Using CDOM optical properties for estimating DOC concentrations and pCO 2 in the Lower Amazon River

: Coloured dissolved organic matter (CDOM) is one of the major contributors to the absorption budget of most freshwaters and can be used as a proxy to assess non-optical carbon fractions such as dissolved organic carbon (DOC) and the partial pressure of carbon dioxide ( p CO 2 ). Nevertheless, riverine studies that explore the former relationships are still relatively scarce, especially within tropical regions. Here we document the spatial-seasonal variability of CDOM, DOC and p CO 2 , and assess the potential of CDOM absorption coefficient (a CDOM (412)) for estimating DOC concentration and p CO 2 along the Lower Amazon River. Our results revealed


Introduction
Terrestrial humic substances are the dominant contributor to the dissolved organic matter pool (DOM) in freshwaters [1]. Dissolved organic carbon (DOC) is the major fraction of DOM active in the global carbon cycle, with fluvial export of DOC providing the largest flux of reduced carbon (0.25 Pg C y −1 ) from land to ocean [2]. The fraction of DOM that absorbs ultraviolet (UV) and visible light, coloured dissolved organic matter (CDOM), is one of the major contributors to the absorption budget of most freshwaters. Freshwater systems also contain large amounts of carbon dioxide (CO 2 ) relative to the atmosphere [3]. In small streams most CO 2 is derived from soils and groundwater inputs [4], whereas in large riverine systems the input and in situ degradation of DOM by both microbes [5] and UV light [6] is a major source of CO 2 , which is subsequently outgassed to the atmosphere. The importance of freshwater systems on controlling the export and transformation of terrestrial DOM and the intimate relationship between these dynamics and the partial pressure of carbon dioxide (pCO 2 ) is well recognized [3, [7][8][9]. However, we lack a detailed understanding of how these dynamics vary in space and time, particularly in the tropics, which contribute disproportionately to global CO 2 emissions and DOM export from inland waters [9,10]. The rivers in the Amazon basin, for example, are estimated to outgas from 0.5 to 1.39 Gt C yr −1 [7,11]. The dynamics of CDOM, DOC and pCO 2 in the Amazon River are largely influenced by the seasonal changes in the river discharge [8,9,[12][13][14][15]. Seasonal variation in the rainfall intensity over the Amazon basin can also induce strong modulation in the quality (lability) of the DOC delivered to the river [4,16,17]. Several recent studies have emphasized trends towards the occurrence of extreme climatic events in some regions of the Amazon basin, illustrating the potential impact of such events on rain and river discharge patterns in the area [18][19][20][21][22].
Sawakuchi et al. [7] estimated that on the order of 0.48 Pg C year −1 of CO 2 could be outgassed just from the lower Amazon River, if the entire spatial extent to the geographical mouth is considered, fueled primarily by the metabolism of DOM [5,23]. But the extreme heterogeneity and vast scales of the region make it difficult to interpolate over space and time, suggesting that suitable means of remote sensing could provide much-needed insight into the dynamics of this complex region. The objective of this work is to investigate the potential use of CDOM absorption for estimating DOC concentration and pCO 2 in the Lower Amazon River along a 900 km transect from the historic downstream gauging station, Óbidos, to the river mouth with samples collected in contrasting water types (clear and turbid waters) under different discharge conditions. This study provides the first bio-optical approach for assessing the carbon content at the Lower Amazon, a region recently pointed out as one of the most active CO 2 emission areas among inland waters [7].
Numerous studies have aimed to relate CDOM absorption properties (a CDOM ) to DOC content in estuarine and coastal waters [24][25][26][27][28][29][30][31] to use the latter optical proxy assessing DOC distribution through in situ or satellite observations. The latter studies have clearly emphasized the presence of highly significant linear relationships between a CDOM and DOC in coastal waters dominated by terrestrial inputs of DOM, further illustrating the strong seasonal and regional dependency in the link between CDOM and organic carbon content. While CDOM absorption at a defined wavelength is providing quantitative information on DOM concentration, the spectral slope of the CDOM absorption spectra in the UV domain [275-295 nm] (S 275-295 ) has been shown to provide relevant insights into DOM composition and origin [26,32]. The potential of S 275-295 to be used as a relevant optical proxy of the DOC-normalized absorption coefficient of CDOM have been further demonstrated emphasizing the interest of this descriptor for better constraining the natural variability in the CDOM to DOC relationship at both seasonal [33] and regional [25,34] scales. High S 275-295 values are usually related to highly (biologically or photochemically) degraded DOM [26,32,35]. Helms et al.
