Ocean-scale patterns of environment and climate changes driving global marine phytoplankton biomass dynamics

Abstract

Effects of marine environment and climate changes on phytoplankton dynamics in global oceans have received increasing attention but remain a mystery. This study used a comprehensive approach combining correlation and information flow to explore relationships among phytoplankton biomass, marine environment, and climate forcing based on global observations over the past multi-decadal period. Correlation and causality between phytoplankton biomass and environmental factors exhibit spatial asymmetry—regions where environmental factors directly drive biomass variations were concentrated in oceanic currents and subtropical circulations. Temperature, light, and mixed layer depth show pronounced influences on global phytoplankton interdecadal variations. Climate forcing over interdecadal timescales directly affects phytoplankton biomass in the equatorial Pacific, South Pacific, and Indian Oceans, with more uncertain biomass variability in the equatorial Pacific due to multiple climate events. Our findings revealed that environment and climate changes directly affect phytoplankton interdecadal variability only in specific regions at the oceanic scale.

Environment and climate changes directly affect phytoplankton interdecadal variability only in specific regions at oceanic scale.

INTRODUCTION

Marine phytoplankton account for half of global primary production and are vital to climate regulation, ecosystem function, and biogeochemical cycling (1–4). Phytoplankton are a fundamental part of the marine food web and play an important role in the productivity of marine ecosystems (5). Variations in marine phytoplankton biomass and communities are of interest in multiple fields. Global phytoplankton growth conditions are influenced by regional variations and seasonal cycles in marine environments, which are also affected by climate change (6). Previous studies have observed the influence of climate change on the marine environment over the past few decades, and climate-related changes in ocean properties have become apparent, such as increased temperatures and ocean acidification (7, 8). Extensive research has revealed a correlation between climate fluctuations and marine ecology with potential effects on phytoplankton dynamics (9, 10). Phytoplankton, which are affected by ecological changes in the marine environment, are sensitive to global sea surface temperature (SST) disturbances driven by climate change (4, 11). Evaluating the effects of oceanography and climate on phytoplankton in different global geographic regions provides the potential to better analyze and predict changes in phytoplankton biomass and communities.

Global marine physical and ecological datasets and long time-series phytoplankton satellite products have been developed to conduct large-scale change analyses of multiple oceanographic and climate factors and determine their effects on phytoplankton biomass (5, 12–16). Several studies have investigated the correlation between certain oceanography factors, climate factors, and phytoplankton biomass changes at global or regional scales and have gradually uncovered the driving mechanisms on phytoplankton biomass variations through experiments or modeling simulations (4, 5, 9, 17–19). The seasonality of phytoplankton biomass has been reported to be controlled by seasonal variations in solar irradiance, wind, and surface stratification (20), and the interannual variability in physical and biological oceanography in some areas is modulated by climate oscillations (21). Tropical and subtropical regions, which cover more than 60% of the global ocean, have a reduced supply of surface nutrients owing to strong, persistent thermal stratification, which results in low phytoplankton biomass and the formation of complex microbial food webs dominated by small phytoplankton (22).

Most studies attempting to explain the year-to-year driving mechanisms of biomass dynamics have predominantly relied on time-lagged correlations (6, 15, 18, 21, 23); however, such approaches are limited by the lack of clear causal relationships between drivers and biomass dynamics, that is, correlation does not necessarily imply causality (24). Therefore, the effect of external factors on phytoplankton biomass under natural conditions at an interannual scale requires further investigation. Information flow (IF), expressed as the time rate of IF from one series to another, provides a natural measure for the causal interaction between dynamical events (24, 25). The combination of correlation and IF methods allows for a view of phytoplankton biomass influenced by multiple oceanographic or climatic factors based on merged multiple simulation or observation datasets.

In this study, we used chlorophyll a (Chla) as a biomass proxy from satellite observations in combination with several classic environmental parameters to explore the effects of the marine environment on phytoplankton biomass over the past multi-decadal period. We also took this opportunity to examine the relationship between global climate change and phytoplankton biomass on an interannual scale. Geographic maps of correlation and IF signals were obtained from multiple global long time-series datasets to address whether variations in phytoplankton biomass were driven by marine environmental or climatic factors. This study analyzed the correlations and cause-effect relationships between biotic and abiotic factors to exhibit potential linkages between global phytoplankton and multiple marine environmental and climatic factors at an oceanic scale.

