Linking phytoplankton absorption to community composition in Chinese marginal seas

Highlights:

• Developing an absorption-based approach to model twenty phytoplankton pigments.

• Characterizing phytoplankton community composition through optical absorption.

• Paving a pathway toward remote sensing of phytoplankton community composition.

Abstract

Phytoplankton pigments significantly affect photosynthesis and play a crucial role in regulating marine ecological and biogeochemical processes. Assessment of phytoplankton pigments through optical means is desirable as it may be extended to satellite remote sensing. Here, using an extensive data set of high performance liquid chromatography (HPLC) phytoplankton pigment concentrations and phytoplankton absorption spectra (a_ph(λ)) collected through five cruise surveys of the Chinese marginal seas during 2016 and 2017, we explore the potentials of using a_ph(λ) to estimate twenty pigments. Specifically, the first and second derivatives of a_ph(λ) are used to construct an a_ph(λ) - pigment model. The validation of the a_ph(λ)-derived pigment classes, specific to individual phytoplankton community and size groups, shows a generally satisfactory model performance. Additionally, hierarchical cluster analysis also exhibits high similarity within the classification results based on the measured and modeled pigments, where only two pigments were decided into different clusters. Although still preliminary in nature, this proof-of-concept study for the Chinese marginal seas shows the potentials of using satellite remote sensing to assess phytoplankton pigment composition once hyperspectral satellite data are available and a_ph(λ) inversion algorithms are developed and validated.

Introduction

In marine waters, phytoplankton assemblages play essential roles in regulating the food web, ecological and biogeochemical processes, and carbon export and cycling (Field et al., 1998, Le Quéré et al., 2005, Guidi et al., 2016, Mouw et al., 2016). It is thus desirable to assess variations in phytoplankton assemblages. While field and laboratory techniques (e.g., microscopic taxonomy, flow cytometry, and high performance liquid chromatography or HPLC) provide accurate measurements of phytoplankton assemblages or pigment composition (Jeffrey, 1997, MacIsaac and Stockner, 1993, Roy et al., 2013, Sosik et al., 2010), they are limited in both space and time. Ocean color remote sensing provides more synoptic and frequent measurements of the surface ocean's optical properties, yet it is not trivial to infer information on phytoplankton assemblages from optical properties.

In the past decades, progress has been made towards retrievals of information on phytoplankton assemblages through remote sensing (Sathyendranath et al., 2004, Brewin et al., 2014, Brewin et al., 2015, Wang et al., 2016, Sun et al., 2019). Chlorophyll a (Chla) pigment is usually regarded as an indicator of phytoplankton (Falkowski et al., 1998, Yoder and Kennelly, 2003, Dierssen, 2010, Harding et al., 2016). While there are correlations in most ecosystems between Chla and other accessory pigments, it is well established that Chla does not carry taxonomic information (Chase et al., 2013). Other methods have been developed for this purpose. For instance, detecting phytoplankton size classes (PSCs) enables us to divide the phytoplankton community into different classes specific to different cell size ranges; three size classes, including pico- (<2 μm), nano- (2 – 20 μm), and microphytoplankton (>20 μm), are generally detected by using those pigment-based methods (Brewin et al., 2015, Devred et al., 2011, Uitz et al., 2006, Hirata et al., 2011). In order to quantify different size classes of phytoplankton, phytoplankton assemblages are strictly divided into three size ranges. Yet, some species are distributed across wide size ranges and even simultaneously belong to two size classes. For instance, diatoms are usually located in micro- and nanophytoplankton classes and prymnesiophytes coexist in nano- and picophytoplankton classes (Uitz et al., 2008, Brewin et al., 2014, 2015). Besides, the same size class sometimes contains multiple phytoplankton communities (Uitz et al., 2008, Brewin et al., 2014, Brewin et al., 2015, Sun et al., 2017, Sun et al., 2019, Sun et al., 2018). The detection of one phytoplankton size class cannot always clearly denote the existence of specific phytoplankton communities in marine waters. Thus, PSCs show a limited capacity in characterizing phytoplankton community composition, despite their superiority to a single Chla indicator.