[32] have further documented the interest of the ratio of two narrow spectral slopes (S R ; S 275-295 /S 350-400 ) for assessing the molecular weight of the DOM within CDOM-rich waters.
Various potential descriptors of pCO 2 variability in inland ecosystems have been recently documented. A strong linear dependency between pCO 2 and DOC content has been for instance emphasized within diverse boreal and temperate inland waters [36][37][38]. The presence of such relationship in tropical regions was conversely not confirmed, including some Brazilian lakes of the Amazon basin [39,40]. CDOM, more likely adapted to the development of ocean colour remote sensing based pCO 2 inversion algorithms, has been also considered as a potential proxy of pCO 2 [41-43]. However, the potential of CDOM for estimating pCO 2 in the Amazon waters still need to be evaluated.

Study area and sampling strategy
Six field campaigns were conducted to cover all hydrological seasons within the Lower Amazon River along a 900 km transect between the upstream boundary at Óbidos (01°55.141′ S, 55°31.543' W) and the Amazon River mouth (Fig. 1). The downstream boundary was the north and south channels near Macapá, which are the last two wellconstrained channels near the Amazon River mouth (00°05.400′ S, 51°03.200′ W and 00°09.415′ S, 50°37.353 W, respectively). The region near the mouth is tidally influenced and experiences semi-diurnal flow reversals [8] but no saline intrusion is observed (Salinity 0) [44]. In addition to the mainstream Amazon River (Óbidos, Almeirim, Macapá stations, Fig.  1), samples were collected in all large tributaries within this section, including the Tapajos, Xingu, Paru, and Jari rivers, as well as the Lago Grande de Curuai floodplain lake. The lowland tributaries are classified as clearwater (CW) [45], with low turbidity and high levels of primary production when compared with the turbid water of the Amazon River and Lago Grande de Curuai [8]. CW rivers were sampled in May-14, Nov-14, Jul-15, Feb-16. During our two final cruises in low water (Nov-16) and high water (Apr-17) we pushed beyond our normal downstream boundary near Macapá to sample closer to the actual river mouth (Table 1, Fig. 1).
In practice, surface water (< 1m) was sampled in the frame of this study for assessing the concentration of the different variables of interest: CDOM, DOC, pCO 2 and chl-a. pCO 2 , chla and temperature data are not available for T6.
The discharge of the Amazon River, which is a parameter translating the seasonal variability in the intensity of the inputs to the river from land-floodplain organic matter flush, was obtained from the National Water Agency of Brazil (ANA). The different in situ measurements performed in the frame of this study have allowed a description of the surface waters of the Lower Amazon River considering all possible discharge conditions over a 3year period. Here we considered the first three months of the year representing the rising water season (Jan, Feb and Mar), followed by High (Apr, May, Jun), Falling (Jul, Aug, Sep) and Low (Oct, Nov, Dec) water seasons.

CDOM absorption
Surface water samples (N = 80) were first pre-filtered through 25 mm Whatman GF/F glass fiber filters (0.7 µm nominal pore size) to remove the larger part of the suspended matter. A second filtration was then performed on 0.2 µm polycarbonate membranes (Whatman nucleopore, 25 mm) under gentle vacuum (< 5 mm Hg) according to the NASA protocol for inherent optical properties [46]. Samples were stored in pre-combusted glass bottle (450°C, 6 hours) wrapped with aluminum foil and kept under refrigeration (4°C) until laboratory analysis. CDOM samples were put to room temperature before spectrophotometric analysis to avoid any bias due the thermal difference between the samples and the reference water (Milli-Q water). CDOM absorbance spectra were measured from 250 to 850 nm, using a Shimadzu, UV 2450 spectrophotometer with a 10-cm quartz cell. The CDOM absorption coefficient (a CDOM (λ)) was calculated from absorbance measurements as followed (Eq. (1): where A(λ) is the absorbance of the filtered water sample at the specific wavelength λ and L is the optical pathway of the quartz cell in meters (here 0.1 m). As recommended by Babin et al. [47] in coastal and estuarine waters, a baseline correction was performed to each spectrum by subtracting the mean absorbance in the range of 680-690 nm from the whole spectrum. The absorption spectral shape of CDOM is estimated using a linear least-squares regression of the logarithm of a CDOM (λ) and reported with units of nm −1 [48] (Eq. (2): where a CDOM (λ) is the absorption coefficient at wavelength λ, a CDOM (λ 0 ) is the absorption coefficient at a reference wavelength λ 0 and S is the spectral slope in the spectral range from λ 0 to λ with λ 0 < λ.
