**2.4. Spectral analysis of water reflectance for hyperspectral remote sensing of water quality in estuarine waters**

Although hyperspectral remote sensing offers an effective approach for frequent, synoptic water quality measurements over a large spatial extent, the optical complexity of brackish estuarine water makes water quality monitoring by remote sensing a challenge. The third goal was to develop algorithms for hyperspectral remote sensing of water quality based on in situ spectral measurement of water reflectance [150].

During a hyperspectral remote-sensing study in the Chesapeake Bay, water reflectance spectra *R*(*λ*) and the discrete water samples were collected at 11 field stations from 2008 to 2011 in the Patuxent River, a tributary of Chesapeake Bay (**Figure 20**). In this study, the relationships between the water reflectance *R*(*λ*) and water quality parameters were examined to establish empirical algorithms that could be used for retrieval of water quality parameters from airborne hyperspectral data for rapid and effective water quality monitoring.

**Figure 20.** Map of the study area and the locations of the field stations for *in situ* measurement on the Patuxent River. The location of the study area in Chesapeake Bay watershed is shown in the upper right map [150].

Also, the aerial hyperspectral imagery of the Patuxent River with 2-m spatial resolution was acquired in the summer of 2005 using AISA-Eagle VNIR remote hyperspectral sensor (Center for Advance Land Management Information Technologies, University of Nebraska, Lincoln, NE). AISA-Eagle is a push-broom hyperspectral system with 1000-pixel swath width which can collect hyperspectral data at 2-m spatial resolution in 97 contiguous bands (2.5-nm bandwidth from 435 to 730 nm and 10-nm bandwidth from 730 to 950 nm). A total of six segments over a 50-km stretch of the Patuxent River were sampled during the summer of 2005.

This research advanced the basis of hyperspectral remote sensing under the optically complex coastal waters. The results provided necessary scientific information for using hyperspectral remote sensing in coastal and estuarine waters to monitor algal blooms. It also provides information leading toward the identification of the predominant bloom organism in estuarine waters. Knowing the type of organism as well as the synoptic view is of critical importance for coastal managers in their decisions relating to blooms, as well as being a more timely and

**2.4. Spectral analysis of water reflectance for hyperspectral remote sensing of water** 

Although hyperspectral remote sensing offers an effective approach for frequent, synoptic water quality measurements over a large spatial extent, the optical complexity of brackish estuarine water makes water quality monitoring by remote sensing a challenge. The third goal was to develop algorithms for hyperspectral remote sensing of water quality based on in situ

During a hyperspectral remote-sensing study in the Chesapeake Bay, water reflectance spectra *R*(*λ*) and the discrete water samples were collected at 11 field stations from 2008 to 2011 in the Patuxent River, a tributary of Chesapeake Bay (**Figure 20**). In this study, the relationships between the water reflectance *R*(*λ*) and water quality parameters were examined to establish empirical algorithms that could be used for retrieval of water quality parameters from air-

**Figure 20.** Map of the study area and the locations of the field stations for *in situ* measurement on the Patuxent River. The

location of the study area in Chesapeake Bay watershed is shown in the upper right map [150].

borne hyperspectral data for rapid and effective water quality monitoring.

economic method for this determination.

spectral measurement of water reflectance [150].

**quality in estuarine waters**

200 Water Quality

The ground-truthing data suggest that water reflectance *R*(*λ*) measured in this study displayed a high degree of variation because of the largely uncorrelated bio-optical properties. However, the influences on the general shape and magnitude of the reflectance by bio-optical constituents were still observed in this study. As shown in **Figure 21**, high chlorophyll-*a* concentration water shows characteristic absorption at the blue (400–500 nm) and red (600–680 nm) wavelength regions. Also, there is a phytoplankton-scattering peak at ~700 nm, which is the result from a combination of chlorophyll absorption and the strong water absorption at wavelength longer than 700 nm. This general influence on spectral reflectance by phytoplankton pigments (i.e., chlorophylls) is also consistent with our previous studies [147, 151]. Our data also suggest that TSS-dominated water shows high reflectance across a wide spectra range from 560 to 700 nm. This influence of TSS in regulating water reflectance spectra was also observed in the previous studies [152, 153]. For the water with high CDOM concentrations, its reflectance was characterized by low reflectance across the blue and green spectral regions, with very small reflectance peaks near 660 and 700 nm [151].

**Figure 21.** The general spectral features for coastal water that are dominated by high concentrations of Chl-*a*, TSS concentrations, and absorption of CDOM [147].

The general features of reflectance spectra observed in this study further provided insights for wavelength bands selection that are likely to be used for retrieval models to retrieve water quality parameters (i.e., Chl-*a*, TSS, and CDOM) [151]. Based on the in situ measurements in this study, band ration algorithms were developed to establish the relationships between water reflectance and selected water quality parameters, and this approach has been suggested as one of the most appropriate methods elsewhere [154]. Several previous studies showed strong relationships between log-transformed water quality data and reflectance values from hyperspectral data [151].