Another recent work estimates concentrations of phytoplankton pigments through optical data, as pigment data have been used widely to characterize phytoplankton community composition. The pigments or pigment groups have been modeled using phytoplankton absorption signatures. For instance, Moisan et al. (2011) made use of in situ data (phytoplankton absorption spectra and HPLC pigments) in a linear inverse calculation to extract pigment-specific absorption spectra, and then established a second linear inverse calculation with the total phytoplankton absorption spectra, and estimated seven pigments with good correlations (R² > 0.5). Chase et al. (2013) reported that the total Chla, chlorophyll b (Chlb), and chlorophyll c (Chlc), as well as photosynthetic carotenoids (PSC) and photoprotective carotenoids (PPC), could be predicted via decomposing the in situ particulate absorption spectra into absorption by the five groups of pigments and non-algal particles, and generated percent errors between 30% and 57%. Note that the pigments predicted here were not individual pigments, just five summed-pigment groups, such as total Chla (representing a sum of Chla, divinyl chlorophyll a (DV-Chla), and chlorophyllide a (Chlide-a)), total Chlb (Chlb and divinyl chlorophyll b (DV-Chlb)), Chlc (chlorophyll c1 (Chlc1) and chlorophyll c2 (Chlc2)), PSC (19′-hexfucoxanthin (19′-hex), fucoxanthin (Fuco), 19′-butfucoxanthin (19′-but), and peridinin (Per)), and PPC (α-carotene (α-Car), β-carotene (β-Car), zeaxanthin (Zea), alloxanthin (Allo), and diadinoxanthin (Diadino)). Liu et al. (2019) compared the performances of two approaches, namely Gaussian decomposition and matrix inversion, and demonstrated that the Gaussian decomposition showed good estimates of total Chl-a, Chlb, Chlc, PPC, and PSC, while the matrix inversion algorithm enabled robust retrievals of specific carotenoids, including Fuco, Diadino, and 19′-hex, unavailable to Gaussian decomposition. However, despite the progress made on absorption-based pigment retrievals, the retrievals are for the five discrete pigment groups rather than all individual pigments, making it difficult to estimate phytoplankton community composition.

Compared with the HPLC analysis, remote sensing algorithms to estimate pigments have an advantage of large-scale applications through satellite data. Pan et al. (2010) established regression models between band ratios of remote sensing reflectance (R_rs) and pigment concentrations. Those models were then applied to MODIS images to map the spatial concentration distributions of Chla, Fuco, Per, and Zea on the northeast coast of the United States. Using in situ hyperspectral R_rs measurements, Chase et al. (2017) developed an inversion algorithm that defined phytoplankton pigment absorption as a sum of Gaussian functions, where amplitudes of the Gaussian functions would be associated with concentrations of Chla, Chlb, Chlc, and PPC. For coastal waters, Craig et al. (2012) retrieved the Chla concentrations and phytoplankton absorption spectra (a_ph) based on the empirical orthogonal function (EOF) analysis and found that the spectral normalization was the key to the models' success. Bracher et al. (2015) applied an EOF analysis to R_rs data to predict a suite of pigments and pigment groups with high quality, which developed multiple linear regression models with measured pigment concentrations and EOF loadings. The EOF method was also utilized later in Palacios et al., 2015, Xi et al., 2015, Xi et al., 2020), Bracher et al. (2020). In recent work, a neural network classifier was used to estimate ten pigments from satellite ocean color observations (El Hourany et al., 2019). However, similar to the existing absorption-based analyses, these R_rs-based analyses also focused on major pigment groups rather than individual pigments, thus facing the same difficulty when estimating phytoplankton community composition. On the other hand, the variations in phytoplankton pigments directly impact phytoplankton absorption signals in water bodies. Therefore, absorption-based methods have a direct advantage in revealing the variability of phytoplankton pigment concentrations.

Thus, this study aims to establish a regionally applicable approach to determine the phytoplankton pigments of Chinese marginal seas through a series of analyses of phytoplankton absorption and pigments. To achieve this goal, the study analyzes an extensive data set of HPLC-derived concentrations of twenty phytoplankton pigments (including Chla and nineteen biomarker pigments), as well as phytoplankton absorption from each of the water samples. The use of phytoplankton absorption spectra is to seek potential remote sensing applications in the future as the former can, at least in principle, be derived from the latter for each remote sensing image pixel. Once individual pigments of a given water sample are estimated from its absorption spectrum, the water sample's phytoplankton community composition can be determined from a hierarchical cluster analysis of HPLC pigment data collected from the study region.