Satellite-Derived Particulate Organic Carbon Flux in the Changjiang River Through Different Stages of the Three Gorges

Remote Sensing of Environment 223 (2019) 154–165

Contents lists available at ScienceDirect

Remote Sensing of Environment

journal homepage: www.elsevier.com/locate/rse

Satellite-derived particulate organic carbon ﬂux in the Changjiang River through diﬀerent stages of the Three Gorges Dam T ⁎ Dong Liua,b, Yan Baib,e, , Xianqiang Heb,e, Delu Panb, Chen-Tung Arthur Chenc, Teng Lib,YiXud, Chaohai Gongd, Lin Zhangb a Key Laboratory of Watershed Geographic Sciences, Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences, Nanjing 210008, China b State Key Laboratory of Satellite Ocean Environment Dynamics, Second Institute of Oceanography, Ministry of Natural Resources, Hangzhou 310012, China c Department of Oceanography, National Sun Yat-sen University, Kaohsiung 80424, Taiwan d Lower Changjiang River Bureau of Hydrological and Water Resources Survey, Nanjing 210011, China e Institute of Oceanography, Shanghai Jiao Tong University, Shanghai 200240, China

ARTICLE INFO ABSTRACT

Keywords: As the largest Asian river and fourth world's largest river by water flow, the Changjiang River transports a Particulate organic carbon considerable amount of terrigenous particulate organic carbon (POC) into ocean, which has experienced sig- Changjiang River nificant pressure from human activities. We conducted monthly sampling (from May 2015 to May 2016) at Three Gorges Dam Datong, the most downstream non-tidal hydrological station along the Changjiang River. To monitor long-term Landsat data POC variations, we developed a two-step POC algorithm for Landsat satellite data and calculated monthly POC Bio-optical model flux during 2000–2016. Monthly POC flux ranged from a minimum of 1.4 × 104 t C in February 2016 to a maximum of 52.04 × 104 t C in May 2002, with a mean of 13.04 × 104 t C/month. At Datong, monthly POC flux was positively exponentially related to water flow (N = 118, R = 0.69, p < 0.01). In the wet season (May to October) with high water flow, a large amount of terrigenous POC was flushed into the Changjiang River. After the Three Gorges Dam (TGD), both POC concentration (R = 0.46, p < 0.01) and POC flux (R = 0.33, p < 0.01) significantly exponentially decreased during 2000–2016. After 2011 with regular operation of the TGD, POC transport at Datong showed relative stability and its seasonal difference became smaller. POC concentration from the TGD to Datong in the Changjiang mainstream (~1000 km) was also derived from Landsat data, which was spatially varied with water flow, slope gradient, and watercourse width. This study improved our under- standing of the spatiotemporal variations of POC transport in the Changjiang mainstream, especially the in- fluences of the TGD construction.

1. Introduction 2012). With wide spatial coverage and high temporal repeatability, satellite remote sensing is ideal and necessary for retrieving dynamic Rivers play crucial roles in connecting land and ocean ecosystems, riverine POC concentration. the two largest active carbon pools in global carbon cycle (Siegenthaler As the largest Asian river according to annual water flow, the and Sarmiento, 1993; Wang et al., 2012). On a global scale, rivers de- Changjiang River exports a great deal of terrigenous POC into the East liver approximately 2.0 × 108 t C/yr of particulate organic carbon China Sea (ECS) every year (Duan et al., 2008; Liu et al., 2014; Wang (POC) into marginal seas (Wang et al., 2012). After coastal transfor- et al., 2012). Over the past several decades, moreover, sediment yield in mation and oxidation consumption, the remaining riverine POC is the Changjiang watershed has been markedly influenced by extensive buried in estuarine and/or coastal sediments, constituting a significant human activities, the construction of Three Gorges Dam (TGD) for ex- carbon sink (Bianchi and Allison, 2009; Saliot et al., 2002; Wollast, ample (Wang et al., 2009; Yang et al., 2006, 2007). In November 1997, 1991). Regarding on global riverine POC flux, 23.74% was transported the Changjiang River was intercepted for TGD construction. On 1 June by Asian rivers (Seitzinger et al., 2005). Asian rivers are usually char- 2003, the TGD began impoundment. Water level reached 135 m on 10 acterized by seasonally varied POC concentration, with high values in June 2003, 156 m on October 2006, and 175 m by the end of October large water flow seasons (Liu et al., 2015a; Ran et al., 2013; Wang et al., 2010, respectively (Zhang et al., 2014). When at 175 m, water storage is

⁎ Corresponding author. E-mail address: [email protected] (Y. Bai). https://doi.org/10.1016/j.rse.2019.01.012 Received 4 June 2018; Received in revised form 5 January 2019; Accepted 10 January 2019 Available online 24 January 2019 0034-4257/ © 2019 Elsevier Inc. All rights reserved. D. Liu et al. Remote Sensing of Environment 223 (2019) 154–165 about 3.93 × 1011 m3 and catchment area is about 1.08 × 106 km2 north and 20 m in the south (Fig. 1c). Rivers upstream of Datong drain (http://news.xinhuanet.com/). 94.3% of the total basin area and account for > 95% of water and se- The TGD has reportedly changed material transport of the diment transports in the whole basin (Duan et al., 2008). Changjiang River, leading to ecological changes in the ECS (Bai et al., 2014, 2015; Feng et al., 2014). Xu and Milliman (2009) reported that 3. Materials and data processing the TGD trapped about 60% of upstream input sediment during 2003–2006. After 2003, the Changjiang River has entered a phase with 3.1. TSM, POC, and other in-situ data sediment flux at Datong hydrological station < 200 mt/yr, compared with average value of ~340 mt/yr during 1986–2002 (Yang et al., In-situ sampling was conducted in the middle of each month from 2006). As part of the suspended sediment, POC transport has also been May 2015 to May 2016, using hydrographic survey vessel, China (No. reduced by the TGD (Wu et al., 2016; Zhang et al., 2014). Based on 138). At each station of P1, P2, and P3 (Fig. 1b), waters at three depths biweekly measurements in Changjiang River estuary (CRE) from 2003 (surface, middle, and bottom) were collected using a Niskin sampler to 2011, Wu et al. (2016) reported that a sharp decrease of POC flux (General Oceanics Inc. USA) within about 15 min. When collecting was observed, especially in the wet season. However, long-term mon- water, the sampler was fixed at 0.5 m to the water surface, half the itoring is still mandatory for assessing POC changes in the Changjiang water depth, and 0.5 m to the water bottom, respectively (Fig. 1c). In River through different stages of the TGD. each month, surface waters were also collected at station P4 several With low cost and labor-consuming, satellite remote sensing is a hours before and/or after the Datong section sampling (Fig. 1b). good alternative tool to explore long-term riverine POC variations. Water was filtered through pre-combusted GF/F membrane Many remote sensing algorithms had been developed for oceanic POC (Whatman, UK, 0.7 μm pore size and Φ47 mm) and pre-weighed cellu- closely related to phytoplankton density (Stramska, 2009; Stramski lose acetate membrane (Sartorius, Germany, 0.45 μm pore size and et al., 1999, 2008), but they were not applicable to riverine POC pri- Φ47 mm) to obtain POC and TSM samples, respectively. All samples marily sourced from watershed soil (Liu et al., 2015a; Wang et al., were stored in a refrigerator ( −18 °C) until laboratory measurement. 2012). In addition, with usual spatial resolution of ~1 km, commonly POC content was determined using high temperature combustion used ocean color data is too coarse to be applied to rivers with widths of method (680 °C) by referring to the Joint Global Ocean Flux Study only several kilometers at most. Alternatively, long-term Landsat sa- (JGOFS) protocols (Knap et al., 1994). As Strickland and Parsons tellite data with a resolution of 30 m was a good choice and had been (1972), TSM content was determined gravimetrically, with a precision frequently used to derive turbidity (Joshi et al., 2017b), total suspended of 0.01 mg. In total, we collected 141 TSM samples and 141 POC matter (TSM) (Islam et al., 2001, 2002; Lymburner et al., 2016; Wang samples (Table 1). et al., 2009), dissolved organic matter (Griffin et al., 2011; Joshi et al., The published in-situ POC and TSM by Wang et al. (2012) were also 2017a; Kutser et al., 2016), etc. in riverine or coastal small waters. used to supplement our in-situ dataset for following analysis. Wang However, due to limited in-situ POC measurement, a 16-day revisit et al. (2012) monthly collected surface, middle, and bottom waters in period for Landsat satellite, and locally dependent relationships be- the central mainstream at Datong in 2009, and mixed them at a 2:1:1 tween POC and TSM (Gao et al., 2002; Ran et al., 2013; Zhang et al., volume ratio. By following JGOFS (Knap et al., 1994), waters were also 2014), etc., few studies about monitoring riverine POC had been re- filtered through pre-combusted GF/F filters to obtain POC samples and ported. These also lead to the importance of validation work and the POC content was determined using a Perkin-Elmer 2400 CHNS Analyzer limited matchup data when using Landsat data. (USA) with an analytical precision of ± 4%. For TSM, Wang et al. To monitor dynamic POC in the Changjiang River from space and (2012) used values by oven drying method from the Ministry of Water determine POC variations after the TGD, we conducted one-year sam- Resources of China (MWRC). pling (May 2015 to May 2016) in each month at Datong, the most In addition to POC and TSM, we also collected Chl-a, colored dis- downstream non-tidal hydrological station. Then, we developed a two- solved organic matter (CDOM), and suspended particle samples step POC satellite algorithm and retrieved sequenced POC concentra- (Table 1). Chl-a concentration was measured using a Turner fluorometer tion from both Landsat TM and ETM+ data during 2000–2016. (Knap et al., 1994). Absorption coefficient spectra of CDOM (aCDOM) Combining water flow, long-term sequenced POC flux was also calcu- and no-algal particle (an) from 250 to 900 nm with an interval of 1 nm lated. Furthermore, we investigated seasonal POC variations at Datong were determined by the spectrophotometric method (Mueller et al., through different stages of the TGD. 2003). Additional details about collection, storage, and measurement of these parameters can be found in the previously published article by Bai 2. Study area et al. (2013). In total, we collected 141 Chl-a, 139 CDOM (two were

