A total of 44 in situ observations were conducted at the SEATS site (18 °N, 116 °E) at regular intervals between 1998 and 2016. These expeditions involved hydrographic seawater sampling and biogeochemical field experiments using research vessels (RV; Ocean Researcher I, Ocean Researcher III, and Fishery Researcher I) (Karl et al., 2003; Wong et al., 2007). On-deck incubation of PP in accordance with ¹⁴C protocols was also conducted during 20 of 25 expeditions between 2003 and 2016 (Figure 1). Conductivity–temperature–depth (CTD) (SBE9/11, SeaBird) and quantum scalar irradiance (QSP-200L, Biospherical) were, respectively, used to obtain vertical profiles of temperature and underwater photosynthetically available radiation (PAR). A Biospherical instrument (QSR-240) was used to obtain daily PAR measurements (i.e., daily surface light intensity). In that study, Z_eu was defined as the depth at which underwater PAR reached 1%. The mixed layer depth (MLD) was defined as the deepest depth at which the temperature was 0.8°C lower than at the surface (Kara et al., 2000; Chou et al., 2006).

Seawater samples were collected at discrete depths to determine Chl concentrations and PP values. Briefly, 2L seawater samples were filtered using 47-mm GF/F filters before obtaining Chl concentrations (via acidification) using a Turner 10-AU-005 fluorometer (Gong et al., 2003; Shiah et al., 2003). PP was measured using the ¹⁴C assimilation method (Parsons et al., 1984) following incubation under an artificial light source (2,000 μE m^–2 s^–1 with full spectrum from 350 to 2450 nm, similar to sunlight) for ∼3 h in a proprietary isolation container using flowing surface seawater for cooling. Following incubation and acidification (0.5 N HCl), radioactive substances collected in 25-mm GF/F filters were transported to a lab for quantification using a scintillation counter (Packard 2200) (Gong et al., 2003; Shiah et al., 2003; Lai et al., 2014). Finally, PP was calculated via trapezoidal integration (i.e. integrated from surface to 1% PAR). Note that SST, Chl and PP measurements from the Ocean Data Bank (Ministry of Science and Technology, Taiwan¹) were used to compensate for deficiencies in in situ observations. Some of the hydrographic records (i.e., SST) and biogeochemical parameters (i.e., PP, Chl_s) of May–October from 2003 to 2006 have been present by Pan et al. (2012).

Between 2003 and 2016, the daily data of level-3 products were obtained using a passive ocean color MODerate resolution Imaging Spectroradiometer (Aqua sensor; MODIS-Aqua) with spatial resolution of 4 km for the region covering 17.5–18.5°N and 115.5–116.5°E. We also collected PP_VGPM, SST, E₀, K_d, and Chl_s values from the Environmental Research Division’s Data Access Program (ERDDAP), National Oceanic and Atmospheric Administration, Department of Commerce, U. S.². Note that 8-day composite data were applied in situations where daily observations were missing. Note also that PP_VGPM values obtained from ERDDAP were estimated using VGPM developed by Behrenfeld and Falkowski (1997).

Plausible deviations between PP_VGPM and PP_{in situ} were examined by comparing data obtained from satellites vs. data obtained in situ, including i.e., SST_sate, (PoptB)_sate, E_0–sate, [E₀/(E₀+4.1)]_sate, K_d–sate, Z_eu–sate, Chl_s–sate; SST_{in situ}, (PoptB)_{in situ}, E_{0–in situ}, [E₀/(E₀+4.1)]_{in situ}, K_{d–in situ}, Z_{eu–in situ}, Chl_{s–in situ}. Note that this analysis was conducted using the VGPM algorithm proposed by Behrenfeld and Falkowski (1997) (with far fewer input variables), as Equation 2–1:

where the D_irr refers to the photoperiod. Z_eu was estimated from the average K_d of the water column (Behrenfeld and Falkowski, 1997 and references therein).

