Spectral Response of Water Under Different Concentrations of Suspended Sediment : Measurement and Simplified Modeling

Understanding the spectral behaviour of water is of the greatest importance to the quality management of water resources. Continuous monitoring by remote sensing is therefore essential for administrators seeking the efficient management of its many uses. The aim of this research was to characterise the spectral response of water submitted to different concentrations of sediments of varying textural properties, organic matter and salts, and to identify the implications of these characteristics using simplified modelling. The experiment was conducted at the Radiometry Laboratory of the Department of Agricultural Engineering of the Federal University of Ceará, Brazil. The soils used in the research came from two areas of irrigated agriculture in Ceará, one in Morada Nova and the other in Pentecoste. Both soils were classified as Fluvic Neosols; the first saline and the second saline-sodic, and presented significant differences in granulometric composition and organic matter content. From the results, it can be concluded that: (i) sediments added at different concentrations cause an increase and deformation of the reflectance curves, and that maximum spectral partitioning occurs at two reflectance peaks; (ii) derivative analysis favours the identification of wavelengths that best differentiate sediment concentration, allowing more-efficient modelling of the process; (iii) the characteristics of texture, organic matter and salt content have little effect on estimating suspended-sediment concentration in the water, making multiple linear regression modelling a viable option for this purpose.


Introduction
Monitoring the quality of surface water is a basic tool in managing this resource (Abdelmalik, 2018;Kim et al., 2017), especially in regions where the supply of water for a wide range of uses is almost entirely dependent on surface reservoirs.Population growth and the increased demand for consumer goods have resulted in a significant increase in the number of artificial reservoirs in various areas of the world, such as Australia (Callow & Smettem, 2009), Brazil (Malveira, de Araújo, & Güntner, 2012), China (Li & Wei, 2008), Romania (Rãdoane & Rãdoane, 2005), Spain (Mamede, 2008), the United States (Minear & Kondolf, 2009) and Canada (Teegavarapu & Simonovic, 2002).Despite efficient storage capacity, surface reservoirs are subject to impacts on the quality and quantity of waterdue to the supply of sediment from natural or anthropogenic processes.
The semiarid regions of Brazil are an example of regions where the supply of water is almost entirely dependent on surface reservoirs (Andrade, Araújo, Rosa, Disney, & Alves, 2007;de Araújo, Güntner, & Bronstert, 2006;Krol, de Vries, van Oel, & de Araújo, 2011).In these environments, the risk of water shortage is imminent due to the growth in demand as well as the water supply being restricted to the rainfall history of the previous years and the accumulated volume in the reservoirs (de Araújo & Bronstert, 2016).Therefore, the availability of water relies on the retention of river flows in the reservoirs, which on its turn is subject to the siltation process and can have serious impacts on the local and regional economy in relation to potable water supply, irrigation and energy supply (Agostinho, Gomes, Santos, Ortega, & Pelicice, 2016).The reduction in water availability is one of the key impact factors on silting in reservoirs on the semiarid regions.It is estimated that the reduction, for the State of Ceará alone, is approximately 388 L.s -1 yearly (90% annual reliability: Campos, 2010;de Araújo et al., 2006).
Still, qualitatively the reservoirs can be affected by silting since they act as filters able to retain a considerable portion of the sediment carried by rivers (Lima Neto, Wiegand, & de Araújo, 2011;Lira, Toledo, & Mamede, 2014).This retained portion, when at high deposition rates, can cause significant increases in nutrient and other water pollutants.The deposition of enriched sediments can have negative environmental consequences, which are detected by changes in indicators of water quality, mainly related to the solubility of salts, nutrients and sediment transport (Andrade et al., 2007).
It is therefore essential to monitor water resources in order to promote measures for controlling and preventing large amounts of sediment in surface reservoirs.
Among the current techniques for monitoring water bodies, remote sensing is important., and has shown great potential for overcoming the spatio-temporal limitations of traditional methods of water monitoring in situ, since it allows the acquisition of information on different spatial and temporal scales (Gin, Koh, & Lin, 2003;Rudorff, Novo, Galvao, & Pereira Filho, 2007;Valerio et al., 2018;Wang & Lu, 2010;Zhang et al., 2016).
With the application of remote-sensing techniques, it is possible to evaluate the responses resulting from disturbances due to anthropogenic activity, so as to predict the impact of such action on the conditions that guarantee maintenance of the environmental quality in the medium and long term (Novo, 2005).Therefore, the relationship between responses in the spectral domain and optically active limnological characteristics are the focus of applying remote sensing to bodies of water.It is sought to understand the dynamics of the composition and biophysical components of water that are relevant to monitoring aquatic ecosystems (Renó et al., 2009).
The difficulty in understanding the dynamics of the composition and optically-active components of water lies in their interactivity, since sediment, due to such characteristics as its granulometric composition, results in changes to spectral response (Lodhi, Rundquist, Han, & Kuzila, 2007;Ostrovsky & Tęgowski, 2010).In addition, this sediment is associated and interacts with other optically active components, such as organic matter.The organic matter acts in reducing the bidirectional reflectance factor of the water (Mantovani & Novo, 1996) in the visible spectrum and promotes increased reflectance in the infrared.Organic matter dissolved in the water causes absorption of the incident energy in the region of the spectrum between 400 and 700 nm (Bricaud, Morel, & Prieur, 2018), with a consequent reduction in reflectance.Increases in the concentrations of organic matter in the water also promote an increase in reflectance amplitude in the red and infrared regions (Mobley, 1994).
In view of the above, the aim of this study was to characterise the spectral response of water submitted to different concentrations of two soils of different texture and concentrations of organic matter.It was also sought to identify the relationship between spectral signature and concentration using multiple regression analysis.