contaminated), and 141 NAP samples (Table 1). Using aCDOM and an The Changjiang River originates from the Qinghai-Tibetan Plateau between 430 and 450 nm, we further calculated CDOM spectral slope at

(Wu et al., 2007; Xu and Milliman, 2009). It is 6300 km long and covers 440 nm (SCDOM(440)) and NAP spectral slope at 440 nm (Sn)(Table 1). a basin area of 1.94 × 106 km2 before finally emptying into the ECS From September 2012 to July 2015, an YSI turbidimeter was de- (Zhang et al., 2014). The part upstream of the Yichang hydrological ployed to measure TSM concentration at station P4 around 10:00 local station is the upper reach (Fig. 1a), and has a complex geomorphology, time every day by MWRC, namely the YSI-measured TSM. Monthly steep gradient, and rapid flow (Wu et al., 2007; Zhang et al., 2014). The water flow and TSM at Datong from 2000 to 2016 were used. Monthly middle reach (between the Yichang and Hukou hydrological stations) water flow and mean TSM at Yichang from 2003 to 2016 were also has a relatively low gradient (Fig. 1a) (Wu et al., 2007). The lower used. These data was sourced from the MWRC (2000–2016). reach (downstream of the Hukou hydrological station) has a wide watercourse (Fig. 1a) (Zhang et al., 2014). Except for some areas in the 3.2. Water reflectance headwater, about 76% of the Changjiang basin is located in the sub- tropical monsoon climate zone. For the whole basin, rainfall mainly When collecting water at Datong, radiance spectra of water surface occurs in the wet season (May to October), accounting for 70–90% of (Lws), skylight (Lsky), and a standard reflecting plate (Lp) were also the annual total (Zhang et al., 2014). measured using a ASD FieldSpec Spectroradiometer (USA), which mea-

Many studies on material transport of the Changjiang River used in- sured radiance from 350 to 2390 nm with an interval of 1 nm. Lws, Lsky, situ data at Datong (Liu et al., 2015b; Wang et al., 2012; Zhang et al., and Lp were determined from above-water measurements by following 2014), which had monitored water and sediment transports since 1950s the Ocean Optical Protocols proposed by NASA (Mueller et al., 2003).

(Duan et al., 2008). Water depth at Datong was around 10 m in the When measuring Lws and Lsky, the probe was positioned at 90–135° to

155 D. Liu et al. Remote Sensing of Environment 223 (2019) 154–165

Fig. 1. Study area and sampling stations. (a) Schematic map of the Changjiang River drainage basin and its sub-catchments. Basin upstream of the Datong can be divided into five parts: ① TGD watershed with the Yichang hydrological station outlet (A); ② Hanjiang River watershed with the Huangzhuang hydrological station outlet (B); ③ Dongting Lake watershed with the Chenglingji hydrological station outlet (C); ④ Poyang Lake watershed with the Hukou hydrological station outlet (D); ⑤ Mainstream region with the Datong hydrological station outlet (E). The Fuling hydrological station is also labeled (F). (b) Datong river reach and sampling stations, with three stations (P1, P2, and P3) across a section vertical to the river bank. Base map is the composite image of Landsat-5 TM data on 28 July 2011 (R: band 7, G: band 4, B: band 3). Region R contains 10 × 10 pixels in the river middle. Only satellite data in region R was used so as to remove “adjacency effect” by land. (c) Sampling profiles at the three stations across the Datong section.

Table 1 3.3. Landsat satellite data Information of in-situ measurements at Datong. For all stations, only surface water reﬂectance was measured, with only 54 measurements from May 2015 to In this study, both data of TM onboard the Landsat-5 satellite May 2016. (Landsat-5 TM) and ETM+ onboard the Landsat-7 satellite (Landsat-7 Parameters Sample size Minimum Maximum Mean ± Std ETM+) was used. Landsat-5 TM recorded data from March 1984 to May 2012. Landsat-7 ETM+ has continuously recorded data since July TSM (mg/L) 141 19.08 125.13 51.11 ± 22.13 1999. Both Landsat-5 TM and Landsat-7 ETM+ scan the earth surface POC (mg/L) 141 0.37 1.47 0.68 ± 0.22 Chl-a (μg/L) 141 0.37 3.56 1.45 ± 0.56 around 10:30 a.m. local time. The Landsat-5 TM and Landsat-7 ETM+

aCDOM(440) (1/m) 139 0.26 0.49 0.38 ± 0.04 had four similar visible light bands setups, with central wavelength at SCDOM (1/nm) 139 0.015 0.019 0.017 ± 0.001 485 nm, 560 nm, 660 nm, and 830 nm, respectively. The scan-line an(440) (1/m) 141 0.622 5.106 2.045 ± 0.803 corrector of Landsat-7 ETM+ failed after 30 May 2003, resulting in Sn(440) (1/nm) 141 0.008 0.012 0.011 ± 0.001 about 22% per scene without being scanned (Singh and Prasad, 2014). Water reflectance 54 / / / Nonetheless, the valid data was still available (Griffin et al., 2011; Wang et al., 2007, 2009). the plane of incident radiation to minimize the effects of sun glint and In total, 59 scenes of Landsat-5 TM data and 118 scenes of Landsat-7 ETM+ data during 2000–2016 were downloaded from the Geospatial ship shadow. View angle to the aplomb direction was 135–150° for Lws, Data Cloud Center (http://www.gscloud.cn/). For the Level 1T data, but 30–45° for Lsky. Based on measured radiance, water reflectance (Rrs, − sr 1) was calculated using Eq. (1) (Mueller et al., 2003). radiometric calibration, geometric rectification, and topographic correction had been processed. Accurate atmospheric correction of inland Rrs()λρλLλrLλπLλ=− ()(ws() sky())/( p()) (1) waters is an unresolved issue (Kutser et al., 2016). For Landsat data, the fast line-of-sight atmospheric analysis of spectral hypercubes (FLAASH) ρ λ fl fl where, ( ) is the re ectance of the standard re ecting plate; r is the was reported valid to remove atmospheric effects and calculate water fl re ectance of the air-water interface, with r = 0.028 for the turbid reflectance for inland waters (Bernardo et al., 2017; Joshi et al., 2017b; Changjiang River waters like those in the Pearl River Estuary (Liu et al., Kutser et al., 2005, Kutser, 2012). Therefore, we used the FLAASH 2015a).

