Mapping Harmful Algae Blooms: The Potential of Hyperspectral Imaging Technologies

Abstract

Harmful algae blooms (HABs) pose critical threats to aquatic ecosystems and human economies, driven by their rapid proliferation, oxygen depletion capacity, toxin release, and biodiversity impacts. These blooms, increasingly exacerbated by climate change, compromise water quality in both marine and freshwater ecosystems, significantly affecting marine life and coastal economies based on fishing and tourism while also posing serious risks to inland water bodies. This article examines the role of hyperspectral imaging (HSI) in monitoring HABs. HSI, with its superior spectral resolution, enables the precise classification and mapping of diverse algae species, emerging as a pivotal tool in environmental surveillance. An array of HSI techniques, algorithms, and deployment platforms are evaluated, analyzing their efficacy across varied geographical contexts. Notably, hyperspectral sensor-based studies achieved up to 90% classification accuracy, with regression-based chlorophyll-a (Chl-a) estimations frequently reaching coefficients of determination ($R^2$) above 0.80. These quantitative findings underscore the potential of HSI for robust HAB diagnostics and early warning systems. Furthermore, we explore the current limitations and future potential of HSI in HAB management, highlighting its strategic importance in addressing the growing environmental and economic challenges posed by HABs. This paper seeks to provide a comprehensive insight into HSI’s capabilities, fostering its integration in global strategies against HAB proliferation.

1. Introduction

Harmful algae blooms (HABs) are a pervasive and escalating concern for marine ecosystems, public health, and economies worldwide [1,2,3]. These phenomena occur when nutrient and temperature conditions cause microalgae to multiply rapidly and dominate a water body [4,5,6]. HABs are characterized by the rapid accumulation of algae cells, which can produce a range of toxins harmful to both marine and terrestrial life [7,8,9,10]. These events manifest in various forms, often named for their appearance, such as “red tides”, “brown tides”, and “green tides” [11,12]. While some algae blooms are nontoxic, their presence can create hypoxic or anoxic conditions by hindering water oxygen exchange, leading to widespread mortality among aquatic organisms [13,14,15]. This results in a eutrophication cycle where decomposing organic matter further feeds the growth of larger bloom events [6,16]. In addition, their opaque composition can prevent sunlight from reaching below the surface, exerting additional euthropic pressure on marine environments [17].

Key factors contributing to these events include nutrient pollution, water temperature, and sunlight availability and intensity, with temperature being reported as a critical catalyst for microalgal growth [18,19]. Additionally, the global supply chain contributes to the spread of microalgae through the movement of nonsterile ballast waters from routine ship operations [20,21]. The increase in sea surface temperatures has been identified as a key factor in the proliferation of HABs in several regions, exacerbating their frequency and geographical spread [22,23]. As a compounding factor, the looming threat of climate change is expected to aggravate these trends, making HABs a critical subject for current and future scientific investigation and monitoring efforts [23,24,25,26].

The consequences of HABs are multifaceted. While they can have significant detrimental impacts on marine and coastal enviroments, they also represent disruptions to human activities, particularly on economic and public health sectors [2,27]. Human exposure to water contaminated by HABs can result in a range of health issues, from dermatological reactions to severe neurological conditions [27,28]. The economic toll is equally severe, with industries such as tourism, shipping, fisheries, and aquaculture suffering considerable losses during HAB events [3,29,30]. For example, the 2018 HAB event in Florida, known as the “Red Tide”, led to widespread fish kills, respiratory issues among residents and tourists, and an estimated economic loss of $8 million per month to the local economy [31,32,33,34]. The ecological impact is also profound, affecting marine mammals, birds, and entire aquatic ecosystems by disrupting the food chain and causing long-lasting damage [1]. These considerations justify and motivate, on both economic and biological terms, the pressing need to develop and establish robust and precise monitoring systems to detect, classify, and predict these events [35,36].

Despite advancements in technology, monitoring HABs remains a daunting task [37,38]. Traditional methods like in situ sampling and lab-based microscopic examination are labor-intensive and time-consuming and provide only a snapshot of conditions at a particular time and location [39,40]. Remote sensing technologies, including satellite and aerial imagery, have been employed to extend the spatial and temporal coverage, but they come with their own limitations, such as reduced effectiveness under cloud cover or an inability to capture fine-scale variations in bloom density [39,41].

Given the urgent need for improved HAB monitoring, this review aims to present a comprehensive evaluation of hyperspectral imaging as a promising alternative. Hyperspectral imaging (HSI) technology captures data across a broad spectrum of wavelengths, offering a much richer and more detailed view of the surveyed area compared to traditional methods [42]. HSI analysis can distinguish between different types of algae based on their unique spectral signatures, enabling more precise identification and monitoring of HABs [43,44]. In this article, we will examine the key characteristics of HSI technology for HAB detection, emphasizing the spectral patterns that can distinguish harmful algae from other phytoplankton or background vegetation. In addition, the primary analytical tools used to leverage these spectral signatures for more accurate bloom identification will be explored. Further, the viability and considerations of various sensing platforms like satellite-based systems, unmanned aerial vehicles (UAVs), and in situ devices will be examined. Finally, the current work provides a comparative analysis of global applications, assessing the effectiveness, adaptability, and limitations of HSI in diverse environmental conditions.

By providing a thorough overview of HSI and its applications in HAB monitoring, this review aims to bridge the gap between current monitoring challenges and the potential solutions offered by advanced imaging and analysis technologies. Our goal is to highlight the strategic importance of integrating HSI into global monitoring and management strategies for HABs, ultimately contributing to better environmental and public health outcomes.

