Correlation Analysis of Water Wave Reflectance and Local TSM Concentrations in the Breaking Zone with Remote Sensing Techniques
Abstract
The coastal zone is a dynamic area in which processes with different origins and scales interact. Several techniques can be used for coastal zone monitoring. Remote sensing is a powerful tool for monitoring coastal processes and managing coastal areas. The quality of coastal water is a very important issue, and remote sensing optical sensors can be used to quantify water quality parameters such as suspended sediments. Therefore, it is possible to estimate the total suspended matter (TSM) concentration with multispectral satellite images. To extract meaningful information, the satellite data need to be validated with in situ measurements. The main objective of this work was to quantify TSM concentrations in the breaking zone with in situ measurements. In situ validation is important for the accuracy of correlations established. A section of the northwest coast of Portugal, near Aveiro city, was chosen as a test area, and all in situ measurements were done in this area. Several in situ techniques have been used to establish a relationship between seawater reflectance and TSM concentration for the range of wavelengths from 400 to 900 nm. Empirical relationships were established for equivalent reflectance values by SPOT/HRVIR (high-resolution visible infrared), Terra/ASTER (advanced spaceborne thermal emission and reflection radiometer), and Landsat TM (thematic mapper) at visible and near-infrared equivalent bands and TSM concentrations. The reflectance values were used to estimate TSM concentration with the use of the relationships established by in situ measurement.
The reflectance of all bands of the satellite images tested showed high correlation with TSM in the wavelengths between 500 and 900 nm. The water leaving equivalent reflectance for each sensor equivalent band in the visible and near-infrared wavelengths was calculated, and a relationship between seawater reflectance and TSM concentration was established. The model coefficients and correlation factors for identical bands on different sensors presented good similarity. The work presented shows that TSM concentration in the breaking zone can be obtained directly from simulated multispectral satellite data. However, in situ measurements are essential to calibrate the process and establish the empirical relationships between TSM concentration and water leaving reflectance.
The same empirical relationships found with in situ measurements will be used to estimate the TSM concentrations directly from real satellite data to try to quantify the sedimentary balance in the study area for the period of the satellite data.
INTRODUCTION
More than a half of the world's population lives close to the sea. The coastal zone represents a comparatively small but highly productive and extremely diverse system, with a variety of ecosystems extending from coastal terrestrial to deep-water regions approaching 200 m depth (Malthus and Mumby, 2003). The importance of understanding the land/ocean interaction, such as how changes on land can influence the biochemical and biophysical processes of the coastal seas, and vice versa, is fundamental to an assessment of the global implications of local changes in coastal seas (Vaughan, 1995).
Remote sensing techniques are a valuable tool to obtain specific information on the spatial and temporal characteristics of the coastal zone. Substances in surface water can significantly change the characteristics of surface water reflectance. Remote sensing techniques depend on the ability to measure these changes in the spectral signature backscattered from water and relate these measured changes by empirical or analytical models to a water quality parameter, such as suspended sediments (Ritchie, Zimba, and Everitt, 2003). Solar radiation reflected from seawater surfaces varies with the amount of suspended sediments and wavelength. In general, reflected solar radiation between wavelengths of 500 and 700 nm increases as the concentration of suspended sediments increases (Muralikrishna, 1983).
The discrimination of suspended sediments or total suspended solids from water reflectance is based on the relationship between the scattering and absorption properties of water and its constituents. Most of the scattering is caused by suspended sediments and the absorption is controlled by chlorophyll a and colored dissolved organic matter. These absorptive in-water components decrease reflectance in a substantial way. However, these absorptive effects occur generally for wavelengths less than 500 nm (Myint and Walker, 2002). Electromagnetic radiation in this range of wavelengths is also significantly affected by atmospheric scattering. Therefore, we will concentrate on the visible and near-infrared electromagnetic spectrum region to study the concentration of total suspended matter (TSM). The spatial variation in the concentration of TSM in near coastal areas or estuaries can be quite difficult to describe because of the effects of tides, coastal currents, and waves (Mikkelsen, 2002).
Various studies have been carried out combining in situ measurements and satellite data to correlate spectral properties and suspended solids sediments. Ritchie et al. (1974) developed an empirical approach to estimate the amount of suspended solids sediments. The general form of these empirical relations is
where X is suspended sediments, Y is spectral reflectance, and a and b are empirically derived coefficients.
