Highlights
What are the main findings?
- There are two main model categories for predicting nutrient and DO concentration: empirical and artificial intelligence (AI). Model accuracy is strongly influenced by the size of the training dataset. Comparative studies conducted in the same study area have shown that ML and NNs achieve higher prediction accuracy than empirical models under the same sample size. However, the exact number of samples needed for either approach is not clearly provided in the literature. ML models (e.g., SVR, XGBoost) often outperform NNs when training data are limited, as they are generally less prone to overfitting under small-sample conditions.
- Current models are site-specific, and their transferability across different regions may be restricted. To improve their transferability, models should be validated across diverse regions and time periods, including a wide range of environmental conditions (e.g., trophic state, seasonality, depth and bottom characteristics).
What is the implication of the main finding?
- In Case II water bodies, eutrophication has emerged as a major water quality issue, calling for prompt management actions. Remote sensing facilitates its monitoring through changes in ocean color linked to chlorophyll a (chl a) fluctuations. While chl a indicates the effects of eutrophication, nutrient levels—its primary drivers—are key to predicting phytoplankton growth and implementing preventive measures.
- Long-term monitoring of nutrients and DO, combined with multiple water quality indicators (e.g., chl a) based on remotely sensed data, could enable more efficient assessment of the trophic state of water bodies and facilitate timely management actions.
Abstract
Eutrophication has emerged as a critical threat to water quality degradation and ecosystem health on a global scale, calling for prompt management actions. Remote sensing enables the monitoring of eutrophication by detected changes in ocean color caused by fluctuations in chlorophyll a (chl a). Although chl a is a crucial indicator of phytoplankton biomass and nutrient overloading, it reflects the outcome of eutrophication rather than its cause. Nutrients, the primary “drivers” of eutrophication, are essential indicators for predicting the potential phytoplankton growth in water bodies, allowing adoption of effective preventive measures. Long-term monitoring of nutrients combined with multiple water quality indicators using remotely sensed data could lead to a more precise assessment of the trophic state. Retrieving non-optically active constituents, such as nutrients and DO, remains challenging due to their weak optical characteristics and low signal-to-noise ratios. This work is an attempt to review the current progress in the retrieval of un-ionized ammonia (NH3), ammonium (NH4+), ammoniacal nitrogen (AN), nitrite (NO2−), nitrate (NO3−), dissolved inorganic nitrogen (DIN), phosphate (PO43−), dissolved inorganic phosphorus (DIP), silicate (SiO2) and dissolved oxygen (DO) using remotely sensed data. Most studies refer to Case II highly nutrient-enriched water bodies. The commonly used spaceborne and airborne sensors, along with the selected spectral bands and band indices, per study area, are presented. There are two main model categories for predicting nutrient and DO concentration: empirical and artificial intelligence (AI). Comparative studies conducted in the same study area have shown that ML and NNs achieve higher prediction accuracy than empirical models under the same sample size. ML models often outperform NNs when training data are limited, as they are less prone to overfitting under small-sample conditions. The incorporation of a wider range of conditions (e.g., different trophic state, seasonality) into model training needs to be tested for model transferability.
1. Introduction
Over the past few decades, eutrophication has become one of the most widespread water quality issues globally. According to the European Marine Strategy Framework Directive (MSFD), eutrophication is defined as “a process driven by enrichment of water by nutrients, especially compounds of nitrogen (N) and/or phosphorus (P), leading to increased growth, primary production and biomass of algae, changes in the balance of organisms and poor water quality” [1]. Biodiversity loss and ecosystem degradation are recognized as major consequences of eutrophication, both in marine and freshwater bodies [2], affecting many sectors of local economies that rely on these natural resources (e.g., fisheries, recreation and drinking water quality) [3].
To determine the trophic state of water bodies, various water quality indicators are employed. Water quality indicators include physical, chemical and biological parameters that are quantitatively measured to present and communicate complex phenomena, such as trends and changes over time [4]. Nutrients, chlorophyll a (chl a), dissolved oxygen (DO), biochemical oxygen demand (BOD), chemical oxygen demand (COD), chemical oxygen demand using permanganate (CODMn) and Secchi disk depth (SDD) are the most widely utilized tropic state assessment indices [5,6,7].
Water quality indicators are traditionally determined by field measurements, sampling and laboratory analysis. Although conventional methods provide highly accurate results, they are costly and time-consuming, providing only point measurements. It is almost impossible to record the spatio-temporal variations in water quality characteristics on a regular basis [8]. This variability could be accomplished with the use of Geographic Information System (GIS), applying spatial interpolation techniques on in situ measurements for the estimation of various physicochemical and biological parameters. The GIS has also been applied in sampling strategies [9,10].
Since the early 1970s, remote sensing techniques have been widely integrated into the study of ocean color. Ocean color refers to the shade or tone of water which is determined by the interactions of electromagnetic radiation with the water itself and its constituents, including molecules and suspended and dissolved matter, both organic and inorganic [11]. It is typically defined as “the intensity and spectral distribution of visible light interacted with the water” [12], and measured variations in satellite imagery provide useful information about water quality indicators [8].
Ocean color remote sensing has been primarily focused on the retrieval of chl a, which is the main pigment found in phytoplankton. Monitoring chl a is crucial, as it serves as an indicator of total phytoplankton biomass and reflects the effects of nutrient overloading in marine and freshwater bodies. However, this condition signals the outcome of eutrophication, rather than its underlying cause. In contrast to chl a, nutrients, the primary “drivers” of eutrophication, could serve as crucial indicators for understanding the loading mechanisms and predicting phytoplankton growth potential in water bodies. Long-term monitoring of nutrients, combined with multiple water quality indicators (e.g., chl a and DO), using remotely sensed data could result in a more efficient assessment of the trophic state and prompt management actions.
Total nitrogen (TN) and total phosphorus (TP) represent the total nutrient load in marine and freshwater bodies and therefore are among the most commonly studied optically inactive water quality parameters (WQPs) in remote sensing-based assessments of eutrophication. Given that several existing review papers have covered TN and TP [8,13,14,15,16], these indicators are beyond the scope of this review.
This literature review aims to summarize the progress and the state of the art in retrieving non-optically active constituents of water such as un-ionized ammonia (NH3), ammonium (NH4+), ammoniacal nitrogen (AN), nitrite (NO2−), nitrate (NO3−), dissolved inorganic nitrogen (DIN), phosphate (PO43−), dissolved inorganic phosphorus (DIP), silicate (SiO2) and dissolved oxygen (DO) using remotely sensed data. Commonly used spaceborne and airborne sensors are mentioned, along with selected spectral bands and derived band indices. Remote sensing methods for the quantitative retrieval of the selected WQPs, as well as their predictive accuracy based on various statistical indices, are also discussed. In addition, there is reference to the potential influence of depth, bottom characteristics, seasonality, trophic state and different concentration levels of nutrients and DO on the accuracy of the retrieval models.
2. Methodology
A systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology [17,18]. The literature search was performed using Scopus, selected for its broad, multidisciplinary coverage, including studies published up to June 2025. Eligible studies comprised peer-reviewed articles and conference papers that employed remote sensing data to estimate nutrients and DO. Reference lists of the selected articles were also checked for additional sources. Studies that focused exclusively on TN and TP were excluded, as they were already covered in prior reviews. Non-English studies, as well as studies unrelated to water quality (e.g., those focused on urban areas, land, crops, soil, groundwater, or agriculture), were also excluded. To ensure comprehensive coverage of relevant studies, the following queries were used in the topic field:
- (“nutrient”) AND (“remote sensing”) AND (“water quality”) AND (“empirical” OR “analytical” OR “semi-empirical” OR “artificial intelligence” OR “machine learning” OR “deep learning”) AND NOT (“total nitrogen” OR “TN” OR “total phosphorus” OR “TP” OR “urban” OR “land” OR “crop” OR “soil” OR “groundwater” OR “agriculture”);
- (“dissolved oxygen”) AND (“remote sensing”) AND (“water quality”) AND (“empirical” OR “analytical” OR “semi-empirical” OR “artificial intelligence” OR “machine learning” OR “deep learning”) AND NOT (“total nitrogen” OR “TN” OR “total phosphorus” OR “TP” OR “urban” OR “land” OR “crop” OR “soil” OR “groundwater” OR “agriculture”).