2. Hyperspectral Imaging Fundamentals

Hyperspectral imaging (HSI) is an imaging technique that enables the capture and processing of information across a wide range of wavelengths in the electromagnetic spectrum. Unlike traditional imaging, which captures data in three primary color bands (red, green, and blue), HSI preserves the spectral distribution of light reflected from a target of interest and collects data in many narrow, contiguous spectral bands, typically ranging from the visible to the near-infrared regions [42]. This high spectral resolution allows for the precise identification and characterization of materials based on their unique spectral signatures. Because each algae species (and, indeed, most biological materials) has a unique chemical composition, its spectral reflectance at each wavelength exhibits characteristic absorption features. Consequently, HSI can discern subtle variations in algal composition—akin to how it differentiates impurities in other materials—enabling species-level classification [45,46].

HSI data are alternatively referred to as hypercubes due to the morphological organization of the information within them [47]. Similar to how an RGB image is composed of red, green, and blue channels, a hyperspectral image is composed of a multitude of individual channels, or spectral bands, with each band corresponding to the perceived light intensity by the sensor at each specific wavelength. A graphical representation of the data structure of an HSI presents each spectral band as stacked on top of each other, organized by wavelength. This morphology is illustrated for clarity in Figure 1.

While HSIs provide an order of magnitude of additional information compared to conventional, visible light imaging, they introduce the necessity of additional consideration to the manipulation and processing of this surplus of information [48,49]. Unlike RGB images that can be visually analyzed using ubiquitous color screens and basic segmentation algorithms, the high dimensionality of HSI data requires researchers to employ more advanced techniques to visualize and interpret this information [50,51]. This often involves the use of machine learning algorithms for tasks such as segmentation, regression, and classification to discern the subtle nuances in the spectral responses of different materials [52,53]. The complexity of handling and analyzing hyperspectral data can be a significant challenge, requiring specialized software and expertise, which may not be readily available in all research settings. In addition, there are considerations related to operating over data in a high-dimensionality vector space, such as HSIs as shown in Figure 1 in order to ensure the accuracy and efficiency of HSI analysis algorithms [54]. In order to provide images with a high spectral resolution, HSI devices construction often requires complex arrangements of optomechanical components, which limits their ease of deployability due to increased weight, volume, and power requirements [55].

In HAB monitoring, HSI offers the advantage of remote data collection, reducing the need for direct water sampling over large areas. While minimal disturbance to the aquatic environment may not always be the primary concern, such non-contact measurement greatly eases logistics and enables faster, broader assessments by research and field data collection expeditions. This is evidenced by the increasing research interest on the spectral characterization and development of HAB detection and monitoring strategies based on HSI data, as shown in Figure 2. For these developments, a single pass of an aerial or satellite hyperspectral sensor can deliver a comprehensive dataset, capturing detailed information about the spectral signatures of different algae species. This method provides a high volume of information with reduced labor compared to traditional in situ sampling and laboratory analyses [56,57]. The thorough spectral analysis enabled by HSI allows for the development of mathematical models that describe the relationships between the spectral behavior of various algae species and key biochemical parameters such as chlorophyll, phycocyanin, and other pigment concentrations, crucial for assessing the health and proliferation potential of algal blooms [58,59,60]. Studies have shown that HSI can effectively map the spatial distribution and density of HABs, providing crucial data for early detection and management efforts [61,62,63,64].

Moreover, HSI has been utilized to evaluate water quality parameters like dissolved organic matter and nutrient levels, which are indicative of conditions conducive to HAB formation [65,66,67,68,69,70]. The technology also offers potential for indirectly identifying the presence of toxins produced by certain algae species, by discriminating between toxin-producing algae species [71,72]. These capabilities introduce important advantages for risk management from the impacts of human activity on both ecosystem and human health through the emergence of bloom events [73]. Finally, the capability to monitor these parameters in detail over large areas makes HSI technology an invaluable tool in environmental management, helping to predict and mitigate the impacts of HABs effectively [44,74,75,76].

3. Sensing Platforms and Data Acquisition

HSI technologies can be deployed on a variety of sensing platforms, each tailored to meet specific monitoring needs [77,78,79,80]. These platforms include satellites, manned aircraft, Unmanned Aerial Vehicles (UAVs), and handheld devices. Each platform provides unique capabilities and trade-offs concerning spatial and spectral resolution, measurement noise, coverage area, and deployment costs. These factors are crucial when choosing the appropriate technology for monitoring harmful algae blooms (HABs).

3.1. Satellites

Satellite platforms represent a consistent, predictable platform for large-scale environmental monitoring. They offer extensive geographic coverage and enable consistent, long-term observation of HAB dynamics across global scales [77,81,82]. This broad coverage is crucial for tracking seasonal variations and understanding long-term changes in algae distribution [83]. Furthermore, recent launches of satellite-mounted sensors such as NASA’s Ocean Color Instrument (OCI) as part of the Plankton, Aerosol, Cloud, Ocean Ecosystem (PACE) mission highlight a broader interest by space agencies to integrate HSI sensors into their Earth observation programs [84].

Among the frequently used satellites for HAB monitoring, multispectral platforms like Landsat offer moderate spatial resolution (30 m in the visible and near-infrared bands) but limited spectral coverage [85]. In contrast, newer hyperspectral satellites such as the Italian PRISMA [86] mission and Germany’s EnMAP [87] instrument collect spectral data across a more extensive range (roughly 400–2500 nm) with spatial resolutions generally near 30 m. Some spaceborne hyperspectral sensors on the International Space Station (e.g., DESIS) [88] also provide high spectral fidelity but operate over a narrower spectral range (approximately 400–1000 nm). The products from these missions can collectively enhance the granularity and accuracy of water-quality and HAB assessments by capturing important spectral absorption features specific to algal species.

However, there remains an inherent tradeoff between spatial and spectral resolution [89]. While hyperspectral satellite sensors resolve more spectral bands than conventional multispectral missions, their spatial resolution is often lower than what is achievable via airborne or UAV-based platforms [90]. This limitation may result in less detailed imagery, challenging the detection of smaller or dispersed algal blooms. Furthermore, satellite revisit frequencies and atmospheric conditions such as cloud cover can restrict data reliability during rapidly evolving HAB events, underscoring the importance of multi-platform observation strategies.

