Satellites reveal hot spots of ocean changes in the early 21st century

Abstract

Ocean change leaves a potentially important imprint on ocean colorimetry. Here, we present an overview and current evaluation of the global ocean color variability from 1998 to 2022, and satellites observe that 36% of oceans (~122 million square kilometers, derived from valid observations) have experienced changes (P < 0.1). In this context, 25% of the area (formerly blue hue) is turning light blue or green, while the remaining 11% becomes bluer, mainly concentrating in the low-latitude oceans. This study further identifies a “direct” notable impact of both sea surface temperature (SST) and climate on ocean colorimetry tendency and anomaly, especially in the low-latitude oceans. Extreme SST events cause “distinct” ocean colorimetry anomalies, although 94% of cases involve relatively small SST fluctuations. Causal analysis reveals important impacts of climate change on equatorial ocean dynamics, particularly ENSO events. Our findings prove the low-latitude oceans as one of the core changing regions that respond to climate change in the early 21st century.

Global ocean dynamics and its climate response are explored through water color changes in the early 21st century.

INTRODUCTION

Current and future variability of global marine ecology and environment amid climate change has garnered extensive attention and research (1–4). Since 2005, a consistent warming trend in global oceans has been observed, leading to increased seawater stratification and ongoing changes in the physical, chemical, and compositional state of the oceans (3–5). Enhanced seawater stratification limits vertical mixing, diminishing the exchange of components such as particulate matter between ocean layers (6). Ocean warming, coupled with intensified stratification, disrupts nutrient cycles, affecting various marine organisms, including phytoplankton (5, 7–10). Previous studies have indicated changes in phytoplankton abundance, community structure, and biogeographic distribution against the backdrop of global warming and climate change, both short and long terms (11–14). The continuous changes in seawater under global ocean warming and climate change have received widespread attention, with the recent development of satellite remote sensing technology, providing an effective tool for monitoring water changes in the global oceans (15–17).

Satellite imagery visualizes the color characteristics of seawater worldwide, and variations and fluctuations in water color serve as a proxy for dynamic changes in water composition (15–17). Water color is contingent upon the inherent optical properties of components in seawater, encompassing rich information about the water body (16, 17). For instance, pure water appears deep blue (low absorption in the blue); increasing phytoplankton biomass shifts the color from blue to green (high absorption in blue and low in green); high colored dissolved organic matter concentration waters tend to “black tea” color (low absorption in the red band); and suspended particulate matter (SPM) induce color changes depending on their composition (16, 17). Assessing the changes in ocean color provides a holistic understanding of both the water quality and the variations in water components, and long–time-series satellite imagery provides support for the analysis of global water color dynamics.

Previous studies have progressively established and refined the mathematical link between the water color and chromaticity coordinates in the CIE (Commission Internationale de l´Eclairage) color system. The conversion between the satellite remote sensing reflectance (R_rs) and the CIE system provides a theoretical basis for the implementation of water color observation based on remote sensing sensors (18–20). The CIE system digitizes color by converting it into RGB coordinates, with the hue angle defined to compress these coordinates, capturing variations in hue (21, 22). Hue angle is widely used in color analysis and serves as an intermediate variable for estimating the Forel-Ule index (a traditional index of water color assessment), enabling the assessment of hue variations (23, 24). The CIE system provides a standard for digitally expressing water color changes of satellite images and is now used to conduct related studies (22, 24–26).

Under the influence of climate change and ocean warming, what variations are unfolding in the global ocean? Previous studies have generally focused on changes in individual or some major components of seawater, with a lack of comprehending overall changes. However, assessing the overall change in seawater provides a reference for forecasting future oceanic changes, and water color allows us to explore ocean variability in the context of climate change. On another front, current studies that use time-series correlation relationships to analyze the impact of climate and sea surface temperature (SST) changes on marine ecological parameters face challenges due to the complex relationships between driving factors and ecological parameters, which prevents a clear delineation of causation (27–29). Therefore, it is vital to investigate how trends and fluctuations in global oceans have altered, as well as whether these changes are due to SST and climate change.

Here, we conduct a global examination of the spatiotemporal patterns of water color using an optical dataset spanning 25 years (1998–2022). Our aim is to identify trends and anomalies in water color variability on a worldwide scale. Leveraging the Liang-Kleeman information flow (L-K IF), a method adept at quantitatively analyzing causality between two time series, together with correlation analysis, we investigate how changes in SST and climate change affect the dynamics of water color in the global oceans. In addition, we use the global in situ phytoplankton data collected during multiple cruises in various seasons to conduct a preliminary analysis of the link between water color and the dominant phytoplankton taxonomic groups (PTGs). Our study targets analyzing the worldwide water color changes over the past 25 years, providing valuable information on the impacts of SST and climate change on global ocean dynamics.

