a process referred to as “series offset.” The authors suggested that for subtropical regions, the optimal threshold for bloom initiation falls within the 10–15% range, while for subpolar regions, a threshold of 15–20% is more appropriate. In this study, we observed minimal sensitivity to threshold selection and consequently adopted a 25% threshold.

3.1.3. Curve-Fitting Methods

When employing curve-fitting methods to determine bloom initiation, it is essential to select an appropriate fitting function based on the characteristics of the Chl-a time series, including its waveform pattern. Commonly used fitting functions include Gauss and sigmoidal functions.
  • Gauss fitting method
Yamada and Ishizaka introduced a method for characterizing phytoplankton blooms based on a Gauss curve, which they applied to the analysis of spring blooms in the Japan Sea [40]. This method has since been utilized by numerous researchers to investigate bloom characteristics across various marine regions [31,42,43,44]. The fitting equation is given by:
B(t) = B0 + Peak × exp [- (t − t_Peak)2/(2σ2)]
where B0 represents the background Chl-a concentration (mg/m3), t_Peak indicates the timing of peak Chl-a concentration (in days), and Peak denotes the maximum Chl-a concentration in the fitted curve. B(t) is the fitted Chl-a concentration at time t, and σ is the standard deviation of the Gauss curve, which defines the width of the peak [25]. The bloom initiation is defined as the time when t = t_Peak − 2σ.
2.
Sigmoidal fitting method
Ueyama and Monger [22] developed a method based on cumulative variance and a sigmoidal curve to explore the relationship between bloom initiation/intensity and wind forcing in the North Atlantic (10°N–70°N, 90°W–10°E). The processing workflow involves the following steps: (1) applying a three-point median filter to Chl-a data for outlier removal; (2) performing gap-filling using linear interpolation; (3) temporally aligning the time series to center the most pronounced seasonal bloom (i.e., positioning the peak at the midpoint of the series); (4) applying three-point smoothing to the processed Chl-a data, followed by the calculation of 5-day moving window variances centered on each daily observation; and (5) cumulatively summing the resulting variance time series. The cumulative variance time series is then fitted with a sigmoidal function:
f(t) = C1/(1.0 + exp (C2 − C3t)) + C4
where t represents time (in days) and C1–C4 are fitting coefficients. Bloom initiation is defined as the time when the curve’s slope reaches one-twentieth of its maximum slope.

3.1.4. Rate of Change Method (RoC)

Alternative methods for determining bloom initiation utilize thresholds based on rates of chlorophyll variation. For example, Sharples et al. [39] identified the initiation of spring blooms as the timing of the maximum daily Chl-a variation rate in surface waters, suggesting that the period of most rapid biomass accumulation marks the initiation of blooms. Brody et al. [24] refined this method by first processing Chl-a time series using Harmonic Analysis of Time Series coupled with Fast Fourier Transform (HANTS-FFT) [45], subsequently identifying the time point of the maximum rate of change (referred to as the Rate of Change (RoC) Method). This preprocessing constrains the initiation to occur prior to Chl-a peaks, thereby minimizing the potential for anomalous detections [24].
More complex statistical techniques have also been employed. Friedland et al. [46] utilized the Sequential T-test Analysis of Regime Shifts (STARS) [45,47,48] to investigate the dynamics of spring blooms and zooplankton biomass responses in the Northeast U.S. continental shelf ecosystem. This method identifies abrupt changes in Chl-a time series to determine bloom initiation, termination, and duration. While some studies have utilized relationships between phytoplankton growth and respiration rates or other biological indicators for bloom detection [49], the present study focuses exclusively on Chl-a time series analyses, thus excluding these alternative methods.

3.2. Bloom Duration

Bloom duration is defined as the time interval from the initiation to the termination of a phytoplankton bloom. While bloom initiation can be determined using the aforementioned methods, accurately identifying bloom termination is crucial for calculating bloom duration. Two primary methods exist for determining termination time: the first employs the same concentration threshold method used for identifying bloom initiation (termed the threshold method) [17,19]; the second approximates termination time as the point symmetric to the initiation time relative to the chlorophyll peak [25], whereby bloom duration is calculated as twice the interval from bloom initiation to peak time (termed the symmetric method).

3.2.1. Threshold Method

Several studies have utilized threshold criteria analogous to those applied for bloom initiation to ascertain bloom termination, thereby determining the total duration of the bloom. A notable example is the method employed by Racault et al. [19], who established a consistent threshold at 5% above the annual median chlorophyll concentration to identify both bloom initiation and termination, subsequently calculating the overall bloom duration.

3.2.2. Symmetric Method

The symmetric method is particularly well-suited for the Gauss and sigmoidal fitting techniques described in Section 3.1.3, as both fitting curves exhibit symmetry around the central point of the time series. Consequently, the bloom initiation and termination times determined by these fitting functions are also symmetric about this central point. In the Gauss fitting method, bloom initiation is defined as the time corresponding to t_Peak − 2σ, while termination corresponds to t_Peak + 2σ, resulting in a total bloom duration of 4σ [25]. For the sigmoidal fitting method, bloom initiation and termination are identified as the time points when the curve’s slope reaches one-twentieth of its maximum slope, with the interval between these points constituting the bloom duration [15].

3.3. Bloom Intensity

Bloom intensity is commonly employed to characterize the magnitude of phytoplankton blooms. Three primary computational methods exist: (1) utilizing the peak Chl-a concentration during the bloom period as the intensity metric; (2) deriving intensity from either the mean or integrated Chl-a concentration over the bloom duration; and (3) calculating a relative intensity index.

3.3.1. Peak Method

Platt et al. [13] defined bloom intensity as the maximum Chl-a concentration within the bloom period in their investigation of North Atlantic spring blooms. This method necessitates the identification of either the observed peak concentration or the fitted maximum value from previously described curve-fitting methods, with the magnitude of this peak serving as the intensity measure [28,29].

3.3.2. Integral/Mean Method

An alternative method quantifies intensity through either integrated Chl-a concentration (the summation of daily spatial average Chl-a concentrations over the bloom duration) or mean concentration. The integral method, sometimes referred to as “bloom magnitude” [41], inherently incorporates both bloom duration and Chl-a concentration levels, thereby providing a comprehensive assessment of bloom conditions while reducing sensitivity to extreme events. However, some studies contend that biomass concentration alone better reflects bloom intensity, leading to the preference for mean Chl-a concentration (integrated concentration divided by the duration in days) as the preferred metric [41].