To evaluate the impact of monsoonal force on time-series variations, all data sets were divided into two groups: SW-monsoon (SWM, Apr. to Oct., including inter-monsoons of Apr and Oct) and NE-monsoon (NEM, Nov. to Mar). Variations between satellite-based observations and in situ measurements were examined by averaging all relevant values and reporting them as mean ± standard deviation (SD). Linear regression was used to assess relationships between any two of the variables. The t-tests were used to compare sets of observations with the significance level set at p = 0.10.

As shown in Figure 2A, PP_{in situ} was ∼2.0 times higher during the NEM (117–528, average = 323 ± 134 mg-C m^–2 d^–) than during the SWM (76–247, average = 159 ± 58 mg-C m^–2 d^–1) (Table 2). PP_VGPM values between 2003 and 2016 were as follows: SWM was 250 ± 36; 154–771 mg-C m^–2 d^–1 and NEM was 443 ± 142; 190–1,153 mg-C m^–2 d^–1 (Figure 2A). Note that PP_VGPM during the NEM was ∼1.8 times higher than during the SWM (p < 0.01) (Table 2). The differences in PP_{in situ} and PP_VGPM between the NEM and SWM are in line with those reported by Ning et al. (2004); Chen (2005), and Hao et al. (2007), all of which were obtained from the same area of the SCS using different methods (Table 1). PP_VGPM exhibited monsoonal variations resembling the curve derived from PP_{in situ}; however, the magnitudes were 37% higher during the NEM and 57% higher during the SWM. On average, PP_{in situ} estimates were ∼50% lower than PP_VGPM estimates, regardless of monsoon.

The lowest SST_{in situ} (25.0 ± 1.4°C) and SST_sate (25.1 ± 1.2°C) values were obtained under the prevailing NEM during the cold season. The highest SST_{in situ} (28.9 ± 0.9°C) and SST_sate (28.8 ± 1.0°C) values were obtained under the prevailing SWM during the warm season (Table 2). The low SST values recorded during the NEM can be attributed to the combined dynamics of NEM-driven vertical mixing of seawater and reduced solar radiation (Tseng et al., 2005; Zhou et al., 2020). By contrast, the PoptB showed a synchronous but opposite distribution (Figures 2B,C). The (PoptB)_{in situ} and (PoptB)_sate values computed from SST for the NEM, respectively, ranged from 4.4 to 6.1 and 4.0 to 6.6 mg-C mg-Chl^–1 h^–1, and the average value was the same (5.4 ± 0.5 mg-C mg-Chl^–1 h^–1). The (PoptB)_{in situ} and (PoptB)_sate values for the SWM, respectively, ranged from 4.0 to 4.4 and 4.0 to 6.0 mg-C mg-Chl^–1 h^–1, with mean PoptB values of 4.1 ± 0.2 and 4.2 ± 0.2 mg-C mg-Chl^–1 h^–1 (Table 2).

E_{0–in situ} and E_0–sate values in the SWM (50 ± 11 and 47 ± 6 Eins m^–2 d^–1) were higher than those in the NEM (33 ± 14 and 36 ± 8 Eins m^–2 d^–1) (see Figures 2D,E and Table 2). From the perspective of variables in the VGPM algorithm, E₀/(E₀+4.1) ratios varied little between the in situ measurement and satellite-based observation. The variable of [E₀/(E₀+4.1)] thereby seemed not an important parameter to influence the PP level. Z_eu was estimated using the expression of –ln(0.01)/K_d (Kirk, 1994; Behrenfeld and Falkowski, 1997), which led to a reversal of monsoonal K_d and Z_eu distributions (see Figures 2F,G). The mean Z_{eu–in situ} was deeper during the SWM than during the NEM (87 ± 10 and 75 ± 14 m, respectively). The Z_eu–sate as well as the Z_{eu–in situ} presented similar trends between SWM and NEM (156 ± 17 and 120 ± 30 m, respectively) (Table 2). Z_eu values were shallower in the NEM than in the SWM, due perhaps to reduced light penetration resulting from more abundant biomass in the water column (Chen, 2005; Tseng et al., 2005; Chen et al., 2008).

