Introduction

Isoprene (2-methyl-1,3-butadiene, C5H8) is an active volatile organic compound (VOC) that has significant impacts on global climate, either through cloud formation1,2,3 or cloud inhibition4,5. While terrestrial vegetation is the primary emission source of atmospheric isoprene6,7, marine-originated isoprene potentially influences secondary organic aerosol (SOA) formation in the remote ocean, especially in association with increased emissions during phytoplankton blooms1,8,9,10. SOA produced by photooxidation of marine isoprene affects the coastal regions most and can reach as far as hundreds of kilometers inland11, altering the air quality of impacted regions12.

Marine isoprene has a strong biogenic origin13,14. Phytoplankton are suggested to be the chief marine isoprene emission source, though heterotrophic bacteria and seaweeds can also emit isoprene15. The dependence of phytoplankton isoprene production on phytoplankton biomass9,16, phytoplankton functional types (PFTs)17,18, and environmental drivers (e.g., light and temperature)19,20,21 have been investigated in laboratory incubation experiments19,22, mesocosm studies23, and in-situ measurements16,24,25.

To better quantify the impacts of marine isoprene on atmospheric chemistry and global climate, marine isoprene emission needs to be better constrained. While bottom-up22,26,27,28 and top-down26,29 methods have been applied to estimate marine isoprene emissions, there are still significant discrepancies (two orders of magnitude) among previous estimates due to the paucity of continuous in-situ measurements and limited understanding of mechanisms involved in marine isoprene production and loss26,30. Although some field measurements31,32 and lab experiments33 proposed that photochemical reactions occurring in the sea surface microlayer (SML) may account for the discrepancies34, underway eddy covariance measurements of marine isoprene emission fluxes in the Northern Atlantic Ocean do not show a statistically significant correlation with shortwave radiation35.

Here, we report marine isoprene emission fluxes estimated from satellite observations and model simulation over a study region covering the western Pacific Ocean and the eastern Indian Ocean (120°E–160°W, 60°S–60°N, Supplementary Fig. 1a) from August 2015 to December 2020. The marine isoprene emission model was established based on a PFT-specific isoprene production module which incorporates both light and temperature dependency (Fig. 1). The 3-D temperature structure in the euphotic zone was reconstructed from remotely sensed sea surface temperature (SST), sea surface height (SSH), and wind stress (WS) measurements. Afterward, dynamic mixed layer depth was derived from the 3-D temperature structure by using the temperature threshold method36. As the light and temperature vary with water depth, the isoprene production in the mixed layer was quantified by integrating the isoprene produced at all water column depths within the mixed layer. Isoprene flux from SML was calculated by estimating the photochemical isoprene production occurring in the SML. The satellite observation data from Himawari-8 L3 (i.e., chlorophyll-a (Chl-a) concentrations, SST, photosynthetically active radiation (PAR)) were used for model simulation. Utilizing high-resolution geostationary satellite observations, this work proposes to: (1) elucidate the spatial and temporal distributions of marine isoprene over the western Pacific Ocean and the eastern Indian Ocean; (2) quantify the trends of marine isoprene emissions and investigate contributions from biological and physical environment parameters; (3) assess the potential impacts from marine isoprene emissions on air quality and climate. We observed significantly high marine isoprene emission flux (7.4 nmol m−2 h−1) in the center of the equatorial Pacific Ocean (CEPO, 7∘S ~ 5∘N, 175∘E ~ 160∘W, Supplementary Fig. 1). At the same time, the trend of isoprene emission was also remarkably high (5.5 ± 0.1% yr−1) during 65 months period in the same region. The results reveal significant positive correlations between isoprene emission flux and some environmental parameters (wind speed, Chl-a concentration, and PAR) in CEPO, while negative correlations between isoprene emission flux and SST. The contribution of marine isoprene emission on aerosol was also investigated, indicating their essential roles in changing the atmospheric oxidation process in the remote ocean.