3.3.3. Relative Intensity Index

Ueyama and Monger [22] developed a relative intensity index (hereafter referred to as the Index Method), defined as the ratio of mean Chl-a concentration during bloom versus non-bloom periods. This index effectively normalizes spatial and interannual variability in background Chl-a levels, providing a more standardized measure of intensity [15,22].

3.4. Data Sources

The study employed Chl-a data from the Yellow Sea, derived from a moderate resolution imaging spectroradiometer (MODIS) remote sensing reflectance dataset covering the period from 2003 to 2020. This dataset was developed by Wang et al. [50] using a Generalized Additive Model (GAM) algorithm, which has been rigorously validated against a comprehensive set of in situ measurements, demonstrating high accuracy. This high level of accuracy is further supported by its extensive application in research related to the Yellow Sea [51,52,53]. The original dataset featured a temporal resolution of 7 days, which we processed using linear interpolation to achieve daily resolution while maintaining a spatial resolution of approximately 4 km × 4 km.
The study utilized the climatological mean Chl-a to quantify the climatological characteristics of the spring bloom. Climatological mean values were calculated by averaging the Chl-a of each day from 2003 to 2020. Subsequently, the daily Chl-a concentration of the climatological mean was used to represent the climatological seasonal variation in Chl-a and to diagnose the climatological characteristics of the spring bloom in the Yellow Sea.

3.5. Correlation Analysis

To investigate the relationships between different characteristics calculated using various methods, a correlation analysis was conducted on the bloom initiation, intensity, and duration derived by these methods. The Pearson correlation coefficient (r) was employed to quantify the strength of the correlations between the variables; an absolute value of r closer to 1 indicates a stronger correlation between the two variables. The p-value from the t-test was used to assess the significance level of the correlations, with p < 0.05 indicating a statistically significant relationship between the variables.

4. Results

Based on the aforementioned diagnostic methods, we conducted comprehensive analyses of the bloom initiation, duration, and intensity in the central Yellow Sea (indicated by the shaded area in Figure 1) using satellite-derived Chl-a data. To minimize potential interference from autumn blooms (September-December) [29], our analysis exclusively utilized Chl-a concentration data from January to August for the central Yellow Sea region.
In our analysis, we found that the CT-H Method produced results nearly identical to those of the CT-S Method. This similarity is likely attributable to the tendency of Chl-a concentrations to remain above the threshold for several subsequent days in our dataset once they exceeded it. This observation may be linked to the relatively low temporal resolution (7 days) of the original Chl-a data. Consequently, the CT-H Method was not employed in subsequent calculations.

4.1. Bloom Initiation of the Yellow Sea

We employed four distinct methods—the CT Method, CCT Method, RoC Method, and curve-fitting methods (Gauss and sigmoidal)—to systematically assess the initiation of spring blooms in the central Yellow Sea. In particular, since the minimum values of the Chl-a time series occur on DOY 1, both the CCT-X and CCT-Y methods yield identical results. Therefore, we represent the results using the “CCT Method” in Figure 2.
Analyses of climatological mean Chl-a time series revealed significant methodological discrepancies, with differences in bloom initiation estimates exceeding three months (Figure 2). The CT-S Method indicated the earliest bloom initiation on Day of Year (DOY) 28, while its modified version (CT-B Method) delayed detection by 12 days (Figure 2a), yet still exhibited substantial divergence from the other methods. Both implementations of the CCT Method consistently identified DOY 70 as the initiation date (Figure 2 b). The RoC Method produced the latest initiation estimates (DOY 93), while the Gauss and sigmoidal fittings yielded intermediate values of DOY 81 and DOY 77, respectively.
Interannual variability analysis (Table 2) indicated that the CT-S Method produced anomalously early initiation estimates (mean: DOY 17). Given that peak Chl-a concentrations typically occur in April (averaging around DOY 111), we imposed an additional constraint in the CT-B Method by limiting the peak detection window to DOY 75–130 (late March to early May). This temporal constraint effectively filtered out spurious chlorophyll maxima occurring outside the typical spring bloom period, thereby enhancing the phenological relevance of the identified initiation. The CT-B Method yielded significantly later initiation estimates compared to the CT-S Method (mean: DOY 68 vs. 17), but with an 8-day larger standard deviation, indicating greater interannual variability in the results. This underscores the influence of threshold selection criteria in the CT Method on bloom initiation estimates. The CCT-X Method exhibited notable temporal stability (mean: DOY 70 ± 8), with minor adjustments (CCT-Y, mean: DOY 75 ± 8) observed when employing series offset initialization.
The results from the RoC Method (mean: DOY 89, or DOY 100 when excluding outliers from 2004 and 2014, Table 2) consistently aligned with the peak timing of chlorophyll concentrations (mean: DOY 111), which is significantly later than the estimates from other methods. The curve-fitting methods demonstrated the highest interannual variability in bloom initiation estimates, with Gauss and sigmoidal fittings exhibiting substantially greater standard deviations (43 and 36 days, respectively) compared to the other methods. These estimates also deviated significantly from climatological mean values. The performance of both fitting algorithms was particularly compromised in years characterized by multiple short-lived bloom events (e.g., 2014, 2016, and 2017), where increased variability in Chl-a time series hindered successful curve convergence, rendering initiation identification ineffective. Additionally, the Gauss fitting method produced non-physical negative initiation values in certain years, an artifact resulting from compressed inter-bloom intervals that led to excessively broad fitted peaks extending beyond the temporal domain of the time series. Collectively, these limitations suggest that curve-fitting methods are more suitable for analyzing smoothed climatological data than for investigating interannual variability in spring bloom phenology.
In summary, comparative analysis indicates that while the CT-B Method and CCT Method yield comparable mean initiation estimates, the former exhibits greater interannual variability. The late initiation estimates from the RoC Method and the excessive variability from the curve-fitting methods limit their applicability for interannual studies.
We conducted a correlation analysis on the results presented in Table 2 (Figure 3), excluding years with outliers (e.g., 2004, 2014) to mitigate their influence on the analysis. The results revealed positive correlations among methods based on the same framework. Specifically, the variants CT-S and CT-B of the CT Method exhibited a significant correlation (r = 0.7, p < 0.01), as did the variants CCT-X and CCT-Y of the CCT Method (r = 0.94, p < 0.01). The high correlation observed among different implementations of the CCT Method indicates that time series offset adjustments have a minimal impact. Consequently, we excluded series offsets from subsequent analyses. Significant positive correlations were also found between the CT-B Method and the CCT-X Method, as well as between the results of the Gauss fitting and RoC Method. No significant correlations were detected among other methodological pairs. These findings underscore the importance of the chosen analytical method, which can significantly influence the derived interannual variability patterns in bloom initiation, potentially leading to varying ecological interpretations depending on the selected method.