Figure 2H presents long-term temporal variations in Chl_s. In situ measurements and satellite-based observations revealed a distinct NEM maximum (0.24 ± 0.16 and 0.20 ± 0.08 mg m^–3, respectively) and a SWM minimum (0.08 ± 0.02 and 0.10 ± 0.02 mg m^–3, respectively). Chl_{s–in situ} and Chl_s–sate concentrations were, respectively, 0.07 to 0.58 and 0.09 to 0.71 mg m^–3 during the NEM, and were, respectively, 0.04 to 0.11 and 0.05 to 0.43 mg m^–3 during the SWM. For the same location, our observations were close to the data reported by Chen (2005) and Tseng et al. (2005). The SWM minimum Chl_s values in the current study were nearly the same as those reported in the oligotrophic ocean time-series studies at HOT (Hawaii Ocean Time-Series) and BATS (Bermuda Atlantic Time-Series Study) during the summer (0.05 mg m^–3); however, NEM values at SEATS exceeded the winter maximum at HOT and BATS (0.1 and 0.3 mg m^–3, respectively) (Karl et al., 2003). The high NEM maximum at SEATS may perhaps be explained by an increase in phytoplankton biomass (particularly larger sizes of >3 μm), which was stimulated by the deepening Z_ML (Table 2). The deeper nutrient-rich water was then efficiently transported to the upper surface water under the influence of NEM given that the nutrient-cline depth was shallower in SCS than in other oceans (Chen, 2005; Tseng et al., 2005).

The PP_VGPM values in this study are in good agreement with PP_{in situ} in terms of amplitude as well as phase. Overall, we observed a maximum PP_{in situ} of 394 ± 190 in January (NEM) and a minimum of 143 ± 70 mg-C m^–2 d^–1 in August (SWM) (Figure 3A). The magnitude of PP_VGPM was higher than the values obtained using in situ measurements; however, the trend over a span of 12 months was similar, with a pronounced NEM peak 619 ± 113 mg-C m^–2 d^–1 in January and the lowest SWM value of 230 ± 28 mg-C m^–2 d^–1 in September. These results are similar to those reported by other researchers for same area using different methods, such as the particulate organic carbon flux re-calculation (NEM: 207, SWM: 149 mg-C m^–2 d^–1) (Chen et al., 1998); on-deck C¹⁴ incubation (NEM: 300–509, SWM: 110–228 mg-C m^–2 d^–1) (Ning et al., 2004; Tseng et al., 2005); on-deck C¹³ incubation (NEM: 190–550, SWM: 190–280 mg-C m^–2 d^–1) (Chen, 2005; Chen et al., 2007), and Chl empirical function (NEM: 148–684, SWM: 86–275 mg-C m^–2 d^–1) (Chen et al., 2006; Table 1).

We observed opposing trends between monthly SST and the corresponding PoptB. The highest (PoptB)_{in situ} values (5.9 mg-C mg-Chl^–1 h^–1) were observed in January and the lowest (4.0 mg-C mg-Chl^–1 h^–1) were observed in August. The (PoptB)_sate values converted from SST_sate (5.8 and 4.0 mg-C mg-Chl^–1 h^–1 in January and June, respectively) were nearly identical to those for the same time slots obtained via in situ measurements (Figures 3B,C). The highest mean monthly E_{0–in situ} in April (63 ± 0 Eins m^–2 d^–1) was higher than the highest mean monthly E_0–sate (53 ± 3 Eins m^–2 d^–1). Overall, E_{0–in situ} and E_0–sate levels in December were very similar (∼26 and ∼27 Eins m^–2 d^–1, respectively). Based on E_{0–in situ}, the highest E₀/(E₀+4.1)_{in situ} ratio (0.94) occurred in April and the lowest E₀/(E₀+4.1)_{in situ} ratio (0.85) occurred in December. E₀/(E₀+4.1)_sate values of 0.93 in April and 0.86 in December were similar with the E₀/(E₀+4.1)_{in situ} values (Figures 3D,E). As for parameterization, it appears that PoptB and E₀/(E₀+4.1) values had little influence on VGPM results (PP), regardless of whether the values were obtained from in situ measurements or satellite-based observation.