Recent works have demonstrated the importance of the slope in the range of 275-295 nm (S 275-295 ) for better constraining the natural variability in the DOC specific absorption coefficient (a* CDOM ) [25,26,32,33]. Furthermore, Helms et al. [32] highlighted the importance of the slope ratio S R = S 275-295 /S 350-400 as a potential proxy for assessing DOM molecular weight. Consequently, S 275-295 , S 350-400 and the slope ratio S R were specifically computed in this study.

pCO 2 and DOC
Water was collected for analysis using a Shurflo submersible pump with a 297 µm mesh at the surface. For the determination of pCO 2 the surface water samples were collected in triplicate (1 L) and stored in polycarbonate bottles closed with a silicone stopper equipped with two stop-cocks and short/long straws to allow creation of a headspace [49]. The bottles were filled with sample water, leaving no headspace. After injecting 60 mL of synthetic gas and removing 60 mL of water, the bottle was sealed and shaken vigorously for 2 min. The headspace was then removed with a 60-mL syringe and directly injected into a Picarro G2201-i Cavity Ring-Down Spectrometer (CRDS). pCO 2 values were corrected based on the common gas law.
DOC samples were collected in triplicate, and filtered through pre-combusted (500°C, 5 hours) GF/F glass fiber filters 0.7 µm nominal pore size (Whatman) and stored in 25 mL precombusted glass vials washed with acid, closed with Teflon lids and preserved in the field with 25 µL of 50% HCl at 0-4°C. DOC concentration (µmol L −1 ) was measured using a Shimadzu total carbon analyzer (Model TOCVCPH). Only samples with a coefficient of variation lower than 10% among the triplicates were considered in this study. Measurements of DOC and a CDOM (412) were used to calculate the DOC specific coefficient absorption (a* CDOM (412) = a CDOM (412)/DOC), expressed here in units of m 2 mmol −1 .

Chlorophyll-a concentrations and ancillary parameters
The chlorophyll-a concentration (chl-a) was measured on a Turner Designs Model 10 AU Fluorometer using triplicate samples filtered in the dark on 25 mm Whatman GF/F glass fiber filters (0.7 μm nominal pore size). Filters were wrapped in aluminum foil and frozen at −20°C in the field and subsequently frozen in liquid nitrogen. Prior to fluorometric analysis, pigments were extracted from the filters using a mixture of dimenthylsulfoxide: 90% acetone (2:3 ratio) for 24 h at 5°C and in the dark.
For each sampling station, the temperature of the surface water was recorded (Thermo Scientific Orion 4-Star instrument).

Statistics
The accuracy of [DOC], a* CDOM (412), S 275-295 and pCO 2 estimates were evaluated using various statistical indicators including the root mean square error (RMSE), the mean relative absolute difference (MRAD) and the mean relative difference (Bias) expressed respectively as: where x i is the in situ data for a define parameter and y i its estimated value.