4.2. Bloom Duration of the Yellow Sea

This section utilized both the symmetric method and the threshold method to calculate the climatological mean duration of spring blooms in the central Yellow Sea from 2003 to 2020 (Figure 2). Analysis of climatologically averaged Chl-a data revealed method-dependent variations in bloom duration estimates: the symmetric method yielded 87 days, the threshold method produced 99 days, the Gauss fitting resulted in 80 days, and the sigmoidal fitting provided the shortest estimate of 73 days. The maximum inter-method discrepancy was one month, with the threshold method generating the longest durations and sigmoidal fitting yielding the shortest.
The analysis of interannual variability (Figure 4), which excluded curve-fitting methods due to their susceptibility to outliers (Section 4.1), demonstrated that durations derived from the threshold method exhibited the lowest variability (mean: 71 ± 21 days). When using initiation dates determined by the CT-B Method, the symmetric method produced substantially longer duration estimates (mean: 92 ± 38 days), representing a 17-day increase compared to direct threshold method calculations (α in Figure 4), along with greater interannual fluctuations. For initiations derived from the CCT-X Method, symmetric method durations averaged 80 ± 26 days, differing by more than 10 days from other estimates. A significant correlation was observed between the threshold and symmetric methods for identical initiation dates (r = 0.87, p < 0.001), although the threshold method exhibited a standard deviation that was 17 days lower. No significant correlations were found when comparing durations derived from different initiation methods.

4.3. Bloom Intensity of the Yellow Sea

This section employed four methods to quantify bloom intensity: the Peak Method, Integral Method, Mean Method, and Index Method. The latter three methods required bloom duration inputs, which were calculated using the three distinct duration estimates presented in Figure 4, with results visualized in Figure 5.
The interannual variability of bloom intensity exhibited notable differences across calculation methods. From 2003 to 2006, the Mean and Integral Methods demonstrated similar patterns of initial increase followed by a decrease and subsequent increase, whereas the Peak and Index Methods displayed an opposite trend of initial decrease followed by an increase. From 2007 to 2010, the Peak and Index Methods produced consistent temporal trends, while the Mean and Integral Methods yielded divergent patterns. After 2014, all methods indicated an overall declining trend in bloom intensity, with the Index Method showing the weakest decline. Since the Index Method incorporates both bloom and non-bloom period Chl-a concentrations, substantial differences in relative intensity indices may arise even when mean concentrations are similar (e.g., 2013 vs. 2014). While the Peak, Mean, and Index Methods all identified 2008 as the year of maximum intensity, the Integral Method indicated 2010 as the peak year. These results demonstrate that, for a given duration calculation method, different intensity estimation methods can yield substantially different results, whereas using the same method with varying duration inputs produces relatively consistent interannual patterns.
Correlation analysis revealed relationships among various methods (Figure 6). The Peak Method demonstrated significant correlations with the Mean Method, Integral Method, and Index Method when using durations derived from the threshold method (α in Figure 4) (r > 0.5, p < 0.05). Additionally, the Integral Method and Index Method exhibited a strong positive correlation (r > 0.5, p < 0.05). In contrast, for durations calculated using the symmetric method (β in Figure 4), only the Integral Method and Peak Method maintained a significant correlation (r = 0.58, p < 0.05). Notably, when employing durations derived from the symmetric method (γ in Figure 4), the Mean Method and Index Method displayed particularly strong correlation (r = 0.85, p < 0.05), but no correlation for other durations. Overall, for varying duration inputs, intensity values derived from the same calculation method usually showed high correlation; for example, the three results obtained from the Integral Method demonstrated significant positive correlations with each other (all r > 0.5, p < 0.05). However, the correlations between different methods were weak and non-significant, suggesting that different methods may capture different interannual variations in spring bloom intensity.

5. Discussion

5.1. Performances of Various Methods

In our study, we summarized the performances of different methods and compiled them into a table (Table 3). The initiation for the CT-S/H calculation was relatively early (average of DOY 17) and was significantly influenced by high Chl-a concentrations during winter. The cause of this issue is explained in detail in Section 5.2. In contrast, the initiation for the CT-B calculation was noticeably later (approximately 50 days later on average) but exhibited a larger standard deviation (19 days). The standard deviation for the initiation calculated using the CCT-X method was the smallest (8 days). The fitting methods (Gauss and Sigmoidal) were prone to outliers (e.g., values of DOY 1 or negative values) and are not recommended for use. Regarding duration, although the threshold method produced a smaller standard deviation (21 days), the symmetric method is more convenient for calculations. For bloom intensity, the different methods reflect varying characteristics of phytoplankton biomass (measured as Chl-a) during the bloom period, with further details presented in Table 3.

5.2. Influence of Winter Chl-a Elevation on the CT Method

The CT Method exhibited anomalous bloom initiation (DOY 1) in specific years (e.g., 2004, 2014), as illustrated in Table 2. This artifact persisted across a range of threshold adjustments (5–30% above the median). An analysis of the Chl-a time series for these years (Figure 7) revealed wintertime peaks that significantly exceeded thresholds, leading to false initiation signals. Two primary mechanisms account for these elevated winter values: (1) ephemeral stabilization of the water column under favorable meteorological conditions may allow minor bloom events to occur, and (2) vigorous winter mixing can induce sediment resuspension, potentially inflating satellite-derived Chl-a estimates [39].
The supplemental criterion of the CT-H Method, which required sustained threshold exceedance for three days, proved ineffective, as winter anomalies typically persisted beyond this duration, yielding results nearly identical to those of the CT-S Method. Conversely, the CT-B Method’s backward-search algorithm—initiating from spring bloom peaks and retroactively identifying threshold crossings—effectively mitigated winter interference. Therefore, the modified CT Method proposed by Brody et al. [24] demonstrates superior reliability for detecting bloom initiation.