We did not observe large monsoonal variations in K_{d–in situ} and K_d–sate; however, the highest K_{d–in situ} (0.072 ± 0.009 m^–1) was obtained in December and the highest K_d–sate (0.052 ± 0.009 m^–1) was obtained in January. These high K_d values resulted in a shallower Z_{eu–in situ} (64 ± 8 m) and Z_eu–sate (92 ± 19 m), compared to the values converted from the K_{d–in situ} (0.053 ± 0.007 m^–1) for August (Z_{eu–in situ} = 89 ± 13 m) and the K_d–sate (0.029 ± 0.003 m^–1) for September (Z_eu–sate = 162 ± 22 m) (Figures 3F,G). The shallow Z_eu observed in December and January can be attributed mainly to an increase in phytoplankton biomass that reduces the light penetration in that region during the NEM (Chen, 2005; Tseng et al., 2005; Chen et al., 2008).

Figure 3H presents temporal variations in Chl_{s–in situ} and Chl_s–sate. Conspicuously high Chl_{s–in situ} values were observed in December (0.40 ± 0.15 mg m^–3) and high Chl_s–sate values were observed in January (0.29 ± 0.09 mg m^–3). The high Chl_s values observed throughout the NEM were triggered by the monsoonal force, which stimulated phytoplankton photosynthesis, thereby increasing the phytoplankton abundance or enriching the area with Chl_s from subsurface waters (Liu et al., 2002; Tseng et al., 2005). The drop in Chl_{s–in situ} and Chl_s–sate values to nearly < 0.10 mg m^–3 during the SWM has previously been reported in studies focusing on the same region of the SCS (Liu et al., 2002; Chen, 2005; Tseng et al., 2005; Shih et al., 2020a). Similar findings were also recorded in oligotrophic time-series studies, such as HOT and BATS (Karl et al., 2003).

PP_{in situ} was generally lower than PP_VGPM; however, we observed a significantly positive correlation between these two parameters; i.e., slope = 0.70, r² = 0.42, p < 0.01 (Figure 4A). This clearly indicates the feasibility of the VGPM for predictions; however, tuning would be required for the study area. As shown in Figure 4, we identified significantly positive correlations between PP_{in situ} and PP_VGPM as well as their respective variables SST, E₀, K_d and Chl_s. As indicated by high r² and low p values with slopes close to 1, the most significant correlations were found in SST and E₀ (Figures 4C–F) : SST (slope = 0.92, r² = 0.92, p < 0.01) and E₀ (slope = 0.62, r² = 0.58, p < 0.01). Taken together, these results indicate that SST_sate and E_0–sate were the variables most strongly correlated with SST_{in situ} and E_{0–in situ}, providing the most accurate estimates of PP when using the VGPM. As indicated by the 1:1 lines in Figures 4B,G,H, the parameters with the greatest variability in terms of slope were Chl_s (slope = 0.41, r² = 0.74, p < 0.01) and K_d (slope = 0.59, r² = 0.50, p < 0.01). This suggests Chl_s–sate and K_d–sate could potentially bias PP estimates obtained using the VGPM.

