**3.1 Chlorophyll (Chl-a) and algal blooms**

Most algal species are nontoxic and are always present in coastal and open oceans. Planktons are the base of the marine food chain [22]. But, algae do not have to produce toxins to be harmful to the environment. The accelerated growth of algae produces a large amount of biomass which blocks sunlight and produces an anoxic or hypoxic condition (dissolved oxygen is depleted from the water column), which is hazardous to marine life. Algal blooms also affect coastal operations such as movement of ships, coastal tourism, and coastal sports (**Figure 3**). Algal blooms can persist from a few days to more than a month and spatially they may extend from a few meters to tens of kilometers.

 The impact of algal blooms on marine life depends largely on the algal species involved. In situ field data collected using vessels are important for determining the algal species and level of toxicity during the bloom. However, field data are always limited for estimating the spatial extent as well as the dispersion. Detection of algal bloom by estimating the Chl-a concentrations using satellite imagery has been well-researched, as remote sensing has been used to observe ocean primary productivity since the launch of CZCS in 1978. High spatial and temporal resolutions are the main requirements of remote sensing data to study the variability in ocean and coastal Chl-a. By comparing a time series of satellite images, researchers can evaluate the spatial and temporal variations in Chl-a concentration during the bloom. This can also help to understand the dynamics of blooms. However, there are still certain conditions for using optical remote sensing to detect Chl-a, including (i) no or low cloud cover, (ii) the bloom should be near to the surface, and (iii) the bloom must cause the coloration of the water.

 Optical remote sensing can observe the coloration of water due to algal pigments. In the open ocean, the color of water is mainly determined by phytoplankton; hence, it is relatively simple to develop algorithms using a bio-optical approach and remote sensing reflectance [22]. In the open ocean, Chl-a can be retrieved from the ratio of blue and green wavelengths as Chl-a absorption is sensitive to blue wavelength and reflectance peak occurs in the green wavelength region [22]. However, in coastal waters, the color of water also depends on organic matter such as NAP, CDOM, and inorganic solids, and consequently it is more complex to determine accurate Chl-a concentrations in coastal/turbid waters. Researchers have demonstrated that waters with increased Chl-a concentrations show a lower

#### **Figure 3.**

*Spread of green algae along the coast of Qingdao in 2008, when summer Olympics was planned in this coast (source: Corey Sheran/Flickr) (right) and algae visible in MODIS false color image (shortwave, NIR, and Red) (source: MODIS rapid response project at NASA/GSFC) (left).* 

 spectral response at short wavelengths especially in the blue wavelength regions [41]. This is due to increased absorption of red and blue wavelengths during photosynthetic process. **Figure 4** shows the reflectance of water with increasing Chl-a concentrations. Thus, in coastal waters, the red/NIR ratio is more effective for retrieval of Chl-a due to the presence of suspended solids and the increased spectral response of Chl-a pigments at longer wavelengths [43].

 Narrow spectral bandwidth is a necessity for accurate retrieval of Chl-a concentrations [7]. The height of the spectral peak between 700 and 710 nm is used as a proxy for phytoplankton biomass [44]. Many researchers have used broad wavelength data (i.e., Landsat, HJ-1A/1B) as input to regression and neural network approaches for estimating Chl-a, achieving reasonable accuracy (70–90%) [9, 19, 45, 46]. **Table 3** shows some studies and datasets used to study Chl-a in marine regions. Lim and Choi [19] found that green and NIR bands of OLI are highly correlated with Chl-a (R = 0.71) in Korean waters. Nazeer and Nichol [46] also used the red/blue ratio to retrieve Chl-a with high accuracy (R = 0.85). Gurlin et al. [43] calibrated three models for Chl-a concentrations from 0 to 100 mg m<sup>−</sup><sup>3</sup> using two bands (red and NIR) of MERIS and MODIS reflectance data. They found that a simple two-band model achieved a higher accuracy than a complex three-band model. Moses et al. [51] also calibrated a red-NIR algorithm for high Chl-a concentrations in productive turbid waters. **Figure 5** shows Chl-a concentrations in highly turbid Pearl River Estuary and connecting rivers, derived using high-resolution MSI data with the method of Moses et al. [51].

 Recently, machine learning approaches taking advantages of reflectance in all bands have also been applied using Landsat [45, 52] and GOCI data [28]. Our work also shows the potential use of Landsat TM, ETM+, and OLI with a machine learning approach to estimate Chl-a in coastal waters (**Figure 6**). We have evaluated three machine learning models to estimate Chl-a in the coastal waters of Hong Kong, of which artificial neural networks (ANN) performed best resulting in higher R (0.91) and lower RMSE (1.4 μg/L) than models based on support vector regression (SVR) and random forest (RF) algorithms. Chlorophyll indices such as the cyanobacteria index [53], maximum chlorophyll Index [54], and maximum peak height algorithm [55] have been demonstrated the robustness for detecting algal blooms and surface scum in coastal waters. Lunetta et al. [56] described the potential of using cyanobacteria index to measure cyanobacteria cell counts in bloom situations using MERIS data. Nazeer et al. [57] used board waveband band data (Landsat TM, ETM+, and HJ-1A/1B CCD) along with meteorological data as inputs to an artificial neural network model to map phytoplankton cell counts