RESULTS AND DISCUSSION

Satellite-observed spatiotemporal patterns of global oceanic colorimetry

Our satellite-observed results capture the distribution and difference in global water colorimetry (hue angle as a proxy, α*; unit, degrees; see Materials and Methods) from 1998 to 2022 across spatial and temporal scales (see Fig. 1). The visible color “fronts” distinguish between the open ocean, continental shelves, and coastal regions, creating a gradient in water color: Oceanic waters exhibit pure blue (α* = ~220°), continental shelf waters appear as light blue (α* = ~210°), and coastal waters mostly related to local marine environment, appearing yellow to green hues (30° < α* < 180°) in some areas (Fig. 1A). High α* appears in the subtropical circulation region of the open ocean, maintaining a deep blue hue throughout the year; coastal waters exhibit a spatially α* difference, with lower values mainly observed along the coasts of the East Siberian Sea, the Laptev Sea, the East China Sea, and the Baltic Sea.

Fig. 1. Global geospatial distribution of interannual average and interdecadal differences in water color in CIE-XYZ 1931 color space from 1998 to 2022.

(A and B) Global and latitude average hue angle [α*; unit, degrees (°)]. (C) Interdecadal differences of α* [(2020–2022) minus (1998–2000)]. (D) Latitudinal variation in interdecadal differences of α* shows the fractions of α* increasing/decreasing pixels within 3° latitude bins.

In comparison to the early 21st century, ~92% of the global marine areas exhibit various extents of color shifts; yet most of the oceanic waters remain in a blue hue, while coastal regions influenced by land-based sources tend to appear yellow-green (Fig. 1C). The interdecadal difference in water color [(2020–2022) minus (1998–2000)] illustrates that the annual average α* increased/decreased by 2.3°/4.5° for ~33%/59% of the global ocean. Increased water hue is mainly displayed along the coast of South Asia and Western Africa, as well as the low to mid-latitudes of the Northern Hemisphere, where the water color transitions from yellow to green or toward a deeper blue. Approximating previous findings (30), we observed a widespread reduction in α* across most global oceanic regions over the past 25 years, indicating that seawater shifted toward a greener hue. Our findings show that the hue decrease appears to be more concentrated in the high-latitude and Southern Hemisphere mid-latitude regions, with a hue increase in the equatorial Pacific Ocean (water now exhibits bluer compared to 25 years ago).

Another question is whether the above-identified differences are evidence of an ongoing shift in global water color. Analyzing the trend component of the global time-series data, we discover that only a limited portion of the global sea area exhibits a notable trend in water color (Fig. 2A, black contours denote confidence intervals at 90% level). Globally, 36% of global oceans exhibit changes in water color, of which 11% of sea areas are increasing in α* at a rate of 0.17° year⁻¹, concentrated at low latitudes, especially in the Indian Ocean, equatorial Pacific, and West African coasts; whereas 25% of the sea area is undergoing a transition to a lighter blue color (α* decreases at a rate of 0.25° year⁻¹) in the eastern South Pacific and the high-latitude region. These findings suggest that trends in water color across most regions of the global ocean have not been statistically significant over the past 25 years, but water color trends appear to be decreasing in a quarter of the ocean while increasing color trends prevail in low-latitude seas.

Fig. 2. Time-series patterns of water color in the global oceans over the past 25 years.

(A) General trend distributions of hue angle (α*) estimated by the Theil-Sen Median method and Mann-Kendall test (area surrounded by black contours, P < 0.1) from global long–time-series data. (B) Latitudinal variation in the fraction of pixels with increasing/decreasing α* trends within 3° latitude bins (P < 0.1). (C) The fluctuation characteristics of the water color anomalies ( $a_{\mathrm{a}}^{*}$ , through SD to quantify amplitudes). (D) Average latitudinal distribution of fluctuations in α*.

Through mapping the global distribution of the SD of anomaly components, we have identified important areas with notable fluctuations in water color. The fluctuation amplitude of α* anomaly ( $a_{\mathrm{a}}^{*}$ ) exhibits more pronounced fluctuations in coastal (SD > 10°) and high-latitude waters (SD, ~5°), whereas mid-low latitude oceans have relatively lower variability (SD < 3°) (Fig. 2C). Coastal waters, influenced by terrestrial inputs, algal blooms, and human activity (26, 31, 32), experience pronounced seasonal fluctuations, with $a_{\mathrm{a}}^{*}$ variations > 20° in certain regions (e.g., coast of the Bering Sea, the Argentine coast, and the marginal seas of China).