The fact that the regression line between PP_VGPM and PP_{in situ} lies above the 1:1 line indicates that PP_VGPM estimates exceeded PP_{in situ}. When using the VGPM to estimate PP, the two main variables are PoptB and E₀/(E₀+4.1), derived from in situ measurements and satellite-based observations of SST (SST_{in situ} and SST_sate) and E₀ (E_{0–in situ} and E_0–sate). Between SST_{in situ} and SST_sate as well as between E_{0–in situ} and E_0–sate, we observed nearly linear relationships (i.e., close to the 1:1 diagonal). This suggests that PoptB and E₀/(E₀+4.1) were not scaling variables governing the magnitude of PP in this study. Influences derived from these two variables [PoptB, E₀/(E₀+4.1)] of the both methods on calculated results were less than 1%. By contrast, the correlations between Chl_{s–in situ} and Chl_s–sate, and Z_{eu–in situ} and Z_eu–sate (converted from K_{d–in situ} and K_d–sate, respectively) had a pronounced impact on VGPM PP estimates. If we considered only the difference between Chl_{s–in situ} and Chl_s–sate, then PP values estimated using VGPM would be slightly higher (∼ 5%, depended on the given case) than those obtained using in situ measurements. If we considered only the difference between Z_{eu–in situ} and Z_eu–sate, then PP values estimated using VGPM would be apparently 1.72-fold higher than those obtained using in situ measurements.

Our analysis of revealed a number of potential uncertainties pertaining to PP estimation using the VGPM algorithm. Most of the discrepancies were due primarily to differences between Z_{eu–in situ} and Z_eu–sate as well as between Chl_{s–in situ} and Chl_s–sate. Overall, the product of Z_eu and Chl_s (i.e., the phytoplankton inventory in the euphotic zone), suggests that the base assumption of vertically distributed standing stock biomass in low latitude waters (SCS) may perhaps be erroneous. If so, then it will be necessary to reformulate methods for the prediction of biomass standing stocks when implementing the VGPM algorithm (Ning et al., 2004; Hill and Zimmerman, 2010).

The fundamental structure of the VGPM is based on a relationship between integrated phytoplankton biomass in the euphotic zone and Chl_s (Behrenfeld and Falkowski, 1997; Hill and Zimmerman, 2010). Thus, obtaining accurate estimates of PP by comparing PP_{in situ} and PP_VGPM results depends on reliable estimates of the integrated phytoplankton biomass in the euphotic zone. However, passive satellites recording the color of the ocean surface are unable to elucidate the situation at arbitrary depths below the surface (Hill and Zimmerman, 2010; Shih et al., 2020b). Contrary to the assumption that Chl decreases exponentially with depth, most observations in the SCS revealed that the subsurface Chl maximum (SCM) was often found at great depths (Liu et al., 2002; Chen, 2005; Shih et al., 2020a, b). This makes it very difficult or even impossible to estimate the integrated biomass in the euphotic zone simply as a product of Z_eu and Chl_s. Enhancing the reliability of the VGPM requires that we increase the number of PP_{in situ} observations and the corresponding variables in order to improve the correlation between our assumptions pertaining to phytoplankton integrated biomass and actual measurements obtained in the field. This is particularly important in phytoplankton populations, dynamics and assemblages in specific locations under specific conditions (Hill and Zimmerman, 2010; Shih et al., 2015, 2020b). Only by increasing the number of observations and enhancing our analysis of water composition will it be possible to reduce the uncertainty associated with Z_eu and Chl_s in estimating PP using the VGPM.