Optical and biogeochemical variability of the Lower Amazon River and tributary waters
Over the studied period, the average values for chl-a (1.3 ± 0.4 mg L −1 ), DOC (306 ± 27 µmol L −1 ), pCO 2 (2777 ± 1719 µatm) and temperature (29.8 ± 0.8 °C) recorded over the Lower Amazon River were in the range of observations previously reported in the region [7,8,14,39]. The average a CDOM (412) (m −1 ) for the Amazon River during these cruises (3.6 ± 1.0 m −1 ) was similar to the a CDOM (400) value (2.97 m −1 ) reported for the Curuai floodplain [50] ( Table 2). Amazon River samples had slightly lower temperature and chl-a values than the CW tributaries, with their higher primary production rates and lower turbidity [51] ( Table 2). All parameters other than temperature generally presented a higher seasonal variability for the CW stations compared to the mainstream. Average a CDOM (412) values were similar between Amazon River and CW samples (3.6 vs 3.3 m −1 respectively). However, higher DOC and pCO 2 were observed in the Amazon River waters when compared to the other rivers. On the other hand, CW rivers generally had higher values of S 275-295 and S R (0.015 nm −1 and 0.91, respectively, p<0.01) indicating differences in DOM quality between these different waters of the Amazon basin. The highest values for S 275-295 and S R were found in CW tributaries when compared to those collected within the Amazon River (0.014 and 0.87 nm −1 , respectively), indicating the presence of lower molecular weight DOM that could be attributed to the higher abundance of algae [32]. These results are similar to molecular level differences in DOM composition previously observed in the Tapajós and Amazon rivers [52].
All parameters other than temperature generally presented a higher temporal variability for the CW stations, emphasizing the seasonal modulation in the intensity of the water mixing of the clear rivers and Amazon River according to the discharge season (Table 2).
Two different seasonal patterns regarding the DOM dynamics could be identified in the Amazon River. First, average CDOM and DOC values for samples corresponding to T1, T2, T3 and T5 cruises are following the discharge patterns with a clear co-variation between CDOM and DOC levels. The highest CDOM and DOC average values (Table 3, 4.6 and 4.7 m −1 , 297 and 337 µmol L −1 for CW and Amazon River, respectively) were found during high river discharges conditions (T1), while the lowest ones (Table 3, 2.8 m −1 and 287 µmol L −1 for the Amazon River) are occurring during low discharge conditions (T2 and T5), especially during T5 due to a severe drought in 2016 related to an El Niño event [20]. Intermediary values (Table 3, 2.6 and 4.2 m −1 , 171 and 313 µmol L −1 for CW and Amazon River, respectively) are found during transition period (falling conditions, T3). The latter pattern tends to suggest that the DOM concentration of the Lower Amazon region is mainly driven by the intensity of the river discharge and therefore the inputs from land-floodplains. Specific situations departing from such general seasonal modulations were, however, observed during T4 and T6. For these two cruises a CDOM (412) values remained relatively low (3.6 and 3.3 m −1 , respectively), when compared to the corresponding DOC contents (381 and 420 µmol L −1 ), leading for these two cruises to the lowest a* CDOM (412) average values (0.009 and 0.008 m 2 .mmol −1 , for T4 and T6, respectively Table 3). The patterns observed for T4 and T6 tend to indicate that the terrestrial inputs of DOC and CDOM to the Lower Amazon are not fully co-varying with the river discharge. As a matter of fact, the correlation between DOC average values and the discharge data are much higher when excluding rising condition measurements (R 2 = 0.16 considering all the cruises and 0.79 excluding T4 and T6, data not shown), with relatively high DOC contents with respect to the corresponding Amazon discharge levels for these two cruises (Table 3).
The latter peculiar feature tends to indicate the presence of strong seasonal modulation in the DOM quality according to the timing of discharge and not only its intensity. T4 is corresponding to the rising season. In this case, the low CDOM-high DOC situation might reflect a difference in the quality (lability) of the DOM accumulated in source areas (e.g. floodplains, flooded forest, seasonally isolated lakes [11],) and then mobilized during the increase of the Amazon waters level. The situation for the T6 samples (April 2017 beginning of the high season) is globally similar to the one found during T4 (February 2016). During the year of 2017 a modification occurs in the general river discharge pattern, probably linked to the El Niño event of 2015-2016 [18,20], as illustrated here by the decay of about one month in the maximal discharge values recorded at Óbidos in 2017 (beginning of May), when compared to the situations observed from 2014 to 2016 (late May, Fig. 1). This suggests that besides seasonal modulation, inter-annual variability in the timing of the hygrometric regime can modulate the quality of the DOM mobilized from land-floodplain flush and delivered to the river. Note that short time scale processes such as strong rain event might also represent another source of variability in the quality of the terrestrial DOM inputs, as emphasized in other tropical environments [53]. Temporal variability in the average S 275-295 and S R are relatively narrow when compared to the variability found at spatial scale between the different water types (CW, Amazon) investigated in the frame of this study (Tables 2 and 3). The slightly lowest S R values found for T4, T6 might suggest the presence of DOM with a higher molecular weight for these two cruises when compared to the other cruises. However, the understanding of the source and sink mechanisms driving this apparent heterogeneity in the DOM characteristics of the Lower Amazon according to the discharge conditions (intensity and timing), would require additional measurements including some specifically dedicated to the characterization of the seasonal DOM dynamics within the diversity of source areas surrounding the Lower Amazon.