156 D. Liu et al. Remote Sensing of Environment 223 (2019) 154–165 based on ENVI 4.8 software (ENVI, 2008). For region R in Fig. 1b, mean Oyama et al. (2009), we also got that phytoplankton contributed only water reﬂectance of each visible band was further calculated by the 0.42 ± 0.25% to TSM. Therefore, water reﬂectance variation was following three steps (Bailey and Werdell, 2006): mainly determined by the NAP content, and we set Chl-a, aCDOM, and SCDOM as the mean in-situ values in the simulation (Table 1). For NAP, 1) For each band, we extracted valid values within the 10 × 10 pixels we used the empirical equations by Ruddick et al. (2006), assuming box (Fig. 1b). Invalid pixels of Landsat-7 ETM+ data after 30 May optics in turbid waters were only described by pure water and NAP

2003 were ignored. (Table 2). From the validation results using in-situ data, the modeled an 2) If valid pixels were > 50% (> 50 pixels), we further calculated their at four visible light bands of Landsat-5 TM were satisfactory (Fig. 2a). mean and standard deviation (SD) and deleted pixels beyond By setting in-situ TSM concentrations as inputs, we simulated water mean ± 1.5SD. reflectance and compared with in-situ data (Section 3.2). For the four 3) If variation coefficient (VC, SD/mean) of the remaining pixels visible light bands of Landsat-5 TM, average relative errors (ARE) were was < 0.15, mean water reflectance of the remaining pixels was 14.4%, 17.01%, 22.26%, and 21.31%, respectively (Fig. 2b). available. With the developed bio-optical model above, we set monthly mean TSM values at Datong during 2000–2016 as inputs to simulate water 4. Algorithm development and validation reflectance (N = 204). From reported data by the MWRC, we got that monthly mean TSM at Datong ranged from 30.5 to 735.4 mg/L during 4.1. Water reflectance simulation using a bio-optical model 2000–2016, with values only in six months larger than 400 mg/L (MWRC, 2000–2016). Hyperspectral water reflectance of turbid inland water can be simulated using bio-optical model (Kutser et al., 2001). From May 2015 4.2. POC satellite algorithm for Landsat data to May 2016, the in-situ surface TSM was within 26.66–96.67 mg/L. Actually, TSM range at Datong is much wider than our measurements, From hyperspectral water reflectance in narrow bands, we could with monthly mean TSM of 30.5–735.4 mg/L during 2000–2016 re- calculate TSM using numerical method such as linear matrix inversion ported by MWRC (2000–2016). To enlarge the data range and develop a (Hakvoort et al., 2002). However, this method cannot be used to re- robust algorithm, we simulated water reflectance using a bio-optical trieve TSM from Landsat data in wide bands. Although POC is not one model. For Lambertian radiance field, water reflectance at wavelength water color constituent like Chl-a, CDOM and TSM, which all have λ (R (λ)) was expressed using Eq. (2) (Lee et al., 1998): rs working bio-optical models (Jiang et al., 2015; Lee et al., 1998), POC is Rrs(λrλrλ )=− 0.518 rs( )/[1 1.562 rs( )] (2) usually related to TSM (Gao et al., 2002; Ludwig et al., 1996; Ni et al., 2008). Therefore, we first retrieved TSM from Landsat data, and then where, rrs(λ) is the subsurface water reflectance. To calculate rrs(λ), we further calculated POC from TSM. To develop a TSM algorithm, we first chose the widely used expression proposed by Gordon et al. (1988), computed equivalent band reflectance (r ) of Landsat data from the shown as Eq. (3). equi 204 simulated water reflectance spectra (Section 4.1). For a specific ruurs(λ)≈+ (0.0949 0.0794 (λ)) (λ) band with central wavelength of λ, requi(λ) was calculated using Eq. (5) ubab(λ)≡+ b(λ)/( (λ) b(λ)) (3) (Liu et al., 2015a). where, u(x) is the intermediate variable; a(λ) is the total absorption λ max λ max ffi λ ffi λ ⎡ ⎤ ⎡ ⎤ coe cient; bb( ) is the total backscattering coe cient. Both a( ) and requi() λ= ⎢ ∫∫ fλR () rs()() λLλdλ⎥ /⎢ fλLλdλ () () ⎥ bb(λ) are determined by clean water, phytoplankton, non-algal particle ⎣ λ min ⎦ ⎣ λ min ⎦ (5) (NAP), and CDOM together, so we got Eq. (4). where, f(λ) is the spectral response function of Landsat-5 TM or aa(λ)=+++ w(λ) a p(λ) a n(λ) a g(λ) Landsat-7 ETM+; L(λ) is the solar irradiance at the top of the atmo- bbb(λ)=+ bw(λ) BbBb p p(λ) + n n(λ) (4) sphere; integral calculation ranges from 400 nm (λmin) to 900 nm (λ ); and R (λ) is the simulated water reflectance. For both Landsat-5 where, b is the scattering coefficient; B is the backscattering ratio (b / max rs b TM and Landsat-7 ETM+, we found that TSM showed power relation to b); footnotes w, p, n, and g refer to clean water, phytoplankton, NAP, the equivalent reflectance ratio of r (band4)/r (band1), shown as and CDOM, respectively. In the bio-optical simulation, the used em- equi equi Eq. (6). pirical equations and parameters for different constituents are shown in Table 2. b ⎧log10 (TSM) =×aRatio For the 54 in-situ water reflectance measurements, CDOM varied in ⎨Ratio= requi(band4)/r equi(band1) a narrow range (Table 1), but corresponding TSM ranged from 22.00 to ⎩ (6) ffi 76.99 mg/L. Using the conversion coe cient of Chl-a to TSM of 0.12 by where, both a and b are fitting coefficients; for Landsat-5 TM, a = 2.1454, b = 0.2945; for Landsat-7 ETM+, a = 2.1012, b = 0.2953. Table 2 The Eq. (6) was applied to retrieve TSM concentration from in-situ The used empirical equations and parameters in the bio-optical simulation. For water reflectance. For both Landsat-5 TM and Landsat-7 ETM+, ARE a specific wavelength λ, a (λ) and a (λ) were listed in Lee et al. (1998). 0 1 were 13.69% and 13.46%, respectively (Fig. 3a). Parameters & equations References For all in-situ TSM and POC concentrations (Section 3.1), POC showed a significant linear correlation with TSM, with R2 = 0.88 and aw(λ) at 400–725 nm Pope and Fry (1997) p < 0.01 (Fig. 3b). The linear relationship between POC and TSM re- aw(λ) at 725–1000 nm Kou et al. (1993) fl ap(λ)=(a0(λ)+a1(λ) × ln(ap(440)) × ap(440) Lee et al. (1998) ected their consistent variations; with an increase in TSM by 100 mg/ λ −0.0123×(λ−443) an( ) = 0.041 × [NAP] × e Ruddick et al. (2006) L, POC increased by 1.12 mg/L (Fig. 3b). Therefore, we calculated POC [NAP] = [TSM] − 0.12 × [Chl-a] Oyama et al. (2009) −SCDOM×(λ−440) from satellite-derived TSM using Eq. (7). aCDOM(λ)=aCDOM(440) × e – −4.32 bbw = 0.00111 × (λ/550) Smith and Baker (1981) 0.62 [POC]=×+ 0.0112 [TSM] 0.1115 (7) bp(λ) = 0.3 × [Chl-a] × (550/λ) Gordon and Morel (1983) λ λ −0.15 bn( ) = 0.51 × [NAP] × ( /555) Ruddick et al. (2006) where, [TSM] is satellite-derived TSM by Eq. (6). In sum, we retrieved Bp and Bn = 0.033 Ma et al. (2008) POC from Landsat TM and ETM+ data through a two-step algorithm by Chl-a, aCDOM, SCDOM This study (Table 1) Eqs. (6) and (7).