5.3. Interrelationships Between Spring Bloom Characteristics

The intrinsic definition of bloom duration as the interval between initiation and termination establishes a fundamental interdependence with initiation. Correlation analyses (Figure 8a) reveal significant negative correlations between initiation and duration across all methods (threshold method with CT-derived initiation [α]: r = −0.91; symmetric method with CT-derived initiation [β]: r = −0.90; symmetric method with CCT-derived initiation [γ]: r = −0.76; all p < 0.01), confirming that earlier initiation is consistently associated with extended bloom durations.
The relationship between bloom intensity and duration exhibits substantial methodological dependence. For threshold durations (α, Figure 8a), both the Integral Method and Index Method show strong duration dependence (|r| > 0.5), indicating a significant influence of temporal extent on intensity metrics (excluding the Peak Method). In contrast, the Mean Method demonstrates a weaker correlation. However, for symmetric durations (β and γ, Figure 8a), all intensity metrics (Mean, Integral, and Index Methods) exhibit significant correlations (r > 0.7, p < 0.01). This suggests that when employing the symmetric method, the calculated results for bloom intensity are more reliant on duration. Notably, for the relative intensity index method, the duration calculated by either method significantly affects the intensity calculations (|r| > 0.5).
The direct influence of initiation on bloom duration subsequently induces secondary correlations between initiation and intensity metrics across various methods (Figure 8b). Notably, initiation dates derived from the CT-B Method demonstrate the strongest association with intensity values obtained from the Integral Method (|r| > 0.7, p < 0.01). For instance, in years characterized by early initiation (e.g., 2004, 2014), the intensity measured by the Integral Method tends to be elevated, primarily due to the extended duration of the bloom resulting from earlier initiation, which in turn increases the integral value.
Figure 8 illustrates significant variability in inter-characteristic correlations depending on the analytical methodology employed for quantifying bloom characteristics. Specifically, bloom intensities derived from the Index Method exhibit a significant negative correlation with durations calculated using the CT-B Method (α in Figure 4), while showing a positive correlation with durations derived from symmetric methods (β/γ in Figure 4). Similarly, intensities calculated using the Mean Method correlate positively with initiation values from CCT-X (Table 2), yet show no significant relationship with initiation from CT-B. These methodological discrepancies highlight how the selection of diagnostic methods fundamentally influences the interpreted relationships between phenological metrics, with direct implications for ecological conclusions drawn from bloom characteristic analyses.
The analysis presented above demonstrates strong correlations between the mean, integral, and index-based intensity methods and the duration of blooms. Based on the correlations, we conducted linear regressions to examine the relationships among bloom duration, intensity, and initiation. The results of these regressions are summarized in Table 4. These linear relations could provide a reference for predicting bloom characteristics in the Yellow Sea.

6. Conclusions

This study systematically evaluates three key spring bloom characteristics—initiation, duration, and intensity—through a comparative analysis of multiple diagnostic methods applied to the central Yellow Sea.
Initiation calculations exhibit substantial methodological dependence. The unmodified CT-S Method produced initiation dates that were, on average, 40 days earlier than those derived from the CT-B Method, albeit with greater variability (σ = 19 days). The CCT Method demonstrated minimal interannual fluctuation (σ = 8 days), although results were sensitive to the initialization of the time series. The RoC Method’s peak-proximity bias (mean 10-day offset from bloom maximum) limits its applicability in this region. Curve-fitting methods (Gauss/sigmoidal) displayed excessive variability (σ > 35 days) due to sensitivity to waveform shapes and frequent outlier generation. Collectively, threshold-based methods (CT-B/CCT-X) offered the most optimal balance between precision and robustness for determining initiation.
Duration calculations revealed a strong correlation between threshold and symmetric methods (r = 0.87) when using identical initiation inputs. However, the threshold method demonstrated superior temporal stability, exhibiting a 17-day lower standard deviation (σ). The duration calculated using the threshold method was the shortest, averaging 70 days, while the symmetric method yielded durations that were, on average, more than 10 days longer than those derived from the threshold method.
In assessing bloom intensity, all analytical methods reveal a consistent interannual trend of rising and then falling bloom intensity, yet the detailed temporal characteristics vary among methods. The Peak Method may be influenced by anomalous chlorophyll peaks resulting from short-term extreme events and is sensitive to the temporal resolution of chlorophyll data. Despite its unique independence from duration, the Peak Method maintained significant correlations (|r| > 0.5) with all other intensity metrics. The Mean Method, Integral Method, and Index Method are all impacted by bloom duration.
For bloom initiation, the CCT-X method is preferred due to the absence of outliers in the results. Regarding bloom duration, the symmetric method is recommended because it can be conveniently calculated based on the initiation of the bloom. For assessing bloom intensity, the relative intensity index method is advisable, as it accounts for Chl-a concentration during non-bloom periods, thereby offering an estimate of bloom intensity based on the complete seasonal variation. Additionally, we advise researchers to select the appropriate calculation method for bloom intensity based on their specific research needs.
Bloom characteristics exhibit interdependencies. An earlier initiation significantly prolongs bloom duration, which directly affects the calculated intensity of the bloom (excluding the Peak Method). These relationships exhibit varying characteristics across different calculation methods, necessitating careful consideration of methodological choices based on research objectives and data characteristics. Different methods may yield entirely different conclusions in the quantitative analysis and research of the spring bloom. Future efforts should focus on developing more unified and comprehensive indicators to quantify the characteristics of spring blooms.

Author Contributions

Conceptualization, K.J. and L.L.; methodology, K.J., C.D. and X.L.; formal analysis, X.L., Y.S. and J.L.; writing—original draft preparation, K.J.; writing—review and editing, K.J. and L.L.; visualization, K.J., Y.S. and J.L.; funding acquisition, L.L. All authors have read and agreed to the published version of the manuscript.

Funding

The study was financially supported by the National Natural Science Foundation of China (42030402), Open Research Fund of State Key Laboratory of Estuarine and Coastal Research (SKLEC-KF202404), Pilot Project of Basic Research Operating Expenses System of Hebei Academy of Sciences (2025PF07), and the Open Research Fund of Shandong Key Laboratory of Marine Ecological Environment and Disaster Prevention and Mitigation (202305).

Data Availability Statement

The satellite data can be downloaded from the U.S. National Aeronautics and Space Administration (NASA) website (http://oceancolor.gsfc.nasa.gov ) (accessed on 15 June 2022).