Influence of climate change on global oceanic colorimetry

This study examined the association between water color (α*) and SST time series at a 90% confidence level, as well as causality analysis using the normalized information flow (NIF; see the Materials and Methods) from SST to α* (Fig. 3). Our findings indicate that SST influenced water color changes over the past 25 years, as evidenced by synchronized correlation and NIF signals across most regions. A time-series analysis over the past 25 years reveals a strong consistency between the trend components of α* ( $a_{\mathrm{t}}^{*}$ ) and SST (SST^t) (Fig. 3, A and B). High correlation signals (|r| > 0.7) occur between $a_{\mathrm{t}}^{*}$ and SST^t in most sea areas, with positive signals primarily detected in low-latitude seas and negative signals spread in mid- and high latitudes of the Southern Hemisphere. The NIF signal from SST^t to $a_{\mathrm{t}}^{*}$ ( ${\mathrm{\tau }}_{{\text{SST}}^{\mathrm{t}}\to a_{\mathrm{t}}^{*}}$ ) supports the long-term influence of SST^t on water color trends. SST variations enhance uncertainty in water color changes in the bulk of Pacific and Atlantic regions, while the opposite is observed in the Indian Ocean. At low latitudes, anomaly components ( $a_{\mathrm{a}}^{*}$ and SST^a) exhibit notable correlation signals (|r| > 0.5), although SST^a has a limited effect on $a_{\mathrm{a}}^{*}$ as judged by NIF signals (Fig. 3, C and D). Positive NIF signals, indicating increasing uncertainty, are primarily recorded in the equatorial ocean, while negative NIF signals appear in the Philippine Sea, implying that seasonal SST variation contributes to stabilizing water color fluctuations. These variances highlight the complex interaction of elements that influence water color dynamics in various marine regions.

Fig. 3. Correlation (r) and causality (NIF, τ) analysis between water color and SST time-series components across global oceans from 1998 to 2022.

(A and B) The correlation and (C and D) the normalized information flow (NIF) signal distributions between the trend ( $a_{\mathrm{t}}^{*}$ and SST^t) and anomaly ( $a_{\mathrm{a}}^{*}$ and SST^a) components of α* and SST, respectively. The colored areas in the figure represent pixels with a P value of less than 0.1.

We observe four main modes of water color fluctuations corresponding to varying levels of SST variability (Fig. 4). Mode 1 occurs when SST^a remains below 4% of the climatology baseline (occurring in 63% of the cases in the past 25 years), and water color maintains relatively stable (median of the absolute bias of α*, $a_{\text{median}}^{*}$ < 1°). Mode 2 emerges with SST^a between 4 and 14% of the climatology baseline ( $a_{\text{median}}^{*}$ < 2°), accounting for 31% of observed cases. During this phase, $a_{\mathrm{a}}^{*}$ initially increases with rising SST^a magnitudes and plateaus when SST^a exceeds 10% of the SST climatology baseline. Mode 3 is characterized by SST^a exceeding 16% of the climatology baseline with the $a_{\text{median}}^{*}$ magnitudes of 2° to 3° yet occurs seldom, representing 5% of cases. Mode 4 ( $a_{\text{median}}^{*}$ > 3°) reflects notable water color fluctuations associated with extreme SST events, with $a_{\mathrm{a}}^{*}$ reaching a maximum of 15°. Our statistical analysis found that water color fluctuations have been relatively small in most cases over the past 25 years. The reason can be related to the relatively low SST^a in most areas and periods (SST fluctuations below 14% in 94% of all cases) and the resilience and adaptability of some oceanic water components (e.g., phytoplankton) to SST-induced disturbances at a certain level (33, 34). Nonetheless, special attention is warranted for the impact of extreme marine heatwave events on water composition changes, given their notable influence on water color.

Fig. 4. The fluctuations in water color (a*) and SST changes calculated from monthly average OC-CCI and OISST datasets from 1998 to 2022.

The colors of the boxes exhibit the absolute bias of α* induced by SST changes at different levels, with “n” denoting the proportion of data represented by each colored box. The absolute biases of SST and α* are determined on the basis of the respective climatology, with the absolute bias of SST further converted into a proportion of the climatological baseline.