157 D. Liu et al. Remote Sensing of Environment 223 (2019) 154–165

Fig. 2. Validations of the bio-optical simulation. (a) Comparisons between modeled and in-situ an. (b) Comparisons between modeled and in-situ water reﬂectance. For Band 1, Band 2, Band 3, and Band 4, we used central wavelengths of Landsat-5 TM. Landsat-7 ETM+ had similar visible light bands setups as Landsat-5 TM (Section 3.3).

4.3. Evaluation of the POC algorithm performance turbidity from Landsat data (Bernardo et al., 2017; Islam et al., 2002; Joshi et al., 2017b; Lymburner et al., 2016; Saliot et al., 2002; Wang Water reflectance variations at Datong were mainly determined by et al., 2007, 2009). For the Changjiang River, Wang et al. (2009) used TSM, especially the NAP (Section 4.1). NAP could not only decrease the single band 4 of Landsat ETM+. At short wavelength (band 1), NAP water reflectance by absorbing irradiance, but also increase reflectance strongly absorbed irradiance and had greatly impacts on decreasing by scattering irradiance (Bricaud and Stramski, 1990; Mueller et al., water reflectance. Therefore, ratio of Rrs(band4)/Rrs(band1) increased 2003). Absorption coefficient of NAP exponentially decreased with in- with increasing TSM concentration. To retrieve TSM in the CRE from creasing wavelength (Bricaud and Stramski, 1990). With negligible space, He et al. (2013) also used a similar index (Rrs(745 nm)/ absorption, water reflectance at long wavelength depended on scat- Rrs(490 nm)). tering by TSM. Therefore, red band (band 3) or near-infrared band By setting TSM = 30 and 120 mg/L, we calculated Rrs(band4)/ (band 4) with long wavelengths were often used to retrieve TSM or Rrs(band1) ratio with different aCDOM and Chl-a values. Results in Fig. 4

Fig. 3. Algorithms for retrieving POC and validations. (a) Comparisons between modeled TSM using in-situ water reﬂectance and in-situ TSM. (b) Correlation between in-situ POC and TSM at Datong. (c) and (d) are comparison results between in-situ and modeled POC calculated from in-situ water reﬂectance using spectral response functions of Landsat-5 TM and Landsat-7 ETM+, respectively.

158 D. Liu et al. Remote Sensing of Environment 223 (2019) 154–165

Fig. 4. Evaluation of the developed algorithm. (a) and (b) are sensitivity analysis of Rrs(band4)/Rrs(band1) under diﬀerent aCDOM and Chl-a conditions by setting TSM as 30 mg/L and 120 mg/L, respectively. Ranges of TSM, aCDOM(440) and Chl-a were set by referring to in-situ values (Table 1). Results in red circles were used in this study. (c) Comparisons between satellite-derived and YSI-measured POC. (For interpretation of the references to color in this ﬁgure legend, the reader is referred to the web version of this article.)

showed that variations of both aCDOM and Chl-a had not significant algorithms were applicable for retrieving POC in the Changjiang River impacts on Rrs(band4)/Rrs(band1). With TSM = 30 mg/L, aCDOM and from Landsat data. Chl-a could vary Rrs(band4)/Rrs(band1) by 6.85% (Fig. 4a); with TSM = 120 mg/L, aCDOM and Chl-a could vary Rrs(band4)/Rrs(band1) 5. Results by only 2.01% (Fig. 4b). In addition, with complex components and aerosol types, precise atmospheric correction of inland waters was still 5.1. Variations in POC concentration not achievable (Bernardo et al., 2017; He et al., 2013; Kutser, 2012). In this case, band ratio algorithm also allowed removal of atmospheric Fig. 6 shows the in-situ POC concentrations at different stations and correction errors to a certain extent (IOCCG, 2018; Kutser, 2012, Kutser depths from May 2015 to May 2016 (Section 3.1). For each station, POC et al., 2016). Therefore, Rrs(band4)/Rrs(band1) ratio was used ratio to at three depths were usually close, especially at the station P1 (Fig. 6). retrieve TSM. At the north station P1 (Fig. 1b), POC ranged from 0.47 to 1.34 mg/L; Using the developed algorithms, we calculated POC concentration mean POC at the surface, middle, and bottom depths were 0.74 mg/L, from in-situ water reflectance (Section 3.2). For Landsat-5 TM, relative 0.77 mg/L, and 0.77 mg/L, respectively (Fig. 6a). At the middle station error (RE) ranged from 0.12% to 32.62%, with ARE of 10.22% (Fig. 3c). P2 (Fig. 1b), POC ranged from 0.44 to 1.15 mg/L; mean POC at the For Landsat-7 ETM+, RE ranged from 0.70% to 31.58%, with ARE of three depths were 0.68 mg/L, 0.67 mg/L, and 0.71 mg/L, respectively 11.56% (Fig. 3d). Islam et al. (2002) and Wang et al. (2009) had pro- (Fig. 6b). At the south station P3, POC ranged from 0.37 to 0.99 mg/L; posed algorithms to retrieve TSM using single bands 3 and 4 of Landsat mean POC at the three depths were 0.6 mg/L, 0.67 mg/L, and 0.67 mg/ data, respectively. For comparison, we calibrated both algorithms using L, respectively (Fig. 6c). For a specific depth across the Datong section our in-situ data and estimated POC. Our band ratio algorithm was (Fig. 1b, c), POC in the north (P1) was slightly higher than that in the better; for Landsat-5 TM, their ARE were 16.23% and 14.32%, re- south (P3). The riverbed was covered by sediment in the north, but rock spectively (Fig. 3c); for Landsat-7 ETM+, their ARE were 16.23% and in the south, and thus sediment resuspension was more significant in 14.33%, respectively (Fig. 3d). the north. However, due to low flow velocity, sediment resuspension in The developed algorithm was also applied to Landsat-derived water the north was limited, especially in winter with low water flow at Da- reflectance to estimate POC concentration. During the sampling year tong (Fig. 6d). For all three stations across the Datong section (Fig. 1b), (Section 3.1), only eight scenes of Landsat-7 ETM+ data were available mean POC concentration was lowest in May 2016 and highest in April and cloud-free at Datong. By following the match-up criteria used by 2016 (Fig. 6d). Moreover, there was no significant relationship between NASA (Bailey and Werdell, 2006), only in-situ POC on May 12, 2015 in-situ POC and water flow (Fig. 6d). In general, POC concentration at had valid match-up. The satellite-derived POC had absolute error (AE) Datong was spatially consistent, but time-varying (Fig. 6). of 0.19 mg/L and RE of 42.22% (Fig. 4c). Using Eq. (7), YSI-measured With temporally varied POC at Datong, it is necessary and achiev- POC was calculated from YSI-measured TSM (Section 3.1). For the YSI- able to monitor POC concentration and flux from long-term archived measured POC, 21 match-ups were found, with ARE = 33.82% Landsat data. The developed algorithms (Section 4.2) were applied to (Fig. 4c). In addition, we found that monthly mean satellite-derived derive mean POC concentration in region R (Fig. 1b) from Landsat re- POC varied consistently with monthly mean TSM from the MWRC, with flectance data (Section 3.3). Results are shown in Fig. 5a. Landsat-de- Pearson's r = 0.41 and p < 0.01 (Fig. 5a). Overall, the developed rived POC concentration was significantly time-varying, with a