Acknowledgments

We thank Yueqi Wang from Yantai Institute of Coastal Zone Research, Chinese Academy of Sciences, for his support regarding the satellite data.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Behrenfeld, M.J.; Randerson, J.T.; McClain, C.R.; Feldman, G.C.; Los, S.O.; Tucker, C.J.; Falkowski, P.G.; Field, C.B.; Frouin, R.; Esaias, W.E.; et al. Biospheric Primary Production During an ENSO Transition. Science 2001, 291, 2594–2597. [Google Scholar] [CrossRef]
  2. Falkowski, P.G.; Barber, R.T.; Smetacek, V. Biogeochemical Controls and Feedbacks on Ocean Primary Production. Science 1998, 281, 200–206. [Google Scholar] [CrossRef] [PubMed]
  3. Kim, Y.; Youn, S.-H.; Oh, H.-J.; Joo, H.; Jang, H.-K.; Kang, J.-J.; Lee, D.; Jo, N.; Kim, K.; Park, S.; et al. Seasonal Compositions of Size-Fractionated Surface Phytoplankton Communities in the Yellow Sea. J. Mar. Sci. Eng. 2022, 10, 1087. [Google Scholar] [CrossRef]
  4. Hutchins, D.A.; Alessandro, T. Feedbacks between Phytoplankton and Nutrient Cycles in a Warming Ocean. Nat. Geosci. 2024, 17, 495–502. [Google Scholar] [CrossRef]
  5. Uth, C.; Asmala, E.; Lewandowska, A.M. Phytoplankton Community Composition as a Driver of Annual Autochthonous Organic Carbon Dynamics in the Northern Coastal Baltic Sea. Mar. Ecol. Prog. Ser. 2024, 745, 13–24. [Google Scholar] [CrossRef]
  6. Record, N.R.; Balch, W.M.; Stamieszkin, K. Century-scale changes in phytoplankton phenology in the Gulf of Maine. PeerJ 2019, 7, e6735. [Google Scholar] [CrossRef]
  7. Legendre, L. The significance of microalgal blooms for fisheries and for the export of particulate organic carbon in oceans. J. Plankton Res. 1990, 12, 681–699. [Google Scholar] [CrossRef]
  8. Salgado-Hernanz, P.M.; Racault, M.-F.; Font-Muñoz, J.S.; Basterretxea, G. Trends in phytoplankton phenology in the Mediterranean Sea based on ocean-colour remote sensing. Remote Sens. Environ. 2019, 221, 50–64. [Google Scholar] [CrossRef]
  9. Sverdrup, H.U. On Conditions for the Vernal Blooming of Phytoplankton. ICES J. Mar. Sci. 1953, 18, 287–295. [Google Scholar] [CrossRef]
  10. Hunt, G.L., Jr.; Stabeno, P.; Walters, G.; Sinclair, E.; Brodeur, R.D.; Napp, J.M.; Bond, N.A. Climate change and control of the southeastern Bering Sea pelagic ecosystem. Deep. Sea Res. Part II Top. Stud. Oceanogr. 2002, 49, 5821–5853. [Google Scholar] [CrossRef]
  11. Atkinson, A.; Rossberg, A.G.; Gaedke, U.; Sprules, G.; Heneghan, R.F.; Batziakas, S.; Grigoratou, M.; Fileman, E.; Schmidt, K.; Frangoulis, C. Steeper size spectra with decreasing phytoplankton biomass indicate strong trophic amplification and future fish declines. Nat. Commun. 2024, 15, 381. [Google Scholar] [CrossRef] [PubMed]
  12. Sari El Dine, Z.; Guinet, C.; Picard, B.; Thyssen, M.; Duforêt-Gaurier, L.; Hourany, R.E. Influence of the phytoplankton community structure on the southern elephant seals’ foraging activity within the southern ocean. Commun. Biol. 2025, 8, 620. [Google Scholar] [CrossRef] [PubMed]
  13. Platt, T.; Sathyendranath, S. Biological oceanography and fisheries management. ICES J. Mar. Sci. 2007, 64, 863–869. [Google Scholar] [CrossRef]
  14. Deng, L.; Zhou, W.; Xu, J.; Gao, W.; Liao, J.; Zhao, J. Estimation of vertical size-fractionated phytoplankton primary production in the northern South China Sea. Ecol. Indic. 2022, 135, 108546. [Google Scholar] [CrossRef]
  15. Liu, F.; Su, J.; Moll, A.; Krasemann, H.; Chen, X.; Pohlmann, T.; Wirtz, K. Assessment of the summer–autumn bloom in the Bohai Sea using satellite images to identify the roles of wind mixing and light conditions. J. Mar. Syst. 2014, 129, 303–317. [Google Scholar] [CrossRef]
  16. Taylor, A.; Geider, R.; Gilbert, F. Seasonal and latitudinal dependencies of phytoplankton carbon-to-chlorophyll a ratios:results of a modelling study. Mar. Ecol. Prog. Ser. 1997, 152, 51–66. [Google Scholar] [CrossRef]
  17. Niu, Y.; Liu, C.; Lu, X.; Zhu, L.; Sun, Q.; Wang, S. Phytoplankton blooms and its influencing environmental factors in the southern Yellow Sea. Reg. Stud. Mar. Sci. 2021, 47, 101916. [Google Scholar] [CrossRef]
  18. Zhai, F.; Liu, Z.; Gu, Y.; He, S.; Hao, Q.; Li, P. Satellite-Observed Interannual Variations in Sea Surface Chlorophyll-a Concentration in the Yellow Sea Over the Past Two Decades. J. Geophys. Res. Ocean. 2023, 128, e2022JC019528. [Google Scholar] [CrossRef]
  19. Racault, M.-F.; Le Quéré, C.; Buitenhuis, E.; Sathyendranath, S.; Platt, T. Phytoplankton phenology in the global ocean. Ecol. Indic. 2012, 14, 152–163. [Google Scholar] [CrossRef]
  20. Malysheva, A.S.; Lobanova, P.V. Interannual Variability of Marine Phytoplankton Spring Bloom Phenology in the North European Basin. Oceanology 2024, 64, S107–S117. [Google Scholar] [CrossRef]
  21. Siegel, D.A.; Doney, S.C.; Yoder, J.A. The North Atlantic Spring Phytoplankton Bloom and Sverdrup’s Critical Depth Hypothesis. Science 2002, 296, 730–733. [Google Scholar] [CrossRef]
  22. Ueyama, R.; Monger, B.C. Wind-induced modulation of seasonal phytoplankton blooms in the North Atlantic derived from satellite observations. Limnol. Oceanogr. 2005, 50, 1820–1829. [Google Scholar] [CrossRef]
  23. Zhai, L.; Platt, T.; Tang, C.; Sathyendranath, S.; Hernández Walls, R. Phytoplankton phenology on the Scotian Shelf. ICES J. Mar. Sci. 2011, 68, 781–791. [Google Scholar] [CrossRef]
  24. Brody, S.R.; Lozier, M.S.; Dunne, J.P. A comparison of methods to determine phytoplankton bloom initiation. J. Geophys. Res. Ocean. 2013, 118, 2345–2357. [Google Scholar] [CrossRef]
  25. Shi, J.; Liu, Y.; Mao, X.; Guo, X.; Wei, H.; Gao, H. Interannual variation of spring phytoplankton bloom and response to turbulent energy generated by atmospheric forcing in the central Southern Yellow Sea of China: Satellite observations and numerical model study. Cont. Shelf Res. 2017, 143, 257–270. [Google Scholar] [CrossRef]