To understand the potential drivers of climate change on water color anomalies in the global oceans, we selected three climate events representing typical variations in the Pacific, Indian, and Atlantic climate systems to analyze the relationship between water color anomalies ( $a_{\mathrm{a}}^{*}$ ) and climate indices (Fig. 5). The α* anomalies exhibit synchronous variations with El Niño–Southern Oscillation (ENSO) events in the equatorial Pacific and Indian Ocean, as reflected by the positive correlation between the multivariate ENSO index (MEI) and water color; in contrast, negative correlations are observed in the mid-latitude regions of the South Pacific Ocean and the Philippine Sea, showing positive MEI signals accompanied by decreasing α*. As evidence of the direct drive of MEI on water color anomalies, NIF signals ( ${\mathrm{\tau }}_{\text{MEI}\to a_{\mathrm{a}}^{*}}$ ) are near-synchronously present in locations with correlation signals, implying that MEI contributes more uncertainty to water color variations. The dipole mode index (DMI), which represents climate events that originate in the Indian Ocean, has a notable impact on water color anomalies in the Pacific and Indian Oceans’ low latitudes. The NIF signals ( ${\mathrm{\tau }}_{\text{DMI}\to a_{\mathrm{a}}^{*}}$ ) indicate that the DMI introduces uncertainty mainly for the low-latitude water color. The corresponding correlation analyses reveal that a positive phase of the DMI leads to bluer water color in the western Indian Ocean, while the opposite for the Sumatran islands. NIF signals indicate that the influence of the Atlantic multi-decadal oscillation (AMO) on water color fluctuation is limited, with most effects occurring in the equatorial Pacific. Long–time-series analysis reveals that climate change directly affects oceanic water color in the low-latitude Pacific and Indian Oceans.

Fig. 5. The link between ocean color anomalies and climate forcing.

Pearson correlation coefficient (r) and NIF (τ) distribution between three typical climate indices [multivariate ENSO index (MEI), dipole mode index (DMI), and Atlantic multi-decadal oscillation (AMO)] and anomalous components of global α* time series ( $a_{\mathrm{a}}^{*}$ ) from 1998 to 2022. The colored areas in the figure represent pixels with a P value of less than 0.1.

Our time-series decomposition analysis reveals a strong link between the trend components of water color and SST trends, further supported by the simultaneous appearance of NIF signals. SST anomalies affect water color anomalies mainly in mid- and low-latitude waters. Furthermore, climate indices exhibit NIF signals in the equatorial regions of the Pacific and Indian Oceans, indicating a clear impact of climate change on the marine environment in these areas over the past 25 years. These findings imply that SST and climate change have a direct impact on the dynamics of water components, particularly in low-latitude oceans.

Water color change inferring phytoplankton dynamics

Phytoplankton, an important component of marine waters, are sensitive to the effects of SST, and light absorption during phytoplankton photosynthesis affects the light signal at the sea surface (35, 36), which is reflected in the contribution of water color (12, 14, 16). We simulated the relationship between phytoplankton biomass [proxy as log-transformed chlorophyll a (Chla)] and water colorimetry by using a satellite-earth synchronization dataset. Water color tends to be lighter with high Chla concentrations, and the trend of Chla with α* varies across three intervals, generally following a decreasing trend [coefficient of determination (R²) = 0.59] (Fig. 6). Transitioning from coastal to continental shelf waters (yellow to green and blue), an increase in α* accompanies a decrease in Chla followed by stabilization, while Chla decreases as α* increases in the transition from the shelf sea to the open ocean. In oceanic waters predominantly composed of phytoplankton, low Chla concentrations lead to a bluish water color, while elevated Chla concentrations impart a greenish hue, attributed to maximal absorption in the blue spectrum and low absorption in the green spectrum (16, 22, 37).

Fig. 6. The linkage between water color and phytoplankton biomass (proxy as Chla).

(A) Graphing the relationship between Chla and water color on the CIE-xy chromaticity diagram. The color bar represents the log-transformed Chla concentration. (B) Scatter plot of the water colorimetry (hue angle, α*) and Chla concentration. The orange solid line denotes the scatter fit line within a 95% prediction interval.

Another question arises: What differences exist in the phytoplankton community structure in regions with different water color? Using the CIE-xy chromaticity diagram (Fig. 7A), we visually depict the color distribution associated with seven dominant PTGs. Diatoms predominate in inner-shelf waters, contrasting with cyanobacteria, which dominate in dark blue waters (e.g., ocean subtropical gyre). Haptophytes are prominent in blue waters, particularly in high hues. Conversely, chlorophytes show a high abundance in waters of varied colors, with a preference for green and light blue waters. Dinoflagellates, pelagophytes, and cryptophytes exhibit relatively low fractions in the global oceans of varying water colors. We also analyze the relationship between the fractions of seven PTGs (in logarithmic space) concerning ocean color variations (α*) (Fig. 7B). Among seven PTGs, Diatoms, haptophytes, and pelagophytes exhibit relatively high R² (>0.3) regarding variations in water colorimetry (Fig. 7B). In the transition from coastal to oceanic waters, α* increases, accompanied by a gradual decline in diatom fraction and simultaneous increase in haptophyte and pelagophyte fractions. Cyanobacteria exhibit distinct patterns: In a diatom-dominated community, the relationship between cyanobacteria and water color changes is weak, whereas, in a cyanobacteria-dominated community, cyanobacteria fractions progressively increase with rising α*, particularly in deep blue waters (R² = 0.19).

