Application of Optical Remote Sensing in Harmful Algal Blooms in Lakes: A Review

Abstract

Harmful algal blooms (HABs) are a critical global issue, severely impacting aquatic ecosystems, public health, and economies. Optical remote sensing (ORS) has emerged as a prominent tool for HABs monitoring, providing operational capabilities for quantifying spatiotemporal dynamics through cost-effective observation platforms. This review systematically synthesizes recent advancements in ORS technologies, encompassing (1) novel sensor development, (2) advanced data analytics frameworks, and (3) the synergistic integration of multi-scale observation platforms (satellite–airborne–ground). The analysis critically evaluates (a) spectral signature identification methodologies and (b) persistent challenges including suboptimal spatiotemporal resolution, atmospheric correction uncertainties, and limited model generalizability across heterogeneous aquatic systems. Emerging technologies, including machine learning, spatial–temporal data fusion, and high-performance sensors, are explored as potential solutions to overcome these challenges.

1. Introduction

Lakes are important water resources [1,2], covering 3% of the global land area [3,4]. Lakes provide habitats for a large number of plants and animals and help maintain biodiversity [5]. However, aquatic systems are increasingly threatened by eutrophication [6], driven by anthropogenic pressures [7]. Since the 1970s, the intensification of agriculture and extensive use of fertilizer have led to a substantial increase in the transport of nitrogen and phosphorus to aquatic systems [8,9], while climate warming has amplified thermal stratification in lakes [10]. These combined factors have pushed 63% of the world’s lakes into eutrophic states [11,12], creating favorable conditions for the proliferation of harmful algal blooms (HABs) [13,14].

In this review, we define HABs as blooms dominated by harmful cyanobacteria in inland lakes, specifically those that produce toxins or cause significant ecological and health impacts. These include species such as Microcystis, Nodularia, and Anabaena, which are known to produce toxins like microcystins and nodularins [15]. It is important to note that not all algal blooms are harmful; general algal blooms may include non-toxic species and do not necessarily lead to adverse effects [16]. Severe ecological impacts are induced by HABs, including hypoxic conditions, toxin production (e.g., microcystins and saxitoxins) [17,18], degraded water quality [19], and threats to human health [20].

Given our focus on inland lakes, this study excludes blooms caused by other groups such as red algae or dinoflagellates, which are more common in marine or brackish environments and are often associated with phycoerythrin-rich species [21]. In this context, HABs are characterized by the formation of surface scum or water discoloration (e.g., green or blue-green hues) due to rapid cyanobacterial proliferation under conditions of nutrient enrichment, elevated temperatures, sufficient light, and stable water flow [22]. The characteristic pseudo-vacuole structure of certain cyanobacteria often results in floating algal mats on the water surface [23]. Although HABs are most common in eutrophic lakes, they have also been observed in oligotrophic environments [24], illustrating their growing prevalence under global warming.

As HABs become more widespread and persistent, there is an urgent need for advanced methods to monitor their spatial extent, duration, and frequency [25]. Optical remote sensing (ORS) has become a useful tool for effectively monitoring and managing HABs. Unlike in situ monitoring, which is limited in spatial and temporal coverage, ORS provides extensive and continuous data, making it the primary method for global HABssurveillance. When HABs occur, they create a cascade of alterations in water color, turbidity, composition, and phytoplankton concentration, resulting in alterations to the spectral reflectance properties of the water surface. By analyzing these spectral features, it is possible to obtain information about the extent, intensity, and frequency of the bloom [26]. Since its inception in the 1970s [27], ORS has evolved into a critical tool for quantifying bloom extent, duration, and frequency [28,29]. Recent studies demonstrate its capacity to track HABs’ intensification in global lakes [12], with Figure 1 documenting a rising frequency in U.S., European, and Chinese waters from 1987 to 2021. Despite significant advancements, ORS still faces major challenges in effectively monitoring HABs in inland waters. Limitations such as spatiotemporal resolution constraints, atmospheric interference, and water turbidity continue to affect accuracy. Furthermore, the lack of robust evaluation and validation for retrieval algorithms across diverse water bodies underscores the need for methodological advancements [30]. Addressing these challenges is critical to improving HABs monitoring in the face of growing pressures from human activity and climate change.

This review follows the evolution in the retrieval of HABs using satellite-based optical sensors, demonstrating the advancements achieved through the application of ORS in HABs monitoring. It also provides a comprehensive examination of its current status and difficulties, including challenges related to spatiotemporal scale, cloud cover, turbid water, lake topography, and wind effects. The paper is organized as follows: Section 2 examines the bibliometric characteristics of HABs–ORS research, including publication trends, key contributors, and thematic shifts. Section 3 describes the data and methods for HABs retrieval, encompassing the development of satellite technologies, spectral-based techniques, machine learning applications, and their implementation across diverse environments. Section 4 addresses the challenges and future opportunities in ORS for HABs monitoring, focusing on limitations such as atmospheric correction and spatiotemporal resolution, and highlighting potential solutions.

2. Bibliometric Characteristics

Many publications limit the topic by using the terms “algal blooms” or “ORS” and “lakes” from the Web of Science Core (Figure 2). There has been gradual progress in the field of ORS studies over the past 30 years (Figure 2a). For example, the employing of ORS methods to retrieve bloom occurrences has revealed a consistent worldwide increase, with an annual growth rate of 11.12%. Notably, 2021 marked a significant milestone with 69 publications. The frequent terms included “chlorophyll-a”, “inland waters”, “algal blooms”, and “phytoplankton”, as shown in Figure 2b. These keywords emphasize the complexities of the studies. Three journals contribute 24% of the publications between them: Remote Sensing, Remote Sensing of Environment, and Science of the Total Environment (Figure 2c). It is noteworthy that the majority of the publications are about ORS. These publications come from many kinds of fields, including biology, hydrology, ecology, ORS, and limnology. The number of publications is increasing, but the quality is also improving. Most of the studies are limited by the spatiotemporal availability of satellites for large lakes. The United States is the country that contributes the greatest amount of studies, with 546 publications, surpassing China (497) and Canada (67). However, there are still countries that have not released research about ORS on HABs.

To further explore the research landscape, Figure 3a maps institutional collaborations from 2016 to 2021. This period (2016–2021) captures the rapid evolution of HABs–ORS research, particularly in methodology and regional focus. The Chinese Academy of Sciences (CAS) emerges as a central hub, with strong ties to U.S.-based institutions such as Indiana University, University of Cincinnati, University of Waterloo, NOAA, University of Toledo, and North Carolina State University. The node sizes reflect publication volume, with CAS and U.S. institutions leading, underscoring a robust trans-Pacific collaboration in addressing HABschallenges, particularly in eutrophic regions. Figure 3b presents the co-authorship network, identifying prominent researchers in HABs–ORS studies. Leading contributors include Zhang Fangfang, Li Yunmei, Loiselle Steven, Duan Hongtao, and Qi Lin (primarily affiliated with CAS and other Chinese institutions), alongside U.S.-based researchers such as Stumpf, Richard P.; Mishra, Deepak R.; Binding, Caren; and Beck, Richard. The color gradient (2016–2021) indicates that researchers like Liu Heng and Zhang Yunlin have been particularly active in recent years, reflecting a focus on advanced methodologies. Figure 3c illustrates the keyword co-occurrence network, revealing thematic shifts from 2016 to 2021. Dominant keywords include “harmful algal blooms”, “remote sensing”, “cyanobacteria”, “phytoplankton”, and “chlorophyll-a”, reflecting core research themes. Region-specific terms like “Lake Erie”, “Taihu”, and “Great Lakes” highlight key study areas, while “eutrophication”, “nutrients”, and “climate change” underscore HABs drivers. Methodological keywords such as “machine learning”, “hyperspectral”, “Sentinel-2”, and “Sentinel-3” have gained prominence since 2018, signaling a shift toward advanced technologies.

3. Data and Methods for HAB Retrieval

ORS detects and monitors HABs (Figure 4) by leveraging their unique spectral signatures, which arise from the interactions between phytoplankton pigments, cellular structures, and light at specific wavelengths [31]. HABs exhibit a reflectance peak at 550 nm and 700 nm and absorption troughs near 440 nm and 675 nm [32], with the latter trough attributed to phycocyanin absorption [33]. A distinct fluorescence peak at 683 nm uniquely differentiates HABs from non-bloom waters [34]. Dense algal cells induce elevated backscattering, producing a reflectance peak near 700 nm analogous to the vegetation “red edge” phenomenon [35]. In intense blooms, this peak may broaden (680–900 nm) or bifurcate into a double-peak at 700 nm and 800 nm [36,37], a variability driven by species-specific differences in pigment composition and cellular morphology [38].

3.1. Evolution of Satellite Technologies

The advancement of satellite-based HABs monitoring has been driven by three technological advancements: improved spatial resolution (4 km to 5 m), enhanced spectral precision (5 bands to >250 contiguous bands), and increased temporal frequency (16-day to hourly observations). HABs can be detected using nearly all optical remote sensing sensors.

Table 1. Comparison of parameters of different satellite series for monitoring HABs.

Sensor Group	Satellite/Sensor	Spatial Resolution (m)	Temporal Resolution (Day)	Band Numbers	Data Availability	Operation	Country/Organization
Multispectral Sensors	AVHRR NOAA	1100	1	5	1978–present	Normal	USA
	MODIS Terra	250–1000	1	36	1999–present	Normal	USA
	MODIS Aqua	250–1000	1	36	2002–present	Normal	USA
	MERIS Envisat	300	3	15	2002–2012	Retired	ESA
	CZCS Nimbus-7	800	6	6	1978–1986	Retired	USA
	SeaWiFS	900	1	8	1997–2010	Retired	USA
	Landsat 1–9	15–80	16	7–11	1972–present	Normal	USA
	Sentinel-2 MSI	10–60	5	13	2015–present	Normal	ESA
	GOCI COMS	500	0.5	8	2010–present	Normal	South Korea
	HJ-1A/1B CCD	30	2	4	2008–present	Normal	China
	HJ-1C/D COCTS	1100	1	8	2018–present	Normal	China
	Sentinel-3 OLCI	300	1	21	2016–present	Normal	ESA
	GF-1/2/3/4	1–50	1–4	4	2013–present	Normal	China
	QuickBird	0.65	1–1.3	5	2001–2015	Retired	USA
	WorldView 1–4	0.31–0.50	1–4	8–16	2007–present	Normal	USA
	IKONOS	1–4	1–3	4	2001–2015	Retired	USA
	ADEOS-II GLI	250–1000	4	36	2002–present	Normal	Japan
	GOES ABI	500–2000	0.01	16	2017–present	Normal	USA
	VIIRS Suomi NPP/NOAA-20	375–750	1	22	2011–present	Normal	USA
	Himawari AHI	500–2000	0.007	16	2015–present	Normal	Japan
	Zhuhai-1 OHS-01/02	2.5	5	256	2017–present	Normal	China
Hyperspectral Sensors	Hyperion EO-1	30	16	242	2000–2017	Retired	USA
	HICO	90	1–3	128	2009–2014	Retired	USA
	EnMAP	30	4	244	2022–present	Normal	Germany
	PRISMA HSI	30	7	239	2019–present	Normal	Italy
	DESIS	30	1	235	2018–present	Normal	Germany
	HySIS	30	5	326	2018–present	Normal	India
	HISUI	20	14	185	2021–present	Normal	Japan
	GF-5	30	4	330	2018–present	Normal	China
	EMIT	60	3	285	2022–present	Normal	USA
	PACE OCI	1000	1–2	250	2024–present	Normal	USA
	PROBA-1 CHRIS	17/34	7	18/62	2001–present	Normal	ESA

compares the spatial resolution and band number of major satellite sensors, highlighting the complementary roles of multispectral and hyperspectral sensors in advancing bloom detection capabilities over the past decades.

3.1.1. Early Multispectral Sensors (1970s–1990s)

The Coastal Zone Color Scanner (CZCS, 1978–1986) and Sea-viewing Wide Field-of-view Sensor (SeaWiFS) gave early data for mapping algal blooms. They were mostly used in oceans. Their spatial resolution was low at 825 m for CZCS and 1.1 km for SeaWiFS. They had only 5–6 bands. This made them hard to use for small inland lakes with cyanobacterial HABs [39,40]. CZCS monitored water quality in large lakes like Lake Superior, but could not detect HABs [41]. SeaWiFS measured chlorophyll-a in the Great Lakes, like Lake Michigan [42], but it was not optimized for specific HABs identification.

In 1979, the NOAA-6 satellite carried the Advanced Very High Resolution Radiometer (AVHRR), which had a 1.1 km resolution and 5 spectral bands [43,44]. AVHRR could detect chlorophyll-a (Chla) in large lakes. For example, it monitored Microcystis blooms in Lake Taihu, but its 1.1 km resolution missed small bloom patches [45]. However, its coarse spectral–spatial resolution and susceptibility to atmospheric interference limited its ability to distinguish cloud cover from blooms [46]. Landsat (1984–present) achieved 30 m spatial resolution with thermal and shortwave infrared bands [47], enabling multi-decadal algal biomass tracking in lakes [48,49,50]. However, the 16-day revisit limited its utility for short-lived bloom events [38,39,51].

3.1.2. Multispectral Sensors with Better Features (2000s–2010s)

The 21st century introduced sensors optimized for temporal continuity and spectral specificity. The Moderate Resolution Imaging Spectroradiometer (MODIS, 2000–present) gave daily global coverage. It had a 1–2-day revisit time. This temporal continuity has proven invaluable for analyzing seasonal bloom patterns [52] and long–term ecological impacts [53,54]. However, its 250–1000 m resolution limited its use for small lakes (<1 km²), where HABs like Microcystis often occur [55]. The Medium Resolution Imaging Spectrometer (MERIS, 2002–2012) addressed this gap with 300 m resolution and enhanced radiometric sensitivity specifically optimized for Case 2 water monitoring [56,57]. Its bands 6 and 7 detected phycocyanin absorption near 630 nm and a reflectance peak near 650 nm [58], distinguishing cyanobacteria like Anabaena and Aphanizomenon from other algae at high biomass levels [59].

Its successor, the Ocean and Land Colour Instrument (OLCI) on Sentinel-3A (2016–present), advanced inland lake monitoring with improved atmospheric correction and 300 m resolution, ideal for large lakes. With 21 spectral bands, including 620, 665, and 709 nm, OLCI enhances chlorophyll-a and phycocyanin quantification over MERIS, capturing subtle bloom signals in turbid, CDOM-rich waters [60]. These bands excel at detecting HABs, such as Microcystis and Planktothrix, in shallow, eutrophic lakes, improving HABs-related water quality retrievals like TSM and CDOM where in situ methods are limited [61].

Sentinel-2 (2015–present) provides 10–20 m resolution, a 5-day revisit time, and 13 spectral bands. Planktothrix rubescens blooms in Sicilian reservoirs during cold winters (<2 °C) shift reflectance to 600–650 nm due to surfaced filaments [62]. Simulations of Sentinel-2’s potential estimated phycocyanin (PC) density at 2.68 × 10⁶ cells L⁻¹ using the B2/B5 ratio, outperforming MODIS’s 1 km limit [62]. In New York lakes, the Maximum Peak Height (MPH) index with Cloud Score+ filtering achieved an R² of 0.87 for PC (>8 µg/L) [63]. In Lake Taihu, persistent Microcystis blooms at <10 °C, with low growth (0.05 day⁻¹) but high overwintering biomass, were detected, improving on MODIS [64]. However, cloud cover, mixed pixels in small lakes (<1 km²), and low-density PC retrieval challenge its use, requiring refined atmospheric corrections.

China’s HuanJing-1A/B satellites (2008–present) bridged this gap by providing 30 m spatial resolution with a 2-day revisit, enabling the dynamic monitoring of large lake systems such as Lakes Taihu and Erhai [65,66]. However, its performance is limited in highly turbid waters [67]. South Korea’s Geostationary Ocean Color Imager (GOCI, 2010–present) provided hourly observations to resolve diurnal HAB dynamics [68]. However, spatial scale mismatches between GOCI and other data introduced retrieval inconsistencies [69]. The Visible Infrared Imaging Radiometer Suite (VIIRS, 2011–present), an upgrade to AVHRR and MODIS, offered daily 742 m global coverage, capturing seasonal HABs in large lakes [70,71], and can be used in collaboration with MODIS to detect HABs. However, VIIRS’s resolution remained suboptimal for sub-kilometer water bodies. Sentinel-2 (2015–present) delivered 10–20 m resolution with 13 spectral bands (including the critical 705 nm red-edge band). They are widely used to monitor HABs in lakes [72,73].

3.1.3. Hyperspectral Sensors (2010s–Present)

Hyperspectral sensors have more bands. They use 5–10 nm sampling. This helps detect pigments like Chla and phycocyanin [74]. Hyperion (2000–2017) pioneered spaceborne hyperspectral monitoring with 220 contiguous bands (10 nm) spanning 400 to 2500 nm, demonstrating exceptional capability for Chla dynamics analysis in medium to large lakes [75]. For instance, it enabled precise water quality assessments in Lake Garda, Italy, highlighting its utility in monitoring Chla and other parameters [76]. Its extensive spectral range provided unparalleled detail, serving as a benchmark for modern hyperspectral models even after its decommissioning in 2017 [77]. However, its 30 m resolution and low signal-to-noise ratio (SNR) limited the sensor’s ability to resolve spectral details in the shortwave infrared (SWIR) region [78].

DLR Earth Sensing Imaging Spectrometer (DESIS) provides 30 m resolution which plays a crucial role in the remote sensing of aquatic ecosystems. Legleiter used DESIS by developing the Multiple Endmember Spectral Mixture Analysis to achieve genus-level discrimination in algal blooms in Lake Owasco [79]. The high spectral resolution of these sensors has enabled the accurate mapping of Chla and phycocyanin concentrations.

Hyperspectral Imager for the Coastal Ocean (HICO) achieved relatively high signal-to-noise ratios [80]. Its spectral fidelity enabled the differentiation of floating algal taxa through feature analysis in the 400–800 nm range.

The bands covered 400–800 nm. HICO had a high signal-to-noise ratio. HICO separated Microcystis and Nodularia blooms in Lake Erie [80]. It used the 620 nm band for phycocyanin. However, HICO did not collect data often. Its coverage was limited [81]. Due to the nature of the sensor and its orbit, inherent limitations prevent HICO from providing anywhere near the frequent global coverage levels offered by standard multispectral ocean color sensors. Despite its capabilities, sporadic data acquisition and restricted coverage limited systematic monitoring [81]. PRISMA (2019–present) provides 30 m resolution with 250+ contiguous bands, enabling the precise quantification of phycocyanin and Chla even in optically complex waters [82]. The Environmental Mapping and Analysis Program (EnMAP) achieves relatively high SNR across red-NIR [83], with maximal Chla correlation at 710 nm enabling accurate quantification in small inland water bodies [84]. Geostationary satellites like the Geostationary Operational Environmental Satellite (GOES) series enable unique diurnal HABs monitoring through hourly thermal and visible-band observations, capturing bloom migration patterns driven by solar insolation and wind forcing [85]. NASA’s Plankton, Aerosol, Cloud, and ocean Ecosystem (PACE) mission, launched in 2024, introduces the Ocean Color Instrument (OCI) with 5 nm spectral sampling (340–890 nm). This hyperspectral configuration enables the precise quantification of pigment absorption coefficients [86], particularly Chla at 440 nm and phycocyanin at 620 nm. However, its 1.2 km spatial resolution restricts its effectiveness for monitoring blooms in lakes smaller than 10 km² [87]. EMIT, with 285 spectral bands (381–2493 nm), allowed for more detailed monitoring of Chla dynamics. EMIT 60 m effectively complemented existing remote sensing systems, such as Sentinel-3 OLCI (300 m resolution), by resolving data gaps in regions like the southern Lower Lake and western Upper Lake of Clear Lake [88]. China’s Zhuhai-1 Orbital Hyperspectral Satellite (OHS, 2017–present) achieves 2.5 m resolution with 256 spectral bands, demonstrating exceptional performance in algal speciation within eutrophic inland lakes [89]. In Dianchi Lake applications, OHS achieved a moderate signal-to-noise ratio (SNR = 59.47) and noise-equivalent chlorophyll concentration (NEChla = 72.86 μg/L), enabling improved Chla detection accuracy [90]. Legacy high-resolution systems—including WorldView-4 (retired in 2019) and China’s Gaofen (GF-1/2/3/4) series—achieved sub-5 m resolution but their application is limited because of expensive data prices and infrequent observation.

3.1.4. Complementary Approaches

Unmanned Aerial Vehicles (UAVs) have emerged as a powerful supplement to satellite-based monitoring [91], particularly for localized HABs detection. UAVs equipped with multispectral or hyperspectral sensors provide ultra-high spatial resolution, making them ideal for capturing short-term bloom events in small lakes [92]. This makes drones good for small lakes. Drones can capture short Microcystis bloom events [93]. Recent studies on HABs detection with RGB images have primarily used UAVs, as camera poses can be preset and images corrected to reduce platform and environmental impacts [94]. However, UAVs are limited to small areas (1–12 km² per flight) and require clear weather, restricting their use compared to broader satellite coverage.

Ground sensors also help monitor HABs [95]. They measure water properties like Chla and phycocyanin. They give continuous data. These data are very accurate. Ground sensors can check satellite data. For example, ground sensors in Lake Taihu measured Microcystis blooms [96]. They helped improve Sentinel-2 results. Some studies suggest that ground-based remote sensing may even replace satellites in small lakes [95]. But ground sensors only work in one spot. They cannot cover large areas.

Special networks combine satellites, drones, and ground sensors [97]. They make a system that works from the sky, air, and ground. This system gives better data. It helps monitor Microcystis blooms in large lakes like Lake Dianchi. For example, a network in Lake Dianchi used Sentinel-2, drones, and ground sensors [97]. It tracked blooms in real time. But these networks are hard to set up. They need a lot of work. Data platforms like Google Earth Engine help analyze HABs data [78]. They combine data from many sources. For example, they can use Sentinel-2 and MODIS together. This helps track Microcystis blooms over large areas. Google Earth Engine can analyze data quickly. It can show how blooms change over time. For example, it mapped Microcystis blooms in Lake Erie [78].

In summary, these methods make HABs monitoring better. Drones and ground sensors help satellites. Networks and data platforms combine all the data. They help us detect HABs in inland lakes more accurately.

3.2. Spectral-Based Methods and Applications

ORS primarily monitors the coverage area and dynamics of HABs but cannot detect below the scum, limiting its ability to provide surface accumulation information [98]. Therefore, most studies quantify the intensity of blooms by assessing the area and frequency of HABs occurrences [99,100,101].

ORS methodologies for HABs detection are classified into two main technical frameworks: spectral-based methods and machine learning (ML) or deep learning (DL) approaches. Each paradigm is fundamentally driven by characteristic variations in the spectral shape and magnitude of water-leaving radiance or reflectance signals detected by satellite sensors. These spectral features can be empirically or analytically correlated with specific biological indicators, such as Chla concentration, phycocyanin content, and cyanobacterial biomass or biovolume [102].

Table 2. Main HABs retrieval methods.

Indices	Equation	Advantages	Limitations and Requirements	References
MPH	${\rho }_{BRmax}$$-{\rho }_{BR664}-(({\rho }_{BR885}-{\rho }_{BR664})\times {\displaystyle \frac{{\lambda }_{max}-664}{885-664}}$ ρBRmax and λmax are the highest value bands at 681, 709 and 753 nm from MERIS	Detects chlorophyll peaks effectively	Sensor-specific (e.g., MERIS); insensitive to low-concentration blooms	[38]
ABDI	$\left[{\mathrm{R}}_{\mathrm{R}\mathrm{e}\mathrm{d}}-{\mathrm{R}}_{\mathrm{R}\mathrm{E}\mathrm{D}}-\left({\mathrm{R}}_{\mathrm{N}\mathrm{I}\mathrm{R}\mathrm{n}}-{\mathrm{R}}_{\mathrm{R}\mathrm{e}\mathrm{d}}\right)\times \left({\lambda }_{\mathrm{R}\mathrm{E}2}-{\lambda }_{\mathrm{R}\mathrm{e}\mathrm{d}}\right)/\left({\lambda }_{\mathrm{N}\mathrm{I}\mathrm{R}\mathrm{n}}-{\lambda }_{\mathrm{R}\mathrm{e}\mathrm{d}}\right)\right]-\left[{\mathrm{R}}_{\mathrm{R}\mathrm{e}\mathrm{d}}-0.5\times {\mathrm{R}}_{\mathrm{G}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{n}}\right]$ ${\lambda }_{\mathrm{R}\mathrm{e}\mathrm{d}}\left(665 \mathrm{nm}\right),{\lambda }_{\mathrm{R}\mathrm{E}2}\left(740 \mathrm{nm}\right)$$,\mathrm{and} {\lambda }_{\mathrm{N}\mathrm{I}\mathrm{R}\mathrm{n}}\left(865 \mathrm{nm}\right)$ represent the central wavelengths of Red, Red Edge-2 and NIRn bands, respectively	Effectively avoids misclassifying thin cloud cover and turbid water as algal blooms, which improves the reliability of algal bloom detection in complex environments	Difficult for ABDI to distinguish between algal blooms and aquatic vegetation	[72]
FAI	Rrc,NIR − R’rc,NIR $\mathrm{R}’\mathrm{rc},\mathrm{NIR}=\mathrm{Rrc},\mathrm{RED}+(\mathrm{Rrc},\mathrm{SWIR}-\mathrm{Rrc},\mathrm{RED}){\displaystyle \frac{{\mathsf{\lambda }}_{\mathrm{NIR}}-{\mathsf{\lambda }}_{\mathrm{RED}}}{{\mathsf{\lambda }}_{\mathrm{SWIR}}-{\mathsf{\lambda }}_{\mathrm{RED}}}}$ ${\mathsf{\lambda }}_{\mathrm{NIR}}-{\mathsf{\lambda }}_{\mathrm{RED}}{-\mathsf{\lambda }}_{\mathrm{SWIR}}$ are the center wavelengths	Resists thin cloud interference	Misclassifies turbid water as algal blooms	[103]
NDVI	${\displaystyle \frac{R_{NIR}-R_{RED}}{R_{NIR}+R_{RED}}}$	Simple, widely applicable	Sensitive to turbidity, overestimates bloom area in high-turbidity waters	[104]
NDCI	${\displaystyle \frac{R_{RE}-R_{red}}{R_{RE}+R_{red},}}$ $R_{R}E$$\mathrm{represents} \mathrm{the} \mathrm{reflectance} \mathrm{at} \mathrm{the} \mathrm{red}-\mathrm{edge} \mathrm{wavelength} \mathrm{and} R_{red}$ represents the reflectance at the red wavelength	Relatively straightforward, allowing for rapid data analysis	Limited by sensor bands, influenced by high turbidity	[105]
MCI	$\mathrm{R}709-\mathrm{R}681-(\mathrm{R}754-\mathrm{R}681) \times {\displaystyle \frac{709-681}{754-681}}$	Sensitive to cyanobacteria at high concentrations	Requires red-edge bands; ineffective for low concentrations	[106]
EVI	$2.5\times {\displaystyle \frac{NIR-RED}{NIR+6\times RED-7\times BLUE+1}}$	Reduces atmospheric and background noise	Requires blue band; limited performance in optically complex waters	[107]
AFAI	$\mathrm{Rrc},\mathrm{NIR}-\mathrm{Rrc},\mathrm{Green}+(\mathrm{Rrc},\mathrm{Green}-\mathrm{Rrc},\mathrm{Red}))\times {\displaystyle \frac{{\mathsf{\lambda }}_{\mathrm{NIR}}-{\mathsf{\lambda }}_{\mathrm{Green}}}{2{\mathsf{\lambda }}_{\mathrm{NIR}}-{\mathsf{\lambda }}_{\mathrm{Green}}-{\mathsf{\lambda }}_{\mathrm{Red}}}}$	Enhances NIR reflection peaks for accurate cyanobacteria detection	Challenged by turbidity; struggles to distinguish turbid water from blooms	[108]
CI	$(-1)\times$$[\mathrm{Rrc} (681\mathrm{nm}-665\mathrm{nm})-(\mathrm{Rrc} 709-\mathrm{Rrc} 665\left)\right] \times$ [(682 − 665)/(709 − 665)]	Exploits absorption peak at ~709 nm	Limited to high cyanobacteria concentrations	[109]
NDPI	(R820 − R690)/(R820 + R690)	Discriminates algae from suspended sediments	Sensitive to water optical complexity; requires NIR bands	[110]
VB-FAH	${(\rho }_{NIR}$$-{\rho }_{Green})+\left({\rho }_{Green}-{\rho }_{RED}\right)\mathrm{\ast }({\mathsf{\lambda }}_{\mathrm{NIR}}-{\mathsf{\lambda }}_{\mathrm{Green}})/(2{\mathsf{\lambda }}_{\mathrm{NIR}}-{\mathsf{\lambda }}_{RED}-{\mathsf{\lambda }}_{\mathrm{Green}})$ $\rho$$\mathrm{is} \mathrm{the} \mathrm{reflectance}, \mathsf{\lambda }$ is the center wavelength of each satellite sensor	Compatible with sensors lacking SWIR (e.g., HJ/GF)	Reduced accuracy in turbid waters	[111]
FLH	$R_{681}-\left(\begin{array}{c}{R}_{665}+{\displaystyle \frac{R_{709}-R_{665}}{709-665}}\cdot \left(681-665\right)\end{array}\right)$	Effective in clear waters	Affected by CDOM and non-algal particles	[112]

summarizes the algorithms, advantages, and limitations of these methods.

3.2.1. Spectral-Based Methods for Pigment Detection

Spectral-based methods detect HABs in inland lakes by analyzing the optical properties of phytoplankton pigments [113]. Cyanobacteria, such as Microcystis and Anabaena, exhibit Chla absorption at 440 nm and 675 nm and phycocyanin absorption at 620–630 nm, with a far-red reflectance peak at 700–720 nm, as observed in Microcystis-dominated blooms in Lakes Taihu, Chaohu, and Dianchi [114]. In contrast, green algae (e.g., Chlorella) show Chl-b absorption at 470 nm [115], while diatoms (e.g., Cyclotella) display carotenoid absorption at 490–550 nm [116], both lacking the far-red reflectance peak typical of cyanobacteria. These optical differences enable species-specific detection through tailored algorithms in inland lake systems.

Spectral methods detect HABs pigments using band differences or ratios, as shown in Table 2. These indices include the normalized difference vegetation index (NDVI), the single band threshold and ratio vegetation index (RVI), the enhanced vegetation index (EVI) [107], the normalized difference peak–valley index (NDPI) [110], the visual cyanobacteria index (VCI) [117], the maximum chlorophyll index (MCI) [106], maximum peak height (MPH) [38], the algal bloom detection index (ABDI) [72], Fluorescence Line Height (FLH) [112], the Virtual-Baseline Floating macroAlgae Height (VB-FAH) index [111], the cyanobacteria index (CI) [109], the floating algae index (FAI) [103], and the adjusted floating algae index (AFAI) [118], widely used for blooms like Microcystis.

Among these, NDVI is highly sensitive to turbidity and atmospheric interference, which can overestimate bloom areas, particularly in high-turbidity waters like Lake Taihu [119]. CI relies on spectral changes between 665, 681, and 709 nm due to the strong absorption of cyanobacteria at around 709 nm, effectively mapping blooms in Lake Balaton [109], though its performance may decline in turbid conditions. FAI, widely considered a robust index, is effective for high concentrations of blue-green algae, but it proves inadequate for estimating Chla concentrations in complex waters [120,121]. The AFAI reduces interference by enhancing reflection peaks in the near-infrared (NIR) range, providing more accurate cyanobacteria detection in turbid waters like Lake Taihu [118], though both FAI and AFAI may misclassify turbid waters as HABs. ABDI achieves >98% accuracy in Lake Hulun by reducing the misclassification of thin clouds and turbid water, improving reliability in complex environments [72]. To address atmospheric interference, VB-FAH maintains performance comparable to FAI while tolerating surface sunlight scintillation, aerosol perturbation, and suspended matter interference, as demonstrated in Lake Dianchi [122]. MCI and FLH utilize MERIS bands to detect intense surface blooms, with MCI measuring the radiance peak at 709 nm and FLH capturing Chla fluorescence near 681 nm [123]. MPH, using the 705 nm band, yields an R² of 0.84 for phycocyanin estimation in New York State’s small- and mid-sized lakes with Sentinel-2 data and Cloud Score+ mitigation, validated by CSLAP 2019–2020 data [63]. O’Shea used a Mixture Density Network (MDN) with HICO and PRISMA images to estimate the cyanobacteria biomass via phycocyanin at 620 nm in Lake Erie, achieving higher accuracy and stability than traditional multispectral indices by exploiting the fine spectral resolution of hyperspectral data [124].

Semi-analytical algorithms model inherent optical properties for precise pigment retrieval. The Enhanced Three-Band Algorithm (ETBA) improves phycocyanin retrieval in low-PC waters like Erhai Lake by incorporating the absorption coefficient of non-PC pigments at 620 nm, achieving an RMSE of 0.37 μg/L and MAPE of 12.76% using Sentinel-3 OLCI bands at 620, 665, and 709 nm [125]. ETBA’s adaptability to low-PC waters and robustness against Chla and total suspended matter interference make it suitable for mesotrophic lakes, though it requires calibration for varying PC-specific absorption coefficients. Similarly, Laneve applied the Jaelani algorithm to Landsat data with ACOLITE atmospheric correction [126], estimating Chla in Lake Owasco with an R² of 0.79, demonstrating the potential for moderate-resolution HABs monitoring, though its performance may vary with water conditions and atmospheric interference. Hyperspectral data can further enhance semi-analytical approaches. For instance, Legleiter developed the Spectral Mixture Analysis for Surveillance of HABs (SMASH) method, using DESIS hyperspectral data and spectral unmixing to identify cyanobacteria genera like Microcystis in Lake Owasco, improving species-specific detection over multispectral methods [79]. However, such applications rely on the availability of high-resolution hyperspectral data, which may limit their use in regions lacking such coverage.

3.2.2. Machine Learning and Deep Learning Methods for HABs Retrieval

Machine learning (ML) and deep learning (DL) methods enhance HABs retrieval by enabling automated detection, classification, and ecological analysis using large datasets of satellite imagery, in situ measurements, and environmental variables. These approaches improve upon traditional Chla threshold methods by modeling complex relationships between reflectance and HABs features, supporting pigment quantification, biomass estimation, and taxonomic differentiation.

ML techniques, such as random forests (RF) and support vector machines (SVM), are widely applied to ORS data for HABs retrieval. For instance, the HABNet model integrates convolutional neural networks (CNNs), long short-term memory (LSTM) networks, and RF classifiers with MODIS and GEBCO bathymetry data, achieving 91% accuracy and a Kappa coefficient of 0.81 [127]. In Lake Erken, Sweden, a two-step workflow employs gradient boosting regressors (GBR) and LSTM networks to first estimate daily nutrient concentrations and then predict Chla variations using meteorological, nutrient, and hydrodynamic inputs, capturing seasonal bloom patterns with mean absolute errors (MAE) of 3.55–3.58 mg m⁻³ and root mean square errors (RMSE) of 5.64–5.77 mg m⁻³ [128]. Phycocyanin (PC) quantification targets cyanobacteria-specific signals at ~620 nm using established ML approaches [129,130], while decision-tree models classify Chla profiles in Lake Taihu, achieving 85% accuracy in distinguishing uniform versus decay-type vertical distributions [130].

DL further refines HABs retrieval through advanced feature extraction. Researchers have explored advanced methods, including integrating ORS data with convolutional neural networks [131], as well as new methods that employ integrated biometrics and ensemble learning algorithms [132]. Yan achieved automatic bloom extraction based on deep learning semantic segmentation [133]. Yang used different ML models for the precise detection of bloom spatial distribution in lakes [134] and the latest developed advanced interpretability of the deep learning model, used to predict and analyze the Chinese freshwater lakes and reservoirs for HABs [135]. Recent advances leverage ML for automated HABs detection and classification, reducing reliance on manual interpretation [94,134,136].

A new automated algorithm recently introduced utilizes the color space established by the International Commission for the detection of HABs. However, significant shortcomings persist, and no universally effective method for precise HABs identification has been established to date [11].

3.2.3. Applications in Different Environments

ORS provides frequent temporal data and extensive spatial coverage for tracking the dynamics of HABs globally. The frequency, severity, and duration of these blooms have been observed to vary over time in different regions, allowing for a more detailed comparison of regional differences in bloom characteristics. Several studies suggest that climate zones and geographical regions influence the occurrence of HABs [52,137]. The global phenological onset of HABs advanced by 2.17 days/decade from 2000 to 2020, with bloom duration extending by 1.14 days/decade. Tropical lakes exhibited faster acceleration (2.4 days/decade) than temperate systems (1.95 days/decade), which is attributed to prolonged warm seasons [52].

HABs are particularly prevalent in tropical and subtropical lakes in Asia and Africa [138]. Blooms in tropical lakes (e.g., Victoria, Tanganyika) persist year-round due to stable warm temperatures (>25 °C) and high nutrient loading from monsoonal runoff [139]. In temperate and tropical climates, HABs occur in over 50% of lakes, with 59% of lakes in temperate regions experiencing significant blooms [140]. In temperate countries, HABs in lakes are generally more frequent in eutrophic lakes [138], while in cold regions, where temperatures are relatively low, ice may occur throughout the winter, but HABs may still occur in the summer [141].

Recent studies have shown that HABs also occur and persist in relatively cold water temperatures (<15 °C), including ice-covered conditions [142,143]. Notably, Planktothrix rubescens, one of the most widespread cyanobacteria in Europe, exemplifies this adaptability by forming blooms in winter under low temperatures and reduced solar irradiance. For instance, in Lake Geneva, a deep temperate lake, P. rubescens thrives at temperatures around 14–15 °C, with machine learning models (e.g., Random Forest) identifying temperature as a key predictor of bloom intensity [144]. Similarly, in Sicilian reservoirs, remote sensing studies have documented P. rubescens blooms triggered by frost events (<2 °C) and low shortwave irradiance during winter, with filaments rising to the surface to form red-colored blooms [145]. These findings highlight that HABs are not confined to warm seasons or climates, as P. rubescens exploits colder conditions in stratified, monomictic lakes across Europe.

Additional region-specific lake studies have further clarified the influence of climate variables on HABs occurrences. In Lake Burragorang (Australia), modeling revealed that climate-induced low water levels combined with high-volume inflows can enhance vertical mixing and trigger nutrient release into surface waters, ultimately initiating algal blooms under warmer scenarios [146]. In China, a national-scale Landsat study of over 400 lakes from 1983 to 2017 found that increasing temperatures and declining wind speeds significantly advanced bloom onset and prolonged duration. Wind speed was identified as a dominant meteorological driver of bloom expansion [99]. More recent research in Lake Taihu confirms that early winter weakening of the East Asian monsoon reduces wind-driven turbulence, increasing underwater light availability and promoting blooms that can persist into spring [147].

However, since 1999, the frequency of HABs in North America has decreased, whereas it has increased in Asia, followed by South America, Africa, and Europe [138]. The highest frequency of HABs in lakes is observed in Asia and North America (Figure 5). Studies on the occurrence of algal blooms in typical lakes across different climatic zones are presented in

Table 3. Optical remote sensing for lake algal bloom monitoring by climate zone.

Climate Zone	Country	Lake	Key Parameters	Sensors	References
Tropical	Guatemala	Atitlán	Chla	Hyperion	[75]
	Kenya Uganda Tanzania	Victoria	Chla	MODIS	[148]
	Nicaragua	Nicaragua	Chla	Sentinel-2	[149]
	Sololá	Atitlán	Chla	Hyperion	[75]
Subtropical	China	Taihu	Chla	MODIS	[150]
	China	Chaohu	Chla	MODIS	[151]
	Japan	Biwa	Chla	Landsat, Sentinel-2	[152]
	Mexico	Chapala	Chla	Landsat, MODIS	[153]
	USA	Okeechobee	Chla	MODIS	[154]
Temperate	USA/Canada	Erie	PC	HICO, PRISMA	[124]
	Estonia/Russia	Peipsi	Chla	PRISMA	[155]
	Hungary	Balaton	Chla	MERIS	[156]
	China	Dianchi	Chla	ground-basedimaging	[157]
	Switzerland	Geneva	Chla	Sentinel-2	[158]
	Canada	Winnipeg	Chla	MERIS	[159]
	Italy/Switzerland	Lake Como, Lake Maggiore, and Lake Lugano	Chla	PRISMA, Sentinel-3	[82]
	Russia	Lake Baikal	Chla	LiDAR	[160]

.

4. Challenges and Future Opportunities

4.1. Limitation of ORS

ORS has made an important contribution to the comparison analysis of HABs frequencies across different climate zones. However, comprehending these data necessitates an in-depth understanding of the essential uncertainties. Climate change causes differences in satellite observation frequencies, and extensive clouds in tropical places usually make satellite data insufficient. For example, satellites like Landsat revisit every 16 days and may miss short-lived HABs, while satellites orbiting the earth, despite their higher revisit times, may still miss short-lived blooms due to limited coverage at higher latitudes, introducing a potential geographic bias. On the other hand, the size and number of lakes within the climate zone can also lead to uncertainty. Larger lakes may be monitored more frequently, providing a comprehensive dataset that can help detect HABs. On the other hand, the frequency of monitoring in small lakes may suffer from insufficient frequency, which, combined with adverse weather conditions, may lead to the underreporting of the presence of HABs. This difference may result in overestimating the frequency of HABs with larger lakes and underestimating the frequency of HABs with smaller lakes.

4.1.1. Atmospheric Correction

Atmospheric interference distorts surface reflectance values by scattering and absorbing solar radiation, necessitating radiometric correction during data pre-processing [161]. Key atmospheric components—like aerosol particles, water vapor, and nitrogen oxides—vary in concentration and composition over time and space, making it harder to retrieve accurate surface signals [162].

Common atmospheric correction methods include the Second Simulation of the Satellite Signal in the Solar Spectrum (6S; [163]), the Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes (FLAASH), and the Gordon algorithm [164]. While these methods demonstrate utility in terrestrial applications [165,166,167], their efficacy diminishes in aquatic environments due to the unique optical properties of water bodies. Lakes exhibit complex interactions between water quality parameters (e.g., turbidity), the surrounding terrain, and atmospheric adjacency effects, demanding specialized correction frameworks [168]. For instance, Vanhellemont developed a water-specific approach using Multiple Endmember Spectral Mixture Analysis (MESMA), which mitigates adjacency and flicker effects while enhancing spectral unmixing accuracy [169].

ACOLITE, tailored for inland and coastal waters, processes Sentinel-2 and Landsat data using a dark spectrum fitting approach [170]. For HABs monitoring, it surpasses land-focused 6S and FLAASH, which overestimate water reflectance, and the Gordon algorithm, which falters in the turbid waters critical for Chla detection [171]. In northern German lakes, ACOLITE excelled in the visible range (R² = 0.79–0.97, RMSE = 0.001–0.002 at 443–490 nm), key for HABs at 665 nm, but weakened in the NIR (R² = 0.21, RMSE = 0.003 at 740 nm) [172]. It effectively mitigates adjacency effects, as seen in Guadiana estuary HABs mapping [172], and handles turbidity well (RMSE ≈ 3 mg/m³ in low-turbidity lakes, rising to 6.6 mg/m³ in extreme cases) [126]. Its scene-specific dark targets adapt to variable aerosols, unlike 6S. ACOLITE shines in turbid waters but requires field calibration for clear, deep lakes [173].

Turbid waters pose additional challenges, as traditional “dark pixel” assumptions fail under high suspended sediment loads. Recent advancements incorporate SWIR or blue–violet wavelengths into correction algorithms, with SWIR demonstrating superior performance in optically complex waters [174,175]. To ensure reliable ORS outputs, future efforts must prioritize the development of lake-specific correction models tailored to regional hydrological and atmospheric conditions.

4.1.2. Cloud Coverage and Shadow

The proliferation of cyanobacteria in lakes during summer is often associated with increased precipitation and persistent cloud cover [176]. However, such meteorological conditions significantly reduce the availability of cloud-free satellite imagery, thereby compromising data reliability for HABs monitoring. Current cloud removal algorithms in ORS can be classified into three categories:

The first category is image restoration techniques leveraging spatial patterns [177]; the second is multispectral data fusion exploiting spectral correlations [178]; and the third is multitemporal reconstruction utilizing temporal continuity [179]. Despite these advances, algorithm accuracy remains constrained by challenges such as unintended HABs signal loss during cloud masking (e.g., Fmask artifacts reported by Zhu [180]. Manual visual inspection, though labor-intensive, remains the most reliable validation approach. Automated workflows (e.g., Google Earth Engine’s CFMask) improve efficiency for large lake groups but risk oversimplification in small water body studies where cloud-free data scarcity exacerbates monitoring uncertainty [2,181]. Emerging solutions integrating synthetic aperture radar (SAR) with optical sensors show potential but require further optimization to address cross-sensor calibration challenges (Figure 6). Future research should prioritize deep learning architectures and multitemporal fusion frameworks to enhance cloud removal precision while preserving ecological signals.

4.1.3. Turbid Water

Turbid water is mostly caused by the amount of small suspended solids, such as silt, sediment organic matter, microorganisms, and other materials [182]. The turbidity alters the reflectance captured by ORS, while optical interference from suspended solids, bottom reflection, and dissolved organic matter can obscure HABs’ spectral signatures. There are many algorithms for the extraction of turbid water, including turbidity or total suspended sediment concentration [183]. For instance, Feng used an empirical algorithm to estimate the distribution of total suspended solids in Lake Poyang from MODIS data [184]. However, there are a few typical problems that may come up when using these algorithms. First, it can be challenging to use ORS because turbid water is often misclassified as clouds, land, or ice. Nonetheless, high-turbidity water can influence the index (NIR/RED), which can result in an overestimation. For example, MODIS bands could become saturated due to significantly turbid water [185,186]. Furthermore, turbid water decreases transparency, which causes the signals from cyanobacteria to become confused. A significant study challenge is still to improve the accuracy of ORS algorithms for the detection of HABs in turbid water. Researchers used multiple bands to improve the ability to detect turbid water and to overcome the influence of turbid water on HABs detection. For instance, Liang developed the Turbid Water Index (TWI) and employed red and SWIR bands to differentiate between suspended solids and HABs in lakes [187].

4.1.4. Wind and Lake Topography

The surface signals of HABs are obtained by ORS, and mixing events caused by wind speed should be considered when studying HABs’ intensity and growth [188]. Both wind regimes and lake bathymetry directly influence HABs’ detectability by altering their spatial aggregation, concentration gradients, and optical signatures [189]. HABs can be detected by ORS and remain on the surface of a calm lake. However, wind-driven turbulence disrupts surface stratification, triggering vertical mixing that dilutes surface HABs concentrations or redistributes the biomass to subsurface layers [190,191]. This vertical heterogeneity, compounded by spectral overlaps among cyanobacterial species [192], introduces uncertainties in satellite-derived HABs quantification—particularly when blooms reside below one Secchi in depth [193]. Strong winds further confound monitoring efforts by inducing the rapid spatial dispersion of HABs and reducing water transparency [194] through sediment resuspension [195]. To mitigate temporal variability, studies often aggregate data to monthly maximum surface extents [196].

The morphology and depth of lakes, as well as their topography, greatly influence HABs [197]. Shallow lakes exhibit higher susceptibility to blooms than deep systems due to enhanced light availability and wind-driven sediment interactions [198]. Bathymetry-driven currents and turbulence regulate cyanobacterial vertical positioning [190,199], while embayments frequently act as accumulation zones due to restricted water exchange.

4.1.5. Aquatic Vegetation and Vegetation on the Lakeshore

The spectral interference of aquatic and lakeshore vegetation poses significant challenges for the ORS of HABs, particularly in optically complex water bodies (Figure 7). While regions with dense aquatic vegetation typically exhibit lower HABs prevalence [200], shadows cast by vegetation can reduce water surface reflectance and mimic cyanobacterial spectral signatures, leading to false-positive HABs detections [201]. Traditional spectral indices often fail to distinguish these targets due to overlapping spectral features (e.g., near-infrared reflectance similarity between submerged macrophytes and Microcystis colonies). Moreover, the growth period of aquatic vegetation has heavy rainfall and it is difficult to obtain cloud-free images. Many methods have been used to deal with these problems and effectively distinguish aquatic vegetation from HABs in ORS.

Analyzing cloud-free multitemporal imagery is necessary to isolate persistent vegetation signals [201]. Existence frequency thresholds can be applied to differentiate seasonally stable vegetation from transient HABs [201]. Manually excluding known vegetated areas is another technique, though residual misclassification necessitates post-processing validation.

Landsat, with its high spatial resolution and long time series, is often the preferred data source for aquatic vegetation extraction [203,204]. Advanced indices such as the Aquatic Vegetation Index (AVI) and FAI improve discrimination in shallow lakes [187], while phycocyanin-sensitive sensors (e.g., Sentinel-3 OLCI) exploit the 620 nm absorption feature unique to cyanobacteria [121]. Hybrid approaches integrating classification decision trees [201,205] and ML-enhanced vegetation presence frequency indices [206] demonstrate cross-lake applicability. Nevertheless, spectral ambiguities persist during seasonal transitions when terrestrial vegetation, HABs, and aquatic plants exhibit overlapping reflectance profiles. Ground-truth data (e.g., in situ vegetation surveys and water quality parameters) remain essential for spectral library construction and model calibration [49].

4.1.6. Validation

A critical operational challenge in HABs detection lies in the absence of universally standardized thresholds, due to their dependence on site-specific lake characteristics (e.g., trophic state, dissolved organic matter content) and sensor-specific spectral configurations. For instance, thresholds based on visible surface scums, Chla concentrations, or cyanobacterial biomass are common, but they may not be universally applicable across different water bodies [207]. Although the implementation of unified threshold approaches enhances cross-lake comparability in large-scale monitoring programs, it inevitably introduces systematic biases that compromise detection accuracy, particularly when addressing lakes with contrasting trophic states or integrating multi-sensor datasets exhibiting spectral response disparities. While some studies attempt to employ common threshold methods to provide stability across lakes [208,209], this often leads to compromised accuracy due to variability in lake conditions and sensor data. When multiple sensors or monitoring individuals are used, results may diverge—one sensor might classify a pixel as a HABs in one lake, but not in another lake.