  26. Friedland, K.D.; Hare, J.A.; Wood, G.B.; Col, L.A.; Buckley, L.J.; Mountain, D.G.; Kane, J.; Brodziak, J.; Lough, R.G.; Pilskaln, C.H. Does the fall phytoplankton bloom control recruitment of Georges Bank haddock, Melanogrammus aeglefinus, through parental condition? Can. J. Fish. Aquat. Sci. 2008, 65, 1076–1086. [Google Scholar] [CrossRef]
  27. Fu, Q.; Yan, X.; Hong, Q.; Lin, L.; Zhang, Y. Variability of the Primary Productivity in the Yellow and Bohai Seas from 2003 to 2020 Based on the Estimate of Satellite Remote Sensing. J. Mar. Sci. Eng. 2023, 11, 684. [Google Scholar] [CrossRef]
  28. Han, J. Numerical Study on the Physical Effect on Phytoplankton Bloom in The Yellow Sea. Ph.D. Thesis, Ocean University of China, Qingdao, China, 2008. [Google Scholar]
  29. Lv, T.; Liu, D.; Zhou, P.; Lin, L.; Wang, Y.; Wang, Y. The coastal front modulates the timing and magnitude of spring phytoplankton bloom in the Yellow Sea. Water Res. 2022, 220, 118669. [Google Scholar] [CrossRef]
  30. Liu, X.; Pu, X.; Qu, D.; Xu, Z. The role of wave-induced mixing in spring phytoplankton bloom in the south yellow sea. Mar. Pollut. Bull. 2025, 211, 117374. [Google Scholar] [CrossRef]
  31. Kim, S.; Lee, D.; Kim, M.; Jang, H.-K.; Park, S.; Kim, Y.; Kim, J.; Park, J.-W.; Joo, H.; Lee, S.-H. Seasonal Patterns and Bloom Dynamics of Phytoplankton Based on Satellite-derived Chlorophyll-a in the Eastern Yellow Sea. Mar. Environ. Res. 2024, 199, 106605. [Google Scholar] [CrossRef]
  32. Henson, S.; Sanders, R.; Holeton, C.; Allen, J.T. Timing of nutrient depletion, diatom dominance and a lower-boundary estimate of export production for Irminger Basin, North Atlantic. Mar. Ecol. Prog. Ser. 2006, 313, 73–84. [Google Scholar] [CrossRef]
  33. Henson, S.A.; Dunne, J.P.; Sarmiento, J.L. Decadal variability in North Atlantic phytoplankton blooms. J. Geophys. Res. Ocean. 2009, 114, 2008JC005139. [Google Scholar] [CrossRef]
  34. Thomalla, S.J.; Fauchereau, N.; Swart, S.; Monteiro, P.M.S. Regional scale characteristics of the seasonal cycle of chlorophyll in the Southern Ocean. Biogeosciences 2011, 8, 2849–2866. [Google Scholar] [CrossRef]
  35. Cole, H.; Henson, S.; Martin, A.; Yool, A. Mind the gap: The impact of missing data on the calculation of phytoplankton phenology metrics. J. Geophys. Res. Ocean. 2012, 117, C08030. [Google Scholar] [CrossRef]
  36. Fleming, V.; Kaitala, S. Phytoplankton Spring Bloom Intensity Index for the Baltic Sea Estimated for the years 1992 to 2004. Hydrobiologia 2006, 554, 57–65. [Google Scholar] [CrossRef]
  37. Greve, W.; Prinage, S.; Zidowitz, H.; Nast, J.; Reiners, F. On the phenology of North Sea ichthyoplankton. ICES J. Mar. Sci. 2005, 62, 1216–1223. [Google Scholar] [CrossRef]
  38. Mackas, D.L.; Greve, W.; Edwards, M.; Chiba, S.; Tadokoro, K.; Eloire, D.; Mazzocchi, M.G.; Batten, S.; Richardson, A.J.; Johnson, C.; et al. Changing zooplankton seasonality in a changing ocean: Comparing time series of zooplankton phenology. Prog. Oceanogr. 2012, 97–100, 31–62. [Google Scholar] [CrossRef]
  39. Sharples, J.; Ross, O.; Scott, B.; Greenstreet, S.P.R.; Fraser, H. Inter-annual variability in the timing of stratification and the spring bloom in the North-western North Sea. Cont. Shelf Res. 2006, 26, 733–751. [Google Scholar] [CrossRef]
  40. Yamada, K.; Ishizaka, J. Estimation of interdecadal change of spring bloom timing, in the case of the Japan Sea. Geophys. Res. Lett. 2006, 33, L02608. [Google Scholar] [CrossRef]
  41. Friedland, K.D.; Mouw, C.B.; Asch, R.G.; Ferreira, A.; Henson, S.; Hyde, K.J.W.; Morse, R.E.; Thomas, A.C.; Brady, D.C. Phenology and time series trends of the dominant seasonal phytoplankton bloom across global scales. Glob. Ecol. Biogeogr. 2018, 27, 551–569. [Google Scholar] [CrossRef]
  42. Jo, C.O.; Lee, J.; Park, K.; Kim, Y.H.; Kim, K.-R. Asian dust initiated early spring bloom in the northern East/Japan Sea. Geophys. Res. Lett. 2007, 34, L05602. [Google Scholar] [CrossRef]
  43. Navarro, G.; Caballero, I.; Prieto, L.; Vázquez, A.; Flecha, S.; Huertas, I.E.; Ruiz, J. Seasonal-to-interannual variability of chlorophyll-a bloom timing associated with physical forcing in the Gulf of Cádiz. Adv. Space Res. 2012, 50, 1164–1172. [Google Scholar] [CrossRef]
  44. Platt, T.; White, G.N.; Zhai, L.; Sathyendranath, S.; Roy, S. The phenology of phytoplankton blooms: Ecosystem indicators from remote sensing. Ecol. Model. 2009, 220, 3057–3069. [Google Scholar] [CrossRef]
  45. Roerink, G.J.; Menenti, M.; Verhoef, W. Reconstructing cloudfree NDVI composites using Fourier analysis of time series. Int. J. Remote Sens. 2000, 21, 1911–1917. [Google Scholar] [CrossRef]
  46. Friedland, K.D.; Leaf, R.T.; Kane, J.; Tommasi, D.; Asch, R.G.; Rebuck, N.; Ji, R.; Large, S.I.; Stock, C.; Saba, V.S. Spring bloom dynamics and zooplankton biomass response on the US Northeast Continental Shelf. Cont. Shelf Res. 2015, 102, 47–61. [Google Scholar] [CrossRef]
  47. Rodionov, S.N. A sequential algorithm for testing climate regime shifts. Geophys. Res. Lett. 2004, 31, L09204. [Google Scholar] [CrossRef]
  48. Rodionov, S.N. Use of prewhitening in climate regime shift detection. Geophys. Res. Lett. 2006, 33, L12707. [Google Scholar] [CrossRef]
  49. Behrenfeld, M.J.; Boss, E.S. Student’s tutorial on bloom hypotheses in the context of phytoplankton annual cycles. Glob. Change Biol. 2018, 24, 55–77. [Google Scholar] [CrossRef]
  50. Wang, Y.; Liu, D.; Tang, D. Application of a generalized additive model (GAM) for estimating chlorophyll- a concentration from MODIS data in the Bohai and Yellow Seas, China. Int. J. Remote Sens. 2017, 38, 639–661. [Google Scholar] [CrossRef]
  51. Lin, L.; Sun, Z.; Lv, T.; Wang, Y.; Tan, W.; Jin, K.; Xiang, D. In-Phase Pacific Decadal Oscillation and El Niño–Southern Oscillation Enhance the Interannual Variability of Spring Blooms in the Yellow Sea. Ocean. Land Atmos. Res. 2025, 4, 0083. [Google Scholar] [CrossRef]