The latter temporal patterns were not observed for the pCO 2 values. An ANOVA of pCO 2 between different seasons, combined with a post-hoc test (Tukey HSD for unequal N), reveals that pCO 2 for the stations of the Amazon was higher during the high (T1) and falling (T3) seasons while the data for the other cruises showed the same mean values (p > 0.05). In contrast with the Amazon samples, no clear seasonal pattern can be found regarding the biogeochemical and bio-optical properties of the CW samples. This might be related to the higher complexity of the factors driving DOM dynamics for the corresponding water masses which result for the balanced effects of autochthonous and allochthonous DOM inputs, mixing processes between CW and Amazon River waters and modulation related to both bacterial and photochemical degradation processes. Low CDOM and DOC values are found during the Low and Falling seasons (T2 and T3, respectively) translating the combined action of low allochthonous (low land flush) and autochthonous (low chl-a) sources of DOM. As observed for the Amazon River, high DOC values recorded for CW during T4 were not associated with an equivalent raise in the CDOM values reinforcing the peculiar characteristics of the DOM input during the rising season. No seasonal difference was found in the pCO 2 recorded for the CW samples (p > 0.05).
In order to illustrate the spatial variability from the Lower Amazon to the river's mouth, average a CDOM (412), [DOC], pCO2, a* CDOM (412), S R and S 275-295 are represented for each cruise in Fig. 2. The spatial distribution show that the carbon content (DOC: Fig. 2(a) and pCO 2 Fig. 2(c)) generally decreased from Óbidos to the river mouth. This general pattern remained consistent for all the seasons sampled, except for DOC during the falling season (T3), which tended to increase towards the mouth of the river (Macapá), while no similar pattern could be observed for CDOM. The increase of DOC along the transect Óbidos-Macapá was reported by Ward et al. [8] as a result of the combination of strong organic matter inputs from tributaries and floodplains and the degradation of particulate carbon into dissolved molecules. Such high export rates of labile organic matter from floodplains to the river during the falling season, was also reported by Moreira-Turcq et al. [14]. However, this labile material does not persist further out into the plume once exported to the ocean, leaving behind a background of recalcitrant DOM [54,55]. Fig. 2. Spatial distribution of biogeochemical parameters along the Amazon mainstream. The parameters were averaged for each station per season. The first station is at Óbidos (~900 km from the mouth). The middle of the transect is at Almeirim (~450 km from the mouth). The transect ends at the Amazon mouth, in Macapá, and T5 and T6 were sampled at the river mouth only.
As observed for DOC, CDOM values show a general decreasing gradient from the Lower Amazon to the river mouth ( Fig. 2(b)) highlighting the impact of DOM degradation processes (especially bacterial) along the river course, as reported by other studies [23,56]. This feature is confirmed here, by the increasing gradient generally found for S 275-295 , with Fig. 2(d) suggesting a gradual decrease in DOM molecular weight (Fig. 2(d)) from Óbidos to Macapá. Interestingly, such changes in the DOM quality within the Amazon River is also illustrated here by the general sharp decreasing patterns in a* CDOM (412), translating differences in the intensity of the degradation rates for DOC and CDOM along the Amazon water course (Fig.  2(e)).
Note that CDOM, DOC and pCO 2 spatial distribution showed a slight inflexion point at Almeirim (Fig. 2), emphasizing a potential influence of inputs of the CW from the Paru River on the optical and biogeochemical characteristics of this area [56] ( Table 2). The decrease in the CO 2 degassing along the Lower Amazon River region reported here, is consistent with previous observations provided by Sawakuchi et al. [7].