159 D. Liu et al. Remote Sensing of Environment 223 (2019) 154–165

Fig. 5. Time-series of satellite-derived POC. (a) Satellite-derived POC concentrations during 2000–2016. Water level of the TGD reached 135 m in June 2003, 156 m in October 2006, and 175 m in October 2010, respectively. (b) Seasonal mean POC concentration. The wet season contained months from May to October, and the dry season contained all other months in the same year. (c) Seasonal mean POC concentrations in different periods. Period I included years during 2000–2005; Period II included years during 2006–2010; and Period III included years during 2011–2016. minimum of 0.26 mg/L on 8 February 2016 and a maximum of 8.1 mg/ were R = 0.33 and p < 0.01; and from 2011 to 2016, the exponential L on 26 September 2001 (Fig. 5a). During 2000–2016, monthly mean fitting results were R = 0.37 and p < 0.01. POC concentration ranged from 0.83 to 2.41 mg/L, with minimum in The monthly decrease trend of POC concentration was mainly April and maximum in September. Moreover, POC concentration de- driven by its significant decrease in wet season. From satellite-derived creased exponentially with respect to time from 2000 to 2016, with POC, we calculated seasonal mean POC concentrations in the wet and R = 0.46 and p < 0.01 (Fig. 5a). Within each stage of TGD operation, dry seasons (Fig. 5b). Seasonal mean POC varied greatly in different it was also the same; from 2000 to 2010, the exponential fitting results years. Along with the TGD construction, seasonal variability in POC

Fig. 6. In-situ POC concentrations at Datong from May 2015 to May 2016. Sampling locations were shown as Fig. 1. Monthly water ﬂow at Datong was from the MWRC (2000–2016).

160 D. Liu et al. Remote Sensing of Environment 223 (2019) 154–165

Fig. 7. Time-series of monthly POC fluxes from 2000 to 2016. (a) Monthly satellite-derived POC fluxes. Due to unavailable Landsat data, POC fluxes in some months were missing. (b) Relationship between monthly POC flux and water flow. (c) Monthly climatologic POC fluxes. Water flow was sourced from the MWRC (2000–2016). concentration decreased from 2000 to 2016 (Fig. 5b). After 2011, mean 4.47 × 104 t C in February to a maximum of 17.61 × 104 t C in Sep- POC concentrations in both wet and dry seasons were very close tember, with a mean of 9.55 × 104 t C/month (Fig. 7c). Monthly cli- (Fig. 5b). According to the TGD construction phase (Fig. 5a), we in- matologic POC flux was also significantly related to water flow, with vestigated three periods: 2000–2005 (Period I), 2006–2010 (Period II), R2 = 0.91 and p < 0.01 (Fig. 7c). and 2011–2016 (Period III). Differences in seasonal POC concentrations were significant in Period I, but not in Period II and III (Fig. 5c). 6. Discussions

5.2. Variations in POC ﬂux 6.1. Relationship between POC and TSM

With spatially consistent POC concentration at Datong section At Datong, POC was linearly related to TSM concentration (Fig. 3b). (Fig. 6), Landsat-derived surface POC concentration in region R could In other form, POC content in TSM (POC(%TSM)) decreased by means represent the mean POC concentration across the entire Datong section. of an inverse proportional function with increasing TSM, with In each month, we took the arithmetic mean of all Landsat-derived POC R2 = 0.48 and p < 0.01 (Fig. 8). At Datong, POC(%TSM) was lower concentrations as the monthly mean value. By multiplying monthly than the fitted values of global rivers (Ludwig et al., 1996) and tropical mean POC concentration by water flow at Datong, we obtained the rivers (Huang et al., 2012). Moreover, even the fitted curve of the monthly POC flux, as shown in Fig. 7a. Changjiang mainstream (Zhang et al., 2014) was not suitable for Da- During 2000–2016, monthly POC flux ranged from a minimum of tong, especially for waters with high TSM concentrations (Fig. 8). 1.4 × 104 t in February 2016 to a maximum of 52.04 × 104 t in May The relationship between POC and TSM concentration was de- 2002, with a mean of 13.04 × 104 t/month. Moreover, POC flux in termined by POC compositions. Riverine POC is categorized as auto- 2009 was 1.35 × 106 t C/yr, only 12.94% lower than the reported value chthonous from phytoplankton and allochthonous from soil (Saliot using monthly in-situ data by Wang et al. (2012). Due to POC con- et al., 2002). Autochthonous POC is composed of phytoplankton and centration decrease in wet season (Fig. 5b), monthly POC flux also related particles with high POC(%TSM), and is an important component showed significant decrease from 2000 to 2016, with R = 0.33 and in rivers with relatively flat gradients (Gao et al., 2002). Allochthonous p < 0.01 (Fig. 7a). Monthly POC flux also seasonally varied, with high POC includes highly decomposed organic matter from soil, vegetation, values in the wet months, and was positively exponentially related to and weathered material with low POC(%TSM) (Gao et al., 2002; Wang water flow (R = 0.69, p < 0.01) (Fig. 7b). About 75.01% of POC et al., 2012). At Datong, in-situ TSM was 51.11 ± 22.13 mg/L transport happened in the wet season. (Table 1). High TSM concentration inhibited phytoplankton growth From monthly water flow and POC flux in Fig. 7a, we also calcu- because of light limitation (Ran et al., 2013). With great water dis- lated monthly climatologic POC fluxes (arithmetic means), as shown in charge (Bai et al., 2013, 2014), high flow velocity was also dis- Fig. 7c. Monthly climatologic POC flux ranged from a minimum of advantageous to phytoplankton growth (Gao et al., 2002). In these

161 D. Liu et al. Remote Sensing of Environment 223 (2019) 154–165

the two-step process allowed for a common approach to identify ap- propriate algorithm. By referring to others (Jiang et al., 2015; Stramski et al., 1999), therefore, a two-step POC algorithm based on the sig- niﬁcant correlation between POC and TSM was developed (Fig. 3b).

6.2. Impact factors on POC transport at Datong

The TGD had great impacts on POC decrease at Datong. Yang et al. (2007) reported that sediment flux across the Datong section was majorly sourced from the upstream of the TGD. Monthly TSM showed significant decrease during 2003–2016 at Yichang, only 40 km downstream of the TGD, with R = 0.51 and p < 0.01 (Fig. 9a). Moreover, there was also a significant linear correlation between monthly TSM concentration at Yichang and that at Datong, with R2 = 0.65 and p < 0.01. It has been reported that the TGD had little impact on water discharge (Bai et al., 2014; Duan et al., 2008), but drastically altered the sediment flux (Xu and Milliman, 2009; Yang et al., 2006). Xu and Milliman (2009) reported that the TGD trapped about 60% of upstream Fig. 8. The relationship between in-situ POC(%TSM) and TSM concentration. input sediment during 2003–2006. Across China, Chu et al. (2009) re- Ludwig et al. (1996) reported a global fitting curve based on data from 60 ported that dam/reservoir construction accounted for 56% of sediment world-wide rivers with relatively flat terrain and small precipitation gradients reduction in nine major Chinese rivers entering oceans during in Europe, America, and tropical Africa. For tropical rivers, Huang et al. (2012) 1959–2007. Globally, Maavara et al. (2017) reported that dam/re- reported that there was power relation between POC(%TSM) and TSM con- servoir buried allochthonous POC by only 9.83 ± 2.97 Tg C/yr in centration. For the Changjiang mainstream, Zhang et al. (2014) reported that 1970, but 20.91 ± 6.46 Tg C/yr in 2000; moreover, the value would be POC(%TSM) exponentially decreased with increasing TSM concentration. 40.17 ± 26.93 Tg C/yr in 2030. The TGD had also decreased the seasonal POC variations at Datong. cases, in-situ Chl-a was only 1.45 ± 0.56 μg/L (Table 1) and majorly After the TGD, TSM at Yichang was significantly decreased, especially POC was allochthonous at Datong. The depleted δ13C value (around in the wet season (Fig. 9a), which was time consistent with POC at −24.0%) also indicated the strong terrigenous POC source in the Datong (Fig. 5b). From 2003 to 2016, both the SD of monthly mean Changjiang River (Wang et al., 2012). As results, POC(%TSM) at Da- TSM concentration at Yichang and the SD of satellite-derived POC tong was low (Fig. 8). Moreover, a great amount of soil was flushed into concentration at Datong decreased significantly and were linearly cor- river in the wet season (Zhang et al., 2014) and POC(%TSM) decreased related (R2 = 0.62, p < 0.01) (Fig. 9b). All these indicated the weak- with increasing TSM due to dilution effects of inorganic mineral matter ening effects of the TGD on seasonal variability of POC concentration at (Fig. 8). Datong. Based on biweekly sampling in the CRE from 2003 to 2011, Wu POC composition had great impacts on POC algorithm. For open et al. (2016) also reported that the TGD had decreased the seasonal oceans, POC was majorly sourced from phytoplankton and POC algo- variations of terrigenous organic matter in the Changjiang River. For rithm usually used remote sensing reflectance at Chl-a sensitive blue or the Mekong River, Lu and Siew (2006) also reported that dams had green band (Stramski et al., 1999), or their ratios (Stramska, 2009; significantly decreased monthly TSM in several hydrological stations. Stramski et al., 2008). The same as the Pearl River Estuary (Liu et al., Climate fluctuation also had impacts on POC transport at Datong. 2015a), POC was majorly sourced from NAP at Datong and near-in- High in-situ POC in April 2016 (Fig. 6) was due to the large precipitation frared band sensitive to NAP was used (Section 4.2). With similar under a strong El Niño event in 2016 (Fig. 10). El Niño would increase source, moreover, relationship between POC and TSM might also be precipitation over the middle and lower reaches of the Changjiang different (Gao et al., 2002; Milliman et al., 1984). Along the Pearl River, River (Yu and Zhai, 2018). Because of strong El Niño events around POC and TSM showed a positive linear relation at the Makou hydro- 2003, 2010, and 2016, water flows were significant high in summer at logical station (Gao et al., 2002), but a power function at eight river Datong, but not at Yichang (Fig. 10). High precipitation would flush a outlets (Ni et al., 2008). Jiang et al. (2015) reported that it was unlikely great deal of terrigenous POC into rivers (Ran et al., 2013). During to develop a POC algorithm that was independent of POC sources, but 2000–2016, maximum monthly Ocean Niño Index (ONI) was