A total of (n = 1147) records were retrieved from the initial search: (n = 1136) records from electronic database (Scopus) and an additional (n = 11) records through reference lists of included articles. After removing duplicates (n = 102), the titles and abstracts of the remaining records (n = 1.045) were screened to exclude irrelevant studies, resulting in the removal of (n = 852). This process resulted in (n = 193) studies being retained for full-text assessment. Studies that focused exclusively on TN and TP were excluded (n = 74) at the full-text stage as not related to the aim of this study. Non-English studies (n = 36), as well as those unrelated to water quality parameters of interest (e.g., those that focused on urban areas, land, crops, soil, groundwater or agriculture) (n = 83), were also excluded. Finally, (n = 66) studies from 2001 to June 2025 were included in the systematic review (Figure 1), the majority of which referred to DO estimation (Figure 2). An in-depth analysis of the studies selected was then conducted. Additional information regarding the research strategy, PRISMA 2020 checklist (Table S1) and PRISMA flow diagram (Figure S1), is available in the Supplementary Materials.
Figure 1.
Temporal trends in publications on nutrient and DO estimation using remote sensing techniques. The points on the line indicate the total number of publications per year.
Figure 2.
Distribution of publications focusing on individual WPQs (DO, NO3−, PO43−, NH3, NH4+, DIN, SiO2, NO2−, DIP, AN). Some case studies employ more than one QWP; hence, the total, which is greater than 66, exceeds the actual number of studies.
For each included study, the following data were extracted:
- Bibliographic details (authors and year of publication);
- Parameters retrieved (NH3, NH4+, AN, NO2−, NO3−, PO43−, SiO2, DIN, DIP or DO);
- Water type (coastal or inland; lake, river, reservoir, wetland, stream/stream network, lagoon or other);
- Location of the study area (country);
- Sensor (spaceborne, airborne, ground-based/multispectral or hyperspectral);
- Number of samples;
- Bands/band indices or equation;
- Retrieval model (empirical, machine learning (ML) or neural networks (NNs));
- Validation metrics (e.g., coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE), etc.).
For studies that reported multiple sensors, all relevant results were extracted. In cases where multiple models were presented, the best performing model was recorded. Missing information or unreported values were recorded as “not available (NA)”. Due to methodological heterogeneity among studies, such as differences in sensor type, sample size and validation metrics, quantitative pooling (meta-analysis) was not feasible. Therefore, a narrative synthesis was conducted. Studies were grouped by retrieval model to identify patterns in retrieval accuracy based on the reported validation metrics.
3. Nutrients—The Primary “Drivers” of Eutrophication
N and P are essential nutrients for plant growth, survival and reproduction. N is typically the limiting factor in marine water bodies, while it is P in freshwater bodies. The limiting factor refers to the nutrient that is most crucial for the growth of primary producers, as it is required in large amounts and is in short supply [3,19]. N and P are found in both dissolved and particulate form, as well as in organic and inorganic form [20]. DIN and DIP refer to the dissolved forms of N and P and constitute immediately bioavailable molecules for primary producers. DIN includes NH4+, NO2− and NO3−, while DIP refers to PO43− [19].
3.1. Un-Ionized Ammonia (NH3), Ammonium (NH4+) and Ammoniacal Nitrogen (AN)
In aquatic environments, ammonia consists of NH3 and NH4+ and depends on temperature and pH values. NH4+ is the predominant form of ammonia in the pH range of the majority of natural water bodies (pH < 8.75). In contrast to NH4+, NH3 accounts for a relatively small fraction of ammonia and increases at a higher pH level and temperature. Although NH3 is the most toxic form of ammonia, high concentrations of NH4+ are toxic too and could affect the welfare of aquatic organisms. High levels of NH4+ could lead to eutrophication, as it is the most preferred N source for phytoplankton growth. AN refers to the portion of N in the form of NH3 and NH4+ [21].
3.2. Nitrite (NO2−) and Nitrate (NO3−)
NH4+ tends to be oxidized to NO3− in a two-step process (NH4+ → NO2− → NO3−) by aerobic chemoautotrophic bacteria, even if DO is found in low concentrations (1.0 mg/L) [22]. During this process, Nitrosomonas oxidize NH4+ to NO2− and Nitrobacter oxidize NO2− to NO3−. NO2− is an intermediate form of inorganic N, usually found in low concentrations, as it is typically converted to NO3− [23]. NO3− is generally present at higher concentrations than NH4+ and NO2− [22]. Aquatic plants assimilate NH4+ and NO3− and then N returns to water via the decomposition of dead organic matter [23]. NO3− is less toxic than NH4+ and NO2− [22].
3.3. Phosphate (PO43−)
PO43− is the only form of P that primary producers can assimilate. In contrast to other nutrients (e.g., NO3−), PO43− is not water-soluble and therefore enters aquatic environments only with soil movement, as it adheres to soil particles [3]. According to Kim et al. [24], PO43− has no toxic effects on aquatic organisms.
3.4. Silicate (SiO2)
Although most studies deal with the effects of eutrophication caused by N and P enrichment, silicon (Si), in the form of SiO2, also, plays a crucial role in algal composition, controlling its growth. In aquatic environment, SiO2 is an essential element for diatoms, a silica-shelled phytoplankton species. As a result, their growth rates are determined by the SiO2 supply. SiO2 enters coastal waters through rivers and therefore its concentration depends on this input load in particular [25]. In natural water bodies, toxicity from SiO2 is not likely to occur [26].
3.5. Sources of Nutrient Enrichment
Eutrophication can occur naturally or be driven by human activities. The main natural sources are atmospheric deposition, river runoff, upwelling and submarine groundwater seepage (SGS) [19]. Atmospheric deposition refers to pollutants and other substances that are carried by wetfall (rain, snow, cloud and fog droplets) and dryfall (aerosols and particulates) to the Earth’s surface [23], while river runoff refers to freshwater that becomes nutrient-enriched by flowing through mineral-loaded rocks. Upwelling describes the “rising of cold waters masses from the deep that are enriched with nutrients, including NO3− and Si(OH)4 that originate from organic matter rematerialized by bacteria on the seafloor” [19]. In other words, it is a mechanism that transports these nutrient-rich deep waters towards the surface. In addition to upwelling, winter mixing and eddies are able to transport significant amounts of nutrients to the water surface [19]. SGS refers to “the continental groundwater that is discharged directly into the ocean, wherever a coastal aquifer is connected to sea” [27]. SGS is recognized as an important transport pathway for a variety of substances, such as nutrients, nearshores, with significant environmental consequences [28]. Upwelling and SGS are primarily associated with marine water bodies, while atmospheric deposition and river runoff affect both marine and freshwater bodies.
Agricultural, industrial and urban wastewater runoffs are the primary anthropogenic sources contributing to increased primary production due to nutrient release [4]. Fertilizers rich in NO3−, NH4+ and PO43− are flushed into rivers and groundwater through irrigation systems, rainfall and floods, eventually reaching marine water bodies. In addition, polluted water originating from domestic and industrial sources enriches marine and freshwater bodies with significant amounts of inorganic nutrients, as sewage treatment is rarely effective in removing N and P in forms of NH4+ and PO43− [19]. Aquaculture is another anthropogenic source of nutrient enrichment caused by the metabolic processes of cultivated organisms and uneaten food [29]. Although it contributes only 1% to nutrient release compared to other anthropogenic activities [30], its impact should not be underestimated.
3.6. Impacts on Aquatic Environment
Algal blooms, oxygen deficiency of bottom water, ecosystem degradation and loss of biodiversity are the main consequences of eutrophication [2]. Algal blooms refer to the proliferation and dominance of specific phytoplankton or macroalgal species that benefit from the rapid nutrients uptake rates. In marine and freshwater bodies, algal blooms can occur naturally. However, anthropogenic activities can significantly increase the intensity and frequency of this ecological process. Harmful substances produced by toxic algal blooms accumulate in filter feeders (e.g., shellfish) and, through them, be transported to higher levels of the trophic chain, which could lead to mortality of upper consumers (e.g., fish and humans).
Non-toxic blooms (e.g., red tides) produce high amounts of biomass in water columns, reducing the transmission of light and thus DO availability. The subsequent microbial decomposition of this biomass can lead to hypoxic or even anoxic conditions (DO < 2 mg/L) in deep waters. This DO reduction changes the redox potential of sediment and increase fluxes of PO43−, NH4+ and SiO2 into the overlaying water. Furthermore, bacterial sulfate respiration could replace DO, generating hydrogen sulfide (H2S), a chemical compound toxic to most benthic macrofauna, leading to changes in species composition and loss of habitats and biodiversity [19].