3.2. Manned Aircraft

Manned aircraft-mounted hyperspectral sensors are highly effective for rapid deployment in response to emerging HAB events. They can swiftly cover large areas with high-resolution imagery, providing detailed data much faster than UAVs and with more detail than satellite-based systems [91]. This capability makes manned aircraft ideal for emergency response and for monitoring regions that are either temporarily cloud-covered or geographically remote due to the fact that they can be operated at variable altitudes in order to eliminate the problem of high-altitude cloud cover [92,93].

Despite their advantages, the use of manned aircraft is significantly more expensive and logistically complex compared to other platforms. Manned aircraft operations require careful planning, flight clearance, and trained personnel, particularly in regulated airspaces [94]. The cost factors and logistical requirements often limit their use to monitoring missions for high-priority events or sparse, multi-sensor flights. Moreover, manned aircraft effectiveness can still be influenced by adverse weather conditions, albeit to a lesser extent than satellite platforms.

3.3. Unmanned Aerial Vehicles (UAVs)

UAVs offer a flexible and accessible means to deploy HSI imagers, particularly suited for detailed studies of localized HAB events. Operating at lower altitudes than manned aircraft, HSIs obtained from aboard a UAV present favorable spatial resolution and data accuracy levels, facilitating the precise mapping of bloom boundaries and density assessments [95,96]. These vehicles can be deployed quickly and maneuvered into positions that are not feasible for larger manned aircraft or satellites, allowing for detailed, site-specific analysis [97,98]. As a sensing platform, the main limitations of UAVs include their shorter operational durations due to battery life and their smaller payload capacities, which restrict the size, weight, and power requirements of onboard equipment [80,99,100]. Additionally, UAV operations are subject to airspace regulations and may require specific permits, which can be a regulatory or logistical barrier in certain regions and overall hinder the deployment speed for the monitoring of idiosyncratically transient bloom events [101,102].

While UAVs can map nearshore or small-lake HABs with high spatial resolution, their offshore applicability depends on flight endurance, regulatory constraints, and advanced imagery techniques to handle uniform water surfaces, such as sun-glint correction and multi-image stitching [103,104]. Nonetheless, modern UAV designs, including fixed-wing and VTOL hybrid platforms, can achieve significantly longer flight times and cover moderate distances from shore, provided that appropriate launch-and-recovery logistics are in place [105].

3.4. Handheld Devices

Handheld hyperspectral cameras provide significant advantages for in situ data collection, especially for immediate, site-specific analysis. These portable devices enable researchers to conduct rapid assessments on the ground, facilitating quick decision-making and thorough investigation of specific areas within a bloom. Furthermore, they facilitate the acquisition of physical samples directly from the event location for further study due to the physical requirement of having personnel on site.

Each of these platforms has its own set of advantages and limitations, which must be carefully considered when designing a monitoring program for HABs. Satellite platforms dominate the reviewed literature due to their extensive spatial coverage and cost-free data availability [63,82,106], although insufficient spectral bands and cloud coverage can reduce sensitivity to toxins or overshadow transient bloom events. UAVs enable flexible, high-resolution data capture at local scales but rely on shorter flight times and have limited payload capacities [59,107,108,109], making them most effective for targeted monitoring rather than continuous regional surveillance.

One noteworthy occurrence is the example of a manned aircraft-based hyperspectral survey [110], illustrating that while HSI acquisition from a manned aircraft platform can cover broad areas more rapidly than UAVs and at higher spatial resolutions than many satellites, the practical challenges of mission planning, airspace regulation, and operational costs can be formidable, as mentioned in Section 3.2. In contrast, handheld and static-mounted systems enable in-depth, close-range hyperspectral measurements, permitting fine-level analyses and direct sample collection [71,111,112,113]. As highlighted in

Table 1. Comparative summary of sensing platforms used in recent HAB studies, highlighting the technological, operational, and environmental limitations encountered.

Study	Year	Location	Sensing Platform	Limitations
[82]	2017	Florida and Ohio, USA	Satellite	Insufficient spectral information for toxin detection.
[106]	2020	San Roque reservoir, Argentina	Satellite	Study relies on a previously developed semi-empirical algorithm to retrieve Chl-a.
[107]	2021	Geum River, Korea	UAV	Further field measurements required for the model to generalize vertical cyanobacteria profiles in other lake environments.
[114]	2021	11 States, USA	Satellite	Further data collection required to model variations in toxin extraction protocols and model accuracy.
[111]	2021	Texas, USA	Handheld	HSI acquisition should be limited to active growth stages for accurate identification due to biological stress variations.
[112]	2021	Virginia, USA	Handheld	None specified.
[108]	2022	Alabama, USA	UAV	All four considered sensors were more sensitive to Chl a concentrations than phycocyanin, likely because phycocyanin absorbs light at 620 nm.
[110]	2022	Geum River, Korea	Manned Aircraft	Using full-dimensionality HSIs may lead to increases in complexity, uncertainty, input noise, and overfitting.
[63]	2022	Klamath River, California, USA	Satellite	Temporal resolution mismatch, unpredictable cloud cover, insufficient temporal resolution for transient bloom events.
[109]	2022	Daecheong Lake, Korea	UAV	Further validation required to estimate various secondary pigments to more accurately model algal phenomena.
[115]	2023	Billings Reservoir, Brazil	Satellite	Spatial resolution of HSI sensor limits the broader applicability of the model.
[113]	2023	Rivers in Korea	Handheld	Study measured only two types of green algae and blue-green algae.
[116]	2023	Geum River, Korea	Satellite	Algal bloom-specific indicators based on remote sensing information might require dominant bands of satellite imagery to account for sensitivity factors.
[59]	2023	UCFR and Gallatin River, Montana	UAV	The results may not be applicable to other times, conditions, and rivers.
[71]	2024	Madrid, Spain	Handheld	The experiment was carried out with one representative species for each genus. Cultures in the lab do not fully represent natural ecosystems.
[117]	2024	Western region of Lake Erie, USA	Satellite	Phycocyanin is present in low concentrations. Microcystin has limited spectral sensitivity. Secchi-depth could be influenced by various factors.