This study used global long-term time-series Chla satellite observations to detect spatiotemporal variations in phytoplankton. We examined the phytoplankton environmental response mechanism by analyzing marine environmental features (MEFs) that directly or indirectly influence phytoplankton biomass variations, including SST, photosynthetically available radiation (PAR), and mixed layer depth (MLD). In addition, climate and phytoplankton biomass were synergistically analyzed using five major SST anomaly indices, the Niño 3.4, Southern Oscillation Index (SOI), Pacific Decadal Oscillation (PDO), Dipole Mode Index (DMI), and Atlantic Multi-decadal Oscillation (AMO).

RESULTS

MEFs versus phytoplankton biomass

We exhibit and review annual spatial patterns of global Chla concentration (a proxy for phytoplankton biomass) and three MEFs (Fig. 1). Overall, the geospatial distribution patterns of these parameters derived from satellite observations or model simulations are determined mainly by latitude, with variations between coastal regions and open oceans. High phytoplankton biomass is primarily observed in coastal and high-latitude waters, whereas subtropical regions, especially the South Pacific Ocean (~25°S), tend to have lower levels. SST and PAR showed similar distribution patterns, with higher values concentrated in the middle- and low-latitudinal regions. The MLD tends to have greater depths in the middle- to high-latitude areas of the North Atlantic and Southern Hemispheres.

Fig. 1. The global annual average distribution of phytoplankton biomass (proxy as Chla) and MEFs from 1997 to 2020.

(A) Chla concentration (unit, milligrams per cubic meter). The subfigures of Chla is shown in log₁₀ space. (B) SST (unit, degrees Celsius); (C) PAR (unit, einsteins square meter per day); (D) MLD (unit, meters).

The combined Chla and MEFs products from 1997 to 2020 were used to characterize the response of phytoplankton biomass to MEFs. Generally speaking, SST exhibited a negative correlation [average correlation coefficient (r) = −0.57, P < 0.05] with Chla in the low- and middle-latitude oceans (~40°N to 40°S), whereas a positive correlation (average r = 0.42, P < 0.05) appeared in the high-latitude areas (>40°N or 40°S) (Fig. 2A). PAR and Chla had a negative relationship in middle-latitude oceans (average r = −0.51, p < 0.05) but a positive correlation (~0.39, P < 0.05) in the high-latitude and equatorial Pacific regions (Fig. 2B). In contrast to SST and PAR, MLD demonstrated a positive connection with Chla in most areas of the middle-latitude oceans (average r of 0.56, P < 0.05), combined with negative trends in high-latitude ocean and the eastern equatorial Pacific Ocean (Fig. 2C).

Fig. 2. Global distribution of correlation coefficients (r) and normalized information flow (NIF).

The global distribution of correlation coefficients between Chla with MEFs (A to C) and NIF from MEFs to Chla (τ_MEFs→Chla) (D to F), with a significance level of 5%. The blank areas in the above subfigures refer to pixels with no valid time series or insignificant signals. The full correlation and NIF distribution maps are shown in fig. S1.

This study explored the causality between MEFs and Chla [using normalized IF (NIF) as a quantification indicator] (Fig. 2), demonstrating various oceanic-scale characteristics for different MEFs driving Chla from a global distribution perspective. SST is an essential driver of Chla changes, as evidenced by the NIF from SST to Chla (τ_SST→Chla) that appears in multiple regions of the global oceans, mainly including negative signals in subtropical areas and positive signals in middle- to high-latitude and equatorial regions throughout the Pacific and Indian Oceans (Fig. 2D). In the subtropical regions, the negative NIF (average τ_SST→Chla = −0.3) indicates that SST stabilizes Chla fluctuations, while SST increases the uncertainty of Chla fluctuations in middle- to high-latitude and equatorial regions, as reflected by the positive NIF signal (average τ_SST→Chla = 0.12, P < 0.05). Similarly, the NIF from PAR to Chla (τ_PAR→Chla) is predominantly positive, with an average τ_PAR→Chla of 0.19 (P < 0.05), in the northern and southern subtropical regions of the Pacific and Atlantic Oceans, as well as the southern Indian Ocean (Fig. 2E). This positive NIF indicates that PAR contributes to Chla variability, leading to more uncertain conditions. Analysis MLD reveals a widespread spatial distribution of regions with a causal relationship to Chla (Fig. 2F), noting that the subtropical regions (~25°S) of the Atlantic and South Pacific Oceans display negative signals (average τ_SST→Chla = −0.23, P < 0.05), while positive signals are present in the Tasman Sea, middle- and high-latitude regions.