Fig. 1: Schematic diagram of marine isoprene emission model in the upper ocean.
figure 1

The model includes modules of phytoplankton isoprene production, chemical loss, biological loss, diffusion to the deeper ocean, seawater isoprene flux, and SML isoprene flux.

Full size image

Results

Marine isoprene emission flux

We calculated the mean values of marine isoprene emission flux, seawater flux, and SML flux for each grid cell over the 65 months between August 2015 and December 2020, as shown in Fig. 2a–c. Comparing different ocean regions (Supplementary Fig. 1a), the marine isoprene emission flux shown in Fig. 2a and seawater isoprene flux shown in Fig. 2b exhibit similar spatial patterns. The mean values of marine isoprene emission flux in coastal regions (6.5 nmol m−2 h−1) were at approximately the same level as that in open ocean regions (6.4 nmol m−2 h−1). Comparing between major open ocean areas, the marine isoprene emission flux in the East Indian Ocean (EIO) was higher than that in the North Pacific Ocean (NPC) and the South Pacific Ocean (SPC), which were 6.9, 6.2, and 6.2 nmol m−2 h−1, respectively (Supplementary Table 3). The contributions from seawater isoprene flux to marine isoprene flux were between 70% and 85% in most ocean areas and can reach as high as 90% in some coastal regions (Supplementary Fig. 2). The mean value of marine isoprene emission flux in the ocean areas is 6.5 nmol m−2 h−1, which was extrapolated for a global estimate (1.2 Tg C yr−1) by multiplying the area ratio in spite of the large uncertainties involved. The comparisons of our estimates with the results reported in previous studies are provided in Supplementary Table 2.

Fig. 2: Mean values of isoprene emission fluxes and major environmental drivers from August 2015 to December 2020.
figure 2

a Marine isoprene emission flux. b Seawater flux. c SML flux. d Chl-a concentrations. e Sea surface temperature. f PAR.

Full size image

Remarkably, there were significantly higher marine isoprene emission flux in CEPO and the Tasman Sea, which were 7.4 and 8.7 nmol m−2 h−1 (Supplementary Table 3). While the marine isoprene flux in CEPO and Tasman sea were about 20% and 40% higher than that in NPC and SPC, the seawater isoprene concentrations in the Tasman Sea (62.7 pmol L−1) were about 26% lower than that in CEPO (84.2 pmol L−1). It suggests the air–sea transfer rate has higher sensitivity of wind speed than SST (see SI Eqs. (2) and (3)), as the wind speeds in CEPO and the Tasman Sea were around 5.6 and 8.3 m s−1, respectively (Supplementary Fig. 2), and SST in two regions were ~28.7 and 16.3 °C (Fig. 2e).

Figure 2c shows that the SML fluxes in subtropical areas of NPC, EIO, and coastal regions of Australia were higher than 1.8 nmol m−2 h−1. In comparison, SML fluxes in the NPC’s equatorial and high-latitude region were merely around 1.2 nmol m−2 h−1. The Chl-a concentrations were lower than 0.1 mg m−3 in most open oceans (Fig. 2d), but the values could reach 0.3 mg m−3 in CEPO and the high-latitude Pacific Ocean. In coastal regions, the Chl-a concentrations can be higher than 0.4 mg m−3, generally due to larger nutrient supplies. In Fig. 2e, the sea surface temperature (SST) was highest (≥30 °C) in the equatorial region and decreased to the lowest value (about 0 °C) with increasing latitude. In Fig. 2f, the photosynthetically available radiation (PAR) value in the equatorial and subtropical ocean (>400 μmol m−2 s−1) was higher than in high-latitude regions. The mean values of the associated parameters in CEPO can be found in Supplementary Table 5. Climatological marine isoprene emissions, seawater isoprene emissions, and SML emissions are displayed in Supplementary Figs. 8, 9, and 10, respectively.

Increasing emission trends over the equatorial Pacific Ocean

The trend significance of the time series was detected at each grid cell using the Mann–Kendall (MK) test37. The magnitude of the linear trend was estimated by the Theil–Sen estimator38,39. While the marine isoprene emission flux in most open ocean areas showed insignificant trends, remarkable increasing trends were discovered in CEPO (Fig. 3). The mean values of absolute and relative marine isoprene emission flux trends were 401 ± 4 pmol m−2 h−1 yr−1 and 5.5 ± 0.1% yr−1 in CEPO (Fig. 3a, d). The mean values of seawater isoprene flux’s absolute and relative trends in CEPO were 342 ± 4 pmol m−2 h−1 yr−1 and 5.9 ± 0.1% yr−1 (Fig. 3b, e). The spatial patterns of seawater isoprene flux trends were very similar to marine isoprene emission flux (Fig. 3a, b, d, e), highlighting its main contribution to marine isoprene emission flux. Although the trends of SML flux in most ocean areas were insignificant, there were notably increasing trends of SML flux in CEPO (Fig. 3c, f). The mean values of the absolute and relative trend of SML flux in CEPO were 75 ± 1 pmol m−2 h−1 yr−1 and 5.0 ± 0.1% yr−1. The complete list of the trends in marine isoprene emissions of the selected open oceans is given in Supplementary Table 4.

Fig. 3: Absolute and relative trends of isoprene emission fluxes from August 2015 to December 2020.
figure 3

a Absolute trends of marine isoprene emission flux. b Absolute trends of seawater isoprene flux. c Absolute trends of SML flux. d Relative trends of marine isoprene emission flux. e Relative trends of seawater isoprene flux. f Relative trends of SML flux.

Full size image

To explore the driving factors of notable increasing trends of isoprene flux in CEPO, we investigated the temporal and spatial variations of biological and physical environment parameters (Fig. 4a–d). The Chl-a concentrations in most open ocean areas were decreasing except for CEPO, where the mean value of the relative trends was 7.1 ± 0.1% yr−1 (Fig. 4a). The mean value of SST relative trends in CEPO was −8.8 ± 0.1‰ yr−1, while in most ocean regions, the relative trends were insignificant (Fig. 4b). The PAR in CEPO had remarkable positive trends, and the mean value was 20.1 ± 0.2 ‰ yr−1 (Fig. 4c). In Fig. 4d, the relative wind speed trend was significant in CEPO, and the mean value was 44.4 ± 0.4 ‰ yr−1. The patterns above suggest prominent impacts from four major environmental parameters on the notable increase in isoprene emission trends in CEPO. The entire list of the absolute and relative trends in environmental parameters in CEPO can be found in Supplementary Tables 6 and 7.

Fig. 4: Relative trends of environmental parameters and their correlation with marine isoprene emission flux from August 2015 to December 2020.
figure 4

a Relative trends of Chl-a concentration. b Relative trends of SST. c Relative trends of PAR. d Relative trends of wind speed. e Correlation coefficients between marine isoprene emission flux and Chl-a concentration. f Correlation coefficients between marine isoprene emission flux and SST. g Correlation coefficients between marine isoprene emission flux and PAR. h Correlation coefficients between marine isoprene emission flux and wind speed.

Full size image

The correlations between marine isoprene emission flux and the primary environmental drivers (i.e., Chl-a concentration, SST, PAR, and wind speed) were calculated for detailed investigation (Fig. 4e–h). Unsurprisingly, the marine isoprene emissions were positively correlated with Chl-a concentrations in most open oceans (see Fig. 4e) because Chl-a concentration is a proxy of phytoplankton biomass for primary productivity. Conversely, the negative correlations between marine isoprene emission flux and Chl-a concentrations were found in some coastal areas, suggesting there could be substantial isoprene consumption in the surface of the coastal regions, which is consistent with previous findings40. While most tropical and subtropical ocean areas had negative correlations between marine isoprene flux and SST, positive correlations were found in high-latitude ocean areas (Fig. 4f). The isoprene production rate of phytoplankton has been found to increase with temperature until optimum temperature and fall thereafter19,21. Ocean areas that show positive correlation might be because the seawater temperature is lower than the optimum temperature for production, while negative correlation might be due to that the seawater temperature is higher than the optimum temperature. In Fig. 4g, marine isoprene emission flux had strong positive correlations with PAR in CEPO and Tasman Sea. Although both seawater isoprene flux and SML flux had a positive relationship with PAR (Eq. (1) and SI Eqs. (4) and (5)), there were negative correlations between marine isoprene emission flux and PAR in the Bay of Bengal, South China Sea, and the Philippine Sea. The variable thermal traits among phytoplankton species and adaptive migrations could alter community composition41,42,43, resulting in the biomass change and shift of dominant PFT. One explanation for the negative correlation is that the migrated PFT with a high isoprene emission factor favors low light intensity conditions or high light intensity cooccurs with lower biomass in some ocean areas. As high-speed wind (≥13 m s−1) tends to disturb the SML formation by breaking waves34, the negative correlation between marine isoprene emission flux and PAR also could result from the coincidence of high-speed wind and high PAR, leading to the disruption of SML and negligible SML flux. Figure 4h shows that the correlation coefficients between marine isoprene emission flux and wind speed were positive in latitudes lower than 30° and negative in latitudes higher than 30°. Sea surface wind enhances air–sea gas exchange, and the gas exchange rate becomes more considerable. As wind can also drive the vertical mixing of the ocean, deepening the mixed layer, the seawater isoprene concentration Cw thus tends to be lower. The two mechanisms result in the increase or decrease of the seawater isoprene flux. The complete list of correlation coefficients between marine isoprene emission flux and main parameters in CEPO is shown in Supplementary Table 8.

Impacts on regional atmospheric chemistry and climate

Marine aerosols alter the Earth’s climate directly by scattering and absorbing solar radiation and indirectly by changing the formation and properties of clouds over the oceans44. Primary sea spray aerosols (SSA) and secondary marine aerosols (SMA) are two major types of marine aerosols. The analysis of aerosol number size distributions from global scale cruise measurements revealed that SSA makes a small contribution to cloud condensation nuclei (CCN) population, especially in tropical regions (around 5.0%)45. Mesocosm experiments also suggested that SMA plays the dominant role over SSA in affecting the cloud properties in marine environment46. Based on the variability analysis of chemical constituents and their ratios during the phytoplankton bloom of the mesocosm experiments, a hypothesis was further made that SMA could be easily affected by the oxidation of non-DMS VOCs such as isoprene46.

The isoprene oxidation chain involves complex reactions and varies considerably with local atmospheric conditions47. Significant positive correlations have been observed between isoprene fluxes at the cruise track and satellite aerosol optical depth (AOD) of the forward trajectories in the western tropical Indian Ocean during the summer monsoon48. To explore isoprene oxidation mechanisms under pristine conditions, we investigated the distribution of OMI tropospheric NO2 and O349 column densities as functions of Himawari-8 AOD (at 500 nm) and the estimated marine isoprene emission flux in CEPO (Fig. 5). Marine isoprene emission flux had general positive correlations with AOD in most of the ocean areas (Supplementary Fig. 4a), while previous studies found isoprene epoxydiols (IEPOX) can be an intermediate of isoprene-derived SOA formation under low nitrogen oxides (NOx = NO + NO2) conditions50.

Fig. 5: Comparisons of marine isoprene flux and AOD with monthly tropospheric column densities of NO2 and ozone in CEPO from August 2015 to December 2020.
figure 5

The data were grouped into 300 × 400 bins with data counts no less than 100 (Supplementary Fig. 7). a NO2 is likely to determine the enhancement rate of AOD from marine isoprene in CEPO. The dashed line represents the regression line of binned data with NO2 column densities higher than 2.5 × 1014 molecules cm−2, and the solid line represents that with NO2 column densities lower than 2.4 × 1014 molecules cm−2. b Lower ozone abundances were observed when marine isoprene emission and AOD were both higher.