Fig. 9. Impacts of the TGD on POC transport at Datong. (a) Variations in monthly total water ﬂow and mean TSM concentration at Yichang during 2003–2016. Data before 2003 was unavailable. (b) SD of satellite-derived POC concentration at Datong and that of monthly mean TSM at Yichang in diﬀerent years.

162 D. Liu et al. Remote Sensing of Environment 223 (2019) 154–165

Fig. 10. Comparisons among ONI and water ﬂows at Yichang and Datong. ONI is used for indicating El Niño (with positive value) and La Niña (with nega- tive value) events in the tropical Paciﬁc. ONI of 0.5, 1.0, and 1.5 are threshold values for weak, moderate, and strong El Niño, respectively. ONI of −0.5, −1.0, and −1.5 are threshold values for weak, moderate, and strong La Niña, respectively. The ONI data sourced from NOAA (http://www.cpc.noaa.gov).

Fig. 11. Landsat-derived POC concentration in the mainstream from the Yichang to Datong in 2011. (a) Landsat-derived POC concentration. Light green filled rectangle denotes coverage area of Landsat data. Four scenes of Landsat-5 TM data were used. (b) Changes of Landsat-derived POC concentration in the mainstream. Due to cloud cover, POC concentrations in some river reaches were unavailable. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) significantly related to maximum monthly water discharge at Datong, (Fig. 5a). On the contrary, suspended sediment would also deposit to with R = 0.57 and p < 0.01. The same as TSM (Feng et al., 2014), bottom and decrease riverine POC. With low water flow velocity, POC variations of POC from the Changjiang River would further change POC was found to be low in the TGD region (Zhang et al., 2014). Based on in- in the CRE and its adjacent coastal waters. situ data, Zhang et al. (2014) reported that surface POC concentration in the Changjiang mainstream (from the Fuling city to the CRE) ranged 6.3. POC concentration along the Changjiang mainstream from 0.2 to 2.4 mg/L in April 2006, but from 0.2 to 1.3 mg/L in April 2008. In these cases, satellite remote sensing could improve our un- POC concentration spatially varied along the Changjiang main- derstanding about spatial variations of POC concentration along the stream. Flow velocity at Datong was usually high in the wet season Changjiang mainstream. (Duan et al., 2008). With high flow velocity, bottom sediment would be Fig. 11 shows the Landsat-derived POC concentration in the re-suspended so as to increase POC concentration. This partly explained Changjiang mainstream from the Yichang to Datong, covering a length the usually high POC concentration at Datong in the wet season of ~1000 km. With large slope gradient, riverbed flushing was