Climate indices versus phytoplankton biomass

The correlation and causality relationships between multi-decadal changes in global phytoplankton and climate were revealed from ~30 years of satellite observations [Coastal Zone Color Scanner (CZCS), November 1978 to June 1986; Sea-viewing Wide Field-of-view Sensor (SeaWiFS), October 1997 to December 2002; and Moderate Resolution Imaging Spectroradiometer aboard the AQUA satellite (MODIS-Aqua), January 2003 to March 2022]. The distribution of the correlation signals between several climate indices (CI; a proxy for SST anomaly indices) and Chla showed a clear regional pattern, with significant signals (P < 0.05) mostly occurring in the Pacific and Indian Oceans (Fig. 3A). In the equatorial Pacific and Indian Oceans, Niño 3.4 showed a negative relationship with Chla (average r = −0.28, P < 0.05), but, in the South Pacific, it was positive. Conversely, SOI displayed the opposite pattern, demonstrating a clear positive correlation in the equatorial Pacific (average r = 0.23, P < 0.05). The PDO exhibited mostly negative signals in several Pacific and Indian Ocean areas, whereas positive signals were seen in the North Atlantic and Northwest Pacific Oceans. DMI and Chla had a negative correlation in the Indian Ocean and equatorial Pacific, and positive correlations were observed in the Sumatra Sea. The AMO signal exhibits a relatively wide global coverage, characterized by a mostly negative correlation (average r = −0.22, P < 0.05) in the northern Pacific and Atlantic Oceans and a positive correlation (~0.28, P < 0.05) in the equator and the Southern Hemisphere.

Fig. 3. Correlation and causality analysis of climate forcing on phytoplankton biomass.

Global distributions of correlation coefficients (r) between Chla with climate indices (CI) (P < 0.05) (A) and NIF from climate indices to Chla (symbolized by τ_CI→Chla) (B). Blank areas represent pixels that have no valid time series, insignificant correlations (A), or insignificant NIF (B). The full correlation and NIF distribution maps are shown in fig. S2.

To assess the potential drivers of climate change on Chla, our NIF analysis of five climate indices (Niño 3.4, SOI, PDO, DMI, and AMO) to Chla (τ_CI→Chla) shows mostly positive signals in the Pacific and Indian Oceans, suggesting that climate change is causing increasingly unpredictable and uncertain changes in Chla (Fig. 3B). However, the causal relationship between climate indices and Chla is only evident in limited spatial areas. The equatorial Pacific is an essential region where Chla is driven by several climate events, with notable NIF signals (~0.2, P < 0.05) from climate indices to Chla. Niño 3.4 and SOI mainly act on neighboring regions of the equatorial Pacific, with almost no causality in other areas. The NIF flow from DMI to Chla occurs mostly in the equatorial Indian Ocean. In addition to the impacts of PDO and AMO on equatorial phytoplankton biomass negative PDO and AMO NIF signals in the middle-latitude regions of the Southern Hemisphere indicate that PDO and AMO variations contribute to the stabilization of Chla fluctuations.