a CDOM (412) to DOC relationships
Considering the whole data set gathered in the frame of this study, the direct relationship between a CDOM (412) and DOC (N = 80; R 2 = 0.17, p<0.05) had a low accuracy (Fig. 3(a)). This absence of general co-variation between CDOM and DOC can be related to two specific features. First, Amazon River water samples for T4 and T6 present a pattern very different from the other cruises, with a general decoupling between CDOM and DOC data (Fig. 3(a)). This absence of co-variation between CDOM and DOC for T4 and T6 can be related to the peculiar characteristics of the terrestrial DOM inputs induced by land-floodplain flush during the rising of the Amazon waters. Here, the decoupling observed between CDOM and DOC could translate a spatial heterogeneity in the DOM sources mobilized during such discharge conditions. Floodplain areas surrounding the Amazon have been for instance showed to be extremely heterogeneous in terms of vegetation-water connections [11,18]. However, no information about the DOM dynamics in the latter water bodies is currently available.
Second, a clear discrepancy in the link between CDOM and DOC exists when considering the CW (whole data set) and Amazon samples (for T1, T3 and T5) separately ( Fig. 3(a)). When splitting these data ( Fig. 3(b)), two different significant linear relationships between a CDOM (412) and DOC can be drawn (Fig. Fig. 3(c), Amazon River: N = 42, R 2 = 0.74, p<0.05 ; CW: N = 13, R 2 = 0.57, p<0.05) underlining the strong biogeochemical heterogeneity of the Amazon basin water masses. The dispersion around these two relationships can be partly attributed to seasonal modulation in DOM quality; however, more data are needed to clarify the impact of these seasonal patterns, especially for the CW data set which is relatively small (N = 13). Note that the presence of such linear dependency between CDOM and DOC is not due, unlike coastal waters, to mixing processes, but more likely to a parallel (biological and photochemical) degradation of both CDOM and DOC along the Amazon River course.
The differences in the slope and offset values for the CDOM to DOC relationships derived for the Amazon and CW samples (22.6 and 224 µmol L −1 , 54 and 91.3 µmol L −1 , respectively) suggest differences regarding the source and sink factors acting on DOM dynamics in these two optically different water types of the Amazon basin. The higher average S 275-295 and S R for the CW samples (Table 2) tend to indicate the presence of DOM of lower molecular weight than for the Amazon samples. Such differences might be attributed to variation in the DOM origin as larger contribution of the phytoplankton derived DOM in CW [8] as well as variation in the intensity of the photo-degradation processes (greater in the less turbid and more stagnating CW samples of the Xingu and Tapajos rivers).
These results suggest that the estimation of DOC from CDOM absorption coefficient using a simple linear model might be possible taking into account specific information about the optical water type, and thus biogeochemical quality (Fig. 3(b), (c)). This might be achieved using optical classification based approaches whose advantages have been demonstrated for diverse applications dealing with the estimation of various biogeochemical products from ocean colour remote sensing (e.g [57][58][59].). However, our results also emphasize that such direct CDOM to DOC models are invalid and would result in inaccurate estimations of DOC (e.g. T4: rising conditions, T6: shift in the discharge timing).
Several studies have emphasized the use of spectral slope S 275-295 for assessing a* CDOM and thus better constraining the natural variability in the dependency between CDOM and DOC at both seasonal [33] and regional [25,34] scales. The interest of using S 275-295 for estimating DOC from CDOM was also evaluated in the frame of this study. As previously observed for the direct CDOM-DOC relationship, a low correlation between a* CDOM (412) and S 275-295 is observed when our whole data set is considered (N = 80; R 2 = 0.26, p<0.05). When excluding the peculiar samples corresponding to T4, T6, a significant nonlinear relationship can however be drawn between a* CDOM (412) and S 275-295 for T1 to T3 and T5 considering both Amazon and CW samples ( Fig. 4(a)). This non-linear model is very close to the one proposed by Vantrepotte et al. [25], from a large data set gathering data collected within highly contrasting coastal waters dominated by terrestrial inputs of DOM. Further, the general validity of this model, when considering the data of Amazon River (T1-T3, T5) and CW rivers, emphasizes the interest of such S 275-295 based approach for avoiding issues related to the use of water masses information as suggested for direct CDOM to DOC linear models. It is worth noting, however, that such a S 275-295 based model might induce a decrease of performance in the DOC retrieval when compared to direct CDOM-DOC relationships, as observed here especially for the samples collected during low water conditions (relative errors reaching 20%, Fig. 4(c)). Our results further indicate that a unique S 275-295 vs a* CDOM (412) model is not sufficient to take into account the occurrence of very specific conditions such as the ones identified for the samples collected during T4 and T6, for which a specific model is required (Fig. 4(a)).