  52. Lin, L.; Liu, D.; Wang, Y.; Lv, T.; Zhao, Y.; Tan, W. Effect of wind on summer chlorophyll-a variability in the Yellow Sea. Front. Mar. Sci. 2023, 9, 1104258. [Google Scholar] [CrossRef]
  53. Lin, L.; Wang, Y.; Liu, D. Vertical average irradiance shapes the spatial pattern of winter chlorophyll-a in the Yellow Sea. Estuar. Coast. Shelf Sci. 2019, 224, 11–19. [Google Scholar] [CrossRef]
Remotesensing 17 03106 g001
Figure 1. Topography of the Yellow Sea. The gray solid lines represent isobaths (in meters), and the shaded area indicates the study region (34–36°N, 123–125°E).
Figure 1. Topography of the Yellow Sea. The gray solid lines represent isobaths (in meters), and the shaded area indicates the study region (34–36°N, 123–125°E).
Remotesensing 17 03106 g001
Remotesensing 17 03106 g002
Figure 2. The calculated results of bloom initiation are presented, with diamonds indicating the initiation and termination of blooms; the shaded area between markers represents bloom duration. (a) The CT, RoC, and Gauss fitting method and (b) the CCT method and sigmoidal fitting method. The gray dashed line in (a) denotes the threshold value established by the CT Method.
Figure 2. The calculated results of bloom initiation are presented, with diamonds indicating the initiation and termination of blooms; the shaded area between markers represents bloom duration. (a) The CT, RoC, and Gauss fitting method and (b) the CCT method and sigmoidal fitting method. The gray dashed line in (a) denotes the threshold value established by the CT Method.
Remotesensing 17 03106 g002
Remotesensing 17 03106 g003
Figure 3. Correlation among different bloom initiation quantification methods. The data is sourced from Table 2. CT-S and CT-B represent the CT Method as implemented by Siegel et al. [21] and Brody et al. [24], respectively, while CCT-X and CCT-Y denote the CCT Method applied to the original time series and offset time series, correspondingly.
Figure 3. Correlation among different bloom initiation quantification methods. The data is sourced from Table 2. CT-S and CT-B represent the CT Method as implemented by Siegel et al. [21] and Brody et al. [24], respectively, while CCT-X and CCT-Y denote the CCT Method applied to the original time series and offset time series, correspondingly.
Remotesensing 17 03106 g003
Remotesensing 17 03106 g004
Figure 4. Estimates of bloom duration (in days) by the three methods. Bars labeled α and β were derived using the initiation from the CT-B method (Table 2) along with the threshold and symmetric methods, respectively, while bars labeled γ were derived using the initiation from the CCT-X method (Table 2) along with the symmetric methods. σ represents the standard deviation of the values across all years.
Figure 4. Estimates of bloom duration (in days) by the three methods. Bars labeled α and β were derived using the initiation from the CT-B method (Table 2) along with the threshold and symmetric methods, respectively, while bars labeled γ were derived using the initiation from the CCT-X method (Table 2) along with the symmetric methods. σ represents the standard deviation of the values across all years.
Remotesensing 17 03106 g004
Remotesensing 17 03106 g005
Figure 5. Bloom intensities of the spring bloom in the central Yellow Sea calculated by the four methods. Subfigures (a–d) correspond to the Peak, Mean, Index, and Integral Methods, respectively. Additionally, α, β, and γ denote the duration metrics specified in Figure 4 that are employed for the intensity calculations. Durations α and β were calculated using initiation data from the CT-B method, whereas duration γ was calculated using initiation data from the CCT-X method.
Figure 5. Bloom intensities of the spring bloom in the central Yellow Sea calculated by the four methods. Subfigures (a–d) correspond to the Peak, Mean, Index, and Integral Methods, respectively. Additionally, α, β, and γ denote the duration metrics specified in Figure 4 that are employed for the intensity calculations. Durations α and β were calculated using initiation data from the CT-B method, whereas duration γ was calculated using initiation data from the CCT-X method.
Remotesensing 17 03106 g005
Remotesensing 17 03106 g006
Figure 6. Correlations of bloom intensities among various methods, with α, β, and γ denoting the different duration metrics referenced in Figure 4 that were employed in the intensity calculations. Durations α and β were calculated using initiation data of the CT-B method, whereas duration γ was calculated using initiation data of the CCT-X method.
Figure 6. Correlations of bloom intensities among various methods, with α, β, and γ denoting the different duration metrics referenced in Figure 4 that were employed in the intensity calculations. Durations α and β were calculated using initiation data of the CT-B method, whereas duration γ was calculated using initiation data of the CCT-X method.
Remotesensing 17 03106 g006
Remotesensing 17 03106 g007
Figure 7. Time series of Chl-a in the central Yellow Sea from January to August in 2004, 2014, 2017, and 2018. The red solid line indicates the concentration threshold. The gray font indicates the initiation of spring blooms in winter diagnosed by the CT method for the corresponding years in Table 2.
Figure 7. Time series of Chl-a in the central Yellow Sea from January to August in 2004, 2014, 2017, and 2018. The red solid line indicates the concentration threshold. The gray font indicates the initiation of spring blooms in winter diagnosed by the CT method for the corresponding years in Table 2.
Remotesensing 17 03106 g007
Remotesensing 17 03106 g008