4. Water Quality Monitoring Using Remote Sensing
4.1. Light–Water Interaction
When sunlight or another light source is directed towards the water surface, a portion of the light is reflected off the surface, while the remainder is transmitted through the water column. Underwater light is either absorbed or scattered by water molecules, particulate and dissolved matter. Only a little part of the incident light is scattered backwards and leaves the water. The back-scattered light gives water its characteristic color and corresponds to the water-leaving radiance detected by remote sensors [31].
Ocean color is determined by the absorption and scattering characteristics of water and its constituents. The processes of absorption and scattering refer to the inherent optical properties (IOPs) and define the intensity and spectral characteristics of water-leaving radiance [32]. IOPs are unique to each constituent [33]. Variations in water-leaving radiance are utilized to derive information about the types of constituents and their concentrations [34].
Remote sensing reflectance (Rrs) is a fundamental apparent optical property (AOP) of water commonly used to quantify ocean color. Rrs depends on the IOPs of the medium and the ambient field light and refers to the ratio of water-leaving radiance to downwelling irradiance, measured just above the water surface [12,31]. In other words, it represents the spectral distribution of the incident light that penetrates the water surface and returns through the air–water interface due to the differential absorption and scattering mechanisms of the water’s constituents. Understanding the optical properties of water and its constituents and hence the relationship between IOPs and the water-leaving radiance is essential for retrieving WQPs using remote sensing [12].
4.2. Optically Active and Inactive Constituents of Water
In ocean color remote sensing, WQPs are classified as optically active or inactive. Phytoplankton, yellow substances, also known as colored dissolved organic matter (CDOM) and inorganic particulate matter are the primary optically active constituents (OACs) that interact with light through absorption and scattering processes, giving water its characteristic color [33,35]. OACs can be successfully detected by remote sensors [34,35]. Chl a, the main phytoplankton pigment, absorbs the blue and red wavelengths of the visible spectrum and reflects green. Therefore, water bodies characterized by high phytoplankton abundance appear in shades of blue-green [31]. CDOM absorbs light in the ultraviolet (UV) and blue regions of the electromagnetic spectrum, giving water a yellow-green to brown tone [12]. A high concentration of inorganic particulate matter gives water a reddish-brown color, as it scatters longer wavelengths (red). On the other hand, water bodies characterized by low concentration of OACs absorb longer-wavelength red light and thus appear blue [31].
Water bodies are classified into Case I and Case II depending on whether their optical properties are dominated solely by phytoplankton or by multiple OACs. Case I refers to optically simple, deep water (open ocean), where the optical properties are primarily determined by phytoplankton. Case II refers to optically complex waters (coastal and freshwater bodies), where ocean color is significantly influenced by several constituents besides phytoplankton, which vary independently of each other (e.g., CDOM and inorganic particulate matter), and often they are accompanied by relatively high levels of scattering. The non-linear nature of Case II waters, along with the similarity in optical properties of various constituents, contributes to their optical complexity [12,34].
Non-optically active constituents, such as nutrients and DO, do not have a well-defined effect on the color of water [16,36], and they appear to exceed weak optical characteristics and low signal-to-noise ratios [8,16]. Several attempts have been made to estimate non-optically active constituents of water by establishing relationships with optically active ones. However, finding significant correlations between optically active and inactive WQPs remains a major challenge and may vary from one study to another [36,37].
4.3. Remote Sensing Imagery
To date, a variety of different remote sensing systems has been used to detect variations in ocean color by measuring the amount of radiation in different wavelengths reflected from the water’s surface [8]. Remote sensed data are available in digital format (images) captured by remote sensing sensors. The choice of a specific sensor depends on its spectral, spatial, temporal and radiometric resolution [16,33,38,39,40].
4.3.1. Optical Remote Sensing
Optical sensors have been widely used in the retrieval of WQPs, collecting data in the visible (VIS) and near-infrared (NIR) regions of the electromagnetic spectrum. Short-wave infrared (SWIR) and thermal infrared (TIR) regions are not directly related to ocean color; however, they are occasionally used for atmospheric correction and sea surface temperature (SST) detection, respectively. Optical sensors usually operate passively, relying on sunlight as the main source of illumination, and can be either spaceborne or airborne, depending on the platforms they are installed on. Spaceborne sensors are mounted on platforms flown outside the Earth’s atmosphere (satellites or aircraft), while airborne sensors are those carried by platforms to areas within the Earth’s atmosphere (unmanned aerial vehicles (UAVs), aircraft, helicopters, balloons and boats). Spaceborne and airborne sensors provide either multispectral or hyperspectral images, depending on the sensor used for image acquisition [16,33].
4.3.2. Microwave Remote Sensing
Microwave radiometers (MWR) and synthetic aperture radar (SAR) operate in the microwave portion of the electromagnetic spectrum. The longer wavelengths used by MWR and SAR, compared to VIS and NIR sensors, enable them to detect electromagnetic radiation under all weather conditions, except during the heaviest rains [16,41]. MWR are passive, while SAR systems operate actively by emitting their own energy. Although MWR and SAR are not directly related to ocean color and water quality monitoring, they provide complimentary data to optical ocean color observations. In particular, MWR provide datasets on SST, sea surface salinity (SSS), wind speed, water vapor, cloud liquid water, rain rate and sea ice [42], while SAR is able to derive data on ocean surface winds, ocean surface currents, sea level, ocean internal waves, sea ice, icebergs, aquatic vegetation, coral reefs, ocean oil spills, ocean tides, eddies and bathymetry [43]. Datasets from MWR and SAR improve the accuracy of ocean color data by accounting for atmospheric conditions and oceanographic processes that influence variations in ocean color.
4.4. Remotely Sensed WQP Retrieval Methods
In general, four different approaches can be used for the retrieval of WQPs through remote sensing: the empirical, analytical, semi-empirical and AI methods [15].
4.4.1. Empirical Method
The empirical method determines statistical relationships between in situ water quality measurements and AOPs of water. The spectral band and/or band combinations that have the highest correlation with WQPs are identified and selected as the remotely sensed data. Then, the statistical relationship between the in situ measurements and reflectance data is determined using different linear or non-linear regression techniques [44]. The empirical method is applied in the retrieval of both optically active and inactive WQPs [13,15].
4.4.2. Analytical Method
The analytical method is based on the light–water interaction, utilizing the radiative transfer equation which defines the relationship between the IOPs and the AOPs of water [12]. The concentration of water constituents can be estimated using a ratio of their absorption and backscattering coefficients [36,44]. Optically active WQPs can be successfully derived using the analytical method [13,36].
4.4.3. Semi-Empirical Method
The semi-empirical approach is a combination of the analytical and the empirical methods, as it integrates the radiative transfer theory with traditional statistical techniques. The IOPs of water are considered to identify appropriate spectral bands and/or band combinations and regression techniques are applied to establish relationships for their retrieval [44]. Semi-empirical models are, in particular, used for the estimation of optically active WQPs [13,45].
4.4.4. AI Method
AI refers to “any technique that aims to enable computers to mimic human behavior and reproduce over human decision-making to solve complex tasks independently or with minimum human intervention” [46]. In recent years, AI methods, such as ML and NNs, have been increasingly applied in various real-world application areas [47], including, among others, ocean color remote sensing [15,16]. ML and NNs aim to identify new insights, patterns, relationships and dependencies in data, make predictions and assist in data-driven decision-making [47].
ML
ML is an interdisciplinary field under AI that combines both statistics and computer science [48]. According to [46], ML aims to “automate the task of analytical model building by applying algorithms that iteratively learn from problem-specific training data, allowing computers to discover hidden insights and complex patterns without being explicitly programmed”. Based on the given problem and the available input data, algorithms can be grouped into three ML techniques:
- Supervised learning;
- Unsupervised learning;
- Reinforcement learning.
In supervised learning, classification and regression are the main methods used. In classification, target prediction (output) is categorical or discrete, while in regression, it is numerical or continuous. In both the classification and regression methods, the building of the predictive model is based on pre-existing data and certain algorithms can be applied, depending on the desired output. On the other hand, in unsupervised learning, there are no pre-existing data labels to train predictive models. Clustering and association are the main methods used in unsupervised learning. In clustering, each data entry is aligned in groups (clusters) that share similar characteristics, while in association, the dependency between one data entry to another is found to map them together [48].