DISCUSSION

Marine environmental drivers of phytoplankton biomass variation

We summarized the spatial distribution of the NIF and corresponding correlation signals obtained according to the driving patterns of different MEFs on phytoplankton biomass (Table 1), showing the main specific response of phytoplankton to interannual variability in the marine environment at the oceanic scale (Fig. 4). Main global NIF distribution patterns reveal pronounced responses of phytoplankton biomass to MEF variations in oceanic currents, gyre areas, and equatorial regions, with physical factors (SST, PAR, and MLD) primarily controlling phytoplankton growth. In addition, we observed an asymmetry between phytoplankton biomass and environmental factors in terms of correlation and causality, suggesting that the direct driving of phytoplankton by MEFs is region specific. The distribution of NIF signals indicated that one or more MEFs clearly influenced biomass dynamics in middle- to low-latitude regions. Conversely, MEFs may indirectly affect phytoplankton biomass changes in regions with correlations but no causal signals.

Table 1. Classification of NIF signal distribution modes of phytoplankton biomass in response to MEFs in the global oceans.

NIF modes	Positive phase	Negative phase
Case I	SST and MLD	/
Case II	PAR and MLD	/
Case III	PAR	SST and MLD
Case IV	SST	/
Case V	SST, PAR, and MLD	/

Fig. 4. The major driving modes of variations in phytoplankton biomass (expressed by Chla) by typical MEFs in the global oceans, as assessed by causality analysis.

(A) Case types and distribution. (B) Average NIF values (bold and underlined style) and correlation coefficients (normal style) of MEFs with Chla for various cases. The red and blue arrows represent warm and cold currents, respectively.

Analyzing the NIF and correlation modes provided evidence of the direct influence of SST on phytoplankton growth across global oceans. This outcome can be attributed to the crucial role of temperature in modulating a diverse array of physiological processes among various phytoplankton taxa and its impact on the structural integrity of biomolecules within their cellular composition (26). The SST NIF signal (Fig. 4, cases I, III, and IV) indicates that SST contributes to heightened uncertainties in biomass variations within oceanic currents (e.g., North Pacific/Atlantic warm current) and equatorial regions while concurrently promoting stability within the subtropical circulation regions. Corresponding negative correlations of SST with Chla indicate that stratification induced by elevated SST levels in tropical and subtropical waters restricts the nutrient supply available to phytoplankton (6), e.g., the drivers of phytoplankton growth via nitrate and phosphate at the equator (27, 28). A distinct phenomenon was the substantial negative SST correlation and NIF signal in the middle-latitude eastern Pacific region (Fig. 4, case III), suggesting that interannual SST variability exerted a stabilizing influence on phytoplankton disturbances in this particular area. The impact of frequent Pacific Ocean heat waves on the phytoplankton community composition warrants diligent monitoring (29).

The impacts of light and MLD on phytoplankton were notably prominent in the middle-latitude regions (Fig. 4, cases II and III). Light controls the variation in phytoplankton biomass in subtropical regions and is one of the leading causes of biomass variation in these regions, contributing to more uncertainty for biomass variability. Light usually promotes phytoplankton photosynthesis, whereas Chla concentrations decrease under high radiation (30, 31). The intensity of light radiation in subtropical regions is higher than that in the northern and southern regions, and high radiation increases SST, leading to a shallow MLD and a chronic state of phytoplankton photoinhibition (32–34). On the basis of the MLD model–derived data, we further revealed that the MLD plays an important role in influencing middle-latitude phytoplankton growth, encompassing several essential factors such as grazing pressure, entrainment of nutrient-rich deep water, light availability, temperature, and short-term environmental variability (3, 35, 36). The impact of vertical mixing on nutrient-driven variations facilitates the upward transport of nutrients from the deep sea to the ocean surface, providing essential support for phytoplankton growth (37, 38).

We observe a specific regional mode in the Tasman Sea in southeastern Australia (Fig. 4, case V). All MEFs consistently showed a positive NIF signal, indicating an increased uncertainty in the perturbation of phytoplankton biomass in the region. The Tasman Sea is sensitive to climate change, especially considering recent intense heat wave events, which may potentially affect ocean stratification, nutrient supply, and even the composition of phytoplankton communities (39–41). Furthermore, exogenous nutrient inputs to the region are also worthy of attention, e.g., the effect of Australian aerosols on iron inputs to phytoplankton biomass (42). In the context of external inputs and global warming, future changes in the phytoplankton community in the Tasman Sea require further monitoring and attention.