Figure 8. Correlations among bloom characteristics. (a) Relationships between bloom duration and intensity/initiation, with duration-α, β, and γ corresponding to the duration metrics in Figure 4, while initiation-CT-B and CCT-X indicate the initiation points from Table 2. (b) Relationships between initiation and bloom intensity. The y-axis represents the initiations corresponding to duration-α, β, and γ, while the x-axis indicates the bloom intensity calculated using different methods.
Figure 8. Correlations among bloom characteristics. (a) Relationships between bloom duration and intensity/initiation, with duration-α, β, and γ corresponding to the duration metrics in Figure 4, while initiation-CT-B and CCT-X indicate the initiation points from Table 2. (b) Relationships between initiation and bloom intensity. The y-axis represents the initiations corresponding to duration-α, β, and γ, while the x-axis indicates the bloom intensity calculated using different methods.
Remotesensing 17 03106 g008
Table 2. The calculated bloom initiation (in days) presents results from four methods: CT-S and CT-B represent the CT Method as implemented by Siegel et al. [21] and Brody et al. [24], respectively, while CCT-X and CCT-Y denote the CCT Method applied to the original time series and offset time series, correspondingly. σ represents the standard deviation.
Table 2. The calculated bloom initiation (in days) presents results from four methods: CT-S and CT-B represent the CT Method as implemented by Siegel et al. [21] and Brody et al. [24], respectively, while CCT-X and CCT-Y denote the CCT Method applied to the original time series and offset time series, correspondingly. σ represents the standard deviation.
YearsCT MethodCCT MethodRoC MethodCurve-Fitting Methods
CT-SCT-BCCT-XCCT-YGauss Sigmoidal
200313687272958344
200417658831−181
2005266974771009381
20061360747414211792
200716346868918783
200811467986918380
2009297273731089764
201020325555120−2537
20114081858510610196
20122394757511210437
201333867378112106101
201417964801−1150
201513476464108104102
20162544758024641
2017186596797831
20181426884106101100
201911536666965238
202022747579838210
Mean ± σ17 ± 1164 ± 1970 ± 875 ± 889 ± 3972 ± 4357 ± 36
Table 3. Features of different calculation methods.
Table 3. Features of different calculation methods.
Bloom CharacteristicsMethodFeatures
InitiationCT-SWhile it is straightforward to calculate, this method is also significantly affected by elevated Chl-a levels during winter, resulting in the presence of outliers.
CT-HSimilar to the CT-S Method.
CT-BIn comparison to CT-S/H, the impact of high concentrations of Chl-a during winter is relatively minor.
CCT-XRelatively stable with few outliers.
CCT-Y Similar to the CCT-X method, but with a more complex calculation.
RoCThe calculated results were very late, approaching the time of peak.
Gauss curve-fittingRestricted by the distribution of the Chl-a time series, outliers (even negative values) occur.
Sigmoidal curve-fittingSimilar to the Gauss curve-fitting method.
DurationThresholdEasily calculated, but influenced by the setup of threshold(s).
SymmetricIt can be conveniently calculated based on the bloom initiation.
IntensityPeakNot affected by calculations of initiation and duration, but strongly influenced by extreme values.
Integral/MeanReflects the overall levels of Chl-a concentration during bloom periods, but is influenced by the Accuracy of duration calculation.
Relative IndexAccounts for Chl-a concentration during non-bloom periods, thereby providing an estimate of bloom intensity based on the complete seasonal variation.
Table 4. Linear regression relationships among the characteristics of blooms. Duration-α, β, and γ correspond to the duration metrics in Figure 4, while initiation CT-B and CCT-X indicate the initiation points from Table 2. Only significant regression relationships are included.
Table 4. Linear regression relationships among the characteristics of blooms. Duration-α, β, and γ correspond to the duration metrics in Figure 4, while initiation CT-B and CCT-X indicate the initiation points from Table 2. Only significant regression relationships are included.
Duration-αDuration-βDuration-γ
Initiation-CT-By = −0.9852x + 133.727 **y = −1.7705x + 204.6519 **
Initiation-CCT-X y = −2.5442x + 257.224 **
Intensity-Integraly = 0.82024x + 14.0847 **y = 0.60759x + 24.7294 **y = 0.47858x + 33.6684 **
Intensity-Mean y = −0.0023191x + 1.1239 *y = −0.0062359x + 1.4479 *
Intensity-Index y = −0.018523x + 3.5654 **
Initiation CT-B (α)Initiation CT-B (β)Initiation CCT-X (γ)
Intensity-Integraly = −0.81034x + 123.9155 **y = −1.0667x + 148.4955 **y = −1.0553x + 145.4369 *
Intensity-Mean y = 0.0048841x + 0.5999 *y = 0.017888x + −0.29735 *
Intensity-Index y = 0.043322x + −0.93355 *
*: p < 0.05; **: p < 0.01.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Jin, K.; Dong, C.; Liu, X.; Sun, Y.; Liu, J.; Lin, L. Phenological Characteristics of the Yellow Sea Spring Bloom: A Comparative Evaluation of Multiple Diagnostic Methods. Remote Sens. 2025, 17, 3106. https://doi.org/10.3390/rs17173106

AMA Style

Jin K, Dong C, Liu X, Sun Y, Liu J, Lin L. Phenological Characteristics of the Yellow Sea Spring Bloom: A Comparative Evaluation of Multiple Diagnostic Methods. Remote Sensing. 2025; 17(17):3106. https://doi.org/10.3390/rs17173106

Chicago/Turabian Style

Jin, Kangjie, Chen Dong, Xihan Liu, Yan Sun, Jibo Liu, and Lei Lin. 2025. "Phenological Characteristics of the Yellow Sea Spring Bloom: A Comparative Evaluation of Multiple Diagnostic Methods" Remote Sensing 17, no. 17: 3106. https://doi.org/10.3390/rs17173106

APA Style

Jin, K., Dong, C., Liu, X., Sun, Y., Liu, J., & Lin, L. (2025). Phenological Characteristics of the Yellow Sea Spring Bloom: A Comparative Evaluation of Multiple Diagnostic Methods. Remote Sensing, 17(17), 3106. https://doi.org/10.3390/rs17173106

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop