Sensors 2012, 12, 7778-7803; doi:10.3390/s120607778 OPEN ACCESS sensors ISSN 1424-8220 www.mdpi.com/journal/sensors Review Satellite Remote Sensing of Harmful Algal Blooms (HABs) and a Potential Synthesized Framework Li Shen 1,2,*, Huiping Xu 1 and Xulin Guo 2 1 State Key Laboratory of Marine Geology, Tongji University, Shanghai 200092, China; E-Mail: [email protected] 2 Department of Geography and Planning, University of Saskatchewan, Saskatoon, SK S7N 5C8, Canada; E-Mail: [email protected] * Author to whom correspondence should be addressed; E-Mail: [email protected]; Tel.: +1-306-966-1488; Fax: +1-306-111-966-5680. Received: 16 May 2012; in revised form: 31 May 2012 / Accepted: 31 May 2012 / Published: 7 June 2012 Abstract: Harmful algal blooms (HABs) are severe ecological disasters threatening aquatic systems throughout the World, which necessitate scientific efforts in detecting and monitoring them. Compared with traditional in situ point observations, satellite remote sensing is considered as a promising technique for studying HABs due to its advantages of large-scale, real-time, and long-term monitoring. The present review summarizes the suitability of current satellite data sources and different algorithms for detecting HABs. It also discusses the spatial scale issue of HABs. Based on the major problems identified from previous literature, including the unsystematic understanding of HABs, the insufficient incorporation of satellite remote sensing, and a lack of multiple oceanographic explanations of the mechanisms causing HABs, this review also attempts to provide a comprehensive understanding of the complicated mechanism of HABs impacted by multiple oceanographic factors. A potential synthesized framework can be established by combining multiple accessible satellite remote sensing approaches including visual interpretation, spectra analysis, parameters retrieval and spatial-temporal pattern analysis. This framework aims to lead to a systematic and comprehensive monitoring of HABs based on satellite remote sensing from multiple oceanographic perspectives. Sensors 2012, 12 7779 Keywords: harmful algal blooms (HABs); satellite remote sensing; multiple-spectral sensors; hyperspectral sensor; spectra analysis; oceanographic parameters; a synthesized framework; multiple oceanographic explanations 1. Introduction Harmful algal blooms (HABs) are deleterious phenomena characterized by the rapid accumulation of biomass in aquatic systems that have escalated worldwide in recent years. HABs have severe impacts on coastal ecosystems, fishery resources, and public health [1,2]. Three primary factors contribute to the occurrences of HABs: phytoplankton species, nutrition sources, and the dispersal mechanism. Eutrophication caused by anthropogenic activities has been determined to be one of the main sources of nutrition of HABs [3–6]. Furthermore, Anderson has pointed out that variation of oceanographic environmental parameters can also stimulate HAB events [7]. HABs can be generally classified into two categories: toxic and non-toxic [8]. The toxic species can directly release poisonous components causing paralytic shellfish poisoning (PSP), amnesic shellfish poisoning (ASP), neurotoxic shellfish poisoning (NSP) and diarrhetic shellfish poisoning (DSP). These toxic species only account for a few dozen of the thousands of known HAB species, but can cause severe diseases in human beings as well as aquaculture moralities [2]. Typical toxic species includes dinoflagellates (Alexandrium spp.), dinoflagellate (Dinophysis spp.) and diatoms (Pseudo-nitzschia spp.). The non-toxic species do not produce toxins, but can lead to aquaculture kills as a result of oxygen depletion or disturbance of the marine food web. That is why they are still called harmful algal blooms even though they produce no deadly toxins. Those phytoplankton are mainly known to include certain types of dinoflagellates (Ceratium spp., Gymnodinium spp.), diatoms (Chaetoceros spp., Rhizosolenia spp., Prymnesiophyte spp., Phaeocystis spp.) and ciliates (Mesodinium spp.), etc. Some of the aforementioned algal species can cause water discoloration when its abundance reaches a certain high level, which is usually referred to as red tides. Most red tide-forming species such as certain dinoflagellates (e.g., Ceratium dens, Ceratium divaricatum, Gymnodinium sanguineum, Protoperidinium), diatoms (e.g., Rhizosolenia setigera), prymnesiophyte flagellates (e.g., Phaeocystis) and ciliates (e.g., Mesodinium rubrum) are non-toxic [8], but some intensely toxic events of low species concentrations only dominating thin layers (subsurface blooms) do not cause the discoloration of water. Therefore, harmful algal bloom (HAB) is used as an obligatory term to encompass all the algal phenomena characterized by high biomass and/or toxin-production [9]. A growing number of global HABs have been reported at different international conferences, workshops and publications on this subject since 1974 [3,7]. Every year many coastal regions throughout the World are threatened by the serious ecological problems associated with huge economic losses and health issues caused by HABs. Hong Kong waters [10], East China Sea [3], Korean South Sea [11], Japanese Sea [12,13], the Gulf of Tokin [14], Arabian Sea [15], the coast of France, the coast of Portugal [16], New Zealand waters [17], the Galican Rias [18], Baltic Sea [19], the Gulf of Mexico [20], Washington [21], the Gulf of California [22], the coast of Florida [23], the Gulf of Maine [24], the coast of Nova Scotia [16], the coast of British Columbia [25] and the South African Sensors 2012, 12 7780 coast [26] are all areas subject to HABs with bewildering tendencies of larger spatial extents and higher frequencies. Therefore, both routine and emergency monitoring of HABs are necessary for those coastal areas, estuaries, bays and gulfs. Countries including the United States, Canada, Norway, Spain, Portugal, Ireland, China, Japan and Korea have invested a large amount of funds and efforts into HABs monitoring programs [11,18,27–30]. Therefore, the complex mechanism of HABs in the context of multiple oceanographic conditions requires a systematic understanding of the effects of different factors as well as their spatial-temporal patterns, which can help monitor and forecast HABs to reduce losses to the marine community [29]. HABs are marine phenomena characterized by large geographic and short temporal scales. Traditional efforts to identify HABs include in situ ship-surveys and laboratory analysis, but these have unavoidable limitations in time, cost, and labor which do not lend themselves to large scale monitoring over a short period [3,5]. As technology developed in 1970s, with the advantages of large-scale, real-time, and long-term monitoring, satellite remote sensing has been widely used to detect HABs as well as the oceanographic environmental characteristics that favor the formation of HABs [29]. Although it is difficult for satellite remote sensing to detect high toxicity HABs existing in thin layers, it still provides an effective tool for identifying high-biomass HABs such as red tides. However, current literature shows that the unsystematic understanding of HABs, the insufficient incorporation of satellite remote sensing, and a lack of multiple oceanographic explanations of HAB mechanisms are the major problems for remote sensing of HABs. A synthesized framework integrated with different remote sensing approaches is necessary to provide a systematical view and explanations of these complicated marine phenomena. In this study, we review the satellites sensors, techniques and algorithms for detecting HABs. Based on the challenges and opportunities found in existing remote sensing of HABs, a potential conceptual framework that combines all solvable strategies with multiple oceanographic explanations is proposed to provide a systematic way to detect HABs. 2. Satellite Remote Sensing of HABs Compared to pure water, most HABs have distinct spectral characteristics (significant absorption bands in around 500 nm, 675 nm, and reflectance peaks in 550 nm and 700 nm) [31], which are caused by the dramatic increase of phytoplankton biomass. In particular, the chlorophyll fluorescence peak at 683 nm is a special characteristic of HABs which can be used to effectively separate it from other types of water. However, for some HABs the reflectance peak is shifted to 700 nm which is not caused by the fluorescence effect, but is contributed to by the elevated back scattering as a result of the increased phytoplankton density, or at least is a combination of the fluorescence and elastic scattering effects [32,33]. Different HAB species have distinct spectral characteristics. Zhao et al. concluded that three main different spectral characteristic types (the single-peak, the double-peak and the wide peak) exist for most HAB species. The single peak is characterized by a single reflectance peak at 680–750 nm (e.g., Heterosigma akashiwo, Ceratium furan) while the double-peak type has a strong reflectance peak at around 700 nm and a weak peak at around 800 nm (e.g., Gymnodinium spp., Pyramimonas spp.). The wide-peak type has a relatively broad reflectance peak distributed from 680 to 900 nm (e.g., Platymonas spp., Nitzschia closterium and Chlorella spp.) [34]. The aforementioned spectral responses are shown more obviously by intense HABs than in water with normal Sensors 2012, 12 7781 phytoplankton concentrations. These different characteristics can allow various satellite system with different spectral resolutions to detect different HABs by developing numerous algorithms. 