Full size image

As shown in Fig. 5a, NO2 plays a critical role in controlling the SOA yields from isoprene oxidation in CEPO. When NO2 column densities were higher than 2.5 × 1014 molecules cm−2, rapid enhancements (dashed regression line. slope: 1.5E-2, R-squared: 0.42, P value: 0) of AOD were observed with moderate isoprene emissions. As NO2 column densities dropped below 2.4 × 1014 molecules cm−2, significantly increasing marine isoprene emissions could only lead to much weaker enhancements (solid regression line. slope: 4.4E-3, R-squared: 0.065, P value: 0) of AOD. It suggests NO2 may act as a catalyst during aerosol formation from isoprene in the remote ocean air. Tropospheric ozone column densities were lower than the average level (about 34.5 DU) when both high marine isoprene emission flux (>8.0 nmol m−2 h−1) and AOD (≥0.15) were observed in CEPO (Fig. 5b). This suggests that the higher isoprene emission suppressed ozone formation since the O3-NOx-VOC sensitivity was NOx-limited in CEPO (Supplementary Fig. 3). The coincidence of high emission flux and significant positive trends of marine isoprene could be the reason for decreasing tropospheric O3 in CEPO (not shown here).

The relative warming in tropical oceans of subsidence (e.g., CEPO) has been linked to decreased lower tropospheric stability and low cloud cover by the proposed mechanism51 and empirical relationships52. Conversely, warmer SST in tropical convection regions is coincident with higher humidity in the upper troposphere, suggesting possible lofting of warm moist air from the surface to the upper troposphere52. As the 2015/16 El Niño was the strongest El Niño event ever recorded and CEPO mostly overlaps the Niño-4 region (5 °S to 5 °N, 160 °E to 150 °W), the high isoprene emissions of CEPO during 2015/16 El Niño were potentially linked to organic aerosol formation in the upper troposphere53. Furthermore, modeling studies demonstrated that marine isoprene-derived SOA and O3 could reach as far as hundreds of kilometers inland11. Consequently, the change in marine isoprene could potentially affect the regional air quality and global climate.

Discussion

Our study discovered significantly high marine isoprene emissions in CEPO, investigated the main driving parameters of remarkable increasing trends of isoprene fluxes in CEPO, and demonstrated important implications for regional atmospheric photochemistry. Furthermore, coastal oceans and the Tasman Sea exhibited high marine isoprene emissions, while emissions decreased in most open oceans during the study period. Although correlations do not necessarily indicate causation, the positive correlations between marine isoprene emission flux and satellite AOD in most ocean areas suggest that isoprene could be important in SOA formation. NO2 may play a critical role during aerosol formation from isoprene in the remote ocean air. This study provides an important initial demonstration of the significance of isoprene emission enhancement in the remote tropical ocean. Further measurements of marine isoprene and environmental parameters, particularly over broader tropical oceans, are needed to determine the distributions and impacts of marine organic gas emissions.

Methods

Isoprene production model

The isoprene production rate p for the predominant phytoplankton species at depth h was established by incorporating a temperature-dependence factor α1/(T2 + α2 ⋅ T + α3) into the log-squared relationship invoked by Gantt et al.22:

$$p=EF\cdot \frac{{\alpha }_{1}}{{(T+\delta )}^{2}+{\alpha }_{2}\cdot (T+\delta )+{\alpha }_{3}}\cdot \ln {(I)}^{2}$$
(1)

where EF (in μmol g Chla−1 h−1) is a constant emission factor for each PFT (Supplementary Table 1), I and T are the ambient solar radiation (in μE m−2 s−1) and temperature (in °C) at the depth h, respectively. The coefficients α1, α2, and α3 were derived as 3.6402, −46.75, and 618.2, respectively, by fitting the rational model to the isoprene production rates of two diatoms species (Supplementary Fig. 6) from the light- and temperature-dependent experiments conducted by Meskhidze et al.21. δ is the difference between 23.375 °C and the optimum temperature calculated from the latitudinal gradient in the optimum temperature discovered by Thomas et al.42. The details on how to derive the coefficients α1, α2, α3 can be found in Supplementary Note 1.4.

