A novel semianalytical remote sensing retrieval strategy and algorithm for particulate organic carbon in inland waters based on biogeochemical-optical mechanisms

Highlights

• A novel bio-optical based POC semi-analytical inversion algorithm for inland water.

• We improved POC estimation strategy by estimating POC terminal members.

• a_ph(674) and a_nap(443) can be used to drive C_endPOC and C_terPOC, respectively.

• Endogenous POC dominance in Lake Taihu increased significantly in summer.

Abstract

The estimation of particulate organic carbon (POC) concentrations from satellite images can provide crucial spatiotemporal continuous observation data for the carbon cycle and ecological environmental governance. Here, we developed a novel inversion algorithm for deriving POC in inland water based on remote sensing and geochemical isotopes, which is summarized as follows. First, we developed empirical relationships between the phytoplankton absorption coefficient and endogenous POC concentration (C_endPOC) and between the nonalgal particulate absorption coefficient and terrestrial POC concentration (C_terPOC). Second, based on the valid relationships, semianalytical retrieval models were established to estimate C_endPOC and C_terPOC. Third, the proportions of endogenous POC (R_endPOC) and terrestrial POC (R_terPOC) to the total POC concentration (C_POC) were derived using a three-band empirical model. Finally, C_POC was obtained by dividing C_endPOC by R_endPOC (R_endPOC ≥ 0.5) or dividing C_terPOC by R_terPOC (R_endPOC < 0.5). Validation with field data shows that our proposed algorithm can accurately derive C_POC (0–20 mg/L), with a root mean square deviation (RMSD), median bias (MB), median absolute percent difference (MAPD), and median ratio (MR) of 1.15 mg/L, −0.05 mg/L, 24%, and 0.98, respectively. Synchronous validation based on Sentinel-3/OLCI images confirmed the accuracy, with RMSD, MB, MAPD, and MR values of 0.41 mg/L, −0.16 mg/L, 28%, and 0.91, respectively. The algorithm was applied to ocean and land color sensor (OLCI) images to reveal the temporal and spatial variations in POC in Lake Taihu.

Introduction

Organic carbon (OC) in water has been deemed to be an important part of the lake carbon pool, and it regulates the pools of inorganic and sedimentary carbon through degradation and deposition (Tranvik et al., 2009; Anderson et al., 2014; Siegel et al., 2014). Particulate organic carbon (POC) generally represents a smaller fraction of OC than dissolved organic carbon (DOC), but it plays a vital role in the element cycle, energy flows and transportation of associated elements and compounds because POC is involved in numerous biogeochemical processes in aquatic ecosystems (Behrenfeld et al., 2013; Jiang et al., 2015; Siegel et al., 2014; Stackpoole et al., 2017; Tranvik et al., 2009). Endogenous production (such as phytoplankton photosynthesis) and terrestrial augmentation (such as runoff input and sedimentary OC upwelling) can increase POC concentrations. Microbial respiration, particle deposition, etc., may transform POC to DOC and dissolved inorganic carbon (DIC), thus decreasing POC concentrations. These physical and biological processes lead to highly dynamic spatial and temporal variations in POC concentrations. Therefore, regional/global monitoring of POC is essential to understand the biogeochemical processes of POC and to further understand the global carbon cycle of aquatic systems (Battin et al., 2009; Bastviken et al., 2011; Jones et al., 2001).