Despite our main focus on the response of Chla variability at the ocean scale, our findings also revealed that some coastal waters exhibited few or no notable causality signals MEFs with Chla. On one hand, our results are derived from analyses of satellite observations and model simulations, where global satellite or model outputs potentially lack sufficient accuracy in capturing the dynamics of complex coastal waters, thereby limiting the correlation and causality signals evaluated in complex coastal waters. On the other hand, this discrepancy can be attributed to the complexity of coastal water systems. For instance, wind-driven upwelling in the Santa Barbara Channel is considered the main driver of seasonal and interannual variations in phytoplankton biomass and community composition (43–45), while this seasonal upwelling is susceptible to the influence of climate change (46–48). Furthermore, diatom variability in this area is potentially affected by freshwater flows and long-term changes in regional advection patterns, among other factors (21). Anthropogenic additional nutrient enrichment, such as fertilizers and the development of marine aquaculture (49, 50), can also potentially drive changes in phytoplankton biomass dynamics and even lead to harmful algal blooms. The proliferation of phytoplankton in some Asian countries can be attributed to the increased use of nitrogen or phosphate fertilizers (50, 51). Therefore, it is necessary to consider various factors, including marine dynamics, climate, and human influences, when analyzing seasonal or interannual driving mechanisms of phytoplankton biomass dynamics in coastal waters.

Response of phytoplankton dynamic to climate forcing over multi-decadal period

Climate forcing drives interannual variations in ocean circulation or seasonal upwelling dynamics, alters marine mixing and the transport of nutrients, and affects the growth of marine phytoplankton (52–54). We conclude by examining the major global pattern of climate-driven phytoplankton biomass variations (Table 2); our results show that the response of phytoplankton biomass variations to climate at long temporal scales has regional characteristics (Fig. 5), with the Pacific and Indian Oceans being the central regions where climate dynamics directly affect phytoplankton biomass. In the equatorial Pacific region, phytoplankton biomass was concurrently affected by multiple climate events, as supported by the notable positive phases of the NIF signals across all climate indices. Seesaw shifts in SST and thermocline depth reflect climate fluctuations, to which local phytoplankton biomass responds, e.g., increased SST leads to increased stratification, reduced nutrients, and decreased phytoplankton growth (6). The clear NIF signals of various climate indices in the equatorial Pacific and Indian Ocean regions also reflect that coupled ocean-atmosphere interactions between the Pacific, Atlantic, and Indian Oceans trigger and regulate global climate change (55), which is a potential element for global phytoplankton biomass forced by different climate events.

Table 2. Classification of NIF signal distribution modes of phytoplankton biomass in response to climate indices in the global oceans.

NIF modes	Positive phase	Negative phase
Case I	Niño 3.4, SOI, AMO, PDO, and DMI	/
Case II	Niño 3.4, SOI, DMI, and AMO	/
Case III	SOI	AMO and PDO

Fig. 5. Main driving modes of changes in phytoplankton biomass (represented by Chla) by climate indices, as evaluated using causality analysis.

(A) Case types and their distribution. (B) Average values of NIF (bold and underlined style) and correlation coefficients (normal style) of climate indices with Chla for different Cases. Red and blue arrows denote warm and cold currents, respectively.

Positive NIF signals between climate indices and Chla indicate that climate change amplifies the unpredictability of biomass variation in the equatorial Pacific region. Correlation analysis revealed that the positive phases of Niño 3.4, PDO, and DMI coincide with a decline in biomass, whereas the positive phase of AMO increased biomass. Furthermore, the middle-latitude regions of the South Pacific and southern Indian Oceans exhibited SOI, PDO, and AMO NIF signals, with interdecadal fluctuations in the PDO and AMO contributing to stabilizing phytoplankton biomass disturbances, and the opposite for SOI (Fig. 5, case III). The AMO has remained positive since 1998, and elevated SST has promoted the dominance of picophytoplankton in the North Atlantic (56). Our findings suggest that the impact of AMO on global phytoplankton biomass appears to be more pronounced, as indicated by the presence of more regions with τ_AMO→Chla signals and a widespread correlation signal. Note that, while the direct drivers of climate on interdecadal phytoplankton changes are primarily observed in limited regions, global climate change can influence the marine environment (e.g., SST and mixing layer), potentially increasing the uncertainty of biomass perturbations. Therefore, continuous monitoring of global phytoplankton community structure is necessary to better understand the ongoing impacts of climate change.