, their limitations include the small spatial extent of sampling, the variability introduced by individual operators or lab conditions, and the environmental differences between field measurements and lab measurements in controlled environments.

For categorical clarity, note that static setups on tripods or rail-mounts, such as those commonly utilized in laboratory settings, are considered “handheld” in the context of Table 1. These solutions, while not offering large-scale coverage, can provide essential calibration or validation data supporting higher-altitude platforms. Ultimately, the selection of an appropriate sensing platform must be tailored to the specific requirements of each study area, the severity or spatial distribution of HABs, and the logistical and financial constraints at hand.

Geographically, the evaluated studies span various regions, including the United States [63,82,108,112,114], Latin America [106,118,119,120,121], China [122,123], and South Korea [116], underscoring the global importance of HAB monitoring and the ubiquitous presence of HABs. For illustrative purposes, the geographical distribution of the evaluated studies is presented in Figure 3.

These findings highlight the diversity of sensing platforms employed in recent HAB-related studies. Manual sensing approaches in laboratory setups, such as those using HSI cameras and spectrometers, provide controlled conditions that are ideal for detailed analysis but are confined to small-scale or experimental scenarios. Satellite-based platforms, in contrast, offer extensive spatial coverage and continuous monitoring capabilities, making them suitable for large-scale observations. Despite their susceptibility to variations in sampling methodologies and meteorological conditions, which can impact data consistency and reliability, satellites remain the predominant sensing platform, involved in 83% of the studies listed due to the widespread availability of remote sensing data in public databases. Manual methods, often paired with advanced sensors with fewer portability, weight, remote operation, and power requirement concerns, still present logistical challenges due to their labor-intensive nature, rendering them impractical for extensive area coverage. UAV-mounted systems offer a middle ground, combining flexibility with specific sensing technologies. However, they face significant challenges such as the integration of HSI sensors onto UAVs, balancing their weight and power requirements, and logistical challenges associated with UAV deployments in remote areas.

Notably, we found no recent, peer-reviewed studies employing manned aircraft specifically for HAB-targeted hyperspectral surveys. This may be a reflection of the costs (pilot, aircraft rental, in addition to the HSI acquisition equipment itself) and airspace constraints discussed in Section 3.2, as well as the growing popularity of UAVs for flexible, high-resolution data capture [124,125].

The trends observed in the evaluated studies indicate a compelling case for future HAB monitoring efforts to be multi-platform to exploit the advantages of each platform. The use of handheld imagers for specific problem areas or lab environments are invaluable for obtaining detailed spectral data with the lowest noise levels, facilitating correlation with individual algae species or water quality parameters [112]. These detailed data are pivotal for enhancing lower-resolution remote sensing data through spectral unmixing techniques, as successfully implemented by other groups in [126,127,128]. This progressive approach, scaling from lab to UAV to satellite data, manages to increasingly trade spectral resolution for spatial coverage effectively, while preserving and utilizing insights from each phase to augment subsequent steps.

4. Data Analysis Techniques and Innovations

HSI acquisition devices generate high-dimensional data that pose unique challenges and opportunities for environmental monitoring [129,130]. As explored in Section 1, the large volume of highly correlated information contained in HSIs makes them unfeasible at best to be analyzed manually by a human researcher in reasonable timeframes [131]. For this reason, effective data analysis methodologies are crucial for translating the large volumes of acquired spectral data into actionable insights, particularly in the context of harmful algal bloom (HAB) detection and monitoring. Determining the precise mathematical relationships between the spectral measurements taken from HSIs of HAB events and the actual target class or magnitude of interest—whether the application focuses on regression or classification—is a highly complex task. Establishing accurate mathematical models to describe the relationship between spectral data and target magnitudes or classes requires careful evaluation of existing methodologies and the development of innovative approaches.

This section will delve into various data analysis techniques used in HSI for HAB monitoring, comparing traditional and advanced methods, and discussing their respective strengths and limitations, which are summarized in

Table 2. Comparative summary of algorithmic approaches in recent HAB studies, highlighting their spectral range, evaluation metrics, and values.

Study	Year	Object of Study	Water Body	Spectral Range (nm)	Methodology	Evaluation Metric	Metric Value
[82]	2017	Cyanobacterial harmful algal bloom (cyanoHAB) frequency	Freshwater	400–900	Classification; Cyanobacteria Index (CI)	Accuracy	0.864
[114]	2021	CyanoHAB presence/absence	Freshwater	400–900	Classification; Cyanobacteria Index (CIcyano)	Accuracy	0.84
[111]	2021	Cyanobacteria detection and identification	Freshwater	400–1000	Classification; Spectral Mixture Analysis (SMA); Spectral Angle Mapper (SAM)	Accuracy	Up to 99%
[108]	2022	Chlorophyll-a and phycocyanin concentrations	Freshwater	400–900	Regression; 26 Vegetation Indices	$R^2$	Up to 0.87
[110]	2022	Phycocyanin (PC) and chlorophyll-a (Chl-a)	Freshwater	400–900	Regression; Artificial Neural Network (ANN)	Coefficient of Determination ($R^2$)	0.80 (PC), 0.74 (Chl-a)
[132]	2022	Microalgae pecies	Saltwater	400–700	Classification; Spectral Angle Mapper (SAM)	p-Value	91–100%
[115]	2023	Phycocyanin (PC) concentration	Freshwater	400–900	Regression; Random Forest; Extreme Gradient Boost; Support Vector Machines	Mean Absolute Error (MAE)	0.45
[116]	2023	Harmful algal bloom (HAB) monitoring	Freshwater	400–900	Classification; Super-Resolution Convolutional Neural Network	Peak Signal-to-Noise Ratio (PSNR)	36.11 dB 1
[59]	2023	Chlorophyll-a and phycocyanin standing crops	Freshwater	400–900	Regression; Spectral Band Ratios	$R^2$	Up to 0.86
[121]	2023	Chlorophyll-a and cyanobacteria concentration	Freshwater	600–730	Regression; Multi-Band Indexes	Root Mean Squared Error (RMSE)	47.6 µg/L (Chl-a), 35.1 µg/L (cyanobacteria)
[71]	2024	Cyanobacteria genera discrimination	Freshwater	400–1000	Classification; Random Forest	Accuracy	Up to 95%
[117]	2024	Bloom proxies: chlorophyll-a, microcystin, phycocyanin, secchi-depth	Freshwater	400–900	Regression; Random Forest	$R^2$	0.55 (Chl-a)
[133]	2024	Algae classification (dense vs. sparse algae presence)	Saltwater	400–900	Classification; SVM, CNN
[134]	2024	Phytoplankton abundance	Saltwater	400–700	Regression; Fourth-Derivative Spectral Similarity Index	Correlation Coefficient	0.542
[135]	2024	Algae bloom detection	Saltwater	450–890	Classification; HAB-Net	Precision	0.901
[136]	2024	Cholorophyll-a forecasting	Saltwater	950–100	Regression; Partial Least Squares	$R^2$	0.9
[137]	2025	Water Fouling Index estimation	Saltwater	604–686	Regression; Random Forest, CNN	Mean Squared Error (MSE)	435.21 (CNN), 2034.22 (RF)