Fig. 7. Phytoplankton community composition reflected by water color changes.

(A) The relationship between seven dominant PTGs and water color based on the CIE-xy chromaticity diagram. The color bar of the scatter points represents the fraction of PTGs. (B) The relationship between α* and the fractions of seven PTGs (in logarithmic space). The orange solid line denotes the scatter fit line within a 95% prediction interval.

Over the past 25 years, low-latitude regions have experienced an upward trend in water color, potentially indicating a sustained decline in total phytoplankton biomass. This decline is also accompanied by a decrease in the fraction of large-sized phytoplankton such as diatoms and an increase in small-sized phytoplankton species (e.g., cyanobacteria and haptophytes), corroborated by previous research (38–42). Ongoing ocean warming has weakened vertical mixing, reducing surface nutrient availability, causing faster sinking of large-sized phytoplankton, and favoring small-sized phytoplankton (38, 40, 41). Conversely, high-latitude water color exhibits a decreasing trend, shifting toward a greener hue, especially in the Southern Ocean, indicating an increase in phytoplankton biomass. The sustained rise in SST leads to shallower mixing layers, promoting light availability (38, 43, 44).

Implications for ocean dynamics

This study reveals global ocean color changes since the early 21st century using satellite imagery based on the CIE-XYZ 1931 system; around 92% of global marine areas show color shifts compared to the early 21st century, with most waters retaining a blue hue. The water color difference between past and present indicates a color shift from deep blue to light blue or green in most of the global ocean (59%), suggesting a global “greening” of the ocean. This greening phenomenon (decrease in α*) by satellite observations is more frequent in mid-high latitude oceans, while low-latitude areas have experienced an increase in α* over the past 25 years. Another notable observation is that trend analysis based on satellite data indicates that 36% of water bodies (~122 million km², derived from valid observations) exhibit long-term trends. This phenomenon seems to be closely associated with variations in water column components. Existing studies, relying on model simulations, suggest that only certain regions may exhibit notable trend signals of water parameters over long time series (e.g., Chla, indicative of phytoplankton changes) (22). It should be noted that the period of our analysis is constrained by data availability, limited to the past 25 years, and the trend estimates and tests presented here specifically capture global water color changes at the beginning of the 21st century. Despite biogeochemical models and machine learning that can provide predictions, ongoing observation is necessary to determine when water color is affected by climate and sea temperature changes.

Influence of SST on water color exhibits distinct patterns, for example, a high positive correlation between $a_{\mathrm{t}}^{*}$ and SST^t appears in the Indian and Pacific Oceans, but SST^t tends to stabilize/uncertainty $a_{\mathrm{t}}^{*}$ in the Indian/Pacific Ocean; nutrient distribution, ocean transport, and changes in the mixing layer induced by SST possibly contribute to these differences (43). Furthermore, climate events, particularly ENSO and DMI, affect marine phytoplankton growth by driving SST changes, seasonal upwelling dynamics, and nutrient transport, leading to notable water color disturbances in equatorial and Indian Oceans (45–49) and possibly indicating uncertainty in the present and future phytoplankton variations (4, 22, 33, 49). Our study also demonstrates a notable impact of extreme SST and climate events on water color fluctuations, and the increasing frequency of marine heatwave events in recent years warrants sustained attention to their influence on the marine environment.

Global ocean change hot spots with regional distribution characteristics can also be attributed to the influence of mesoscale and sub-mesoscale processes on marine environments (8, 50). On the oceanic scale, phytoplankton biomass fluctuations are one of the important contributors to water color changes. Sub-mesoscale dynamics drive nutrient transport into the euphotic zone and affect the time phytoplankton spend in well-lit euphotic layers, potentially altering fractions of PTGs through mixing processes (46, 51–53). For example, in the Southeast Pacific and high-latitude regions, phytoplankton biomass increases potentially because of two important processes (hue angle decrease): upwelling that vertically transports nutrients to the euphotic layer, and ocean currents and eddy dynamics that expand the distribution of these nutrients, collectively supporting phytoplankton growth (46, 51–55). However, beyond oceanic physical processes, extreme SST events and climate change have exerted notable effects on the mid- and low-latitude oceans, particularly in the northeastern and equatorial Pacific. Extreme SST fluctuations and ENSO events have increased uncertainty in phytoplankton biomass changes (statistically significant SST and MEI NIF signals), especially with the frequent warm blobs in the northeastern Pacific affecting mixed layer depth fluctuations (56, 57), and reduced equatorial upwelling during ENSO events has disrupted nutrient transport (48, 49, 58). Under the context of global climate change and frequent marine heatwave events, we recommend maintaining continuous concern about the low-latitude oceans’ dynamics.