Extended analysis of PTGs

Our analysis primarily explored the linkages among total phytoplankton biomass, MEFs, and climate indices using correlation and IF methods for global oceans. Advancements in remote sensing technology have facilitated the production of long-term datasets for typical phytoplankton taxa globally, offering valuable data support for understanding the changes in phytoplankton community structure. An absorption-based model for the Chla concentration (milligrams per cubic meter) of the dominant phytoplankton taxonomic groups (PTGs) was developed, and global distribution records were obtained from July 2002 to March 2022 (see the “Methods” section) (16). Four dominant PTGs, namely, chlorophytes, diatoms, haptophytes, and cyanobacteria, explained ~73% of the assemblage variation, and we provide a preliminary analysis of the correlation and causality of their variations with MEFs and climate (Figs. 6 and 7). The four PTG biomass types present nearly identical relevance signals to MEFs, and the calculated NIF (τ_MEFs→PTGs) shows similar spatial distribution characteristics among different PTGs. In the correlation and causality analyses, the PTGs both exhibited spatial patterns similar to those of Chla. Conversely, this study has also conducted correlation and causality analyses between four dominant PTG Chla concentrations and climate indices. The spatial patterns of the correlation between the PTGs and climate indices were similar to those of Chla, and the same characteristics were observed in the NIF flowing from climate indices to PTGs, with notable signals mainly distributed in the Pacific region. The effects of climate change on the different dominant phytoplankton biomass were similar over the past multi-decadal period, showing a clear influence of climate events on phytoplankton community changes in equatorial regions, requiring sustained attention.

Fig. 6. Correlation and causality analysis of marine environmental factors on PTG Chla concentrations.

Global distribution of correlation coefficients (r) between four dominant PTGs with MEFs (A), and NIF from MEFs to PTGs (τ_MEFs→PTGs) (B), with a significance level of 5%. Blank areas represent pixels that have no valid time series, insignificant correlations (A), or insignificant NIF (B).

Fig. 7. Correlation and causality analysis of climate forcing on PTG Chla concentrations.

Global distribution of correlation coefficients (r) between four dominant PTGs with climate indices (CI) (A), and NIF from climate indices to PTGs (τ_CI→PTGs) (B) (P < 0.05). Blank areas indicate pixels with no valid time series, insignificant correlations (A), or insignificant NIF (B).

Future implications

Applying the IF methodology yielded valuable insights into the driving mechanisms of marine phytoplankton biomass dynamics (24, 25). By analyzing global oceanography and climatology concerning phytoplankton biomass with correlation and NIF methods, the driving mechanisms of phytoplankton were found to be diverse at the ocean scale and exhibit complex interannual variability. The results revealed notable patterns such as the dominant influence of SST, PAR, and MLD on phytoplankton biomass variations at middle and low latitudes (e.g., equator and subtropical circulation) and the additional impact of climate events in the equatorial region. Furthermore, the regional sporadic distribution of NIF signals deserves further attention in the future, e.g., the presence of positive SST and PAR NIF signals at high latitudes suggests an impact on biomass: Increased light and warming trends enable phytoplankton growth rate to reach an optimal level (11, 57, 58). Our findings emphasize the need for a comprehensive investigation into the mechanisms governing the interannual dynamics of phytoplankton biomass, integrating multiple factors to more accurately analyze the drivers of marine primary productivity and carbon cycling. The complexity of factors controlling the variability of phytoplankton biomass in coastal waters demands more attention, especially regarding the impacts of human activities.