Note that a significant relationship between S 275-295 vs a* CDOM (412) was found for these samples (N = 25, R 2 = 0.78, p<0.05, RMSE = 0.0004, Bias = 0.001, MRAD = 0.016) ( Fig.  4(b)) whereas a general scattering was conversely observed when looking to the direct link between CDOM and DOC for the corresponding data ( Fig. 3(b)). The better result for T4, T6 in comparison to Amazon and CW (R 2 = 78 and R 2 = 58, respectively, Fig. 4(b)) can be explained by the data set of similar bio-optical characteristics. Amazon and CW gather samples from contrasted turbidity while T4 and T6 where sampled only on Amazon waters with similar optical constituents.
Interestingly, a sharp non-linear relationship was found between a CDOM (412) and S 275-295 (Fig. 5) as reported by other authors [25]. This unique relationship derived from our whole data set indicates the link between CDOM quality and quantity, and suggests the possible assessment of S 275-295 from a CDOM (412) for remote sensing applications.
The expressions and coefficients used to develop the relationship of CDOM-DOC are documented in Table 4.

a CDOM (412) and pCO 2 relationship
Various studies have documented significant relationships between pCO 2 and DOC in boreal and temperate inland waters [36][37][38]. Based on results from this study, we have not observed such a link between pCO 2 and DOC in the Lower Amazon River waters (N = 69, R 2 = 0.04, p>0.05, Fig. 6(a)). This result is consistent with other studies in tropical inland waters [39,40,60], also emphasizing an absence of correlation between these two parameters in various Brazilian lakes, including some located in the Amazon basin. Rather, the pCO 2 has been related to rainfall and temperature [40,60,61]. In contrast, a strong linear positive relationship was found here between a CDOM (412) and pCO 2 ( Fig. 6(b) and 7(a), (pCO 2 = 1240 * a CDOM (412) −1845, N = 69, R 2 = 0.65, RMSE = 979, p<0.05). This general relationship between a CDOM (412) and pCO 2 seems to be less affected by the optical water type or river discharge conditions, when compared to the results previously obtained regarding CDOM to DOC relationship. Several studies have demonstrated the use of chl-a and surface temperature data (T) as a predictor for pCO 2 in diverse river-dominated coastal areas [62,63]. CDOM has also been used in combination with the aforementioned parameters as a proxy for salinity or humic substances [42, 64,65]. The chl-a measurement indicates the presence of phytoplankton and therefore primary production that fixes CO 2 , leading to lower values of pCO 2 [66]. Inversely, higher temperatures decrease the solubility of CO 2 [67] and DOM availability enhances bacterial respiration leading to higher pCO 2 [7,9,60,68].