Current studies frequently employ visual criteria (e.g., surface scum presence or water discoloration) to define HABs [210]. However, greenish water coloration does not exclusively indicate HABs, as similar chromatic changes may result from blooms of Chlorophyta, Euglenophyta, Bacillariophyceae, or Dinophyta [211]. Additionally, vertically stratified water columns may harbor cyanobacterial biomass within thermocline layers, further complicating visual validation [212]. These limitations necessitate the development of quantitative detection protocols integrating spectral signatures and biomarkers to replace subjective visual assessments.

The validation of ORS typically relies on cross-comparisons with ground-based measurements [213]. However, spatial–temporal variability in water quality parameters and sampling artifacts (e.g., insufficient sample size, geolocation mismatches, or sensor calibration drift) may introduce systematic biases [214]. Short-term monitoring campaigns often fail to capture seasonal or episodic HABs dynamics [215], leading to the potential overestimation of algorithm accuracy when validated against incomplete field data.

Although multi-sensor cross-validation approaches mitigate these issues to some extent, they face inherent challenges: spectral resolution disparities between platforms, inconsistent radiometric calibration protocols, and temporal asynchrony can obscure error attribution in complex aquatic systems [216]. To enhance robustness, we recommend adopting hybrid validation frameworks that combine adaptive field sampling (e.g., event-driven campaigns), multi-platform harmonization standards, and ML-driven uncertainty quantification.

4.2. Current Directions

An increasingly important direction in current HABs research is the advancement of vertical bloom monitoring, which is essential for accurately capturing bloom dynamics and their ecological impacts—particularly in stratified lake systems. In these environments, cyanobacteria often exhibit vertical migration in response to light availability, temperature gradients, and nutrient distributions, forming dense subsurface layers that escape detection by conventional surface-only observation methods [217].

The vertical monitoring of HABs has become increasingly critical for accurate biomass estimation in eutrophic lakes, particularly where surface accumulation does not fully represent the total algal load. Traditional surface-only remote sensing methods often fail to capture subsurface biomass due to vertical heterogeneity in chlorophyll distribution. Recent advances address this limitation by integrating surface reflectance data with empirically derived algorithms that estimate biomass both within and below the euphotic zone. For example, a novel Algal Biomass Index (ABI) has been developed and validated in Lake Chaohu, China, to retrieve column-integrated algal biomass using MODIS reflectance data while accounting for vertically stratified chlorophyll profiles [218]. This approach substantially improves the accuracy of biomass estimation across time and space, and demonstrates transferability to other remote sensors like Sentinel-3 OLCI. It provides a scalable solution for large-scale monitoring and supports the long-term trend analysis essential for eutrophication management and water safety.

These integrated vertical observation frameworks not only enhance the ecological accuracy of current detection systems but also lay the groundwork for future innovations.

4.3. Future Opportunities

4.3.1. Defining Operational Thresholds for HABs Detection

Future research should focus on developing automated segmentation thresholds based on image features of bloom boundaries, potentially improving the accuracy of HABs detection. However, these methods require a substantial number of training samples and need to account for differences in sensors used across regions and lakes. A promising direction would be the establishment of region-specific thresholds that could better reflect the unique characteristics of individual lakes, while also considering the dominant cyanobacterial taxa (e.g., Microcystis vs. Dolichospermum) and ecosystem resilience. This approach should work towards unifying threshold methodologies that are adaptable yet reliable across various geographic and climatic conditions. By validating these thresholds across a wide variety of hydroclimatic settings and sensors, future monitoring could achieve a balance between consistency and accuracy in HABs detection.

4.3.2. Multi-Sensor Synergies and Data Fusion

The integration of multiple remote sensing technologies, including hyperspectral sensors, UAVs, SAR, and LiDAR, can significantly improve the spatiotemporal coverage for HABs monitoring, particularly in challenging conditions like cloud cover [219]. Spatiotemporal fusion, which merges high spatial-resolution data with high temporal resolution, offers a promising solution to overcome the limitations of single-sensor reliance [220]. Additionally, advances in satellite technology, such as expanded spectral ranges (e.g., SWIR bands), will be crucial in resolving complex water spectral features, further enhancing monitoring accuracy. The synergy of these technologies, alongside the development of sensor networks through small satellite clusters and ground-based systems, will pave the way for more efficient, accurate, and large-scale monitoring strategies for HABs [221,222,223].

4.3.3. Machine Learning

The integration of artificial intelligence (AI) into remote sensing has revolutionized HABs monitoring by enabling the efficient processing and analysis of large, heterogeneous datasets [224]. Machine learning algorithms, such as CNNs and RNNs, are improving the temporal resolution, reliability, and predictive capabilities of HABs monitoring systems. These technologies address issues like data gaps caused by cloud cover and enhance detection sensitivity, which are critical for analyzing the subtle optical signals associated with HABs. Furthermore, interdisciplinary cooperation and the establishment of international data repositories for validated algorithms and in situ datasets will be essential for standardizing methodologies and promoting innovation, thus fostering effective global monitoring efforts.

4.3.4. Future of HABs Detection

The future of HABs detection hinges on overcoming existing limitations through advancements in real-time forecasting, small-lake adaptability, and climate-specific response modeling. These developments, driven by cutting-edge technologies and interdisciplinary collaboration, promise to transform HABs monitoring into a more predictive, precise, and globally applicable science. A key frontier is the development of real-time HABs forecasting systems, which leverage dynamic environmental data to deliver early warnings. Recent research highlights the potential of deep learning models—such as ConvLSTM, iTransformer, and hybrid SSA-TCN frameworks—to achieve accurate short-term predictions, even amidst incomplete datasets or fluctuating signals [225,226,227]. By integrating remote sensing, weather, and in situ measurements, these models enhance both timeliness and accuracy. Innovations like Bloomformer-2, which embeds WHO alert frameworks, further bridge the gap between prediction and operational decision-making, offering practical tools for water management [228]. For example, such systems could prove transformative in regions like Lake Erie (USA) or Lake Taihu (China), where rapid bloom escalation demands swift action.

For small or optically complex lakes, traditional detection methods often falter due to coarse resolution or spectral noise. Tailored deep learning approaches are emerging as a solution, with studies in JinJi Lake (China) and Baltic state lakes demonstrating that deep neural networks and adaptive modeling can accurately detect and track blooms despite limited training data [229,230]. Moreover, explainable deep learning with fine-tuned transfer learning has shown promising results in improving predictive performance in data-scarce lakes and reservoirs across China, demonstrating potential applicability in under-monitored systems globally [135].

Together, these developments herald a new era of HABs detection—adaptive, regionally attuned, and capable of real-time response under diverse and evolving environmental conditions. By bridging technological innovation with ecological insight, future systems can better mitigate the growing threat of HABs worldwide.

5. Conclusions

ORS has become a vital tool in monitoring HABs in lakes worldwide, driven by the increasing prevalence of cyanobacteria. While numerous methods have been developed to effectively extract and analyze HABs, no single approach is without limitations. Challenges such as atmospheric interference (e.g., clouds and wind), lake topography, and aquatic vegetation continue to impact monitoring accuracy, underscoring the need for ongoing advancements in the field. The evolution of HABs monitoring is shifting towards a more comprehensive three-dimensional framework. This approach goes beyond surface-level observations to account for the biomass and vertical distribution of cyanobacteria, providing a deeper understanding of their ecological dynamics. The long-term objective is to develop efficient and accurate methods for HABs extraction that not only enhance detection capabilities but also provide novel insights into the ecological processes and impacts of HABs.