**Figure 4.**  *Changing spectral response of water with different levels of chlorophyll concentration [42].* 


#### *Detection and Monitoring of Marine Pollution Using Remote Sensing Technologies DOI: http://dx.doi.org/10.5772/intechopen.81657*

#### **Table 3.**

*Methods used to retrieve Chl-a using remote sensing data in the river and marine waters.* 

#### **Figure 5.**

*Chl-a concentration observed in the Pearl River Estuary and its connecting rivers on 31 December 2017.* 

during a bloom in the complex coastal waters of Hong Kong and validated the model in two lakes in the United States and Japan.

Synthetic aperture radar (SAR) data can also be used to detect large algal blooms in cloudy weather as algal blooms may appear as an area of low backscatter compared to surrounding water surfaces [50].

**Figure 6.** 

*Comparison of measured and predicted values from three machine learning models. (a) Chl-a concentration using artificial neural network, (b) Chl-a concentration using support vector regression, and (c) Chl-a concentration using random forest.* 

#### **3.2 Turbidity, total suspended sediments, and stormwater runoff plumes**

Turbidity is an optical property of water and is highly influenced by concentrations of suspended and dissolved organic and inorganic materials in water, including Chl-a, SS, and CDOM. SS is mainly responsible for the light scattering, whereas CDOM and Chl-a control the light absorption properties of water [58].

Turbidity and TSS are two important variables of marine systems studies because of their direct linkages with photosynthetically available radiation, which affects the growth of plankton and other algae [41]. Turbidity has also been used to measure fluvial SS concentrations in rivers and river plumes [59]. These fluvial SS loads are rich in nutrients and considered a cause of eutrophication. So, it is vital to have time series records of suspended sediment concentrations for better understanding of land-ocean interactions. High SS loads negatively affect aquaculture [59] and are hazardous to benthic invertebrates [60]. These parameters are also associated with the diffuse attenuation coefficient (penetration of light, in the blue-green region of the spectrum, through water column) and Secchi disk depth (a measure of water transparency) [41]. For all these reasons, turbidity and TSS concentrations are considered to be critical parameters in the study of marine systems.

 Ocean color remote sensing techniques are widely used to monitor spatiotemporal variations in SS concentration and for mapping of water turbidity. **Figure 7**  shows the changes in ocean color due to high sediment loads in the Yangtze River Estuary [60] and the Pearl River Estuary [61]. It is suggested that an algorithm using single bands provides a good estimation of TSS concentrations if an appropriate band is used [62]. Moreover Novo et al. [63] and Curran et al. [64] have demonstrated that a single-band approach may be adopted when water reflectance in the single band has a linear relationship with TSS concentrations. However, coastal water often consists of a complex mixture of substances and results in large variations in reflectance. In this case, multiple spectral bands should be adopted for TSS retrieval [62, 65, 66]. These methods using band arithmetic can achieve high accuracy around 80% for retrieving TSS concentrations in complex waters [67, 68]. The peak of the reflectance curve shifts from the green region to the red region with increasing concentration of dissolved and suspended matter; and water starts reflecting significantly in NIR region [21] (**Figure 8**). For water with high TSS concentrations, the spectral region between 600 and 900 nm should be used. Several studies using Landsat TM, ETM+, and OLI show that the blue, green, red, and NIR bands are useful for the determination of TSS [8, 19, 68–70]. Literature also shows that TM, ETM+, OLI, and MODIS are the most frequently used sensors for developing algorithms to study seasonal TSS variability in coastal and estuarine

*Detection and Monitoring of Marine Pollution Using Remote Sensing Technologies DOI: http://dx.doi.org/10.5772/intechopen.81657* 

#### **Figure 7.**

*Terra-MODIS true color image, captured on 16 September 2000, shows the sediment plume of the Yangtze River Estuary (left). The Sentinel-2 true color image, captured on 31 December 2017, shows high sediment concentrations in the Pearl River Estuary (right).* 

#### **Figure 8.**

*Remote sensing reflectance (Rrs) spectra of water containing different suspended solid concentration (mg/L) [21].* 

areas, due to the large amount of archived remote sensing data [24, 71, 72]. The recently launched MSI sensor onboard Sentinel-2A and Sentinel-2B provide high spatial resolution of 10–20 m with a high temporal resolution of 5 days. The high spatial resolution (10 m) red and NIR bands are capable of routine monitoring of TSS concentration and turbidity in narrow bays, rivers, and inlets. **Figure 9** shows the suspended matter concentrations, and **Figure 10** shows turbidity in the Pearl River Estuary and connecting rivers using MSI data with algorithms of Nechad et al. [62] and Nechad et al. [73], respectively.