NNs
NNs are a subfield of ML which refer to “a collection of connected units (neurons) organized in layers”. The architecture of NNs consists of three types of layers, including the input, the hidden (middle) and the output layer. Based on the number of hidden layers, NNs are divided into shallow (a single hidden layer) and deep (multiple hidden layers). A larger number of hidden layers enables deep NNs to learn high-level features with more complexity than shallow NNs [49].
5. Narrative Literature Synthesis
Because of substantial heterogeneity among studies, including differences in study area, sensor type, sample size and validation metrics, a quantitative synthesis (meta-analysis) was not feasible. Therefore, a narrative synthesis was conducted. Studies were grouped by WQP, study area (inland or coastal waters) and retrieval model to identify patterns in retrieval accuracy based on the reported validation metrics. The estimation of nutrients and DO relied on proxy relationships, rather than on the direct optical sensitivity of the parameters, as these constituents are non-optically active. A summary of studies that have retrieved nutrients and DO using remotely sensed data is provided in Table 1, Table 2, Table 3 and Table 4.
Table 1.
Remote sensing-based retrieval of nutrients and DO in inland waterbodies using empirical models.
Table 2.
Remote sensing-based retrieval of nutrients and DO in coastal waterbodies using empirical models.
Table 3.
Remote sensing-based retrieval of nutrients and DO in inland waterbodies using ML, shallow and deep NN models.
Table 4.
Remote sensing-based retrieval of nutrients and DO in coastal waterbodies using ML, shallow and deep NN models.
5.1. Types of Water Bodies
The findings of the literature review suggest that the total number of studies on the retrieval of nutrients and DO pertain to Case II waters, with the majority focusing on highly nutrient-enriched water bodies. Freshwater bodies account for 77% of the total studies, with the majority of them (25% and 24%) referring to lakes and rivers, respectively. Reservoirs, wetlands and stream/stream networks account for 14%, 4% and 3%, respectively. Marine water bodies, which specifically refer to coastal regions, represent 20%, while brackish waters account for only 3% of all studies (Figure 3).
Figure 3.
Types of Case II waters studied in the literature. Marine waters refer to coastal regions.
5.2. Remote Sensing Systems for the Retrieval of Nutrients and DO
Different satellite sensors have been used to retrieve nutrients and DO. Since the early 2000s, spectral data from Landsat 4 and Landsat 5 (TM) have been utilized, and over time, additional satellites have been incorporated into research, including IKONOS, Landsat 7 Enhanced Thematic Mapper Plus (ETM+), Moderate Resolution Imaging Spectroradiometer (MODIS) and Satellite pour l’Observation de la Terre 5 (SPOT 5). In recent years, Indian Remote Sensing Linear Imaging Self-Scanning Sensor-III (IRS LISS III), Worldview-2, Landsat 8 Operational Land Imager (OLI), Huang Jing-1 (HJ-1), Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), Sentinel-2 Multi-Spectral Instrument (MSI), Landsat 9 Operational Land Imager 2 (OLI-2), Geostationary Ocean Color Imager (GOCI), Ziyuan-3 (ZY-3), Zhuhai-1 OHS-2A (OHS), PlanetScope (SuperDove) and Sentinel-3 Ocean and Land Colour Instrument (OLCI) imagery have also been employed (Table A1).
The choice of an appropriate sensor depends on the spectral, spatial, temporal and radiometric resolution of the remotely sensed images. The cost of image acquisition is also crucial [16,33]. In small water bodies, high-resolution imagery is required [51,104]. High-resolution satellites (e.g., IKONOS, Worldview-2, PlanetScope (SuperDove), Sentinel-2 MSI) provide detailed spatial information in optically complex water bodies, where highly variable features are present. Moreover, they have the potential to minimize the mixed-pixel effect. In small water bodies, mixed could be observed when low-resolution imagery is utilized. On the other hand, high-resolution satellites are not suitable for large water bodies due to their limited spatial coverage and lower temporal resolution. For regional and global scale studies, moderate- and low-resolution satellites (e.g., MODIS and GOCI) are preferred [80,110,112,113].
Several studies compared the accuracy of different satellite sensors in retrieving nutrients and DO. Satapathy et al. [78] employed data from Landsat 7 (ETM+) and IKONOS to quantitatively retrieve PO43− in the West Coast of Mumbai (IN). The results indicated high prediction accuracy using either set of satellite data.
Khattab and Merkel [57] analyzed remote sensing data obtained from Landsat 5 (TM) and Landsat 7 (ETM+) imagery to retrieve NO2−, NO3− and PO43− in Mosul Dam Lake (IQ). The models based on Landsat 7 (ETM+) spectral data showed more precise results compared to Landsat 5 (TM). Pizani et al. [75] compared the performance of Sentinel-2 (MSI) and Landsat 8 (OLI) for the retrieval of DO in Três Marias Reservoir (BR). The utilized reflectance values were based on averaging kernels of 30 m and 90 m. Sentinel-2 (MSI) achieved higher predictive accuracy for both kernel sizes compared to Landsat 8 (OLI). Fu et al. [96] used Sentinel-2 (MSI) and Zhuhai-1 OHS-2A (OHS) reflectance data to quantitatively retrieve NH4+ in Poyang Lake (CN). Hyperspectral images from Zhuhai-1 OHS-2A (OHS) achieved better predictive results compared to the multispectral images from Sentinel-2 (MSI).
Researchers have also utilized data from multiple satellite sensors to achieve higher predictive accuracy. Peterson et al. [107] used a harmonized virtual constellation of Landsat 8 (OLI) and Sentinel-2 (MSI) to create a high temporal frequency dataset at a relatively fine spatial scale. Elhag et al. [60] used a spectral correspondence between Sentinel-2 (MSI) and Sentinel-3 (OLCI). The correlation between all bands was determined and the bands with similar central wavelengths were selected to construct the retrieval models (Table 1, Table 2, Table 3 and Table 4).
Recently, with the rapid development of Unmanned Aerial Vehicle (UAV) technology, drones have also been used in the retrieval of nutrients and DO using either multispectral [77,90,91,93,108] or hyperspectral imaging [94,95] (Table A2). Tian et al. [93] used the reflectance values of DJI M300 RTK UAV multispectral platform, Sentinel-2 (MSI) and Landsat 7 (ETM+) to estimate NO3−, NH3 and DO in the Quinwu, Longjing and Nanshan Reservoirs (CN). The accuracy of the model derived from the airborne UAV dataset was relatively higher compared to spaceborne satellite sensors. Unlike satellites, drones can capture the continuous spectral signatures of water bodies, enabled by their flexible and on-demand deployment capabilities [15]. In addition to satellite and UAV sensors, ground-based spectrometers [29,92,98] have been used as an alternative source of remotely sensed data (Table 1, Table 2, Table 3 and Table 4). UAV and ground-based sensors provide data with higher spectral and spatial resolution, while are much less affected by the atmosphere compared to satellites. However, they have the limitations of high cost and limited capacity to observe large water bodies [15,33].
5.3. Remote Sensing Methods for the Retrieval of Nutrients and DO
As mentioned in Chapter 1.5, there are four different approaches by which WQPs can be extracted using remotely sensed data. However, in the case of nutrients and DO, researchers have focused on two types of methods, the empirical and the AI method. Based on the existing literature, the empirical approach is the most commonly used, representing 56% (58 studies) of the total. AI-based methods follow (48%), comprising 32 ML and 21 NN studies, respectively (Figure 4). Analytical and semi-empirical approaches appear not to be used in establishing inversion models for nutrients and DO. Table 1, Table 2, Table 3 and Table 4 include a detailed list of the empirical and AI methods used for the retrieval of nutrients and DO, as well as their prediction accuracy based on various statistical indices.
Figure 4.
Distribution of approaches (empirical, ML and NNs) used for nutrient and DO estimation based on remotely sensed data. Some case studies employ multiple approaches; hence, the total, which is greater than 66, exceeds the actual number of studies.
5.3.1. Empirical Models
Empirical models are simple and easy to operate. However, they require a large number of in situ measurements to assure a higher accuracy. In addition, they have poor generalization abilities in the spatial and temporal scales [45,116] and usually are not suitable for applications in Case II waters due to the various constituents that affect the optical properties of the water [44]. Empirical approaches employ regression analysis using various polynomial and non-polynomial equations. Among these, multiple linear regression (MLR), stepwise MLR and simple linear regression (SLR) are the most widely applied techniques for nutrient and DO estimation (Figure 5; Table 1 and Table 2).
Figure 5.
Overview of individual models (empirical, ML and NNs) for nutrient and DO estimation via remote sensing. Some case studies employ multiple models; hence, the total, which is greater than 66, exceeds the actual number of studies.
Based on Sentinel-2 (MSI) data, Dong et al. [51] applied SLR to establish inversion models for NH3 estimation in Danjiangkou Reservoir (CN). The quadratic model had the highest prediction accuracy (R2 = 0.85, MAE = 0.01, RMSE = 0.01) compared to the other constructed empirical models. The band ratio between the green peak (560 nm) and the blue (490 nm) band was formed to develop the retrieval model. A quadratic model was, also, developed by Abdelmalik [65] to estimate PO43− in Quaroum Lake (EG), using the SWIR (2145–2185 nm) band of ASTER. By using Landsat’s 8 (OLI) data, Markogianni et al. [54,55] applied several empirical models to estimate NH4+ concentration in Trichonis Lake (GR). The cubic model showed a higher prediction accuracy compared to the other models, with weak (R2 = 0.47, SEE = 0.00) prediction results. The band combination of the coastal blue (433–453 nm), green (525–600 nm) and red (630–680 nm) bands was used to establish the inversion model. For NO3−, NH4+, SiO2 and PO43− estimation in the Xiangxi River (CN), Liu et al. [53] built several empirical models utilizing HJ-1 satellite imagery. The cubic and the quadratic model provided moderate predictive accuracy for NO3− (R2 = 0.75) and SiO2 (R2 = 0.56), respectively, utilizing a band combination of the blue (430–520 nm), green (520–600 nm) and NIR (760–900 nm) bands. The retrieval models of NH4+ and PO43− exhibited weak predictive accuracy (R2 < 0.5), using a band combination of the green, red (630–690 nm) and NIR bands and a band combination of the blue and NIR bands, respectively. Qui et al. [69] established an exponential model for the inversion of DO in the Huangpu River (CN), with moderate predictive accuracy (R2 = 0.68), using the band ratio between the NIR (760–900 nm) and the green (520–600 nm) bands of Landsat 5 (TM) (Table 1 and Table 2).
5.3.2. ML Models
The most commonly applied regression-based ML algorithms in studies related to nutrient and DO estimation are the following:
- Support Vector Regression (SVR) [37,85,86,87,89,92,93,94,99,102,104,105,106,107,111,114,115];
- Extreme Gradient Boosting (XGBoost) [87,90,93,95,96,97,102,103,105,112,115];
- Categorical Boosting (CatBoost) [96,105];
- Gradient Boosting (GB) [91,92,96,99,103];
- Random Forest (RF) [87,88,89,90,91,92,96,99,102,103,112];
- Adaptive Boosting (AdaBoost) [90,92,96];
- Regression Tree (RT) [99];
- Elastic Net Regression (ENR) [92];
- k-Nearest Neighbor Regression (k-NNR) [92];
- Gaussian Process Regression (GPR) [92,111];
- Extra Trees Regression (ETR) [92];
- Extreme Learning Machine Regression (ELM) [107,108].
5.3.3. NN Models
The most widely acknowledged NN architectures employed in the retrieval of nutrients and DO are the following:
- Backpropagation Neural Network (BPNN) [89,91,98,106,110];
- Artificial Neural Network (ANN) [87,99,100,101,108];
- Multilayer Perceptron (MLP) [92,102,109,115];
- Pixel based-Deep Neural Network Regression (pixel-DNNR) [94];
- Patch based-Deep Neural Network Regression (patch-DNNR) [94];
- Binary Whale Optimization Algorithm-Artificial Neural Network (BWOA-ANN) [101];
- Transformer (TR) [103,105];
- Deep learning model that integrates Transformer and LSTM Networks (TL-Net) [105];
- Convolutional Neural Network (CNN) [91,105,109];
- Fully Connected Neural Network (FCNN) [109];
- Recurrent Neural Network (RNN) [109];
- Long Short-Term Memory (LSTM) [105];
- Vanilla-Long Short-Term Memory (V-LSTM) [109];
- Stacked-Long Short-Term Memory (S-LSTM) [109];
- Bidirectional-Long Short-Term Memory (Bi-LSTM) [109];
- Convolutional-Long Short-Term Memory (Conv-LSTM) [109];
- Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) [109];
- Mixture Density Network (MDN) [93,103,115];
- Progressively decreasing Deep Neural Network (pDNN) [107];
- Dynamic Network Surgery-Deep Neural Network (DNS-DNNS) [108];
- Deep Neural Network (DNN) [90,93];
- Spatiotemporal-Deep Belief Network (ST-DBN) [113];
- Message Passing Neural Network (MPNN) [113];
- Generalized Regression Neural Network (GRNN) [113].
5.3.4. Comparative Analysis of Predictive Accuracy: Empirical, ML and NN Models
The accuracy of different retrieval models, including empirical, ML and NN approaches, has been compared in several studies.
Empirical vs. ML Models
Similarly to empirical models, ML models are only applicable within the range and settings of the training data. However, this approach uses iterative learning to reduce the overall error and maximize the model fit. Furthermore, the ML is able to operate in multidimensional space and capture complex, non-linear relationships [45].
Based on SPOT 5 imagery, Wang et al. [85] developed a MLR and a semi-supervised SVR (SS-SVR) model to retrieve NH3 and DO in Weihe River (CN). SS-SVR showed higher prediction accuracy (R2 = 0.98, MSE = 0.05 for NH3 and R2 = 1.00, MSE = 0.00 for DO) compared to MLR (R2 = 0.48, MSE = 4.57 for NH3 and R2 = 0.81, MSE = 0.92 for DO). Similarly, Wang et al. [86] developed a MLR model and a genetic algorithm (GA) SVR model (GA-SVR) to quantitatively retrieve NH3. The SVR model indicated better prediction accuracy (R2 = 0.98, MSE = 0.77, MAD = 0.43) compared to the MLR (R2 = 0.76, MSE = 4.46, MAD = 3.03). In both studies, the single bands from the green (500–590 nm) to SWIR (1580–1750 nm) were used as independent variables to construct the inversion models. As indicated by the authors, the results from a small number of in situ measurements showed that SVR could provide more accurate predictive results compared to traditional MLR. Wang et al. [95] applied various empirical models and the XGBoost model to estimate NH4+ in Shahu Port and Xunsi River (CN), using UAV hyperspectral imagery. Spectral data in the 400 to 900 nm range of the electromagnetic spectrum were used to construct the retrieval models. XGBoost demonstrated improved prediction accuracy compared to traditional empirical models (Table 3).
Empirical vs. NN Models
Amanollahi et al. [100] used an ANN model and a linear regression (LR) model to retrieve NO3− and PO43− in Zarivar International Wetland (IR) based on Landsat 8 (OLI) image data. The results showed that for both the ANN and LR model R2, RMSE and MAE values exhibited low predictive accuracy. On the other hand, Wu et al. [113] established a DBN, MPNN, GRNN and a MLR model in both original and spatio-temporal-integrated forms to estimate DIN and DIP in the Zhenjiang Coastal Sea (CN). MODIS’s single bands from red (620–670 nm) to SWIR 3 (2105–2155 nm), except SWIR 2 (1628–1652 nm), were extracted and used as input features to the retrieval models. The spatio-temporal-integrated deep learning framework achieved significantly higher R2 values and reduced estimation errors compared to the original forms. The ST-DBN model showed the highest predictive accuracy for both DIN (R2 = 0.83, RMSE = 0.21, MAE = 0.14) and DIP (R2 = 0.64, RMSE = 0.01, MAE = 0.01), while the ST-MLR model was the worst (R2 = 0.62, RMSE = 0.32, MAE = 0.20 for DIN, R2 = 0.42, RMSE = 0.02, MAE = 0.01 for DIP) (Table 3 and Table 4).
ML vs. NN Models
Theoretically, in the case of large and high-dimensional data, NNs outperform conventional empirical and ML approaches [46]. However, the complex architecture of NNs combined with the high computational performance and the huge amount of training data required make their use more complicated compared to ML models [48]. On the other hand, in the case of limiting and low-dimensional data, ML tends to provide better results than those generated by NNs [46].
Based on Landsat 8 (OLI) data, Sharaf El Din et al. [106] established a BPNN model and an SVR model for estimating DO, in Saint John River (CA). The BPNN model showed higher accuracy (R2 = 0.94, RMSE = 0.19) compared to the SVR model (R2 = 0.89, RMSE = 0.57). Single bands, from the coastal blue (433–453 nm) to SWIR 2 (2100–2300 nm), except for the green (525–600 nm) band, were selected to form the input layer of the inversion model. Li et al. [99] used four different ML algorithms (SVR, RF, RT and GBM) and an ANN model to retrieve AN in Nandu River (China), using Landsat 8 (OLI) imagery. The spectral data from coastal blue to SWIR 2 were used as input parameters for the retrieval models. The results demonstrated that the ANN model had the highest R2 values, which were close to 1.00 for the training and 0.44 for the testing dataset. The RMSE values were 0.01 and 0.19, respectively. Other researchers compared the prediction accuracy of the XGBoost algorithm with the SVR, MLP, MDN, RF and ANN models for DO and NH3 estimation, using the single bands from the blue (490 nm) to NIR narrow (865 nm) of Sentinel-2 (MSI) in Shenzhen Bay and Q Reservoir (CN), respectively. In both case studies, the results indicated that the XGBoost outperformed the other ML and NN models and can provide better prediction accuracy when limited sample data are available [87,115]. Lo et al. [91] used the RF, GB, BPNN and CNN models to estimate DO and NH3 in Yuandang Lake (CN), using UAV multispectral imagery. In the case of DO, RF provided higher accuracy results both for the training (R2 = 0.96, RMSE = 0.22) and the testing (R2 = 0.67, RMSE = 0.62, MAE = 0.41) datasets, using the band combination of the green (560 nm) and NIR (840 nm) bands. For NH3 estimation, GB showed the best prediction accuracy (R2 = 0.84, RMSE = 0.06 for training and R2 = 0.55, RMSE = 0.12, MAE = 0.09 for testing dataset) compared to the RF and the DL models. The band combination of the red band (650 nm) and red edge (730 nm) was selected as the independent variable to establish the retrieval model (Table 3 and Table 4).
Empirical vs. ML vs. NN Models
Using the spectral reflectance data derived from the harmonized virtual constellation of Landsat 8 (OLI) and Sentinel-2 (MSI), Peterson et al. [107] applied a pDNN model for the estimation of DO, in Mississippi River, Lake Decatur and Lake Carlyle (USA). The developed model was evaluated against MLR, SVR and ELM. The pDNN model outperformed all the other retrieval models (R2 = 0.91, RMSE = 2.06, MAPE = 9.01 for training dataset, R2 = 0.89, RMSE = 1.81, MAPE = 9.08 for testing dataset), while the MLR model generated the lowest prediction accuracy (R2 = 0.93, RMSE = 13.65, MAPE = 8.90 for training dataset, R2 = 0.44, RMSE = 3.31, MAPE = 13.98 for testing dataset). Based on Sentinel-2 (MSI), Landsat 7 (ETM+) and UAV multispectral imagery, Tian et al. [93] constructed an unclustered band combination (UC-BC) empirical model, fuzzy C-means band combination (FCM-BC) empirical model, XGBoost, SVR, MDN and DNN to quantitatively retrieve NO3−, NH3 and DO in the Quinwu, Longjing and Nanshan Reservoirs (CN). The prediction accuracy of FCM-BC improved compared to the BC model; however, it lagged behind the ML and DL models. The MDN model showed the best prediction accuracy. Chen et al. [90] compared the prediction accuracy between GA-XGBoost, RF, GA-RF, AdaBoost, GA-AdaBoost, DNN and several empirical models for NH3 estimation in the Nanfei River (CN), using UAV multispectral imagery. The performance of GA-XGBoost (R2 = 0.69, MAE = 0.14, RMSE = 0.16) was better compared to the empirical, DNN and other ML models. The band combinations of the blue (475 nm), green (560 nm) and red (670 nm) bands and red edge (720 nm) were used as input features of the GA-XGBoost model (Table 3).
5.4. Parameters That Influence the Accuracy of the Retrieval Models
In optically shallow waters (depth less than 30 m) the water-leaving radiance may be influenced by the light reflected off the bottom of the water body. This influence could vary depending on water clarity and bottom composition (e.g., sand, mud, rock, coral reefs, seagrass and macroalgae). The bottoms of Case II water bodies often exhibit strong influences and more complex models may be required to account for the effects of depth and seabed on ocean color [12,34,35]. In order to establish the relationship between NH4+, NO3−, DO and reflectance data from Landsat 7 (ETM+) and Landsat 8 (OLI), Theologou et al. [52] developed three simple linear regression models, each corresponding to a specific depth of a shallow freshwater body. The first model referred to relatively deep parts, the second to very shallow parts and the third to all lake depths. In general, the single linear regression models did not establish well-defined relationships between WQPs and remotely sensed data and only the first model provided accurate predictive results. Moreover, the shallow parts of the water body appeared to follow different regression patterns.
Seasonal variations may influence the spectra detected by a remote sensor [16,34,65,75]. Seasonal in situ measurements and Landsat 8 (OLI) spectral data were used by Khalil et al. [72] to quantitatively retrieve DO in Bardawil Lagoon (EG). Stepwise multiple linear regression models were developed, indicating different prediction accuracy between the four seasons. Similar, Vatitsi et al. [104] employed two different experimental setups to determine the impact of seasonality in the retrieval of DO using PlanetScope (SuperDove) spectral data. One predictive model was developed, considering data from multiple seasons, while two models were developed focusing on spring and summer, respectively. The accuracy between the two experimental setups varied, while the two seasonal models demonstrated different levels of performance. Besides seasonal variations, different concentration levels of nutrients and DO could affect the accuracy of the retrieval models. Indeed, Xu et al. [79] mentioned that it is not feasible to retrieve DIN via remote sensing when concentration is below 70 μg/L (Table 1, Table 2 and Table 3).
6. Discussion and Conclusions
In the present study, a systematic review was conducted to highlight recent advances and the current state of the art of quantitative nutrient and DO concentration retrieval using remote sensing techniques. A narrative literature synthesis was undertaken due to the heterogeneity among studies, and findings were summarized qualitatively. The findings were limited to Case II waters, where both empirical and AI models were applied. Studies were grouped by WQP, study area (inland or coastal waters) and retrieval model (empirical, ML or NNs) to identify patterns in retrieval accuracy based on the reported validation metrics.
Differences in sample size, validation processes and validation metrics limited our ability to make direct comparisons between studies. Some studies used entire datasets for model training without employing independent testing, while others did not clarify whether the reported metrics were derived from independent training or testing data. In addition, validation metrics varied across studies, and in some cases, sample size information was not reported. This lack of consistency constrained the strength of general conclusions.
Patterns in retrieval accuracy were assessed solely by comparing results within the same study area with the same number of samples and using comparable validation processes and metrics. In such cases, ML and NNs outperformed empirical models [85,86,93,94,95,105,107,113]. In general, empirical models require a large number of in situ measurements to achieve high accuracy [45,116], while NNs also demand substantial training datasets [46]. However, the exact sample size needed for either approach is not clearly defined in the literature. ML models (e.g., SVR and XGBoost) often outperform NNs when training data are limited, as they are generally less prone to overfitting under small-sample conditions [46].
Although several studies report high predictive accuracy, current models are site-specific, and their transferability across different regions may be restricted. To improve their transferability, existing models should be validated across diverse regions and time periods. Moreover, they should be applied across a wide range of environmental conditions, considering trophic state (e.g., oligotrophic, mesotrophic, eutrophic and hypereutrophic conditions), as well as seasonal changes in nutrient and DO concentration. Additionally, measurements of the water column profile might be helpful. In optically shallow waters, depth and bottom characteristics could also be incorporated into models due to their influence on the signals detected by remote sensors [117]. In almost all cases, regression models were used; however, if a wide range of training datasets is available, cluster-based models could also be tested in future.
Remote sensing shows strong potential as a complementary tool for monitoring non-optically active WQPs such as nutrients and DO. However, several challenges remain, and therefore further research is required. Future work should emphasize transparent reporting of sample size and validation metrics to improve the reliability and comparability of results.