2.1. Data Sources and Their Suitability for Monitoring HABs 2.1.1. Multiple-Spectral Sensors Since the first ocean remote sensing instrument, Coastal Zone Color Scanner (CZCS), was launched in 1978, a number of ocean remote sensing missions including Sea-viewing Wide Field-of-view Sensor (SeaWiFS), Moderate Resolution Imaging Spectrometer (MODIS), Medium Resolution Imaging Spectrometer (MERIS), Ocean Color Monitor(OCM) series and Hyperion, were developed to measure various marine biophysical and biochemical parameters (Table 1). These remote sensors supply a series of ocean color imagery which have been successfully applied in pigment concentration estimation and Sea Surface Temperature (SST) retrieval, playing a vital role in marine environmental management. The Advanced Very High Resolution Radiometer (AVHRR), a sensor carried on National Oceanic and Atmospheric Administration (NOAA) and mainly designed for climate change study, can also provide SST for HABs detection. Previous studies have shown great potential of these satellite data for remote sensing of HABs monitoring. CZCS (1978–1986) CZCS, operated from 1978 to 1986, has six spectral bands (443 nm, 520 nm, 550 nm, 670 nm, 750 nm and 11.5 µm), five of which are located in the characteristic region of HABs spectra, providing the first opportunity for satellite observation of HABs by quantifying phytoplankton pigment concentrations [20]. In 1978, CZCS was successfully utilized to detect Karenia brevis blooms by identifying a highly chlorophyll discolored anomaly in the Gulf of Mexico [35]. Several subsequent studies on monitoring HABs by CZCS were conducted by [23,36,37]. However, as [38] mentioned, due to the infrequency and short living period of HABs, CZCS could not be used to routinely monitor HABs because of its delays in data collecting and processing, which led to relatively little research being found in publications [20]. SeaWiFS (1997–2010) As the second generation of ocean sensors, SeaWiFS was launched in SeaSTAR in 1997, which stopped collecting data in 2010. Compared with CZCS, SeaWiFS has more potential for initiating routine monitoring of chlorophyll concentration which is considered an effective means to identify HABs. The advantages of SeaWiFS can be demonstrated by the additional four spectral bands (412 nm specific to the absorption of yellow substances, 490 nm sensitive to the chlorophyll variation, 765 nm and 865 nm more suitable to atmospheric correction), and daily imagery can also enable the real-time detecting of variation in short-lived HABs [31]. SeaWiFS data has contributed significantly to global HAB monitoring in the past 15 years. SeaWiFS imagery integrated with other ancillary datasets played an important role in monitoring Karenia brevis blooms of the CoastWatch program in September 1999, Florida, initiated by the National Oceanic and Atmospheric Administration's (NOAA). Sensors 2012, 12 7782 Table 1. Characteristics of historical and current ocean-color sensors (International Ocean Color Coordinating Group. http: http://www.ioccg.org/sensors). Sensor Agency Satellite Operating Dates Spatial Resolution (m) Bands Spectral Coverage (nm) Obit CZCS NASA (USA) Nimbus-7 (USA) 24/10/78–22/06/86 825 6 433–12,500 Polar SeaWiFS NASA (USA) OrbView-2 (USA) 01/08/97–14/02/11 1,100 8 402–885 Polar MODIS-Terra NASA (USA) Terra (EOS-AM1) Launch 18/12/99 250/500/1,000 36 405–14,385 Polar MODIS-Aqua NASA (USA) Ma’an Aqua (EOS-PM1) Launch 04/05/02 250/500/1,000 36 405–14,385 Polar Polder CNES (France) ADEOS (Japan) 17/8/96–29/6/97 6,000 9 443–910 Polar Polder-2 CNES (France) ADEOS-II (Japan) 14/12/02–24/10/03 6,000 9 443–910 Polar Polder-3 CNES (France) Parasol Launch 08/12/04 6,000 9 443–1,020 Polar MOS DLR (Germany) IRS P3 (India) 21/03/96–31/05/04 500 18 408–1,600 Polar MERIS ESA (Europe) ENVISAT (Europe) Launch 01/03/02 300/1,200 15 412–1,050 Polar OCTS NASDA (Japan) ADEOS (Japan) 17/08/96–29/06/97 700 12 402–12,500 Polar GLI NASDA (Japan) ADEOS-II (Japan) 14/12/02–24/10/03 250/1,000 36 375–12,500 Polar OCI NEC (Japan) ROCSAT-1 (Taiwan) 27/01/99–16/6/04 825 6 433–12,500 Polar OSMI KARI (Korea) KOMPSAT-1/Arirang-1 (Korea) 20/12/99–31/1/08 850 6 400–900 Polar GOCI KARI/KORDI (South Korea) COMS Launch 26/6/10 500 8 400–865 Geostationary CMODIS CNSA (China) SZ-3 (China) 25/03/02–15/09/02 400 34 403–12,500 Polar CZI CNSA (China) Hy-1A (China) 15/05/02–01/04/04 250 4 420–890 Polar COCTS CNSA (China) Hy-1A (China) 15/05/02–01/04/04 1,100 10 402–12,500 Polar CZI CNSA (China) Hy-1B (China) Launch 11/04/07 250 4 433–695 Polar COCTS CNSA (China) Hy-1B (China) Launch 11/04/07 1,100 10 402–12,500 Polar OCM ISRO (India) IRS-P4 (India) Launch 26/05/99 360/4,000 8 402–885 Polar OCM-2 ISRO (India) Oceansat-2 (India) Launch 23/09/09 360/4,000 8 400–900 Polar MMRS CONAE (Argentina) SAC-C (Argentina) 21/11/00 ~ 2009 175 5 480–1,700 Polar HICO ONR and DOD JEM-EF Launch 18/09/09 100 124 380–1,000 51.6°, Space Test Programme Int. Space Stn. 15.8 orbits p/d Sensors 2012, 12 7783 Such an effort allowed the distribution of “Harmful Algal Bloom Bulletins” nationwide for effective responses to HABs in the Gulf of Mexico [20]. The potential of SeaWiFS data has also been explored in the G. catenatum blooms in New Zealand waters [39], the South China Sea [14], the Pearl River Estuary [10], and the Baltic Sea [40] based on the abnormal chlorophyll concentration shown by the imagery. Unfortunately, SeaWiFS has no bands designed in the 683 nm region (the florescence peak of chlorophyll) which is a significant spectral indicator of HAB occurrence [33]. In addition, inappropriate atmospheric correction resulting from poor knowledge of aerosol conditions, absorption by colored dissolved organic matter (CDOM), scattering by inorganic suspended components, and reflection by a shallow bottom can all impact the optical properties detected from the imagery [40]. Additionally, lack of contemporaneous in situ validation data can restrict the accuracy of applying SeaWiFS in HAB detection [3]. Reinart and Kuster in particular have emphasized the limitations of SeaWiFS data in the detection of heavy algal blooms because of the high water-leaving radiance in near infrared regions [40]. The relatively coarse spatial resolution of SeaWiFS data limits the study area of HABs to a large spatial scale (>1,000 km2) which is often found in coastal areas such as the East China Sea, Bohai, the Gulf of Mexico and the Korean coastal waters [3,20,41]. MODIS (Terra/1999–Present and Aqua/2002–Present) MODIS, a third generation sensor for oceanic satellite observations, was launched on both Terra and Aqua satellites in 1999 and 2002, respectively. It can provide daily imagery of 36 bands at three spatial resolutions (250 m for bands 1 and 2, 500 m for bands 3 through 7, and 1,000 m for bands 8 through 36). Especially bands 8 to16 in the 405–877 nm spectral region are specifically designed for studying ocean color, phytoplankton concentration as well as biogeochemistry [31]. Compared with SeaWiFS, the primary advantage of MODIS is that the particularly designed fluorescence band (676 nm) can be used to detect HABs based on a fluorescence line height (FLH) calculation for the coastal optically-complex water, the optical spectra of which is dominated by CDOM [34]. Such exploration has been conducted in the Gulf of Mexico and in the Bohai Sea, and the results showed a good correlation between the satellite-derived information and in situ measurement due to the little impacts of atmosphere and suspended sediments [33,42]. Tomlinson et al. applied MODIS FLH imagery in a K. brevis bloom in the Gulf of Mexico and found 71% of the blooms could be identified [43]. SST information can also be obtained at the same time as chlorophyll concentration is retrieved for the same HAB occurrence which greatly improves the accuracy of HAB detection by integrating analysis of multiple satellite information [44]. Furthermore, MODIS also provides the opportunity to estimate the primary production for algal bloom water because of the availability of all the necessary parameters including chlorophyll concentration, SST, daily photosynthetically active radiation, and daily diffusion attenuation coefficient derived from MODIS. The spatial resolution of MODIS data can guarantee the accuracy for HABs with an area more than 1,000 km2 [44]. However, MODIS imagery suffers severely from the sunglint problem. Because the sensor was designed for observations of atmosphere, land and ocean, it does not tilt toward the track to avoid the solar flare influence. Another limitation is that the fluorescence region (676 nm) in MODIS is a little farther from the actual chlorophyll peak (683 nm), especially when the chlorophyll concentration is higher [34]. Sensors 2012, 12 7784 MERIS (2002–Present) MERIS, another popular third generation satellite sensor, was launched on the ENVISAT-1 satellite in 2002 by the European Space Agency (ESA). It has 15 spectral bands (350–1,040 nm) at 300 m spatial resolution covering all the regions for ocean studies. Compared with MODIS, the fluorescence bands (681 nm and 709 nm) are closer to the actual chlorophyll peak position, and are, therefore, more suitable for detecting HABs based on FLH methods. MERIS can extract 78% of fluorescence information while MODIS can extract a mere 57% [45,46]. Additionally, the 620 nm band is more sensitive to the suspended materials and the 900 nm band responds better to the water content in the atmosphere. All of these advantages can improve the accuracy of radiometric correction of satellite imagery for obtaining reliable information of HABs [47]. In addition, MERIS band 6 (620 nm) and band 7 (665 nm) are respectively more sensitive to the absorption region (603 nm) and reflectance peak (650 nm) of cyanobacterial blooms while MODIS does not have this characteristic spectral band [48]. The suitability of MERIS for cyanobacterial bloom identification from other types of HABs compared to MODIS has also be demonstrated by Kuster et al. and Koponen et al. [49]. MERIS was also found to have more reasonable band design than MODIS for identifying other HABs including Dicrateria zhanjiangensis Hu, Pyramimonas sp. and Nitzschia closterium species. Furthermore, the spatial resolution of MERIS is superior to that of SeaWiFS and MODIS, which allows for more accurate detection of HABs with an area of less than 1,000 km2 in comparatively small water areas such as lakes and rivers. The disadvantage of using MERIS data for HAB studiers is that since ENVISAT is a commercial satellite providing no free data to researchers, the availability of data is limited, which restricts the operational observation of HABs [47]. Many potential uncertainties about the future status of the MERIS instrument pose another threat to its data availability for use in HAB detection [29]. AVHRR (1978 to Present) AVHRR was aimed to study the global climate and environmental change with high temporal resolution (daily) and moderate spatial resolution (1.1 km × 1.1 km). There are four bands in the first AVHRR carried on the TIROS satellite (1978) and AVHRR/2 was enhanced to five bands (0.6, 0.9, 3.5, 11 and 12 um, respectively) initially aboard NOAA-7 (1981) [31]. Due to the operational real-time capability and two visible bands sensitive to phytoplankton scattering in coastal turbid water, AVHRR data were greatly explored in studying large-area HABs as shown by the amount of published literature [20,25,50]. More frequently, AVHRR data are exploited for SST information retrieval due to the thermal bands allowing information extraction of water mass movement associated with HABs [10,20]. However, compared with MODIS and MERIS, fewer spectral channels and lower spatial resolution limit the operational monitoring of HABs, AVHRR can merely be used to detect large-scale HABs (more than 1 km2) and is incapable of discriminating specific phytoplankton species within the HABs [47]. OCM (IRS-P4)/ OCM-2(Oceansat-2) Ocean Color Monitor (OCM) was launched onboard the Indian Remote Sensing Satellite IRS-P4 (Oceansat) in 1999 and completed its mission in 2000. It was specifically designed for oceanic Sensors 2012, 12 7785 observation including chlorophyll distribution, phytoplankton blooms, suspended matter movement, and atmospheric aerosol identification. It had eight spectral channels (404–882 nm) providing imagery with a spatial resolution of 360 × 236 m every two days for the same study area [51]. OCM-2, a sensor with the same configuration as OCM, was carried by Oceansat-2 in 2009 as the following mission of Oceansat. OCM series imagery have the common characteristic spectral bands (414 nm, 440 nm, 510 nm, 556 nm, 668 nm) of HABs, so it can be applied to general HAB detection. Saragngi and Mohammed detected dinoflagellate algal blooms in the Kerala coastal and Calicut waters by exploring OCM imagery based on the OC2 empirical algorithm, showing good correlation with the in situ data [52]. Utilization of OCM-2 data for HAB detection is rarely found in the published literature, perhaps due to the relatively short period since its launch in 2009. Although the spatial resolution of OCM series imagery is superior to SeaWiFS and MODIS (bands 3 through 36), it is still incapable of accurate identification of HABs due to the lack of a specific fluorescence band for developing FLH methods. 