Dynamic variations in global marine phytoplankton biomass and communities have garnered continuous attention, particularly concerning the modeling or prediction of past and future biomass using ecological mechanisms or machine learning approaches (5, 12–15). Our results emphasize the asymmetry in correlations between environmental factors and climate with phytoplankton biomass, which is mirrored by the asymmetry of causality relationships between environmental factors or climate and phytoplankton biomass. Furthermore, our results highlight the importance of considering the effects of multiple causally related factors on biomass. Understanding global marine biological engines, specifically the causal relationships among the environment, climate, and phytoplankton, plays a crucial role in quantifying biogeochemical fluxes and forecasting future changes in marine plankton ecosystems (4, 5, 9, 17). We recommend considering the causality relationships between environmental or climatic factors and changes in biomass in future modeling or predictive research and selecting factors that effectively influence phytoplankton dynamics to develop more authentic models encompassing phytoplankton responses to environmental or climate changes.

MATERIALS AND METHODS

Data sources

The time-series datasets used in this study include satellite observations or model simulations that document global variations in phytoplankton biomass, MEFs, and climate indices. The spatiotemporal variations of phytoplankton biomass (proxy as Chla; unit, milligrams per cubic meter) were observed by the Chla monthly level-3 satellite products of the CZCS, SeaWiFS, and MODIS-Aqua, available at https://oceancolor.gsfc.nasa.gov/ from the NASA Goddard Space Flight Center (GSFC) covering November 1978 to June 1986, October 1997 to December 2002, and January 2003 to December 2022 with a raw spatial resolution of 9 km.

The global SST and PAR time-series records were also derived from satellite observations. The monthly ¹/₄° resolution high-resolution optimum interpolation SST (unit, degrees Celsius), produced by integrating multi-platform observations (satellites, ships, buoys, and Argo floats) using optimal interpolation methods, was distributed by NOAA Physical Sciences Laboratory (https://psl.noaa.gov/data/gridded/data.noaa.oisst.v2.highres.html) (59). Similarly, global long time-series variability data for PAR are obtained from the monthly level-3 mapped 9-km resolution PAR (unit, einsteins per square meter per day) products of MODIS-Aqua, downloaded from NASA GSFC (https://oceancolor.gsfc.nasa.gov/) (60). Long time-series global distribution results for MLD were derived from model simulations from September 1997 to December 2020. The monthly MLD data were assimilated from the Geophysical Fluid Dynamics Laboratory’s Modular Ocean Model (version 3) in a quasi-global configuration, distributed from the NCEP Global Ocean Data Assimilation System (https://psl.noaa.gov/data/gridded/data.godas.html), with 1° × 0.3° spatial resolution (61).

Climate indices in an anomaly format covering 1978–2022 are available from the Global Climate Observing System website (https://psl.noaa.gov/gcos_wgsp/) (Fig. 8). These indices include the Niño 3.4, which measures the anomaly SST in the Equatorial Pacific region between 5°S to 5°N and 170°W to 120°W (62); SOI, which represents the normalized pressure difference between Tahiti and Darwin (63); PDO, which characterizes a long-lived El Niño–like pattern of Pacific climate variability at 20°N in the Pacific basin (64); DMI, which is defined by 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) (65); and AMO, which describes SST anomalies in the North Atlantic region with basin-scale, multiyear intergenerational oscillations (62).

Fig. 8. Indices of the dominant modes of global climate variability from 1978 to 2022.

(A) Niño 3.4; (B) PDO; (C) SOI; (D) AMO; (E) India Ocean DMI.

Methods

Data preprocessing

Climatological errors in the satellite observations were eliminated by adjusting the monthly average characteristics of the SeaWiFS and CZCS data to match those of MODIS-Aqua. To minimize the error and default values from the global observation data, spatial interpolation was performed on long time-series data at 3 × 3 pixel space dimensions using the median filtering method, and missing values near the valid ones were further interpolated according to the temporal dimension of each pixel using the nearest method. The effective latitude range for all observational data was specified as 60°S to 60°N because of the seasonal absence of high latitudes and polar regions. All global data were uniformly resampled to a spatial resolution of ¹/₄° by the bilinear method, and the time series of each image pixel was removed the long-term linear trend and normalized (standard score method) before analysis. The correlation and causality analysis between the factors was conducted on a per-pixel basis, ensuring that the valid values of the time series of each pixel, excluding outliers or missing values, constituted more than two-thirds of the total observations.