The mean PP in the euphotic zone (PP/Z_eu, mg-C m^–3 d^–1) presented a positive linear relationship between (PP_VGPM/Z_eu–sate) and (PP_{in situ}/Z_{eu–in situ}); i.e., slope: 0.51, r² = 0.39, p < 0.01 (Figure 5A). Higher values were observed during the NEM (4.7 ± 2.4 and 3.8 ± 2.2 mg-C m^–3 d^–1 of PP_{in situ}/Z_{eu–in situ} and PP_VGPM/Z_eu–sate, respectively) and lower values were observed during the SWM (1.9 ± 0.8 and 1.7 ± 0.4 mg-C m^–3 d^–1 of PP_{in situ}/Z_{eu–in situ} and PP_VGPM/Z_eu–sate, respectively). The slope of 0.51 for PP_VGPM/Z_eu–sate and PP_{in situ}/Z_{eu–in situ} was lower than that of PP_VGPM and PP_{in situ} (slope = 0.70), such that most data fell on the right side of the 1:1 line. Ratios of PP_VGPM/Z_eu–sate were nearly 20% lower than those of PP_{in situ}/Z_{eu–in situ}, indicating that Z_eu–sate was deeper than Z_{eu–in situ}, particularly during the SWM. This also indicates that the VGPM parameter Z_eu–sate indeed substantially affected PP estimates in this study. It has been proposed that satellite-based K_d values have to be calibrated against the zenith solar angle during the data processing in accordance with the methods outlined by Lee et al. (2005) and Li et al. (2015). However, users may not to confirm the processing from the downloaded or retrieved satellite-based products of PP and its relevant variables.

The significant linear correlation between (PoptB)_sate and (PoptB)_{in situ} (slope = 0.97, r² = 0.94, p < 0.01; Figure 4D) is simply a reflection of the relationship between the two SST records, resulting from the fact that PoptB is computed as a function of SST (i.e., a seventh-order polynomial, Equation 2–2) (Behrenfeld and Falkowski, 1997). Thus, any potential deviations between SST and PoptB had only a negligible influence on PP estimates obtained using VGPM. Many studies have nevertheless concluded that the accuracy of VGPM-based estimates of PP are poor, when using the 7th order polynomial of SST (Equation 2–2) to calculate the input of PoptB (Mizobata and Saitoh, 2004; Kameda and Ishizaka, 2005; Yamada et al., 2005; Siswanto et al., 2006; Ishizaka et al., 2007; Tang et al., 2008; Tang and Chen, 2016).

Researchers have reported that much of the uncertainty in estimating PP_VGPM is related to the computation of PoptB under the effects of phytoplankton physiology, growth conditions, abundance, size, and productivity (Gong and Liu, 2003; Kameda and Ishizaka, 2005; Yamada et al., 2005). It has been suggested that PoptB is influenced by SST as well as E₀ and various biological parameters, such as Chl concentration. PoptB represents an optimal daily carbon fixation rate in the water column previously described as a 7th order polynomial of SST, however, when SST exceeds 28.5°C, PoptB remains fixed at a constant 4 mg-C mg-Chl^–1 h^–1 (Behrenfeld and Falkowski, 1997; Mizobata and Saitoh, 2004), indicating that PoptB tends to be underestimated when SST exceeds 28.5°C. In low-latitude regions of the SCS, the constant PoptB mentioned above usually occurred in late spring, summer, and early fall, during which SST_{in situ} and SST_sate both exceeded 28.5°C. This increased the margin between actual PP values and the estimates obtained by inputting PoptB derived using SST_{in situ} or SST_sate. The reliability of PoptB estimates obtained using the VGPM seventh-order polynomial SST algorithm is not universally applicable (temporally or spatially). This has prompted oceanographers to tune existing methods or devise new methods for the precise estimation of PoptB, especially for ocean water at low latitudes.

Our results reveal that the satellite-derived primary production (e.g., PP_VGPM) may significantly affect global carbon export flux to deep waters, but what is the overall significance and impact of these PP values on POC fluxes? For example, Dunne et al. (2005) used the empirical model expression (Equation 4–1) to estimate carbon flux (or sequestration) in oceans. PP (mg-C m^–2 d^–1), Z_eu (m) and SST (°C) are three major factors affecting the estimated values of POC flux which is almost linearly proportional to PP.