¹ Although dB commonly refers to decibels of sound intensity, it is also the conventional unit of measure for signal ratios such as PSNR in image processing contexts. A higher PSNR (in dB) generally indicates better reconstruction quality or less distortion from the original signal.

.

4.1. Model-Based vs. Learning-Based Approaches

A variety of algorithmic techniques have been applied for HSI processing in HAB monitoring applications. We divide exploration of these techniques into Section 4.1.1 and Section 4.1.2.

4.1.1. Model-Based Approaches

Model-based methods rely on explicit mathematical relationships between measured reflectance at particular wavelengths, and biophysical or biochemical parameters. One straightforward example is the use of spectral indices such as the Normalized Difference Vegetation Index (NDVI) or Normalized Difference Chlorophyll Index (NDCI). Structurally speaking, a generic normalized difference index can be computed as follows:

$$\mathrm{Index}={\displaystyle \frac{R\left({\lambda }_{\mathrm{A}}\right)-R\left({\lambda }_{\mathrm{B}}\right)}{R\left({\lambda }_{\mathrm{A}}\right)+R\left({\lambda }_{\mathrm{B}}\right)}},$$

(1)

where $R\left({\lambda }_{\mathrm{A}}\right)$ and $R\left({\lambda }_{\mathrm{B}}\right)$ are reflectances at any two given wavelengths ${\lambda }_{\mathrm{A}}$ and ${\lambda }_{\mathrm{B}}$, chosen with the objective of maximizing contrast in absorption features of target pigments (e.g., chlorophyll-a, phycocyanin). Variations of this formulation, or more advanced linear/nonlinear spectral unmixing models, allow researchers to infer the presence of specific algae or predict algal biomass from the observed abundance of their chemical components. A representative example for Chl-a detection might focus on the red-edge region (665–705 nm), where chlorophyll absorption is pronounced, yielding a targeted red-edge index for HAB detection.

4.1.2. Learning-Based Approaches

In contrast, learning-based methods (e.g., Random Forest, Support Vector Machine, and various neural network architectures) treat the mapping from HSI data to bloom classification or concentration as a general function $f(\mathbf{x};\mathit{\theta })$. The function f is learned from labeled training data rather than being specified a priori. For example, a simple feed-forward neural network can be described as follows:

$$f(\mathbf{x};\mathit{\theta })=\sigma \left(W^{\left(L\right)}\sigma \left(W^{(L-1)}\cdots \sigma (W^{\left(1\right)}\mathbf{x}+{\mathbf{b}}^{\left(1\right)})\cdots +{\mathbf{b}}^{(L-1)}\right)+{\mathbf{b}}^{\left(L\right)}\right),$$

(2)

where $\mathbf{x}$ is the input reflectance vector (e.g., a pixel spectrum across multiple bands, such as the colored HSI pixels shown in Figure 1), $\mathit{\theta }=\{W^{\left(l\right)},{\mathbf{b}}^{\left(l\right)}\}$ are the trainable weights and biases for each layer $l\in \{1,\dots ,L\}$, and $\sigma (·)$ denotes a nonlinear activation function (e.g., ReLU or sigmoid). The parameters are optimized by minimizing a loss function, such as mean squared error (for regression of algae concentration) or cross-entropy (for bloom vs. non-bloom classification):

$$\underset{\mathit{\theta }}{min}\sum _{i=1}^N\mathcal{L}\left(f({\mathbf{x}}_i;\mathit{\theta }),y_i\right),$$

(3)

where $\mathcal{L}$ could be the mean squared error if the goal is to predict Chl-a concentration, or a cross-entropy term if the goal is to classify algae bloom presence.

Model-based approaches often have lower data requirements but may fail to capture complex spectral variability outside their predefined relationships. Learning-based methods are more flexible but risk overfitting and typically demand larger training datasets for robust HAB detection.

From the summarized findings in Table 2, it is evident that a wide range of algorithms are employed to address the challenges posed by HSI data in HAB monitoring. The use of both model and learning-based approaches across various studies indicates a diverse range of possibilities for tackling the problem of high-dimensional data interpretation. Although satellites collect global data, high-quality labeled HAB samples may be sparse, especially under frequent cloud cover or limited ground-truth campaigns. Consequently, for training complex deep learning models, the representative dataset of HAB vs. non-HAB pixels can indeed be too small or patchy, increasing the risk of overfitting.

Traditional methods are often constrained by their simplicity, offering less flexibility in handling the complex nature of hyperspectral data effectively [138,139]. However, their relative computational simplicity enables these methodologies to be employed as part of independent insight extraction systems aboard a given sensing platform [140,141]. In addition, they have the capability of extracting insights from lower data volumes without being exposed to the higher risks of overfitting that more complex methodologies do [130,142].

Alternatively, advanced techniques based on deep neural networks, although computationally intensive, provide a greater depth of analysis through sophisticated pattern recognition and learning capabilities [138,143,144]. These advanced methods, such as CNNs and GANs, can uncover subtle spectral differences that traditional techniques may miss [138,142]. However, the increased data requirements, extensive and power-hungry training operations, and the need for large, well-annotated datasets pose significant challenges [142,145]. The computational burden of these advanced techniques often necessitates high-performance hardware, which may not always be feasible for onboard processing on remote sensing platforms [146]. Furthermore, the risk of overfitting is heightened in these complex models, particularly when training data are limited or not representative of all possible conditions encountered in the field [147,148]. This is of particular interest to the topic of HAB event detection and monitoring applications following the practice of utilizing satellite data for model training, as the consistency of quality data cannot always be ensured due to varying atmospheric conditions and the time delay between satellite passes over a given region [149,150].

In terms of algorithm performance, Table 2 provides a concise comparative overview of advanced techniques such as Artificial Neural Networks, Random Forests, and SVMs. For instance, ANN-based models in [110] reported $R^2$ values around 0.74–0.80, whereas ensemble methods like Extreme Gradient Boosting in [115] achieved a lower mean absolute error in phycocyanin estimation. The observed disparities highlight that the best-performing approach often depends on the specific water body, target pigment, and available labeled data. Hence, it is of utmost importance to establish that no single advanced technique is universally optimal. Each individual technique offers distinct advantages in classification or regression tasks based on data volume, computational resources, and the complexity of the HAB environment of interest.

Despite these challenges, the application of advanced machine learning techniques has shown promise in enhancing the accuracy and reliability of HSI data analysis for HAB monitoring [135,151]. The integration of these sophisticated algorithms can significantly improve the precision of HAB detection and classification, leading to more effective automated environmental monitoring and management strategies. Additionally, innovations in algorithm efficiency and the development of low-complexity machine learning models are crucial for enabling the deployment of advanced techniques in resource-constrained environments, such as onboard satellite systems or UAVs [141,152,153].

Furthermore, the development of hybrid approaches that combine traditional and advanced methods is emerging as a promising direction. By leveraging the strengths of both types of algorithms, future research efforts can produce more robust and versatile models. For example, traditional indices such as NDVI and NDCI could be used to pre-process and reduce the dimensionality of hyperspectral data, which are then further analyzed using advanced machine learning techniques. This hybrid approach can help mitigate some of the computational challenges associated with high-dimensional data while still benefiting from the enhanced analytical capabilities of machine learning models. Recent work has demonstrated how hybrid approaches that integrate lower-level satellite imagery with higher-resolution UAV measurements can improve retrieval accuracy in applications such as maize chlorophyll estimation [154], intertidal seaweed biomass mapping [155], grassland aboveground biomass monitoring [156], and soil salinity inversion [157]. Although these studies primarily focus on terrestrial or nearshore use-cases rather than HAB detection, they illustrate the broader feasibility and advantages of hybrid pipelines to handle diverse spectral signatures, bridge differences in spatial resolution, and reduce computational overhead.

The evaluation metrics used in these studies, ranging from precision and accuracy to more complex measures underscore the critical aspects of algorithm performance and their evaluation. These metrics not only reflect the effectiveness of each method in accurately detecting and classifying HABs but also highlight the ongoing need for robust, reliable analysis techniques that can operate under varied environmental conditions. The continuous improvement of these evaluation metrics is essential for advancing the field and ensuring that the most effective techniques are adopted in practice.

One of the consistent themes across the relevant studies is the trade-off between computational complexity and analytical precision. This trade-off is particularly relevant for applications requiring real-time data analysis on onboard systems, where computational resources are limited. Innovations in algorithm efficiency, particularly in deep learning, are critical to advancing the field, allowing for more sophisticated analyses to be conducted directly on the platforms where data are collected, thereby reducing the time between data acquisition and decision-making. In addition, continued developments on edge computing systems could provide a solution for onboard HSI processing. In this manner, a “best-case” scenario for HAB monitoring would seamlessly integrate advanced neural networks with embedded or cloud-edge computing to achieve rapid, high-accuracy analysis of hyperspectral data. For example, a UAV carrying a lightweight hyperspectral sensor could transmit compressed data or key spectral features to an onboard GPU [158,159,160] or a nearby ground station. By running a carefully optimized deep learning model, this pipeline could deliver near-real-time classification results with minimal latency, enabling operators to detect nascent algal blooms immediately and respond proactively. Such an approach balances the high analytical precision offered by deep neural networks against the limited computational and power budgets typical of mobile platforms, maximizing the operational utility of HSI in resource-constrained, real-world applications. This synergy—uniting advanced algorithms, efficient hardware, and low-latency connectivity—epitomizes how ongoing innovations can resolve the central trade-off between performance and practicality in hyperspectral data analysis for HAB detection.

A potential cost-saving strategy for HSI data generation is data-driven reconstruction, wherein the contents of an HSI are approximated from lower-resolution multispectral data. Although methods exist that learn a mapping from a few discrete bands to a denser spectral range, these reconstructions can be limited when the multispectral sensor lacks coverage in critical wavelengths. In these cases, the model can produce spectral details that are not originally measured, raising concerns about reliability and interpretability in HAB detection of these generated HSIs. As a result, while these techniques may help reduce costs, they are prone to inaccuracies if the sensor is blind to relevant algal absorption features.