163 D. Liu et al. Remote Sensing of Environment 223 (2019) 154–165 significant from Yichang to Hankou (MWRC, 2000–2016). As results, Province (BK20181102). POC concentration showed an increase along the short river reach downstream of Yichang and then tended to be stable on 30 May 2011, References but increased gradually after Chenglingji on 8 June 2011 (Fig. 11). Besides different topography, this might due to low water flow of Bai, Y., Pan, D., Cai, W.-J., He, X., Wang, D., Tao, B., Zhu, Q., 2013. Remote sensing of 2.43 × 1010 m3/month at Yichang in May 2011, but 4.19 × 1010 m3/ salinity from satellite-derived CDOM in the Changjiang River dominated East China – – Sea. J. Geophys. Res. Oceans 118, 227 243. month in June 2011 (MWRC, 2000 2016). In June 2011, about Bai, Y., He, X., Pan, D., Chen, C.T.A., Kang, Y., Chen, X., Cai, W.J., 2014. Summertime 8 3 8.05 × 10 m /day of water was discharged into the Changjiang Changjiang River plume variation during 1998-2010. J. Geophys. Res. Oceans 119, mainstream from the Dongting Lake (MWRC, 2000–2016). This con- 6238–6257. Bai, Y., Cai, W.J., He, X., Zhai, W., Pan, D., Dai, M., Yu, P., 2015. A mechanistic siderable water flow increased riverbed erosion and continually in- semi#Y7SL051001(Grant #201505003), National Natural SciepCO2 in river (2015). creased the POC concentration from Chenglingji to Hankou (Fig. 11). J. Geophys. Res. Oceans 120, 2331–2349. Xu and Milliman (2009) also reported that riverbed erosion mainly took Bailey, S.W., Werdell, P.J., 2006. A multi-sensor approach for the on-orbit validation of – place in mainstream from Yichang to Hankou. In July 2011, water flow ocean color satellite data products. Remote Sens. Environ. 102, 12 23. Bernardo, N., Watanabe, F., Rodrigues, T., Alcântara, E., 2017. Atmospheric correction at Yichang was higher than that in June 2011 (MWRC, 2000–2016), issues for retrieving total suspended matter concentrations in inland waters using resulting in the high POC concentration downstream of the Hankou OLI/Landsat-8 image. Adv. Space Res. 59, 2335–2348. “ ” (Fig. 11). On 28 July 2011, POC concentration showed no obvious Bianchi, T.S., Allison, M.A., 2009. Large-river delta-front estuaries as natural records of global environmental change. Proc. Natl. Acad. Sci. U. S. A. 106, 8085–8092. changes from Hukou to Datong (Fig. 11), likely due to the diminished Bricaud, A., Stramski, D., 1990. Spectral absorption coefficients of living phytoplankton riverbed erosion (Xu and Milliman, 2009). Moreover, with small slope and nonalgal biogenous matter: a comparison between the Peru upwelling area and – gradient and wide watercourse, suspended POC might even deposit to the Sargasso Sea. Limnol. Oceanogr. 35, 562 582. Chu, Z.X., Zhai, S.K., Lu, X.X., Liu, J.P., Xu, J.X., Xu, K.H., 2009. A quantitative assess- bottom. In summary, POC concentration in the Changjiang mainstream ment of human impacts on decrease in sediment flux from major Chinese rivers en- is dependent upon the combined effects of water flow, slope gradient tering the western Pacific Ocean. Geophys. Res. Lett. 36. and watercourse width. Duan, S., Liang, T., Zhang, S., Wang, L., Zhang, X., Chen, X., 2008. Seasonal changes in nitrogen and phosphorus transport in the lower Changjiang River before the construction of the Three Gorges Dam. Estuar. Coast. Shelf Sci. 79, 239–250. 7. Conclusions and outlook ENVI User's Guide, 2008. ENVI On-line Software User's Manual. ITT Visual Information Solutions, Boulder, CO, USA. fl Feng, L., Hu, C., Chen, X., Song, Q., 2014. Influence of the Three Gorges Dam on total At Datong, water re ectance was primarily determined by TSM suspended matters in the Yangtze Estuary and its adjacent coastal waters: observa- concentration. Due to limited in-situ data, we simulated water re- tions from MODIS. Remote Sens. Environ. 140, 779–788. flectance based on a bio-optical model and developed a two-step POC Gao, Q., Tao, Z., Shen, C., Sun, Y., Yi, W., Xing, C., 2002. Riverine organic carbon in the fl remote sensing algorithm using the reflectance ratio of near-infrared to Xijiang River (South China): seasonal variation in content and ux budget. Environ. Geol. 41, 826–832. blue band (requi(band4)/requi(band1)) for Landsat data. From 2000 to Gordon, H.R., Morel, A.Y., 1983. Remote Assessment of Ocean Color for Interpretation of 2016, both POC concentration and flux at Datong were significantly Satellite Visible Imagery. Springer US. seasonally varied, with high values in the wet seasons. Under the in- Gordon, H.R., Brown, O.B., Evans, R.H., Brown, J.W., Smith, R.C., Baker, K.S., Clark, D.K., fl fl fi 1988. A semianalytic radiance model of ocean color. J. Geophys. Res. Atmos. 93, uences of the TGD, both POC concentration and ux signi cantly 10909–10924. decreased by following exponential functions. After regular TGD run- Griffin, C.G., Frey, K.E., Rogan, J., Holmes, R.M., 2011. Spatial and interannual varia- ning, satellite-derived POC transport showed relative stability during bility of dissolved organic matter in the Kolyma River, East Siberia, observed using – ff satellite imagery. J. Geophys. Res. 116. 2011 2016 and seasonal di erence of POC transport became smaller. Hakvoort, H., Haan, J.d., Jordans, R., Vos, R., Peters, S., Rijkeboer, M., 2002. Towards From the TGD to Datong, the POC concentration in the Changjiang airborne remote sensing of water quality in the Netherlands-validation and error mainstream was dependent upon the synthesis effects of water flow, analysis. ISPRS J. Photogramm. Remote Sens. 57, 171–183. He, X., Bai, Y., Pan, D., Huang, N., Dong, X., Chen, J., Chen, C.T.A., Cui, Q., 2013. Using slope gradient, and watercourse width. geostationary satellite ocean color data to map the diurnal dynamics of suspended A series of Landsat satellites had been launched since 1972, which particulate matter in coastal waters. Remote Sens. Environ. 133, 225–239. could be applied to monitor variations in POC transport from the Huang, T.H., Fu, Y.H., Pan, P.Y., Chen, C.T.A., 2012. Fluvial carbon fluxes in tropical rivers. Curr. Opin. Environ. Sustain. 4, 162–169. Changjiang River. In addition to the long-term archived Landsat data, IOCCG, 2018. Earth observations in support of global water quality monitoring. In: Greb, charge coupled device onboard the Chinese environmental and disaster S., Dekker, A., Binding, C. (Eds.), IOCCG Report Series, No. 17, International Ocean monitoring satellites (HJ-CCD) also has recorded Earth's radiance at Colour Coordinating Group, Dartmouth, Canada. near-infrared and blue bands with a spatial resolution of 30 m and Islam, M.R., Yamaguchi, Y., Ogawa, K., 2001. Suspended sediment in the Ganges and Brahmaputra Rivers in Bangladesh: observation from TM and AVHRR data. Hydrol. temporal resolution of 1 day since 2008. The HJ-CCD data could be Process. 15, 493–509. applied to monitor daily variations of POC concentration in the Islam, M.R., Begum, S.F., Yamaguchi, Y., Ogawa, K., 2002. Distribution of suspended ff Changjiang River. sediment in the coastal sea o the Ganges-Brahmaputra River mouth: observation from TM data. J. Mar. Syst. 32, 307–321. Jiang, G., Ma, R., Loiselle, S.A., Duan, H., Su, W., Cai, W., Huang, C., Yang, J., Yu, W., Acknowledgement 2015. Remote sensing of particulate organic carbon dynamics in a eutrophic lake (Taihu Lake, China). Sci. Total Environ. 532, 245–254. Joshi, I.D., D'Sa, E.J., Osburn, C.L., Bianchi, T.S., Ko, D.S., Oviedo-Vargas, D., Arellano, We thank the Datong hydrological station and MWRC for providing A.R., Ward, N.D., 2017a. Assessing chromophoric dissolved organic matter (CDOM) the survey assistance and long-term water and sediment transport data. distribution, stocks, and fluxes in Apalachicola Bay using combined field, VIIRS ocean This study was supported by the National Key Research and color, and model observations. Remote Sens. Environ. 191, 359–372. Joshi, I.D., D'Sa, E.J., Osburn, C.L., Bianchi, T.S., 2017b. Turbidity in Apalachicola Bay, Development Program of China (Grant #2017YFA0603003), National Florida from Landsat 5 TM and field data: seasonal patterns and response to extreme Basic Research Program of China (973 Program) (Grant events. Remote Sens. 9, 367. #2015CB954002), Public Science and Technology Research Funds Knap, A.H., Michaels, A., Close, A.R., Ducklow, H., Dickson, A.G., 1994. Protocols for the Joint Global Ocean Flux Study (JGOFS) Core Measurements; IOC Manuals and Guides Projects for Ocean Research (Grant #201505003), National Natural No. 29. United National. Educational, Scientific and Cultural Organization Science Foundation of China (Grants #41825014, #41825014, (UNESCO), Paris, France. #41676172, #41676170, and #41621064), Global Change and Air-Sea Kou, L., Labrie, D., Petr, C., 1993. Refractive indices of water and ice in the 0.65- to 2.5- – Interaction Project of China (GASI-02-SCS-YGST2-01, GASI-02-PAC- um spectral range. Appl. Opt. 32, 3531 3540. Kutser, T., 2012. The possibility of using the Landsat image archive for monitoring long YGST2-01, and GASI-02-IND-YGST2-01), Project of State Key time trends in coloured dissolved organic matter concentration in lake waters. Laboratory of Satellite Ocean Environment Dynamics (Grant Remote Sens. Environ. 123, 334–338. #SOEDZZ1801), the Research Startup Project of Nanjing Institute of Kutser, T., herlevi, A., Kallio, K., Arst, H., 2001. A hyperspectral model for interpretation of passive optical remote sensing data from turbid lakes. Sci. Total Environ. 268, Geography and Limnology, Chinese Academy of Sciences (Grant 47–58. #Y7SL051001), and the Natural Science Foundation of Jiangsu Kutser, T., Pierson, D.C., Kallio, K.Y., Reinart, A., Sobek, S., 2005. Mapping lake CDOM by