Global climate change alone may not fully account for the observed changes in coastal water color (lack of correlation and NIF synchronization signals in most coastal regions), suggesting that it possibly requires more attention to complex coastal environments and anthropogenic drives, e.g., inputs of terrestrial sediment and eutrophication-driven algal blooms. Variations in concentration and type of SPM affect water color characteristics in coastal and estuarine regions (16, 59), such as the Yangtze, Yellow, Amazon, and La Plata Rivers (31). Riverine inputs are one of the major sources of SPM in the coastal zone, reflecting both natural and anthropogenic influences on coastal water quality via lakes and rivers (60, 61). In addition to the effects of wind and seawater dynamics on matter resuspension, mixing, and transport (62, 63), human activities including river damming or alterations in land use affect the amount of sediment entering the ocean, which, in turn, affects the water quality of coastal rivers (31, 60, 64–67). Moreover, seasonal variations in phytoplankton biomass and blooms may potentially have an impact on anomaly fluctuations in the hue of coastal waters. Relatively high phytoplankton biomass, coupled with pronounced seasonal dynamics, influences coastal water hues, resulting in a tendency to shift toward green hues (16, 23), e.g., the East China Sea, and the Argentinean coastal waters. Noticeable color changes can be seen on satellite imagery during phytoplankton blooms, which are caused by large amounts of algae floating on the sea surface for a set amount of time. The reported distribution of algal blooms through satellite analyses is similar to identified coastal areas with high hue fluctuations (25), underscoring the potential impact of phytoplankton blooms on water color anomalies. Natural triggers, aquaculture, and fertilizer use in coastal regions seem to contribute to algal blooms (25, 26, 68–70).

MATERIALS AND METHODS

Data sources

The European Space Agency (ESA) Ocean Color Climate Change Initiative (OC-CCI) provides global binned multi-sensor time series of satellite ocean-color data and has been applied to a large number of climate and marine ecological and environmental research. The global geospatial distribution of water color during the past 25 years was generated on the basis of the monthly level 3, 4-km, version 6.0 OC-CCI remote sensing reflectance (R_rs) data from January 1998 to December 2022 from ESA Climate Office website (https://climate.esa.int/en/projects/ocean-colour/) (71). The 8-day 4-km R_rs datasets were also matched to the in situ phytoplankton datasets for R_rs values at the corresponding stations, enabling further estimation of water color parameters. The satellite-earth synchronization dataset matching rules refer to previous studies (72): (i) At least five or more valid values exist in the nearest 3 by 3 image pixels to the target station; (ii) if the coefficient of variation of the valid pixels is higher than 0.15, then the matchup was removed; and (iii) matchups with negative values in the red bands (665 nm) were excluded.

The linkage between water color and climate change was explored on the basis of satellite observations merged with SST and climate indices. In this study, the monthly ¹/₄° resolution high-resolution optimum interpolation SST (unit, degrees Celsius) was used as the analysis standard data and can be downloaded from National Oceanic and Atmospheric Administration (NOAA) Physical Sciences Laboratory (https://psl.noaa.gov/data/gridded/data.noaa.oisst.v2.highres.html) (73). Three classic climate events, namely, ENSO, Indian Ocean Dipole, and AMO, were examined, concerning their ability to drive water color fluctuations, and we chose the MEI, DMI, and AMO to represent each of these three types of climate events. The MEI provides a more comprehensive depiction of the atmospheric and oceanic anomalies observed during ENSO events. DMI is defined as the anomalous SST gradient between the western equatorial Indian Ocean (50°E to 70°E and 10°S to 10°N) and the southeastern equatorial Indian Ocean (90°E to 110°E and 10°S to 0°N). AMO characterizes SST anomalies in the North Atlantic region, exhibiting basin-scale, multiyear intergenerational oscillations. These three climate indices can be accessed through the NOAA Physical Sciences Laboratory website (https://psl.noaa.gov/enso/mei.ext/index.html) and the Global Climate Observing System website (https://psl.noaa.gov/gcos_wgsp/).

The in situ phytoplankton high-performance liquid chromatography (HPLC) pigment datasets were gathered from various sources, including NASA SeaBASS archive (https://seabass.gsfc.nasa.gov/), PANGAEA Data Center (https://pangaea.de/), Australian Ocean Data Network Portal (https://portal.aodn.org.au/), as well as our seven cruise surveys conducted in the marginal seas of China from 2016 to 2018 [refer to details in the study by Li et al. (72)]. These datasets encompass a wide range of water color conditions, spanning both oceanic waters and coastal waters, providing insights into the phytoplankton community. The quality control procedures for measured datasets followed three rules (72): (i) Samples with total Chla concentrations < 0.001 mg m⁻³ were removed; and (ii) pigments with default values or < 0.001 mg m⁻³ were adjusted to 0.001 mg m⁻³, and samples with values equal to or > 0.001 mg m⁻³ in more than three cases were rejected; (iii) samples collected in water depths > 10 m were omitted from the analysis.