Supplementary Materials
The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/rs17244044/s1, Figure S1: PRISMA flow diagram illustrating the study selection process; Table S1: PRISMA 2020 Checklist.
Author Contributions
Conceptualization, investigation and, writing—original draft preparation, A.D.; review and editing, C.D., D.V. and G.M., supervision, review and, editing, N.N. All authors have read and agreed to the published version of the manuscript.
Funding
The research is conducted in the operating framework of the University of Thessaly Innovation, Technology Transfer Unit and Entrepreneurship Center “One Planet Thessaly”, under the “Scholarship Grants to University of Thessaly Doctoral Candidates (5600.03.06.04)” and was funded by the Special Account of Research Grants of the University of Thessaly.
Conflicts of Interest
The authors declare no conflicts of interest.
Abbreviations
The following abbreviations are used in this manuscript:
| AAE | Average Absolute Error |
| ABR | Adaptive Boosting Regression |
| Acc. | Accuracy |
| AdaBoost | Adaptive Boosting |
| Adj. R2 | Adjusted R2 |
| AI | Artificial Intelligence |
| ANN | Artificial Neural Network |
| AOP | Apparent Optical Property |
| ARE | Absolute Relative Error |
| ASTER | Advanced Spaceborne Thermal Emission and Reflection Radiometer |
| Bi-LSTM | Bidirectional-Long Short-Term Memory |
| BGT | Brightness index |
| BOD | Biochemical Oxygen Demand |
| BPNN | Backpropagation Neural Network |
| BR | Brazil |
| BR | Coastal/NIR of the Landsat 8 (OLI) |
| BWOA-ANN | Binary Whale Optimization Algorithm-Artificial Neural Network |
| CA | Canada |
| CatBoost | Categorical Boosting |
| Chl a | Chlorophyll |
| Chlam-1 | Chlorophyll a in One Month Prior |
| CDOM | Colored Dissolved Organic Matter |
| CMEMS | Copernicus Marine Environment Monitoring Service |
| CN | China |
| CNN | Convolutional Neural Network |
| CNN-LSTM | Convolutional Neural Network-Long Short-Term Memory |
| CO | Colombia |
| COD | Chemical Oxygen Demand |
| CODMn | Chemical Oxygen Demand using Permanganate |
| Conv-LSTM | Convolutional-Long Short-Term Memory |
| DBN | Deep Belief Network |
| DIN | Dissolved Inorganic Nitrogen |
| DIP | Dissolved Inorganic Phosphorus |
| DL | Deep Learning |
| DNN | Deep Neural Network |
| DNS-DNN | Dynamic Network Surgery-Deep Neural Network |
| DO | Dissolved Oxygen |
| DST | Disturbance index |
| EE | Estonia |
| EG | Egypt |
| ELM | Extreme Learning Machine |
| ENR | Elastic Net Regression |
| ETM+ | Enhanced Thematic Mapper Plus |
| ETR | Extra Trees Regression |
| EVI | Enhanced Vegetation Index |
| FCM-BC | Fuzzy C-Means Band Combination |
| FCNN | Fully Connected Neural Network |
| FPR | Fraction of Photosynthetically active radiation |
| GA | Genetic Algorithm |
| GA-SVR | Genetic Algorithm-Support Vector Regression |
| GA-XGBoost | Genetic Algorithm-Extreme Gradient Boosting |
| GB | Gradient Boosting |
| GBM | Gradient Boosting Machine |
| GIS | Geographic Information System |
| GOCI | Geostationary Ocean Color Imager |
| GPP | Gross Primary Productivity |
| GPR | Gaussian Process Regression |
| GR | Greece |
| GREON | Great Rivers Ecological Observation Network |
| GRN | Greenness index |
| GRNN | Generalized Regression Neural Network |
| GVF | Green Vegetation |
| H2S | Generating Hydrogen Sulfide |
| HJ | Huang Jing |
| IN | India |
| IOA | Index Of Agreement |
| IOP | Inherent Optical Property |
| IQ | Iraq |
| IRS LISS | Indian Remote Sensing Linear Imaging Self-Scanning Sensor |
| k-NNR | k-Nearest Neighbor Regression |
| KR | (South) Korea |
| LAI | Leaf Area Index |
| LASSO | Least Absolute Shrinkage and Selection Operator |
| LOOCV-GB | Combining Leave-One-Out Cross Validation with Gradient Boosting |
| LR | Linear Regression |
| LSTM | Long Short Term Memory |
| MA | Morocco |
| MAD | Mean Absolute Deviation |
| MAE | Mean Absolute Error |
| MAPE | Mean Absolute Percentage Error |
| MDN | Mixture Density Network |
| ML | Machine Learning |
| MLP | Multilayer Perceptron |
| MLR | Multiple Linear Regression |
| MNB | Mean Normalized Bias |
| MODIS | Moderate Resolution Imaging Spectroradiometer |
| MPNN | Message Passing Neural Network |
| MRE | Mean Relative Error |
| MSE | Mean Square Error |
| MSFD | Marine Strategy Framework Directive |
| MSI | Multi-Spectral Instrument |
| MWR | Microwave Radiometers |
| N | Nitrogen, Number of Samples |
| NA | Non-Available |
| NDV | Normalized Difference Vegetation index |
| NDVILP | Normalized Difference Water Index After Applying Low-Pass Filter |
| NDWI2 | (NIR − SWIR)/(NIR + SWIR) of Landsat 8 (OLI) |
| NH3 | Un-Ionized Ammonia |
| NH4+ | Ammonium |
| NIR | Near-Infrared |
| NLP | Natural Language Processing |
| NLR | Non-Linear Regression |
| NN | Neural Network |
| NO2− | Nitrite |
| NO3− | Nitrate |
| NPV | Non-Photosynthetic Vegetation |
| NRMSE | Normalized Root Mean Square Error |
| OACs | Optically Active Constituents |
| OCR | Ocean Color Radiometry |
| OHS | Orbita Hyperspectral Satellite |
| OLCI | Ocean and Land Colour Instrument |
| OLI | Operational Land Imager |
| P | Phosphorus |
| PC4 | Principal Component 4 |
| PC4_SR2 | Principal Component 4 of Surface Reflectance, Using Post DOS |
| PC5_SR1 | Principal Component 5 of Surface Reflectance, Using Post FLAASH |
| PCP | Ocean Physio-Chemical Properties |
| p-DNN | Progressively Decreasing Deep Neural Network |
| Patch-DNNR | Patch-Based-Deep Neural Network |
| PH | Philippines |
| pH | Potential of Hydrogen |
| pixel-DNNR | Pixel-Based-Deep Neural Network Regression |
| PK | Pakistan |
| PLSR | Partial Least Squares Regression |
| PO43− | Phosphate |
| PS | Palestine |
| R | Correlation Coefficient, Reflectance |
| R2 | Coefficient of Determination |
| Rrs | Remote Sensing Reflectance |
| RF | Random Forest |
| RF-SHAP | Random Forest-Shapley Additive Explanation |
| RMSE | Root Mean Square Error |
| RMSLE | Root Mean Squared Log-Error |
| RNN | Recurrent Neural Network |
| RPD | Residual Prediction Deviation |
| ROD | Rate of decrease |
| ROI | Rate of increase |
| RT | Regression Tree |
| RTOAB7 | Top Of Atmosphere Reflectance Band 7 |
| SA | Saudi Arabia |
| SAR | Synthetic Aperture Radar |
| SDD | Secchi Disk Depth |
| SE | Standard Error |
| SEE | Standard Estimated Error |
| SGS | Submarine Groundwater Seepage |
| Si | Silicon |
| Sig. | Significance |
| SiO2 | Silicate |
| SLR | Simple Linear Regression |
| S-LSTM | Stacked-Long Short-Term Memory |
| SOI | Soil |
| SPOT 5 | Satellite pour l’Observation de la Terre 5 |
| SR2B1 | Surface Reflectance Band 1 calibration Using Post-DOS |
| SSS | Sea Surface Salinity |
| SS-SVR | Semi-Supervised Support Vector Regression |
| SST | Sea Surface Temperature |
| SSTm-1 | Sea Surface Temperature in One Month Prior |
| ST-DBN | Spatiotemporal-Deep Belief Network |
| STI | Spatio-Temporal Information |
| ST-MLR | Spatiotemporal-Multiple Linear Regression |
| SVR | Support Vector Regression |
| SWIR | Short-Wave Infrared |
| TDS | Total Dissolved Solids |
| TIR | Thermal Infrared |
| TL-Net | Deep learning model that integrates Transformer and LSTM Networks |
| TM | Thematic Mapper |
| TN | Total Nitrogen |
| Tol. | Tolerance |
| TP | Total Phosphorus |
| TR | Turkey, Transformer |
| UAV | Unmanned Aerial Vehicle |
| UC-BC | Unclustered Band Combination |
| URB | Urban |
| USA | United States of America |
| UV | Ultraviolet |
| VIIRS | Visible Infrared Imaging Radiometer Suite |
| VIS | Visible |
| V-LSTM | Vanilla-Long Short-Term Memory |
| WET | Wetness index |
| WQP | Water Quality Parameter |
| WSN | Wireless Sensor Network |
| XGBoost | Extreme Gradient Boosting |
| xPLSR | Extended Partial Least Squares Regression |
| ZY3 | Ziyuan-3 |
Appendix A
Appendix A.1
Table A1.