Time-series correlation and causality analysis

The Spearman’s rank correlation coefficient (r) was used to assess the potential correlation between the phytoplankton biomass and various environmental and climatic factors. The calculated correlation coefficients should also have a P value of less than 0.05. After obtaining the global correlation distribution results between the factors, the images were smoothed using the median filtering method.

Causality analysis was based on the Liang-Kleeman (L-K) IF method (24). The L-K IF method has been rigorously established from the first principles for quantitative analysis of causality between two time series (24). Causality in terms of IF can be regarded as a real physical notion ab initio (24, 66, 67). 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 (66, 68–70). According to Liang (71), the IF from time series X2 to X1 (T_2→1) can be defined as

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

where C_ij and C_i,dj denotes the covariance of X_i and X_j and of X_i and ${\dot{X}}_j$ (${\dot{X}}_j$ = $\frac{X_{j,n+1}-X_{j,n}}{\text{∆}t}$, Δt denotes time step), respectively. If T_2→1 > 0, then X2 will introduce additional uncertainty to X1; if T_2→1 < 0, then X2 will reduce the entropy of X1 and stabilize it; when T_2→1 = 0, X2 is not the cause of X1. The NIF allows a better comparison of the contributions of different factors to phytoplankton (25)

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

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 in X1, whereas $\mid {\mathit{dH}}_1^{\text{noise}}/\mathit{dt}\mid$ expresses the contribution of noise to H₁. τ_2→1 is the proportion of X2 to X1 IF in the total cause contribution. If τ_2→1 > 0, then X2 creates more uncertainty or unpredictability for X1; if τ_2→1 < 0, then X2 reduces X1 uncertainty and stabilizes X1; else τ_2→1 = 0, X2 does not affect the change in X1. This study primarily discussed the unidirectional IF of MEFs or climate indices on phytoplankton biomass, with a significance level of 5%.

Satellite estimation concentrations of PTGs

Global Chla concentrations of dominant PTGs were estimated using a customized absorption-based model (16) through the quasi-analytical algorithm (QAA) model (72) applied to long-term satellite data. The phytoplankton absorption spectra (a_ph) characterize the complex absorption of solar radiation by phytoplankton during photosynthesis and can be represented as a linear sum of multiple Gaussian function curves (73)

$$a_{\text{ph}}\left(\mathrm{\lambda }\right)=\sum _{i=1}^n{a}_{\text{gaus}}\left({\mathrm{\lambda }}_i\right)\times e^{-\frac{1}{2}{\left(\frac{\mathrm{\lambda }-{\mathrm{\lambda }}_i}{{\mathrm{\sigma }}_i}\right)}^2}$$ (3)

where a_gaus(λ_i) represents the phytoplankton absorption coefficient at the ith Gaussian band (λ_i) and σ_i refers to the full width at half maximum of each Gaussian peak. The Chla concentrations of PTGs can be modeled using a_gaus(λ_i) through a stepwise regression method, and the best relationship between the PTG concentrations (C_PTGs) and a_gaus(λ_i) in logarithmic form was as follows

$${\text{log}}_{10}\left(C_{\text{PTGs}}\right)=a_0+\sum _{i=1}^n{a}_i\times {\text{log}}_{10}\left[a_{\text{gaus}}\left({\mathrm{\lambda }}_i\right)\right]$$ (4)

where a₀ and a_i are model coefficients (as shown in table S1). We applied the QAA model (72) to derive a_ph from MODIS-Aqua remote sensing reflectance data and subsequently used the PTG models to estimate PTG Chla concentrations. The global monthly satellite dataset of Chla concentrations for dominant PTGs from 2002 to 2022 can be downloaded from https://doi.org/10.17632/c9d852g8j9.1 (16).

Supplementary Materials

The PDF file includes:

Figs. S1 and S2

Table S1

Legend for code S1

Other Supplementary Material for this manuscript includes the following:

Code S1