Collection of Experimental Data
The experiment was conducted at the Radiometric Laboratory of the Department of Agricultural Engineering of the Federal University of Ceará (DENA/UFC), Brazil.To minimise interference from variations in the angular position of the sun, all readings were taken under a clear sky from 10:00 to 14:00.
The experiment was carried out in a blue polyethylene container, painted internally with matte black paint.The container had a total capacity of 310 litres; the volume of water used in the experiment was 200 litres (Figure 1).An ASD FieldSpec®3 Hi-Res spectroradiometer, manufactured by Analytical Spectral Devices, was used to measure the radiance.The spectral resolution of this equipment is 1 nm, with a spectral range of from 350 to 2,500 nm.

Figur
The wavel optically a used in me

Data Analysis
Derivative analysis is a technique that makes use of the derivative value of reflectance with respect to wavelength.This allows identification of the points at which the spectral curve displays abrupt changes in behaviour due to the presence of components that favour absorption and scattering of the target.Derivative analysis was therefore used to evaluate more accurately the wavelength at which spectral changes occurred in relation to the sediment concentrations under investigation.
The first derivative (dRλ) of the reflectance spectrum with respect to wavelength x at a given point i can be numerically approximated by a finite-difference scheme using central or symmetric approximation, expressed by Equation 2. ( where, Δx is the separation between two successive bands.Since Δx=x j -x k , it is assumed that x j > x k and the interval between the bands is constant.From the results obtained with the derivative analysis, a multiple regression analysis was carried out in order to establish empirical models for estimating the suspended-sediment concentration in the water from the reflectance data. In order to evaluate the performance of the resulting models, the following criteria based on statistical indices were used: coefficient of determination (r²), which indicates the degree of correlation between the independent variables and the dependent variable; Pearson correlation coefficient (r) (Equation 3); and the concordance index (d) proposed by Willmott (1981), that indicates the degree of precision of the equation (Equation 4) and can assume values from 0 (zero) to 1, for the no agreement and perfect agreement, respectively.The confidence index of the model (c) was also analysed, which includes the indications of the two coefficients, r and d, (c = r•d), where c = 1 means perfect confidence and c = 0 zero confidence, as proposed by Camargo and Sentelhas (1997).
Finally, the measured and simulated data were evaluated using the efficiency coefficient proposed by Nash and Sutcliffe (1970) (Equation 5).The NSE coefficient varies from -∞ to 1, and the greater the coefficient, the better the performance of the model.If the NSE is less than 0 (zero), the predictive capacity of the model is less than the measured mean value.
Different reflectance data were used in this analysis to those used in calibrating the models, thereby avoiding errors resulting from autocorrelation of the data, which could result in the models giving skewed results.The reflectance data for the selected wavelengths that were used in the calibration step constitute a mean of a set of three values obtained for each concentration of suspended sediment (0 to 1500 mg L -1 ).The reflectance data for the same wavelengths used to model the SSC from the regression models evaluated correspond to the mean obtained from the two remaining values of the readings repetitions (as previously described, at each stage of addition of sediments were performed a sequence of five repetitions of readings).
Thus, it was possible to perform the SSC simulations from the empirical models with reflectance data not used in the elaboration of the equations, allowing the comparison with the measured suspended sediment concentration.
where, E: SSC value number of