According to the expression 4–1, the POC fluxes estimated from in situ measurements (PP_{in situ}, SST_{in situ} and Z_{eu–in situ}) of Hung et al. (2000); Hung and Gong (2007), and Shih et al. (2015) were from 12 to 319 mg-C m^–2 d^–1, an average of 20% less than the trap POC fluxes (25–274 mg-C m^–2 d^–1) (Figure 6). As described above, the inputs of SST_{in situ} and Z_{eu–in situ} to the expression were fixed, the PP_{in situ} was replaced by the PP_VGPM, an average difference between estimated and trap POC fluxes (estimated POC flux: 16–781 mg-C m^–2 d^–1) was a factor of 2. It has been suggested that the uncertainty of these POC fluxes is quite large if satellite-based PP is used to estimate carbon sequestrations in oceans. If the discrepancy between in situ measurements and satellite-based observations of PP can be diminished and the reliance on them (e.g., PP_VGPM) can be increased, it is to dedicate the potential importance and goal of the present study.

In both the Atlantic and Pacific oceans, increased phytoplankton biomass is observed at low latitudes during the boreal warm season. Other than equatorial upwelling, there are no other conspicuous seasonal trends in biogeochemical activities (Dandonneau et al., 2004). In the Arabian Sea, enhanced biogeochemical responses are also observed at low latitudes during the SWM (Banse and English, 2000). In this study, long-term and monthly variations in PP_{in situ} and PP_VGPM demonstrated influential monsoonal system at the SEATS (Figures 2, 3 and Table 2). The concurrence of low SST, deep Z_ML, high Chl_s, shallow Z_eu, and increased PP suggest that increased phytoplankton biomass or specific phytoplankton communities dominating were triggered during the NEM. It has been reported that the average nitrate+nitrite (N+N) concentration, one of the key nutrients for phytoplankton growth in the SCS, in the MLD in the seasons of NEM (0.1–0.3 μM) is ∼10 times higher than of SWM (∼0.03 μM) (Tseng et al., 2005). Moreover, the inventory of N+N and the depth of nitracline in the NEM (30 ± 19 mmol m^–2 and 28–62 m, respectively) have been observed a ∼ 4.5 fold higher and a ∼ 25–50% shallower than those reported in the SWM (7 ± 4 mmol m^–2 and 52–82 m, respectively) (Chen, 2005; Shih et al., 2020a). Evidences of abundant nutrient (e.g., N+N) and shallow nitracline depth favoring biological activities, imply that the growth of phytoplankton communities and associated photosynthesis were affected mainly by vertical advection providing inorganic nutrients from deeper waters, under the NEM system prevailing in the SCS (Liu et al., 2002; Bai et al., 2018; Chen et al., 2020; Zhou et al., 2020).

The global decrease in PP is particularly pronounced in high-latitude waters, which lose roughly 2,000 Mt-C y^–1 (Mt = 10¹² g), accounting for a 70% decline in carbon fixation via photosynthesis (Gregg et al., 2003). To compare annually reductions in PP_{in situ} (−11 mg-C m^–2 d^–1 y^–1) vs. the annual PP_{in situ} (241 mg-C m^–2 d^–1; mean PP_{in situ} of NEM and SWM; Table 1), it indicated that the annual reduction in carbon fixation via photosynthesis was ∼ -5% y^–1. Based on the 200 m isobath boundary of oligotrophic waters in the SCS (2.76 × 10¹² m²; Lin et al., 2003), the e-ratios were 5–16% in the SCS (Hung and Gong, 2010; Shih et al., 2019) and the decreased in carbon fixation was roughly 11 Mt-C, thereby accounting for 30–90% of the export production. This suggests a gradual decrease in the efficiency of photosynthetic carbon fixation by phytoplankton. Nevertheless, satellite-based observations do not show the signs of global warming on carbon fixation and sequestration.