Dimensionality reduction techniques play a crucial role in managing the high-dimensional nature of hyperspectral data, making it more manageable for subsequent analysis. Methods such as Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and advanced approaches like autoencoders and feature selection algorithms significantly enhance computational efficiency and reduce the risk of overfitting [161,162,163,164,165,166]. Efficient data preprocessing not only ensures that algorithms can extract accurate information but also maintains computational complexity at accessible levels, enabling deployment on mobile platforms or ground-based systems with computational complexity, cost, and power consumption constraints [158,167,168]. This can aid in the democratization of HSI technology deployment, allowing for effective monitoring over larger areas and the integration of diverse data sources, which is critical for comprehensive environmental monitoring strategies [169,170,171].

Accuracies in HAB or chlorophyll-a (Chl-a) estimation using HSI-based techniques typically range between 70% to 90% under field conditions, as reported in [59,108,110]. These studies have shown that relatively simpler spectral indices (e.g., NDVI, NDCI) can often achieve $R^2$ values around 0.70–0.80 for moderate Chl-a concentrations (about 5–20 µg/L). However, complex or hybrid algorithms, such as Artificial Neural Networks or advanced Random Forests, may be required when extremely low (<5 µg/L) or very high (>50 µg/L) Chl-a values are present, or when distinguishing multiple algal species in a highly mixed bloom. These robust algorithms can improve accuracy up to 90–95%, but necessitate larger, well-annotated datasets to avoid overfitting. In freshwater lakes with dense blooms and relatively turbid water, performance tends to be higher due to the stronger reflectance signal from algae. By contrast, in coastal or open-ocean waters, achieving similar accuracy can be more difficult since background signals and water optical properties are more variable. These findings underscore that the choice of algorithm can hinge on the bloom intensity, water type, and data availability for training.

Ultimately, the exploration of data analysis techniques in HSI for HAB monitoring reveals a dynamic and evolving field. The continuous improvement of algorithms and adaptation to new computational technologies are paving the way for more precise, efficient, and actionable environmental monitoring strategies. It is the aim of the authors that the insights gained from this comparative analysis serve as a foundation for future research and development efforts in hyperspectral data analysis, guiding them towards the optimization of these techniques for better environmental stewardship.

5. Current Challenges in HSI Implementation

The implementation of hyperspectral imaging (HSI) for environmental monitoring, particularly for harmful algal bloom (HAB) detection, faces several significant challenges. These challenges span technical limitations, scalability issues, and the need for standardization across different applications. Addressing these challenges is crucial for the widespread adoption and effective utilization of HSI technologies [151,172].

One of the primary technical limitations is the capacity of current processing systems. HSI generates vast amounts of data, requiring substantial computational resources for processing and analysis. Advanced machine learning algorithms, which are necessary for extracting meaningful insights from HSI data, further exacerbate these demands. While cloud computing offers a potential solution by providing scalable processing power, reliance on remote servers introduces latency and dependency on stable internet connections or private communication links, which may not be feasible in all monitoring scenarios [173,174,175,176]. Furthermore, the increasing global demands on resources such as power, water, and minerals for the development of large-scale computational models highlight the importance of focusing on efficiency improvements. Rather than solely increasing performance through greater computational requirements, it is essential to develop more efficient algorithms that can achieve high performance with lower resource consumption [177,178,179]. This approach is particularly crucial in the context of accelerating global climate change and the need for sustainable solutions in environmental monitoring.

Another challenge is the temporal inconsistency of satellite data. Satellite-based HSI systems are constrained by their orbital paths, leading to periodic data collection that may miss critical events [180,181]. For instance, the infrequent satellite passes can result in gaps in monitoring, failing to capture rapid changes in algal bloom dynamics. Moreover, atmospheric conditions such as cloud cover can obstruct satellite sensors, leading to incomplete datasets [182]. This inconsistency necessitates the integration of data from multiple sources, including UAVs and ground-based sensors, to ensure comprehensive and continuous monitoring [183].

The high cost of hyperspectral equipment is another significant barrier to widespread HSI implementation [184]. Hyperspectral sensors are typically more expensive than their multispectral counterparts, making it challenging for many organizations to invest in this technology. This cost factor often limits HSI applications to research and development rather than operational deployment in production environments. To address this, a potential strategy is to use HSI for the initial study and model development phases and then deploy multispectral systems for routine monitoring. Multispectral sensors, though less detailed, can provide sufficient data for ongoing monitoring once robust models have been established using HSI [59,121,185].

Scalability is also a critical issue. Implementing HSI at a large scale requires significant infrastructure, including high-performance computing facilities and extensive data storage capabilities. Additionally, the deployment of HSI systems over large areas necessitates coordinated efforts and substantial logistical support, which can be challenging in remote or resource-limited regions. These scalability challenges highlight the need for innovations in data compression, efficient algorithms, and decentralized processing methods that can operate on less powerful hardware [170,186,187].

Moreover, there is a pressing need for standardization in HSI applications. Currently, the lack of standardized protocols and methodologies across different studies hampers the ability to compare results and integrate data from diverse sources. Standardization would facilitate the development of universal models and algorithms that can be widely adopted, enhancing the reproducibility and reliability of HSI-based monitoring systems [44,188,189]. Efforts to establish common frameworks and guidelines are essential for advancing the field and ensuring consistent and accurate application of HSI technologies.

The overcoming of these processing capacity limitations, temporal inconsistencies, high equipment costs, scalability issues, and the lack of standardization are critical steps towards the effective wide-scale implementation of HSI-based monitoring systems. Continued research, collaboration, and innovation in these areas will be pivotal in advancing the capabilities and accessibility of hyperspectral imaging for environmental monitoring.

6. Future Directions and Emerging Technologies

The continuous advancement of hyperspectral imaging (HSI) technology and its application to environmental monitoring initiatives necessitates ongoing research and interdisciplinary collaboration. This section discusses potential advancements in HSI technology and approaches to improve its effectiveness, precision, and applicability.