Relatio
In It can also be seen that for the same sediment concentration, spectral reflectance values was slightly higher for the soil with a greater sediment content of smaller particle size, i.e. clay (Figure 2B).In fact, for suspended particles with diameters smaller than 63 μm, the maximum reflection coefficient is 11% at 570 nm (Gin et al., 2003), while for sediment smaller than 10 μm, the maximum reflectance at the same wavelength is 15%.
For the saline soil (Figure 2A), reflectance values was similar between the peaks (700 nm and 800 nm), in contrast to the saline-sodic soil (Figure 2B), where the values of the first peak was higher than those of the second.These findings support the hypothesis that organic matter present in larger proportion in saline soil has the effect of reducing reflectance in the 400 nm to 700 nm range, and at the same time promotes a substantial increase in reflectance around 800 nm.
It is known that the presence of organic matter reduces the reflectance of both silt and clay particles, which, despite being fractions of smaller diameter, undergo changes due to organic matter.Clay particles in the presence of more organic matter have a darker colour and result in lower reflectance values at all wavelengths compared to silty soil (Lodhi et al., 2007).
The greatest reflectance value, which occurred around 800 nm for the saline soil (soil with a higher concentration of organic matter) confirms the findings of (Mantovani & Novo, 1996).Those authors report that the greatest increase in the bidirectional reflectance factor as a function of organic matter concentration occurs at around 800 nm.
Increases in the concentration of dissolved organic matter found in water systems decreases reflectance, especially in the blue to red range.In the spectral range between 400 and 570 nm (blue and green), concentrations of humic substances promote a reduction of approximately 40% in water reflectance.In the red and infrared region, 670 and 900 nm, there is a proportional increase in reflectance amplitude throughout the range (Mobley, 1994).
However, it is important to highlight the uncertainties that may occur due to factors in the field, such as the variation in days and times of measurements; variations in irradiance and in solar illumination geometry during data collection; specular reflectance; and movement of the water surface caused by the wind, which intensifies the effects of the specular reflectance (Rudorff et al., 2007).
The most relevant characteristic of the spectral curves is the predominance of two reflectance peaks.These two crests tend to become narrower and more prominent as more sediment is added.Maximum spectral partitioning (rate of increase of peak reflectance at each concentration) for both soils can be seen at the wavelength bands corresponding to the reflectance peaks.The formation of two reflectance peaks in the spectral profile when sediment is added to the water is a fact resulting from this study, and is also seen in other studies, such as those of (Lodhi et al., 2007;Rudorff et al., 2007).
With the reflectance curves for 1,250 mg L -1 and 1,500 mg L -1 of sediment, an increase in suspended-sediment concentration does not result in an increase in reflectance, i.e. there is low spectral separation, in such a way that an increase in sediment concentration results in no significant increase in reflectance.Cloud cover and energy intensity in the environment (Gin et al., 2003) are factors that cause these variations, however experimental conditions during this research minimised the action of these factors.Another important factor, and one that may have exerted greater influence, is the granulometric and mineralogical composition of the soil (Curran & Novo, 1988), especially sediment properties.The influence of granulometric composition should explain the result described here, associating water saturation by the suspended-sediment load with the deposition rate, which tends to be greater at high concentrations.

Relationship Between the First-Order Derivative and Sediment Concentration in the Water
After differentiating the reflectance curves for both soils (Figure 3), it was found that the first spectral derivative applied to reflectance reduced variations from the constant addition of radiation at both levels of data acquisition, highlighting certain spectral features.With these curves, the variation pattern was similar for each simulated concentration of suspended-sediment in the water.First there was a negative peak around 725 nm, followed by a positive rise to 800 nm, with the subsequent formation of another, more-pronounced negative peak around 825 nm.These findings for spectral behaviour are explained by the combined effect that the spectral properties of the suspended sediments and dissolved organic matter together promote in the water.For Rudorff et al. (2007), these joint properties of the dissolved elements translate into standard spectral behaviour of changes in the reflectance and absorption peaks.