164 D. Liu et al. Remote Sensing of Environment 223 (2019) 154–165

satellite remote sensing. Remote Sens. Environ. 94, 535–540. overview of global nutrient export from watersheds (NEWS) models and their ap- Kutser, T., Casal Pascual, G., Barbosa, C., Paavel, B., Ferreira, R., Carvalho, L., Toming, K., plication. Glob. Biogeochem. Cycles 19. 2016. Mapping inland water carbon content with Landsat 8 data. Int. J. Remote Sens. Siegenthaler, U., Sarmiento, J.L., 1993. Atmospheric carbon dioxide and the ocean. 37, 2950–2961. Nature 365, 119–125. Lee, Z., Carder, K.L., Mobley, C.D., Steward, R.G., Patch, J.S., 1998. Hyperspectral remote Singh, R., Prasad, P.R.C., 2014. Interpolation of data gaps of SLC-off Landsat ETM+ sensing for shallow waters. I. A semianalytical model. Appl. Opt. 37, 6329–6338. images using algorithm based on the differential operators. J. Appl. Comput. Sci. Liu, Q., Pan, D., Bai, Y., Wu, K., Chen, C.T.A., Liu, Z., Zhang, L., 2014. Estimating dis- Methods 6, 93–100. solved organic carbon inventories in the East China Sea using remoteand Yellow Smith, R.C., Baker, K.S., 1981. Optical properties of the clearest natural waters River. J. Geophys. Res. Oceans 119, 6557–6574. (200–800 nm). Appl. Opt. 20, 177–184. Liu, D., Pan, D., Bai, Y., He, X., Wang, D., Wei, J.A., Zhang, L., 2015a. Remote sensing Stramska, M., 2009. Particulate organic carbon in the global ocean derived from SeaWiFS observation of particulate organic carbon in the Pearl River Estuary. Remote Sens. 7, ocean color. Deep-Sea Res. I Oceanogr. Res. Pap. 56, 1459–1470. 8683–8704. Stramski, D., Reynolds, R.A., Kahru, M., Mitchell, B.G., 1999. Estimation of particulate Liu, D., Pan, D., Bai, Y., He, X., Wang, D., Zhang, L., 2015b. Variation of dissolved organic organic carbon in the ocean from satellite remote sensing. Science 285, 239. carbon transported by two Chinese rivers: the Changjiang River and Yellow River. Stramski, D., Reynolds, R.A., Bain, M., Kaczmarek, S., Lewis, M.R.R.R., Sciandra, A., Mar. Pollut. Bull. 100, 60–69. Stramska, M., Twardowski, M.S., Franz, B.A., Claustre, H., 2008. Relationships be- Lu, X.X., Siew, R.Y., 2006. Water discharge and sediment flux changes over the past tween the surface concentration of particulate organic carbon and optical properties decades in the Lower Mekong River: possible impacts of the Chinese dams. Hydrol. in the eastern South Pacific and eastern Atlantic Oceans. Biogeosciences 5, 171–201. Earth Syst. Sci. 10 (2 (2006-03-21)), 181–195 (10). Strickland, J.D.H., Parsons, T.R., 1972. A practical handbook of seawater analysis. Ludwig, W., Probst, J.L., Kempe, S., 1996. Predicting the oceanic input of organic carbon Bulletin 167. by continental erosion. Glob. Biogeochem. Cycles 10, 23–41. Wang, J., Lu, X.X., Zhou, Y., 2007. Retrieval of suspended sediment concentrations in the Lymburner, L., Botha, E., Hestir, E., Anstee, J., Sagar, S., Dekker, A., Malthus, T., 2016. turbid water of the Upper Yangtze River using Landsat ETM+. Chin. Sci. Bull. 52, Landsat 8: providing continuity and increased precision for measuring multi-decadal 273–280. time series of total suspended matter. Remote Sens. Environ. 185, 108–118. Wang, J.J., Lu, X.X., Soochin, L., Yue, Z., 2009. Retrieval of suspended sediment con- Ma, R., Song, Q., Li, G., Pan, D., 2008. Estimation of backscattering probability of Lake centrations in large turbid rivers using Landsat ETM+: an example from the Yangtze Taihu waters. J. Lake Sci. 20, 375–379. River, China. Earth Surf. Process. Landf. 34, 1082–1092. Maavara, T., Lauerwald, R., Regnier, P., Van Cappellen, P., 2017. Global perturbation of Wang, X., Ma, H., Li, R., Song, Z., Wu, J., 2012. Seasonal fluxes and source variation of organic carbon cycling by river damming. Nat. Commun. 8, 15347. organic carbon transported by two major Chinese Rivers: the Yellow River and Milliman, J.D., Xie, Q., Yang, Z., 1984. Transfer of particulate organic carbon and ni- Changjiang (Yangtze) River. Glob. Biogeochem. Cycles 26. trogen from the Yangtze River to the ocean. Am. J. Sci. 284, 824–834. Wollast, R., 1991. The coastal organic carbon cycle: Fluxes, sources and sinks. In: Ocean Ministry of Water Resources of China, 2000–2016. Sediment Communique of Chinese Margin Processes in Global Change, Dahlem Workshop Reports. Rivers. (In. Beijing). Wu, Y., Zhang, J., Liu, S.M., Zhang, Z.F., Yao, Q.Z., Hong, G.H., Cooper, L., 2007. Sources Mueller, J.L., Fargion, G.S., McClain, C.R., 2003. Ocean optics protocols for satellite and distribution of carbon within the Yangtze River system. Estuar. Coast. Shelf Sci. ocean colour sensor. In: In Special Topics in Ocean Optics Protocols and Appendices. 71, 13–25. Goddard Space Flight Center, Greenbelt, MD, USA. Wu, Y., Bao, H., Yu, H., Zhang, J., Kattner, G., 2016. Temporal variability of particulate Ni, H.G., Lu, F.H., Luo, X.L., Tian, H.Y., Zeng, E.Y., 2008. Riverine inputs of total organic organic carbon in the lower Changjiang (Yangtze River) in the post-Three Gorges carbon and suspended particulate matter from the Pearl River Delta to the coastal Dam period: links to anthropogenic and climate impacts. J. Geophys. Res. Biogeosci. ocean off South China. Mar. Pollut. Bull. 56, 1150–1157. 120, 2194–2211. Oyama, Y., Matsushita, B., Fukushima, T., Matsushige, K., Imai, A., 2009. Application of Xu, K., Milliman, J.D., 2009. Seasonal variations of sediment discharge from the Yangtze spectral decomposition algorithm for mapping water quality in a turbid lake (Lake River before and after impoundment of the Three Gorges Dam. Geomorphology 104, Kasumigaura, Japan) from Landsat TM data. ISPRS J. Photogramm. Remote Sens. 64, 276–283. 73–85. Yang, Z., Wang, H., Saito, Y., Milliman, J.D., Xu, K., Qiao, S., Shi, G., 2006. Dam impacts Pope, R.M., Fry, E.S., 1997. Absorption spectrum (380–700 nm) of pure water. II. on the Changjiang (Yangtze) River sediment discharge to the sea: the past 55 years Integrating cavity measurements. Appl. Opt. 36, 8710–8723. and after the Three Gorges Dam. Water Resour. Res. 42, W04407. Ran, L., Lu, X.X., Sun, H., Han, J., Li, R., Zhang, J., 2013. Spatial and seasonal variability Yang, S.L., Zhang, J., Xu, X.J., 2007. Influence of the Three Gorges Dam on downstream of organic carbon transport in the Yellow River, China. J. Hydrol. 498, 76–88. delivery of sediment and its environmental implications, Yangtze River. Geophys. Ruddick, K.G., Cauwer, V.D., Park, Y.-J., Moore, G., 2006. Seaborne measurements of Res. Lett. 34. near infrared wateric carbon transpance: the similarity spectrum for turbid waters. Yu, R., Zhai, P., 2018. The influence of El Niño on summer persistent precipitation Limnol. Oceanogr. 51, 1167–1179. structure in the middle and lower reaches of the Yangtze River and its possible me- Saliot, A., Parrish, C.C., Sadouni, N.M., Bouloubassi, I., Fillaux, J., Cauwet, G., 2002. chanism (in Chinese). Acta. Meteor. Sin. 76, 408–419. Transport and fate of Danube Delta terrestrial organic matter in the Northwest Black Zhang, L., Xue, M., Wang, M., Cai, W.J., Wang, L., Yu, Z., 2014. The spatiotemporal Sea mixing zone. Mar. Chem. 79, 243–259. distribution of dissolved inorganic and organic carbon in the main stem of the Seitzinger, S.P., Harrison, J.A., Dumont, E., Beusen, A.H.W., Bouwman, A.F., 2005. Changjiang (Yangtze) River and the effect of the Three Gorges Reservoir. J. Geophys. Sources and delivery of carbon, nitrogen, and phosphorus to the coastal zone: an Res. Biogeosci. 119, 741–757.

165