2.1.2. Hyperspectral Instruments Hyperspectral instruments are deemed a promising tool for future harmful algal bloom detection due to their continuous spectrum which allows for more accurate quantification of phytoplankton characteristics [29]. Previous literature has shown the potential of both in situ spectroradiometers (e.g., ASD Fieldspec) and onboard sensors in identifying HABs. Lee and Carder used field collected hyperspectral remote sensing reflectance to derive absorption spectra of phytoplankton pigments with an accuracy of 78.6%, which provided great possibility of retrieving more pigments information besides chlorophyll concentration from multiple-spectral sensors [53]. In addition, Randolph et al. showed that in situ hyperspectral reflectance collected by ASD Fieldspec can effectively estimate chlorophyll concentration and phycocyanin absorption characteristics of cyanobacteria HABs [54]. The most frequently used hypespectral satellite sensors are Hyperion and the Compact High Resolution Imaging Spectrometer (CHRIS). As the first civilian hyperspectral imaging spectrometer initiated by NASA’s New Millennium Program (NMP), Hyperion was launched on the Earth Observing 1 (EO-1) satellite in November 2000 [55]. Compared to the aforementioned discrete bands of multispectral satellite sensors, Hyperion can provide spectrally continuous data in 196 spectral bands (355–2,577 nm) with each 10 nm width band comprising the visible through shortwave infrared region. In addition, the spatial resolution of Hyperion (30 m) is as high as that of the Landsat Thematic Mapper (TM); however, the spectral resolution for the latter is far inferior to the former for HAB detection. Although Hyperion was designed specifically for land applications, the spectral channels can cover the entire region for water remote sensing as well [55]. As shown by [56], Hyperion has been used to monitor cyanobacterial blooms in the western part of the Gulf of Finland in 2002 by estimating chlorophyll concentration based on a bio-optical model. This study also indicated that the chlorophyll derived from multispectral sensors yields an underestimated value due to the limitation of their spatial resolutions. In spite of its potential designed for coastal water monitoring, Hyperion still suffers from several shortcomings. The relatively longer revisit period (16 days) as well as the small coverage (7.7 × 185 km) do not allow for routine real-time HAB monitoring. Besides, the poor signal to noise ratio also restricts its wide application in HAB communities. However, Hyperion data can still Sensors 2012, 12 7786 serve as important ancillary data for HAB detection by other superior instruments [40]. CHRIS, onboard the Project for Onboard Autonomy (Proba) satellite launched by the ESA in October 2001, is capable of acquiring both hyperspectral and multi-angular data at a spatial resolution of 18 m in a wavelength range of 415–1,050 nm with a revisit period of seven days. Simultaneous observation in 19 bands out of the total 62 bands can provide environmental information for both land and coastal monitoring [57]. The potential of CHRIS in retrieving chlorophyll concentration and estimating phytoplankton biomass has been proven by [58] in monitoring cyanobaterial blooms based on an empirical model. Despite the fact that airborne hyperspectral instruments with higher spatial and spectral resolutions can improve HAB detection, they are still far from offering routine and real-time monitoring due to the expensive flight costs and the limits of the geographical scope that the sensor can cover. Therefore, airborne hyperspectral instruments such as Airborne Imaging Spectrometer for Applications (AISA) and Push-broom hyperspectral imager (PHI) are more frequently used to validate satellite-derived information or detect HABs in small area such as bays, lakes or along the shore [49]. 2.2. Available Remote Sensing Techniques for Monitoring HABs The principle remote sensing techniques for detecting HABs are interpretation of discoloration, spectral analysis and oceanographic parameters retrieval. 2.2.1. Interpretation of Discoloration True-color and false-color satellite imagery generated by combining different spectral layers can be used to identify the presence of water discoloration caused by HABs [59,60]. True-color composite imagery has more advantages for visual interpretation since it can reflect the actual color of algal blooms, which allows for identification of specific phytoplankton species directly based on some empirical knowledge of the species. This has been successfully proven by previous studies on remote sensing of HAB to detect Skeletonema costatums [44,61]. In addition, harmful algal blooms caused by other phytoplankton species including coccolithophores, trichodesmium and cyanobacteria can also been effectively identified by observing the discoloration of waters [19,62–64]. However, it is difficult to obtain quantitative information for a HAB merely based merely on the observation of discoloration. Besides, not all HABs produce water discoloration since some color anomalies are caused by other materials such as sediment or CDOM. Therefore, examining discoloration is not totally reliable, particularly when the study area is located in an optically complex coastal area without sufficient field data for validation, but the discoloration method can still provide some general information about a potential bloom such as the location and extent of the event [60]. 