Islam et al. (2001) used a linear relationship between reflectance and suspended sediments concentration, which was developed for the Ganges and Bralimaputra rivers (India). The general form of this model is where X is
suspended sediments concentration (mg/L), Y is spectral reflectance (%), and a and c are the regression coefficients.
Forget and Ouillon (1998) established a relationship between satellite equivalent reflectance (Rk) and TSM concentration from the Rhone River (France) with the use of SPOT high-resolution visible (HRV) and Landsat thematic mapper (TM) data. Linear, polynomial, log-linear, and log-log relationships were applied. The log-linear relationship gave the best correlation. This form is
where ak and bk are empirically determined factors.
Doraxan et al. (2002) developed an experimental method to determine the water composition in the Gironde estuary (France) using SPOT data. They used in situ measurements of water reflectance and collected water samples. They established empirical relationships between remote sensing reflectance of SPOT HRV bands and TSM concentration.
The objective of this work is to evaluate the potential of multispectral satellite images to quantify the TSM concentration in the breaking zone. Images from different sensors were tested to choose the most adequate for the description of different breaking patterns. For each sensor (Landsat TM, advanced spaceborne thermal emission and reflection radiometer [ASTER], and SPOT high-resolution visible infrared [HRVIR]), the equivalent reflectance for the visible and near-infrared bands was calculated with the use of the sensor spectral normalized response and in situ reflectance measurements. The main characteristics of the analyzed satellite sensors are listed in Table 1.
STUDY AREA
The test area is located on the Portuguese west coast, facing the Atlantic Ocean. This area is limited to the north by the mouth of the Douro river and to the south by Mira lagoon, (Figure 1). It has a linear extension of ca. 80 km and a general orientation of NNE–SSE.
Citation: Journal of Coastal Research 23, 6; 10.2112/05-0482.1
Coastal scenery includes important environmental features: the lagoon of Aveiro and a small residual lagoon, the Barrinha de Esmoriz, and the Mira lagoon.
From a geomorphologic point of view, this segment is mostly a sandy and low-lying coast composed of alluvium sands and dune systems.
Coastal morphodynamics is mainly shaped through wave action. Waves are indeed the dominant force, driving the littoral processes on this coast. Although the wave climate changes seasonally, it can be characterized by medium significant wave heights from 2 to 3 m, with periods ranging from 8 to 12 seconds (Veloso Gomes et al., 2003).
Seasonal storms, particularly between October and March, can produce significant storm surges when they coincide with astronomical tides. Waves can reach heights of more than 8 m, with periods reaching 16 to 18 seconds (Veloso Gomes et al., 2003). The tide range can vary between 2 and 4 m at spring tides twice a day because of the semidiurnal characteristics of the tide. Furthermore, local wave phenomena, especially refraction, diffraction, shoaling, breaking, and bathymetry, can influence tremendously local wave conditions.
Another important process is the littoral drift currents. Along the western Portuguese coast, these currents have a dominant North–South direction, except for local deviations from specific hydrodynamic processes (e.g., near river mouths). The dominant direction of the littoral drift can be directly shown by sediment accretion to the north updrift of obstacles (e.g., groynes) and by erosion to the south downdrift. Indirectly, this can also be demonstrated by analysis of wave direction frequencies reaching the Portuguese west coast, which exhibit higher frequencies and intensities in the northwest quadrant. The wind climate also affects coastal morphodynamics. Beyond its indirect influence on waves and currents—local wind climate generates currents and small waves with intensities and directions that can be related to the velocity, persistence, and direction of the wind that caused them—it also has a direct effect on the formation of dunes. Sea currents end up having negligible importance when compared with the other actions involved. Another characteristic of this coastal area is its high vulnerability to erosion. Through various physical processes, the shoreline has been eroded and shaped and the landscape modified. Most shoreline changes are natural responses to these processes, either at a time scale of days (e.g., between tides) or of years (e.g., global climate change). These natural dynamics are, in some cases, incompatible with the increase of human development along the coast (Veloso Gomes et al., 2004).
METHODOLOGY
Different methods (maritime platforms, aerial platforms, simulation on the beach, and water samples collected in the breaking zone) were used to determinate a relationship between TSM concentration and spectral response of the sea-water.