The use of cloud-based platforms for data processing and analysis is a significant innovation in the field. Cloud computing offers scalable resources that can handle the computational demands of advanced machine learning algorithms. This approach can not only reduce the cost, power, and weight constraints on onboard systems but also allows for the integration of large datasets from multiple sources, facilitating more comprehensive and accurate analysis to be performed [190]. In addition, advances in the development of faster and more power-efficient communication systems, such as those based on 5G and further mobile technologies, play a crucial role in this context [191,192,193]. Leveraging these technologies in conjunction with cloud computing can effectively allow for the outsourcing of complex processing methodologies off-device, without sacrificing inference speed or quality. The ability to process data in near real time using cloud infrastructure, enabled by high-speed communication networks, can significantly improve the responsiveness of HAB monitoring systems, enabling quicker decision-making and intervention without introducing additional hardware and power usage constraints to systems aboard a given sensing platform. This synergy between cloud computing and advanced communication technologies offers a promising path forward for enhancing the efficiency and effectiveness of environmental monitoring systems.

Another critical area of development is the use of transfer learning and domain adaptation techniques. These methods can help address the challenge of limited annotated data by transferring knowledge from one domain (e.g., a different geographic region or type of algal bloom) to another. Transfer learning can reduce the amount of training data required and improve model performance in new or underrepresented areas [194,195]. This approach is particularly valuable for HAB monitoring, where conditions can vary widely across different bodies of water and regions.

Additionally, the integration of multi-sensor data fusion techniques is enhancing the capabilities of HSI for HAB monitoring. By combining data from different sensors, such as optical, thermal, and radar, researchers can create more comprehensive models that capture a wider range of environmental variables and alleviate the limitations of optical-based systems subject to cloud interferece [190]. Multi-sensor fusion can improve the robustness and accuracy of HAB detection and classification by providing complementary information that single-sensor approaches might overlook or underperform with [196,197].

Further research is needed to address the remaining challenges and fully realize the potential of HSI in HAB monitoring. This includes developing more efficient algorithms, enhancing data acquisition and preprocessing techniques, and improving the integration of multi-sensor data, as elaborated on in Section 5. In addition, the continued technological development in HSI acquisition device cost reduction should result in an increasing number of applications that will contribute to a growing knowledge base in this field. Collaboration across disciplines, including remote sensing, computer science, and environmental science, will be crucial for advancing the state of the art and ensuring that HSI technology can be effectively applied to mitigate the impacts of harmful algal blooms.

7. Policy and Economic Implications

Evaluating the economic relevance of HSI applications for monitoring HABs and discussing how effective implementations can mitigate their consequences are crucial. Effective strategies and recommendations for policymakers on utilizing this technology for the development of more effective environmental and public health strategies must be provided. The high cost of hyperspectral equipment is a significant barrier, as previously discussed in Section 5, making it imperative for policymakers to consider funding and incentives for research and development in this area [198,199].

Beyond the important but often regarded as economically inviable aspect of preserving marine biodiversity and ecological health [200], investing in HSI technology research and development can lead to substantial long-term savings by preventing the adverse effects of HABs on public health, fisheries, tourism, and water quality [201,202,203]. Both policymakers and researchers should promote the standardization of HSI applications, ensuring that data collected from different sources and regions can be integrated and compared effectively. This standardization can enhance the development of universal models and algorithms, increasing the reliability and reproducibility of HSI-based monitoring systems. Encouraging collaborative synergy between governmental agencies, research institutions, and private sectors can also facilitate the advancement and deployment of HSI technologies [204].

Moreover, the adoption of HSI technology in environmental monitoring can create new economic opportunities. The development and commercialization of HSI sensors and data analysis services can drive job creation and economic growth. Policymakers should support the creation of public–private partnerships to foster innovation and accelerate the adoption of HSI technologies in various sectors.

Finally, the policy and economic implications of HSI technology are significant. By addressing the challenges and leveraging the opportunities presented by HSI, policymakers can develop effective strategies to monitor and mitigate the impacts of HABs, protect public health, and promote sustainable economic development [205]. Continued investment and collaboration in this field will be essential for realizing the full potential of hyperspectral imaging in environmental monitoring.

8. Conclusions

Hyperspectral imaging has emerged as a powerful tool for monitoring harmful algal blooms (HABs) in both freshwater and marine environments. This review examined fundamental HSI concepts, highlighted major sensing platforms (satellites, manned aircraft, UAVs, handheld systems), and synthesized modern data analysis techniques ranging from straightforward spectral indices to advanced deep neural networks. While each approach offers unique trade-offs in cost, coverage, and accuracy, collectively they underscore how multi-platform, hybrid solutions can mitigate limitations in spectral resolution, spatial coverage, or data sparsity. Conversely, the additional information volume contained in HSIs in contrast to RGB images introduces the requirement of developing robust methodologies for data sanitation and feature extraction in order to provide optimal solutions.

Main Findings

• HSI-based HAB detection frequently achieves 70–90% accuracy (or R22 values > 0.7) in moderate conditions, with complex or hybrid algorithms needed for extreme cases of algal concentration or highly mixed species.
• Satellite sensors remain the most popular due to wide coverage and open-access data, yet their spectral limitations and revisit constraints can hamper real-time detection of transient events.
• UAVs and handheld devices excel at fine-scale analysis, but their narrower coverage and logistical requirements restrict broader operational use.
• Both model-based (e.g., spectral indices) and learning-based approaches (e.g., convolutional neural networks) demonstrate substantial potential for HAB monitoring applications, with each showing different sensitivities to data volume and environmental complexity.

Future Directions

• Improved synergy between UAV, satellite, and lab-based HSI data can yield significantly improved cross-scale detection.
• More robust machine learning algorithms need to address domain adaptation, data scarcity, and edge-deployment efficiency for real-time HAB alerts.
• Innovations in hardware miniaturization and cost reduction remain vital for expanding HSI to resource-limited regions.
• Standardization of measurement protocols, data calibration, and reporting formats is critical for consistent inter-study comparison.

By embracing these strategies and harnessing the convergent progress in sensor design, data analysis, and policy-driven research, HSI can play a decisive role in the proactive assessment, management, and mitigation of harmful algal blooms worldwide.