The advantages of remote sensing technology in large-area synchronous observation can provide important support for POC monitoring. Much progress has been made in deriving POC using remote sensing approaches during the past two decades, greatly expanding our knowledge of POC spatiotemporal dynamics in oceans and freshwater (Allison et al., 2010; Cetinic et al., 2012; Hu et al., 2015; Huang et al., 2017a). Among them, the water constituent-based algorithm was developed to remotely derive POC from chlorophyll a (Chla) or suspended particulate matter (SPM), with the assumption that POC is significantly correlated with these constituents. A significant correlation between POC and Chla was observed in the continental shelf seas of China (Wei et al., 2019), the Arabian Sea, and Tokyo Bay (Sathyendranath et al., 2009), indicating that POC in these waters could be estimated through satellite-derived Chla products. The relationships between POC and SPM were successfully built and applied to coastal and inland waters because a substantial fraction of POC is correlated with particulate matter in turbid waters (Liu et al., 2019). However, the fact that the relationship between POC and SPM and Chla usually varies with the composition of water particles results in large uncertainties in deriving POC using this approach. Moreover, there are still substantial errors in satellite-derived Chla and SPM in the open ocean and inland waters. Apparent optical property (AOP)-based empirical models were developed to directly retrieve POC from R_rs(λ) or some spectral indexes. A single-band approach using normalized water-leaving radiance at 555 nm and diffuse attenuation at 490 nm was developed to estimate POC in the South Atlantic, the South Pacific Ocean and the Gulf of Mexico (Mishonov et al., 2003; Gardner et al., 2006). A large number of previous studies have shown that the ratio of R_rs(λ) in the blue and green bands can be used to derive POC with satisfactory accuracy in the open ocean (Stramski et al., 2008; Stramska and Cieszynska, 2015; Stramski et al., 2022). However, the limitations of blue and green bands to capture the variations in POC in optically complex inland waters resulted in the formulations of several regional POC algorithms (Huang et al., 2017a; Jiang et al., 2015; Lyu et al., 2017). The inherent optical property (IOP)-based algorithm, which uses a bio-optical mechanism, was developed based on optical closure theory to model the relationships between R_rs(λ) and IOPs, as well as the links between POC and relevant IOPs. In particular, the backscattering coefficient at 510 nm was the best proxy of POC and was utilized to derive POC from remote sensing in the open ocean (Stramski et al., 1999). Recently, an absorption-based model was developed to estimate POC based on the local absorption characteristics and POC in inland and coastal water areas (Jiang et al., 2019). However, accurately relating IOPs to POC and deriving IOPs from R_rs(λ) are critical and challenging to successfully derive POC from satellite data using the IOP-based algorithm.

Although it is evident that POC can already be accurately derived from satellite data with various algorithms, it is still a challenge for inland waters because POC in such waters usually has complex sources. Recently, the empirical algorithm was improved for inland waters through consideration of the different POC sources. Lin et al. (2018) divided water into two categories—phytoplankton-derived POC dominated and terrestrial POC dominated—by subtracting the R_rs(560) of pure water from OLCI-R_rs(560), comparing it with OLCI-R_rs(709), and then establishing two-step algorithms to estimate the POC concentration in inland water. The critical steps are to differentiate between POC sources and to quantify POC terminal member concentrations. Jiang et al. (2019) found that a_ph(443)/a_nap(443) could be used to identify phytoplankton-dominated POC and detritus-derived POC in inland water. Based on R_rs(560), R_rs(674) and R_rs(709), Xu et al. (2021) proposed a POC-source optical index to estimate the percentage and concentration of endogenous POC in Lake Taihu. These previous studies suggest that the optical approach provides an effective way to identify POC sources and to quantify POC terminal member concentrations and thus has critically important implications for improving POC satellite retrieval algorithms, but it has not been fully explored. Although these strategies have improved the retrieval accuracy of POC to a certain extent, their classification indicators are only based on absorption or reflectance spectra and are not supported by the biochemical indicators of POC itself.

Due to the differences in isotopic fractionation of different carbon sources, POC sources can be distinguished effectively according to their geochemical characteristics (Bouchez et al., 2014; Chen et al., 2018; Jiang and Ji, 2013; Meng et al., 2021; Galy et al., 2015; Guo et al., 2015). The aims of our study are to (1) explore the relationships between the absorption coefficients of water constituents and POC terminal member concentrations quantified by isotope-tracing technology, (2) propose a novel POC satellite retrieval strategy and algorithm for inland waters according to the absorption-POC terminal member concentration relationships, and (3) reveal the pattern of spatial and temporal variations in POC in a large shallow eutrophication lake (Lake Taihu in China) using the proposed algorithm.