Figure 5B illustrates the significantly negative relationship between PP_{in situ} and SST_{in situ} at the SEATS site. The slope of −37 mg-C m^–2 d^–1°C ^–1 was exceptionally close to the −36 mg-C m^–2 d^–1°C ^–1 reported in previous studies (Chen and Chen, 2006; Chen et al., 2007, 2008). Scaling factors related to the increase in SST_{in situ} caused by global changes are extremely complex (Sarmiento et al., 1998; Hung et al., 2010; Bai et al., 2018). Notwithstanding the complexity of factors governing SST_{in situ}, they can still be used to estimate PP values. The differences between daytime and nighttime SST_{in situ} values were statistically insignificant, during the NEM as well as the SWM (one-tail t test: p = 0.15 and 0.24, respectively (Figure 6). We therefore surmise that SST sampling schedules had no effect on the overall results. Furthermore, we observed a strong statistically significant correlation between SST_sate and SST_{in situ} (Figure 4C), indicating the efficacy of SST in estimating PP distributions over a broad horizontal area, regardless of the method used for derivation.

The straightforward relationship between PP_{in situ} and SST_{in situ} is important when seeking to predict PP values and estimate new or export production in euphotic zones. Under the environmental conditions described above, the reduction in carbon fixation due to photosynthesis would be ∼−15% °C^–1, and the decrease in carbon fixation would be roughly 37 Mt-C, thereby accounting for a 1- to 3-fold quantity of export production. The negative correlation between PP_{in situ} and SST_{in situ} in the current study matched the findings observed in mid-latitude tropical/subtropical regions of the Pacific and Atlantic oceans (Chen, 2000; Tilstone et al., 2009), but differed drastically from those reported in high-latitude regions (Kudryavtseva et al., 2018).

Generally speaking, the sampling resolution of in situ time-series is too low to eliminate temporal uncertainty over all timespans. Fortunately, we can use SST_sate to compensate for deficiencies in SST_{in situ} coverage (Bai et al., 2018; Chen et al., 2020). Only one daily SST_sate reading can be obtained at any given location; however, it would be perfectly reasonable to substitute that value with one obtained SST_{in situ}. We observed a statistically significant linear relationship between SST_sate with SST_{in situ} (Figure 4C); however, differences between daytime and nighttime SST_{in situ} measurements did not reach the level of significance (Figure 7). This suggests that SST estimates obtained using either method could be used to assess the influence of temperature on biogeochemical phenomena, such as PP.

Asynchronous variations between PP_{in situ} and PP_VGPM and their related variables in the VGPM algorithm indicate the following: (1) Satellite-based evaluations depend primarily on the assumption of an exponential decrease in the vertical distribution of PP from the surface to the euphotic zone base. Nonetheless, remote sensing cannot penetrate beyond the surface, and therefore cannot reflect the true vertical distribution of PP at depth (Hung et al., 2010; Buitenhuis et al., 2013; Shih et al., 2013, 2020b). This assumption is the primary cause for uncertainty between PP values obtained from satellite-based observations (i.e., PP_VGPM) and those derived from in situ measurements (i.e., PP_{in situ}). (2) When using satellite-based observations to estimate horizontal PP distributions, mathematical extrapolation/interpolation is commonly used to compensate for gaps in data coverage resulting from cloud coverage, heavy rains, rough seas, extreme weather events, natural episodes, suspended particles, and/or chromophoric dissolved organic matter. Thus, this approach cannot reflect “true” or “in situ” biogeochemistry responses to PP in the oceans. (Boyd and Trull, 2007; Shang et al., 2008; Tang et al., 2008; Hung et al., 2009, 2010; Shih et al., 2020a). For decades, SEATS has been used as a natural laboratory for studies of prolonged environmental changes and reciprocal biogeochemical responses. It is time to tune existing models or develop more reliable models if we are to gain meaningful estimates of biogeochemical phenomena in the oceans. The proposed calibration method aimed at improving PP estimates for the VGPM is expected to enhance our understanding of changes in the SCS.