The fractions of different dominant PTGs within the total biomass (represented by Chla) are determined using a rapid and accurate method, namely, diagnostic pigment analysis (DPA). DPA is a widely adopted approach for estimating phytoplankton abundance in marine environments by summing the weighted diagnostic pigments (associated with specific PTGs) and their contribution to the total Chla concentration (74–78). The weight coefficients used in this study were referenced from the study by Li et al. (72), which were derived from multiple linear regressions based on large numbers of global HPLC samples. Seven diagnostic pigments were used to determine the fractions of the corresponding PTGs, including total chlorophyll b (chlorophytes), fucoxanthin (diatoms), peridinin (dinoflagellates), 19′-butanoyloxyfucoxanthin (pelagophytes), 19′-hexanoyloxyfucoxanthin (haptophytes), alloxanthin (cryptophytes), and zeaxanthin (cyanobacteria). The in situ PTG samples were further used to match the corresponding satellite R_rs values, creating a satellite-earth synchronization dataset for subsequent analyzes.

Water chromaticity in CIE-XYZ 1931 color space

This study uses the CIE-XYZ 1931 color space, established by the CIE in 1931, to simulate the true color (the color of reflected light from the sea surface as captured by the human eye) of global sea surface waters. The CIE-XYZ color space characterizes a specific color through a two-dimensional chromaticity coordinate system [color chromaticity xy diagram (CIE-xy)], where the coordinates x and y are computed from three tristimulus values: X, Y, and Z. The mathematical relationship between chromaticity coordinates and the three tristimulus values, along with R_rs values, is as follows (21, 22, 24, 79–81)

$$\begin{array}{l}\left\{\begin{array}{c}X=k{\displaystyle \underset{{\mathrm{\lambda }}_1}{\overset{{\mathrm{\lambda }}_2}{\int }}}{R}_{\text{rs}}\left(\mathrm{\lambda }\right)\overline{x}\left(\mathrm{\lambda }\right)d\mathrm{\lambda }\hfill \\ Y=k{\displaystyle \underset{{\mathrm{\lambda }}_1}{\overset{{\mathrm{\lambda }}_2}{\int }}}{R}_{\text{rs}}\left(\mathrm{\lambda }\right)\overline{y}\left(\mathrm{\lambda }\right)d\mathrm{\lambda }\\ Z=k{\displaystyle \underset{{\mathrm{\lambda }}_1}{\overset{{\mathrm{\lambda }}_2}{\int }}}{R}_{\text{rs}}\left(\mathrm{\lambda }\right)\overline{z}\left(\mathrm{\lambda }\right)d\mathrm{\lambda }\end{array}\right.\\ x=X/(X+Y+Z)\\ y=Y/(X+Y+Z)\end{array}$$ (1)

where k is the illumination and is defined as $k=Y_{\text{CIE}}/$ $\left[{\int }_{{\mathrm{\lambda }}_1}^{{\mathrm{\lambda }}_2}{R}_{\text{rs}}\left(\mathrm{\lambda }\right)\overline{y}\left(\mathrm{\lambda }\right)d\mathrm{\lambda }\right]$ [the CIE specifies a light source Y-stimulus (Y_CIE) value of 100]; $\overline{x}\left(\mathrm{\lambda }\right)$ , $\overline{y}\left(\mathrm{\lambda }\right)$ , and $\overline{z}\left(\mathrm{\lambda }\right)$ denote the color-matching functions (79, 80). λ signifies the wavelength of R_rs and color-matching functions. In this study, the specific integrals of the three tristimulus values within the interval [λ₁, λ₂] are computed using the summation of Riemann integrals in the OC-CCI wavelength range after linear interpolation (21, 81).

We use hue angle (α*), signifying the angle of the hue, to describe the water color of each pixel in OC-CCI data. Constructing a coordinate system within the xy chromaticity system with the equivalent white point (x_w = y_w = 1/3) as the origin, where α* is the angle between the ray from the target point (x, y) to the origin and the coordinate system’s horizontal axis (21, 22, 80). The conversion of x and y to α* is performed as follows

$${\mathrm{\alpha }}^{*}= \text{arctan}[(y-y_{\mathrm{w}})/(x-x_{\mathrm{w}})] \text{modulus} 2\mathrm{\pi }$$ (2)

where α* converts from radians to degrees in this study. It is worth noting that, as the chromaticity of the sea surface falls within the saffron to blue range in most cases, we do not consider cases of complementary colors. The CIE-xy diagram and the relationship between the hue of color and α* are shown in Fig. 8. In natural waters, α* ranges from ~40° to ~235°, transitioning from brown (~40°) to green (~100°) and blue (~200°) waters and then to deep blue oceanic waters (~235°), as shown in the chromaticity diagram (22, 24).