Individual regressions (Table 5) indicate that degassing fluxes related to CDOMdegradation processes are the main factor driving pCO 2 dynamics in the Lower Amazon explaining 65% of pCO 2 variability through a linear relationship (Fig. 7(a)). Conversely, temperature and chl-a explain 49% and 30% of pCO 2 variability, respectively, and pCO 2 exhibit a non-linear dependency with the latter two parameters (Table 5). The value added when including these two additional descriptors for estimating pCO 2 was specifically assessed using polynomial multivariate regression considering a CDOM (412), T and chl-a as descriptive variables. Results indicate that the inclusion of both CDOM and temperature provide better pCO 2 estimates (N = 69; R 2 = 0.80; RMSE = 757 ppm, Bias = −13, MRAD = 26) than a model based on CDOM and chl-a (N = 69; R 2 = 0.72; RMSE = 875 ppm, Bias = −11%, MRAD = 32%). A chl-a and temperature based model provides the lower pCO 2 retrieval accuracy (N = 69; R 2 = 0.54; RMSE = 1136 ppm, Bias = −16%, MRAD = 39%). Further, the consideration of CDOM, T and chl-a only slightly improve pCO 2 retrieval (N = 69; R 2 = 0.83; RMSE = 749 ppm, data not shown) when compared to the CDOM and T based model ( Fig. 7(b)). The estimation of chl-a concentration from ocean colour remote sensing in very turbid waters such as those of the Amazon River remains very challenging [69,70]. The performance of models based on reflectance bands more suitable to retrieve chl-a in turbid environments (i.e. red and near infra-red) has been shown to provide a relevant alternative at least when chla concentration is sufficiently high (> 2mg m −3 ) [70][71][72][73]. Such chl-a levels were never reached in this data set, with Amazon River samples showing an average chl-a concentration of 1.29 ± 0.38 mg m −3 (Table 2). Conversely, various studies have emphasized the possibility of using remotely-sensed temperature estimations in inland waters with a relatively high accuracy [74][75][76][77]. Considering these issues, a formulation based on CDOM and T for estimating pCO 2 in the Lower Amazon region could be developed: Similarly, various CDOM absorption inversion algorithms adapted to coastal or freshwaters environments have been also recently documented (e.g [78].). Considering the latter feature, a formulation based on CDOM and T for estimating pCO 2 in the Amazon waters is here proposed for assessing pCO 2 content, potentially applicable for further applications of ocean colour remote sensing.
Interestingly, in contrast with the previous results on DOC, these results suggest that a unique relationship can be used without distinguishing Amazon and CW data, or considering specific rain conditions (T4, T6), highlighting a greater stability in the factors driving CO 2 production, consumption, and outgassing.

Conclusions
The major aim of this study was to assess the potential use of a CDOM (412) for estimating DOC and pCO 2 content within the poorly-investigated region of the Lower Amazon River. To achieve the main goal, the spatial-seasonal variability was documented for the study period.
The heterogeneity of the DOM dynamics between clear and turbid water bodies of the Lower Amazon River region was clearly observed. The prevalent impact of degradation processes more likely related to bacterial activity has been further highlighted by the changing optical properties of CDOM along the course of the Amazon River that agree well with past observations of more degraded DOM molecular composition towards the river's mouth. Besides these general spatial features, a strong seasonal variability was found for both CDOM and DOC values. While the seasonality in the Amazon River discharge represents a major controlling factor for the DOM annual variation, results also illustrate the complexity of the DOM temporal dynamics in the Lower Amazon. Significant modulations in the DOM quality were specifically observed according to the discharge timing at either seasonal (rising conditions) or inter-annual (El Niño event) scales. The understanding of the CDOM and DOC dynamics depicted here in the Amazon mainstream would however need additional information specifically regarding the seasonal variation in the quality of DOM inputs, considering the diverse source areas surrounding the Lower Amazon (floodplains, flooded forest, seasonally isolated lakes).
The potential for the assessment of DOC loads from a CDOM inversion algorithm was investigated. Results demonstrate the strong heterogeneity of the water masses of the Lower Amazon region with its clear and turbid waters, suggesting that assuming a single direct CDOM to DOC relationship is problematic. While the use of S 275-295 based approach has been shown to represent a relevant alternative for assessing DOC from CDOM in coastal waters, results here tend to indicate that such optical proxy can only partly describe the natural variability of the CDOM to DOC ratio in the inner Amazon. A single S 275-295 based model is not able to capture the impact of DOM inputs potentially related to the seasonal modulation in the quality of the DOM delivered to the river from land-floodplain flush.
Finally, our results have demonstrated that CDOM absorption can be considered as a relevant proxy of the pCO 2 in the Lower Amazon. A model based on CDOM and temperature seems to provide the most reliable pCO 2 estimates (relative error of 26% on average). The assessment of the CO 2 flux from the entire Amazon River, including the lower reaches that are tidal-influenced, is crucial to understand the role of inland waters to the carbon budget. For example, past global estimates of CO 2 emissions do not include tidal rivers, and including the tidal reaches of the Amazon River, alone, increase global CO 2 outgassing estimates by ~43% [7]. Widely applying the optical approach used here will allow a broader assessment of the lower reaches of rivers worldwide and allow for more persistent monitoring of alterations to aquatic carbon cycling under a changing climate.