These find
As a result L -1 , and th

In addition suspended
In addition of a genera analysis of The adjust R² adjusted g representa and partial   Note.SSC = suspended-sediment concentration (mg L -1 ); r = Pearson correlation coefficient; d = concordance index; c = confidence index of the models; NSE = Nash-Sutcliffe efficiency coefficient.
Despite the good results presented in this research, it is necessary to discuss that the laboratory conditions to which the models were developed may not offer the conditions of applicability in the monitoring of the bodies of water.This is due to the fact that the suspended sediments originate in different sources and the proportion that each source contributes in the blend is variable in the time and the space, consequence of the erosive processes that are in course in the contribution basin.Even growth methods that are based on measures and comparisons between measures; which admit the fundamental principle that suspended sediments maintain some of their acquired geochemical properties in the sources, and these properties are then used as tracers for efficiency; have limitations on the applicability in watersheds and differentiation of sediment sources.An example of this is the fingerprinting technique (Collins & Walling, 2002;Minella, Merten, & Clarke, 2009;Poleto, Merten, & Minella, 2009;Tiecher et al., 2018).
Thus the temporal dynamics of sediment in the catchment as well as the different sources (e.g.roads, crops, pastures, fluvial channel), induce the necessity of larger models as opposed to specific models for different types of soil.Although general models have lower efficiency under laboratory conditions, they may overcome the problem of the innumerable sources of variation that sediments may present.For example, the sites where the soil samples were collected are characterized as being of intense agricultural use and in some cases without the adoption of conservation practices, which can cause high rates of soil loss and sediment production that, when they reach the bodies of water cause significant damage.Despite this logic of establishing that agricultural areas contribute more to sediment production, studies by Croke, Mockler, Fogarty, and Takken (2005); Foltz, Copeland, and Elliot ( 2009); Navarro-Hevia, Lima-Farias, de Araújo, Osorio-Peláez, and Pando (2016) e Ziegler et al. (2007) showed that unpaved roads were significant sources of runoff and sediment production, albeit to a lesser extent.In addition, erosion rates produced on roads were much higher (or at least similar) than the usual erosion rates dating back to agricultural fields (Cao, Zhang, & Zhang, 2009;Rijsdijk, Sampurno Bruijnzeel, & Sutoto, 2007).
These findings suggest that the monitoring conditions of the bodies of water from specific models for soils of agricultural area may not represent the share of sediments in the water from significant sources, such as unpaved rural roads.These questions highlight the need for monitoring and continuity of research in order to overcome the limited applicability of the empirical models that, for Lopes et al. (2014), are useful only within the context from which they were determined and with the aid of real field data.

Conclusions
The results obtained in this research, under controlled laboratory conditions, allowed to conclude that: Different concentrations of suspended sediments in the water causes an increase and deformations in the reflectance curves for the bands corresponding to the optically active components, with maximum spectral partitioning correlated with the predominance of two reflectance peaks that were more intense at higher sediment concentrations.
The technique of derivative analysis was promising in identifying points on the spectral curve for differentiating between sediment concentrations, thereby allowing the elaboration of more-efficient regression models for the wavelength where the spectral changes occurred in relation to the sediment concentrations under study.
The use of multiple linear regression models was a viable alternative for estimating sediment concentration in surface waters, in such a way that the characteristics of texture, organic matter and salt content do little to interfere in modelling the concentration of suspended-sediments in the water.
It is necessary to use the empirical models in field conditions, in order to verify the applicability of the method in the estimation of the sediment concentration in the bodies of water and in the monitoring of the water quality from the reflectance data.Thus, allowing to evaluate the interference that soil heterogeneity in a catchment, the temporal dynamics of suspended sediments and field conditions different from those in controlled laboratory experiments, can generate in the estimates.
and dissolved organic matter (close to the 450 nm band), and by rotation of the water molecules (750 nm and 900 nm).

Table 4 .
the grouping of a larger number of non-explanatory random variables, makes it feasible and applicable in certain situations.Results of the evaluate the performance of the models for Fluvic Neosol