The seawater temperature T at water column depth can be estimated from the satellite-based sea surface observations (i.e., SST, SSH, and WS) by performing the multiple linear regression (MLR) analysis. ISAS temperature gridded fields and climatology, based on the Argo network of profiling floats, were used to train and evaluate the MLR model. The SSH and WS were interpolated to the same grids with the PAR product of Himawari-8 by using the linear interpolation method. More details on training and evaluation of the MLR model can be found in Supplementary Note 1.1.

The reliability of fitted PAR depth profile derived from Beer-Lambert’s Law has been verified by the directly measured downwelling PAR profiles14. Hence, the light radiation I at depth h of the water column can be estimated from the solar radiation at the sea surface I0 as follows:

$$I={I}_{0}\cdot {e}^{-{k}_{d}\cdot h}$$
(2)

The satellite PFT identification was performed on the MODIS-Aqua data using the PHYSTWO algorithm (Supplementary Fig. 1b), which is applicable to coastal and open oceans54. The six retrieved PFTs are Coccolithophorids bloom, Phaeocystis-like, Diatoms, Synechococcus, Prochlorococcus, and Nanoeukaryotes (see Supplementary Fig. 1b for an example). Their corresponding emission factors and references are listed in Supplementary Table 1.

Cruise measurements demonstrated that isoprene production mainly occurs in seawater deeper than 5 m55,56. To better represent the Chl-a concentrations in the whole euphotic layer, our study uses the mean Chl-a concentrations within the euphotic layer instead of the satellite-retrieved sea surface concentrations. The mean Chl-a concentration \(\overline{[Chla]}\) can be computed as follows:

$$\overline{[Chla]}=\frac{Chl{a}_{tot}}{{H}_{max}}$$
(3)

where Hmax is the euphotic layer depth, Chlatot is the column-integrated Chl-a content in mg m−2. Chlatot can be derived from the remotely sensed Chl-a concentration [Chla] by using the statistical relationship given by Morel and Berthon57:

$$Chl{a}_{tot}=38.0\times {[Chla]}^{0.425}\quad \quad if\,[Chla]\le 0.5$$
(4)
$$Chl{a}_{tot}=40.2\times {[Chla]}^{0.507}\quad \quad if\,[Chla]\, > \,0.5$$
(5)

where [Chla] is the remotely sensed Chlorophyll concentration in mg m−3.

The euphotic zone depth Hmax, within which isoprene production can occur in our model, was assumed to extend from the surface to the depth at which the light levels are reduced to 2.5 W m−2, which is the level at which the photosynthesis ceases. This value is the lowest light level for Prochlorococcus19, here we assume it is applicable for all the PFTs. The depth which represents the maximum possible extent of the planktonic euphotic zone can be determined by using the following equation:

$${H}_{max}=-\ln \left(\frac{2.5}{{I}_{0}}\right)\cdot {k}_{d}^{-1}$$
(6)

where kd is the diffuse attenuation coefficient for PAR. To compute kd, here we use the revised empirical model provided in ref. 14, Morel and Maritorena58,

$${k}_{d}=\frac{4.6}{426.3}\times Chl{a}_{tot}^{0.547}\quad \quad if\,Chl{a}_{tot}\le 13.6222\,{\mathrm{mg}}\, {{\mathrm{m}}}^{-2}$$
(7)
$${k}_{d}=\frac{4.6}{912.5}\times Chl{a}_{tot}^{0.839}\quad \quad if\,Chl{a}_{tot}\, > \,13.6222\,{\mathrm{mg}}\, {{\mathrm{m}}}^{-2}$$
(8)

Isoprene concentration and flux

The isoprene concentration variation rate P caused by phytoplankton production can be established by integrating the isoprene production rate within the depth H22:

$$P=\frac{\beta }{{D}_{ML}}\cdot \overline{[Chla]}\cdot \int\nolimits_{0}^{H}p\,dh$$
(9)

where β is the prefactor set to 86.8798, which was derived by constraining the mean of modeled isoprene seawater concentration to equal that of the in-situ measurements of T/S Osyoro-maru Cruise No. 56 (sampling locations are shown in Supplementary Fig. 1a, comparison between estimates and measurements can be found in Supplementary Fig. 5). DML is the depth of surface mixed layer in which the isoprene is considered well mixed, and hence the concentration can be treated as uniform, H is the less of Hmax and DML. The DML was calculated by applying the temperature threshold method on the temperature-depth profile, using temperature threshold ΔT = 0.2 °C as the criteria59.