A FieldSpec spectroradiometer was used in all cases to determinate seawater reflectance. This optical sensor operates at between 350 and 2500 nm, with 10 nm of spectral resolution. Data were collected through a fiber optic cable input with 1.2 m length and a 25° full-angle cone field of view. Data quality depends critically on the precision at the calibration stage. On the basis of the responsivity of the FieldSpec spectroradiometer, the maximum radiance values measurable are well in excess of twice those for a 0° solar zenith angle and a 100% reflectance lambertian panel (Kuester et al., 2001).
Quantification of the TSM (mg/L) of the samples collected was made through a filtering process. The 1-L water samples collected were refrigerated in the dark and processed after a few hours in the chemistry laboratory. A cellulose nitrate filter was used, with a pore diameter of 0.45 μm. The quantification of TSM was made by subtracting the final and initial weights of the filters.
Maritime Platform
Three campaigns were carried out. The purposes of the field campaigns were to simultaneously collect water samples and measure the seawater leaving reflectance. The first two campaigns were scheduled so that they would coincide with clear-sky satellite overpasses by ASTER, but unfortunately, atmospheric conditions invalidated image acquisition. A boat from the Portuguese Marine Institute was used 10 July 2003 to collect water samples and simultaneously measure the reflectance. The position and depth of each location was also measured with a Global Positioning System (GPS) receiver and an echo-sounding lead. The main problems with this method were the difficulty in immobilizing the boat and the impossibility of collecting data close to the breaking zone. Water depth was not taken under consideration because the reflectance measurements and water sample collections were done at the surface. The quantification of depth was not relevant to establish the empirical models. It was only taken under consideration to demonstrate reflectance attenuation in the open sea.
Aerial Platform
Helicopters provide an efficient platform for the collection of water samples and for reflectance measurements. A large area can be sampled rapidly, and the platform can be immobilized over specific areas in the breaking zone.
The helicopter was stabilized about 2–3 m above the sea surface; 30 samples of water were collected, and reflectance was measured simultaneously. The position of the helicopter was identified with a GPS receiver. This was the only method that allowed measurement over the breaking zone.
Simulations on the Beach
Several field campaigns were carried out on different beaches to simulate different concentrations of suspended sediments and different types of sand (texture, color, and grain size). The sediment type was found to be very similar throughout the study area. A container of 0.8 m height and 0.4 m diameter was used. To minimize the effect of reflection from the sides and bottom of the container, it was lined with a black and completely opaque plastic. Reflectance was measured and water samples were taken simultaneously for a range of sediment concentrations. The seawater was mixed to put the sedimentary particles in suspension.
The container conditions lead to signal attenuation in the reflectance values. This attenuation was quantified by measuring seawater sample reflectance inside the container and with a wet sand background. The reflectance measurements made in the container were intended to simulate deep zones or the open sea. The reflectance measurements made in the transparent containers placed above wet sand were intended to simulate what happens when breaking waves occur in places of low depth, for which the reflectance values are strongly affected by the sea bottom. The relationship between container and wet sand background was found to be nearly constant for all concentrations of TSM tested. Figure 2 shows a typical example of this attenuation for a TSM concentration of 30 mg/L.
Citation: Journal of Coastal Research 23, 6; 10.2112/05-0482.1
Water Sample Collection in the Breaking Zone
To get samples directly from the breaking and swash zones, water was collected from a surfboard in small containers (about 1 L each). The reflectance was measured for each water sample collected by putting the transparent small container in the wet sand to simulate sea bottom conditions.
The average value of TSM concentration found in the breaking zone was 32 mg/L and in the swash zone 50 mg/L. Average reflectance values for these two areas are similar (Figure 3).
Citation: Journal of Coastal Research 23, 6; 10.2112/05-0482.1
RESULTS
The water leaving equivalent reflectance for each sensor band in the visible and near infrared (RES) was calculated by
where Rm is reflectance measured by the spectroradiometer, ϕ is the sensor spectral normalized response, and λ is the wavelength.