2.2.2. Optical HAB Algorithms Based on the premise that a unique spectral characteristic corresponds to a specific harmful bloom, remote sensing optical approaches for detecting HABs can be categorized into two major types. One is aimed at exploring the optical properties (absorption, backscattering and reflectance) of each component (CDOM, suspended sediment, water and chlorophyll) present in the HAB water to Sensors 2012, 12 7787 establish equations which can indicate the reliable relationship between the optical characteristics of each component and the total sensor signals [65–67] as follows: where R, , , and refer to the irradiance reflectance of water, the spectral band, the total absorption and the total backscatter at spectral respectively. and are represented by: + + + , and + + where w, c, d, p, s, are water, CDOM, detritus, phytoplankton and suspended sediment, respectively [68]. This relationship can be obtained generally by developing empirical, semi-analytical or radiation transfer models. However, as Stumpf and Tomlinson indicated, the performance of those algorithms is determined by the stability of each component’s spectral characteristics [29]. For this optical method, many in situ data (both the spectral characteristics and component percentage) are required for the equation establishment and validation. This method is only feasible for investigating HABs in Case 1 waters (deep ocean with chlorophyll pigments as the dominant component, rarely influenced by organic and inorganic components) but not applicable in turbid coastal waters (Case 2 water) because of the complex optical signals contributed by CDOM and particulate inorganic materials (POM). Also, a significant error may be yielded if the accuracy of the atmospheric correction of the imagery cannot be guaranteed, which indicates that such satellite data without good radiometric correction is not suitable for HAB detection using this method [3]. Numerous spectral band algorithms have been developed to overcome the limitation of the standard optical algorithms for HABs detection. Those algorithms include the single band model, two bands difference/ratio model, and multiple bands difference/ratio model [47]. The single band method has been exploited for detecting coccolithophore HABs in the northeast coast of the Atlantic by [69] who set a threshold for the reflectance of the AVHRR first band (580–680 nm). When the threshold is reached, there is a potential risk of a coccolithophore bloom. The same method based on AVHRR data applied in the Baltic Sea by Kahru et al. also showed effectiveness in identifying the scope and frequency of nodularia blooms [19]. For MODIS data, Kuster et al. demonstrated that band 1 (620–670 nm) and band 2 (841–876 nm) are sensitive to the variation of cyanobacterial blooms. This was further supported by Duan et al. who found that a threshold of 0.1 for MODIS band 2 reflectance could be used as the indicator of cyanobacterial blooms [48,70]. The two bands difference/ratio model, the earliest of which for CZCS was developed by [36], who suggested that a threshold for the ratio of MODIS band 1 and band 2 had significant effectiveness in coccolithophore bloom detection. Also Stumpf and Tyler demonstrated that a threshold for the ratio of AVHRR band 1 and band 2 is capable of identifying HABs on the west coast of Canada when the chlorophyll exceeds 5 mg/m3 [50]. By taking the idea of Normalized Difference Vegetation Index (NDVI) in vegetation remote sensing, a NDVI algorithm was developed for HAB detecting by utilizing the reflectance of AVHRR band 1 and band 2 [71]. Multiple bands difference/ratio algorithms are more established for SeaWiFS data. Mao and Huang established a model C = (R(band1) − R(band3))/(R(band5) − R(band3)) for detecting gymnodinium HABs in the East China Sea while Gu et al. developed C = (R(band5) − R(band4))/(R(band4) − R(band3)) for detecting Skeletonema costatum HABs [72,73]. To overcome the inefficiency of optical models in coastal turbid waters, Ahn et al. utilized three water-leaving radiances