Due to no significant isoprene accumulation in the depth profile during weeks time frame60 and the total turnover times of isoprene in the upper ocean varied from 1.4 days (productive waters) to 16 days (oligotrophic waters) in seawater incubation experiments40, the monthly mean seawater isoprene concentration Cm can be derived by using the steady-state model27:

$$P-({k}_{bio}+{k}_{chem})\cdot {C}_{m}-\frac{{F}_{ocean}}{{D}_{ML}}-{L}_{mix}=0$$
(10)

where kbio and kchem is the biological loss rate and chemical rate constant for all possible loss pathways, their values (\({k}_{bio}=\mu \,{\overline{[Chla]}}^{1.28}\), μ is equal to 0.14, kchem is 0.05 day−1) are given according to the incubation experiments in ref. 40. Focean is the seawater isoprene flux. Lmix is the loss due to the diffusion downward to the deep ocean, calculated from the following equation:

$${L}_{mix}={k}_{mix}\cdot {C}_{m}$$
(11)

where kmix is the vertical mixing rate at the bottom of the mixed layer. The value of kmix is set to −0.005 day−1, which is suggested by a modeling study61 and is the median value of the cruise measurements40.

The marine isoprene flux at the ocean–atmosphere interface is described as the sum of the seawater isoprene flux and the SML flux. Seawater isoprene flux is due to the phytoplankton production and biochemical loss in the water column, whereas the SML flux results from the photosensitized reactions in the SML. The detailed calculation processes for calculating seawater isoprene flux and the SML flux are provided in Supplementary Note 1.2 and Supplementary Note 1.3.

Trend analysis

We evaluated trends of the time series in marine isoprene flux, the environmental drivers (Chl-a concentration, SST, PAR, wind speed), AOD, ozone, formaldehyde, and NO2 at each grid cell. Trend analysis was performed on the time series using the Mann–Kendall (MK) test37, a robust non-parametric test for trend detection. The slope of linear trends was determined using the Theil–Sen estimator38,39. Here, we set the significance level at 0.05, which means the trend is significant when Z statistic is larger than 1.96 or less than –1.96. The uncertainties on the estimate of the non-zero trend slope were quantified using the Exclude-one and Estimate Slope (EES) method62, with a significance level of 5%. The statistical analysis of absolute and relative trends in selected ocean regions was only performed upon significant trends. The confidence levels for the isoprene emission fluxes are shown in Supplementary Fig. 11.

Sensitivity test

A set of sensitivity analyses were designed to assess the model response to six driving parameters, namely Chl-a, SST, PAR, EF, Hmax, and DML. One base case was performed in which all six parameters were held spatially and temporally constant to the median values in the study domain, and other six simulations were made in which all but one of these parameters were held constant. The differences between each of these six simulations and the base case for March 2020 are shown in Supplementary Figs. 12 and 13 as an example to quantify the contribution of each parameter to the spatial distributions of the marine isoprene emissions and seawater isoprene concentrations, respectively. The Chl-a, SST, and PAR have the greatest impact on the estimates of marine isoprene emissions in most of the study domains (Supplementary Fig. 12a–c). In Supplementary Fig. 12d, the contribution of EF to the estimates of isoprene emissions ranges from –55% to 37%, highlighting the importance of a good measurement of EF for specific PFT. As Hmax and DML are related to the assumptions of our model, the low percentage differences (lower than 10%) of Hmax and DML in most of the study domains suggest the robustness of our results to the assumptions (Supplementary Fig. 12e, f). Therefore, our work indicates Chl-a, SST, PAR, and EF have higher requirements in their accuracy due to their significant contributions to marine isoprene emissions.