Fig. 8. Color chromaticity xy diagram in CIE-XYZ 1931 color space (CIE-xy).

The relationship between the coordinates of the target point, the equivalent white point, the dominant wavelength, and the hue angle (α*) (79, 80).

Time-series correlation and causality analysis

The correlation and causality relationship between water color and climate change were analyzed using the Pearson correlation coefficient and L-K IF. This study was performed using MATLAB version R2021b (MathWorks, 2021) and its extension libraries for data processing and figure production. In this study, each global observation product was down-sampled to a spatial resolution of ¹/₄°. Addressing missing pixel values involved spatial interpolation within a 3-by-3 pixel space using the bilinear method, and any remaining gaps were filled on the basis of the temporal dimension of each pixel using the nearest-neighbor method. The water color raw time series (X_wc) underwent decomposition into seasonal (S_wc), interannual (I_wc), and anomaly (A_wc) components by using the singular spectrum analysis method (82), i.e., X_wc = S_wc + I_wc + A_wc. Additionally, the Theil-Sen median method was applied to evaluate trends in the interannual components of long–time-series data. Significance tests for the trend values were conducted using the Mann-Kendall test (significance level P < 0.1 in this study). For each pixel, if the number of effective values comprises less than one-third of the entire time series, then that pixel is excluded from the analysis. The time-series data were also normalized for correlation analysis.

The L-K IF method has been rigorously established for quantitative analysis of causality between two time series (27). Causality in terms of IF can be regarded as a real physical notion ab initio (27, 29, 83). The validation and successful application of the L-K IF to problems in various disciplines are now underway, e.g., CO₂ emission–global warming relation, the long-term El Niño prediction, light absorption of phytoplankton, and current dynamics (83–86). According to Liang (87), for a set of d time series X₁, X₂, …, X_d, the maximum likelihood estimator for the rate of information flow, referred to as information flow, from one series X_j to another series X_i (in nats per unit time), is

$$T_{j\to i}=\frac{1}{\text{det}\mathrm{C}}\cdot {\displaystyle \sum _{k=1}^d}{\mathrm{\Delta }}_{\mathit{jk}}{C}_{k,\mathit{di}}\frac{C_{\mathit{ij}}}{C_{\mathit{ii}}}$$ (3)

where C is the covariance matrix and C_ij represents the covariance between X_i and X_j; ∆_jk denotes the cofactors of the covariance matrix of C = (C_ij); C_k,di signifies the sample covariance of X_k and the Euler forward differencing estimate of dX/dt. Considering the IF between two time series (d = 2), Eq. 3 can be reduced as

$$T_{2\to 1}=\frac{C_{11}{C}_{12}{C}_{2,d1}-C_{12}^2{C}_{1,d1}}{C_{11}^2{C}_{22}-C_{11}{C}_{12}^2}$$ (4)

The IF in normalized form (NIF) allows a better comparison of the strength of the causal relationship, as follows (28)

$$\left\{\begin{array}{c}{Z}_{2\to 1}=\mid T_{2\to 1}\mid +\mid \frac{{\mathit{dH}}_1^{*}}{\mathit{dt}}\mid +\mid \frac{{\mathit{dH}}_1^{\text{noise}}}{\mathit{dt}}\mid \hfill \\ {\mathrm{\tau }}_{2\to 1}=T_{2\to 1}/Z_{2\to 1}\end{array}\right.$$ (5)

where Z_2→1 is the normalizer and H₁ is entropy. $\mid {\mathit{dH}}_1^{*}/\mathit{dt}\mid$ denotes the contribution of H₁ due to the lack of randomness of X1, while $\mid {\mathit{dH}}_1^{\text{noise}}/\mathit{dt}\mid$ expresses the contribution of noise to H₁. τ_2→1 is the proportion of X₂ to X₁ information flow in the total cause contribution. It is noted that causality and correlation are not necessarily symmetrical; causation implies correlation, but the reverse is not always true. If τ_2→1 > 0, then X₂ increases the entropy of X₁ and creates more uncertainty or unpredictability; if τ_2→1 < 0, then X₂ will reduce the entropy of X₁ and stabilize it; if τ_2→1 = 0, then X₂ is not the cause of X₁.