A relationship between seawater reflectance and TSM concentration was established for each equivalent satellite band. These correlations were made with the TSM concentrations and reflectance values obtained by in situ measurements. No real satellite data were used in this stage of the work. Equations of linear (R = aTSM + b), polynomial (R = aTSM2 + bTSM + c), logarithmic (R = a log(TSM) + b), power (R = aTSMb), and exponential (R = aebTSM) models were tested for all equivalent satellite image bands. The coefficients of determination (r2) were also calculated for each model. For the visible and near-infrared (VNIR) equivalent bands of Landsat TM (bands 1, 2, 3, and 4), the linear and polynomial models tested presented high determination coefficients (r2 ≥ 0.96). For power and exponential models, the obtained coefficients were smaller (Table 2).
The results obtained for equivalent bands of SPOT HRVIR (bands 1, 2, and 3) for all models tested are comparable to those obtained with TM data. Figure 4 shows a graphical representation of the linear model for equivalent bands of SPOT HRVIR.
Citation: Journal of Coastal Research 23, 6; 10.2112/05-0482.1
The coefficients of determination for all models established for equivalent bands of ASTER VNIR are of the same sort as those obtained for equivalent bands of SPOT HRVIR and TM. Table 3 shows the model coefficients and determination factors for linear and exponential models of equivalent bands (visible and infrared) of ASTER VNIR data.
The linear models established for the equivalent bands of all images tested present a determination coefficient (r2) greater than 0.95. The nonlinear models present a lower but acceptable coefficient of determination factor. A great similarity was verified between the model coefficients and determination factors for spectrally comparable bands on different sensors. These parameters are presented in Table 4 for the green bands of ASTER, TM, and HRVIR sensors.
The reflectance of the tested type of satellite image band range showed good correlation with the TSM for wavelengths between 500 and 900 nm. A reflectance peak appears between 550 and 600 nm, and above 900 nm, the reflectance is practically null (Figure 5).
Citation: Journal of Coastal Research 23, 6; 10.2112/05-0482.1
CONCLUSIONS
The great similarity observed between the correlation and determination coefficients for identical bands of different sensors indicates a high confidence level in the established models, and the small variations in the wavelength interval did not affect these coefficients significantly.
Measurement conditions affect the results. The influence of sea bottom and open ocean in the reflectance measures and the distribution of sediments in the water column need to be considered. The purpose of taking measurements under different conditions was to try to address this issue. The container aimed to simulate deep-sea conditions, without the influence of the sea bottom, and the measurements made with the wet sand background attempted to simulate the influence of the sea bottom. However, the breaking zone depth is usually shallow. Moreover, because the main objective of this work is to obtain TSM concentrations directly from satellite images, distribution of sediments in the water column is not a factor of extreme relevance. Nevertheless, this issue will be the aim of another study based on the results presented in this paper.
Breaking waves at the ocean's surface inject bubbles and turbulence into the water column. These bubbles can significantly change the optical properties of water, depending on their concentration and size distribution, introducing potentially significant errors into remote sensing measurements. The empirical models established were based on reflectance measurements and TSM concentrations of water samples collected at the breaking zone. In the simulations, the water was mixed to put particles in suspension and to generate bubbles. Therefore, the established models take under consideration the effect of the bubbles. It was not corrected, nor quantified, but it was considered in the correlations established.
This study shows that multispectral data can be used to obtain the TSM concentration in the breaking zone. However, in situ measurements are essential to calibrate the process to establish the empirical relationships between TSM concentrations and water leaving reflectance.
A set of satellite images were acquired: five from Landsat TM, two from ASTER, and one from SPOT HRVIR. All satellite image bands (visible and near infrared) were calibrated for radiance values and, after, for reflectance values. The atmospheric correction procedure was implemented and is based on an improved dark object subtraction technique c(Chavez, 1988). The same empirical relationships found with the in situ measurements will be used to estimate the TSM concentrations directly from these satellite images. The final aim of this work is to apply the techniques developed in this study made with in situ measurements to the satellite images acquired to quantify the sedimentary balance in the study area for the period of satellite data acquisition.
Study area located between the River Douro mouth and Mira Lagoon (Veloso Gomes et al., 2003).
Attenuation factor.
TSM average concentration for two water samples collected: 32 mg/L for the water samples collected in the breaking zone and 50 mg/L for the water samples collected in the swash zone.
Empirical model for reflectance functions vs. TSM concentrations for SPOT HRVIR.
Relationship between TSM concentrations and seawater reflectance.