Chlorophyll-*a* model: As one of the most commonly measured water quality parameters, Chl*a* has a significant correlation with the reflectance ratio of 700–670 nm in our dataset (**Figure 22A**). This was also found to be the case in several previous studies [155]. This relationship is due to the backscattering by phytoplankton at 700 nm and the strong absorption at 670 nm. The regression model is of the following form:

**Figure 22.** Scatter plots of *in situ* log-transformed water quality parameters versus corresponding band ratio values: (A) Chl-*a*; (B) CDOM; and (C) TSS [150].

Relationship between Land Use and Water Quality and its Assessment Using Hyperspectral Remote... http://dx.doi.org/10.5772/66620 203

$$LN(\text{Chl} \cdot a) = 1.29 \left( \frac{R700}{R670} \right) + 1.83, \quad \text{with } R^2 = 0.81 \tag{1}$$

CDOM model: Because of low CDOM concentration in lot oceanic areas, the impacts of CDOM on spectral variability have been largely neglected in some previous studies [156, 157]. This study shows that CDOM concentration strongly suppressed the reflectance at blueto-green wavelength region, and then this effect declined exponentially toward the longer wavelength (**Figure 21**). Based on this CDOM spectral feature, the regression model (**Figure 22B**) was developed as the following form:

appropriate methods elsewhere [154]. Several previous studies showed strong relationships between log-transformed water quality data and reflectance values from hyperspectral data [151]. Chlorophyll-*a* model: As one of the most commonly measured water quality parameters, Chl*a* has a significant correlation with the reflectance ratio of 700–670 nm in our dataset (**Figure 22A**). This was also found to be the case in several previous studies [155]. This relationship is due to the backscattering by phytoplankton at 700 nm and the strong absorption at 670 nm.

**Figure 22.** Scatter plots of *in situ* log-transformed water quality parameters versus corresponding band ratio values: (A)

The regression model is of the following form:

202 Water Quality

Chl-*a*; (B) CDOM; and (C) TSS [150].

\*\*22B) was developed as the following form: 
$$LN(\text{CDOM}) = 0.89 \binom{R700}{R450} \text{--} 0.15, \quad \text{with } R^2 = 0.83 \tag{2}$$

TSS model: TSS represents both living organic solids (mainly phytoplankton) and inorganic suspended solids which mainly contribute to scattering of light. So, as shown in **Figure 21**, water with high TSS concentration tends to have high reflectance, especially at 500–600-nm regions. Remote-sensing algorithms for TSS reported in literature are less consistent and more dependent on the specific bio-optical conditions [158]. In this dataset, the relationships between log-transformed TSS concentration showed good correlation with the reflectance ratio of 650:420 nm. The regression model (**Figure 22C**) is of the following form:

$$\text{ratio of } 650: 420 \text{ nm. The regression model (Figure 22C) is of the following form:}$$

$$LN(\text{TSS}) = 0.23 \left(\frac{\text{R650}}{\text{R420}}\right) + 1.98, \quad \text{with } R^2 = 0.75 \tag{3}$$

Water quality mapping: The ASIA-Eagle hyperspectral imagery acquired in this study was processed and classified by using EXELIS ENVI 4.7 and ESRI ArcGIS 10.1. The ASIA imagery was first corrected geospatially and radiometrically to "at platform reflectance," then atmospheric correction was conducted on the image using the FLAASH in ENVI using the band at 820-nm wavelength to produce the image of "water reflectance." This "water reflectance" imagery is further classified by using a supervised classification in ENVI based on the selected water region of interests (ROIs). The terrestrial features were masked out to create the water-only image for further analysis. Finally, the band ratio models developed in this study were applied to these radiometric data in the image to create the pixel-level water quality maps. **Figure 23** shows the pixel-level maps of water quality (Chl-*a*, TSS, and CDOM) in a small section of the flight segment at the mouth of the Patuxent River. These maps clearly demonstrate the heterogeneity of water quality over a relatively small area. A patch of phytoplankton bloom (dinoflagellate *Prorocentrum minimum*, based on field observation) with high Chl-*a* concentration was observed in the middle of the river, while the Chl-*a* concentration in other area was relatively low. The water quality maps also show that the water with high TSS concentration was mainly located in the small creeks, coves, and river banks, suggesting that the land runoff might be the major source for TSS in water column. The CDOM absorption was also found high in the small creeks and coves around the river, as well as in the area with phytoplankton blooms. This suggests that the origin of CDOM could be both terrestrial or produced during the algal bloom.

Overall, the feasibility of hyperspectral remote sensing that can capture the fine-scale variation of water quality parameters is illustrated in **Figure 23**. In these maps, the algal bloom could be very patchy, and there was a large variation in these water quality parameters over a relatively small body of water. These techniques could provide vast water quality information that would not be seen by conventional monitoring program which probably only involve several sampling stations in the same area. However, the retrieval models in this study were derived from the dataset that is specific to the study area, which do not necessarily represent the coastal waters in other areas. The accuracy of such algorithms is always subject to the location of ground-truthing dataset.

**Figure 23.** Maps of Chl-*a*, TSS concentrations, and CDOM absorption at the mouth for the Patuxent River, summer 2005 [150].