Methods and algorithms for estimation of TSS and turbidity have been evolved from simple methods such as linear/nonlinear regression and principal component analysis (PCA) to relatively complex techniques such as genetic algorithms and ANN. Nazeer and Nichol [68] initially developed a regression model resulting

#### **Figure 9.**

*High levels of suspended matter concentration were observed in the Pearl River Estuary and its connecting rivers on 31 December 2017.* 

 in an RMSE of 2.60 mg/L. Later, Nazeer et al. [52] evaluated the potential of a machine learning model for estimating TSS in the complex coastal area of Hong Kong achieving an RMSE of 4.59 mg/L. Our work of machine learning models with Landsat TM, ETM+, and OLI data in the same area also shows promising results for estimation of TSS (**Figure 11**). In our work, ANN outperformed the other two machine learning approaches, SVR (support vector machine) and RF (random forest), resulting in the lowest RMSE of 2.8 mg/L. **Table 4** includes some studies and methods used to study TSS in rivers, bays, estuaries, and relatively open coastal waters.

Stormwater runoff is also a large source of marine pollution as runoffs and pollutants from the urban watershed enter into the coastal environment after rainstorms. Stormwater runoff and municipal wastewater plumes may sometimes be overlooked due to persistent cloud cover in optical imagery. These types of

*Detection and Monitoring of Marine Pollution Using Remote Sensing Technologies DOI: http://dx.doi.org/10.5772/intechopen.81657* 

#### **Figure 11.**

*Comparison of measured and predicted values from three machine learning models. (a) TSS concentration using artificial neural network, (b) TSS concentration using support vector regression, and (c) TSS concentration using random forest.* 


#### **Table 4.**

*Methods used to retrieve TSS using remote sensing data in marine waters.* 

 runoff are often detectable via SAR as they deposit surfactants on the sea surface, smoothing the small gravity waves and thus producing an area of low backscatter in comparison to the surrounding sea surface [74]. DiGiacomo et al. [74] used highresolution SAR to monitor such plumes in the Southern California Bight. In their study, the dynamics of runoff plume was modeled using SAR images together with meteorological data as a function of cumulative event discharge, timing of the peak flow, and total storm precipitation. Holt et al. [75] used multi-platform SAR data along with MODIS and precipitation data to study a stormwater plume and its flow direction.

### **3.3 Oil spill**

A large oil spill from tankers causes not only significant economic loss but also destruction to the aquatic ecosystem. After the spill, oil undergoes several processes such as spreading, evaporation, dissolution, drifting, photolysis, biodegradation, and the formation of oil-in-water and water-in-oil emulsions [76].

Owing to the dynamic spreading nature of the spill, both remote and stationbased sensors are essential for comprehensive and effective monitoring. Airborne survey of an oil spill can be carried out by side-looking airborne radar (SLAR), laser fluorosensor (LF), and ultraviolet and thermal infrared video cameras. Ultraviolet, microwave, thermal, and optical airborne sensors all exhibit the ability to detect oil spills [6]. Ultraviolet sensors are sensitive to oil thickness of 0.01–0.05 μm. Oil

#### *Detection and Monitoring of Marine Pollution Using Remote Sensing Technologies DOI: http://dx.doi.org/10.5772/intechopen.81657*

appears as a bright target in this region of the spectrum, and brightness increases with the thickness of the oil. Optical sensors can measure thicker oil (2–500 μm) and are able to detect oil dispersed in water, whereas thermal infrared sensors measure oil with a thickness of about 10–50 μm [34]. Airborne LF and microwave radiometers (MWR) are considered to be the most appropriate sensors for oil spill detection. SLAR, ultraviolet, and thermal video cameras were used to identify areas of thick oil during the Sea Empress oil spill in 1996. Oil also undergoes weathering and aging. Multispectral satellite images, taking advantage of fluorescence characteristics of oil, can detect spills and assess the levels of weathering of the oil [31].

 Spaceborne synthetic aperture radar (SAR) is commonly used for ocean pollution monitoring, especially oil spills. **Table 5** includes some SAR-equipped satellites used for oil spill detection. The advantage of SAR is the capability to take measurements during all day and all-weather conditions. Therefore, they are considered superior to optical sensors in this application [5]. The spreading trend of oil highly depends on wind direction and speed. An oil spill would break up and disperse if the wind speed is greater than 10 m/s [74]. DiGiacomo et al. [74] used ERS-2 SAR and RADARSAT-1 SAR images to map oil spills in the Southern California Bight. Shirvany et al. [77] evaluated the potential of different polarizations using RADARSAT-2 data for oil spill detection in the Gulf of Mexico. In another study, ENVISAT data was used effectively as an input to a hydrodynamic model to track the fate of oil after the Kerch Strait oil spill in 2007 [78]. **Figure 12** shows an incident of large oil spill on the Galicia coast [79] and the Korean coast [80] for which spaceborne SAR data was used to access the coverage areas and the damage caused by the spills.