Overview of spaceborne sensors applied to nutrient and DO estimation.
Appendix A.2
Table A2.
Overview of UAV sensors applied to nutrient and DO estimation.
References
- Ferreira, J.G.; Andersen, J.H.; Borja, A.; Bricker, S.B.; Camp, J.; da Silva, M.C.; Garcés, E.; Heiskanen, A.S.; Humborg, C.; Ignatiades, L.; et al. Marine Strategy Framework Directive Task Group 5 report eutrophication. In JRC Scientific and Technical Reports EUR; Zampoukas, N., Ed.; Scientific and Technical Research Series; European Union: Brussels, Belgium, 2010. [Google Scholar]
- Boesch, D.F. Challenges and Opportunities for Science in Reducing Nutrient Over-Enrichment of Coastal Ecosystems. Estuaries 2002, 25, 886–900. [Google Scholar] [CrossRef] [Scilit]
- Khan, M.N.; Mohammad, F. Eutrophication: Challenges and Solutions. In Eutrophication: Causes, Consequences and Control; Ansari, A.A., Gill, S.S., Eds.; Springer: Dordrecht, The Netherlands, 2014; Volume 2, pp. 1–45. [Google Scholar] [CrossRef] [Scilit]
- Ferreira, J.G.; Andersen, J.H.; Borja, A.; Bricker, S.B.; Camp, J.; Cardoso da Silva, M.; Garcés, E.; Heiskanen, A.S.; Humborg, C.; Ignatiades, L.; et al. Overview of Eutrophication Indicators to Assess Environmental Status within the European Marine Strategy Framework Directive. Estuar. Coast. Shelf Sci. 2011, 93, 117–131. [Google Scholar] [CrossRef] [Scilit]
- Zhang, Y.; Li, M.; Dong, J.; Yang, H.; Van Zwieten, L.; Lu, H.; Alshameri, A.; Zhan, Z.; Chen, X.; Jiang, X.; et al. A Critical Review of Methods for Analyzing Freshwater Eutrophication. Water 2021, 13, 225. [Google Scholar] [CrossRef] [Scilit]
- Yan, T.; Shen, S.L.; Zhou, A. Indices and Models of Surface Water Quality Assessment: Review and Perspectives. Environ. Pollut. 2022, 308, 119611. [Google Scholar] [CrossRef] [Scilit]
- Yao, L.; Wang, W.; Yu, K.; Song, Z.; Lei, K.; AN, L. Assessment of Marine Eutrophication: Challenges and Solutions Ahead. Mar. Pollut. Bull. 2025, 216, 117977. [Google Scholar] [CrossRef] [Scilit]
- Gholizadeh, M.H.; Melesse, A.M.; Reddi, L. A Comprehensive Review on Water Quality Parameters Estimation Using Remote Sensing Techniques. Sensors 2016, 16, 1298. [Google Scholar] [CrossRef] [Scilit]
- Neofitou, N.; Papadimitriou, K.; Domenikiotis, C.; Tziantziou, L.; Panagiotaki, P. GIS in Environmental Monitoring and Assessment of Fish Farming Impacts on Nutrients of Pagasitikos Gulf, Eastern Mediterranean. Aquaculture 2019, 501, 62–75. [Google Scholar] [CrossRef] [Scilit]
- Dimoudi, A.; Karampetsou, P.; Domenikiotis, C.; Tziantziou, L.; Klaoudatos, D.; Skordas, K.; Panagiotaki, P.; Neofitou, N. A Spatial Interpolation Approach for Environmental Assessment of Aquaculture Effects on Water Quality of Pagasitikos Gulf (Eastern Mediterranean). Mar. Environ. Res. 2023, 188, 106036. [Google Scholar] [CrossRef] [Scilit]
- Ocean Biology Processing Group. Ocean Color Web. Available online: https://oceancolor.gsfc.nasa.gov/ (accessed on 26 January 2025).
- Dierssen, H.M.; Randolph, K. Remote Sensing of Ocean Color. In Earth System Monitoring; Orcutt, J., Ed.; Springer: New York, NY, USA, 2013; pp. 439–472. [Google Scholar] [CrossRef] [Scilit]
- Wang, X.; Yang, W. Water Quality Monitoring and Evaluation Using Remote-Sensing Techniques in China: A Systematic Review. Ecosyst. Health Sust. 2019, 5, 47–56. [Google Scholar] [CrossRef] [Scilit]
- Hassan, N.; Woo, C.S. Machine Learning Application in Water Quality Using Satellite Data. IOP Conf. Ser. Earth Environ. Sci. 2021, 842, 012018. [Google Scholar] [CrossRef] [Scilit]
- Yang, H.; Kong, J.; Hu, H.; Du, Y.; Gao, M.; Chen, F. A Review of Remote Sensing for Water Quality Retrieval: Progress and Challenges. Remote Sens. 2022, 14, 1770. [Google Scholar] [CrossRef] [Scilit]
- Adjovu, G.E.; Stephen, H.; James, D.; Ahmad, S. Overview of the Application of Remote Sensing in Effective Monitoring of Water Quality Parameters. Remote Sens. 2023, 15, 1938. [Google Scholar] [CrossRef] [Scilit]
- Page, M.J.; McKenzie, J.E.; Bossuyt, P.M.; Boutron, I.; Hoffmann, T.C.; Mulrow, C.D.; Shamseer, L.; Tetzlaff, J.M.; Akl, E.A.; Brennan, S.E.; et al. The PRISMA 2020 Statement: An Updated Guideline for Reporting Systematic Reviews. Br. Med. J. 2021, 372, n71. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Page, M.J.; Moher, D.; Bossuyt, P.M.; Boutron, I.; Hoffmann, T.C.; Mulrow, C.D.; Shamseer, L.; Tetzlaff, J.M.; Akl, E.A.; Brennan, S.E.; et al. PRISMA 2020 Explanation and Elaboration: Updated Guidance and Exemplars for Reporting Systematic Reviews. Br. Med. J. 2021, 372, n160. [Google Scholar] [CrossRef]
- Jessen, C.; Bednarz, V.N.; Rix, L.; Teichberg, M.; Wild, C. Marine Eutrophication. In Environmental Indicators; Armon, R.H., Hänninen, O., Eds.; Springer: Dordrecht, The Netherlands, 2015; pp. 177–203. [Google Scholar] [CrossRef] [Scilit]
- Devlin, M.; Brodie, J. Nutrients and Eutrophication. In Marine Pollution—Monitoring Management and Mitigation; Reichelt-Brushett, A., Ed.; Springer: Berlin, Germany, 2023; pp. 76–100. [Google Scholar] [CrossRef] [Scilit]
- Lin, K.; Zhu, Y.; Zhang, Y.; Lin, H. Determination of Ammonia Nitrogen in Natural Waters: Recent Advances and Applications. Trends Environ. Anal. Chem. 2019, 24, e00073. [Google Scholar] [CrossRef] [Scilit]
- Camargo, J.A.; Alonso, A.; Salamanca, A. Nitrate Toxicity to Aquatic Animals: A Review with New Data for Freshwater Invertebrates. Chemosphere 2005, 58, 1255–1267. [Google Scholar] [CrossRef] [Scilit]
- Smith, T.M.; Smith, R.L. Elements of Ecology, 9th ed.; Pearson Education: London, UK, 2015. [Google Scholar]
- Kim, E.; Yoo, S.; Ro, H.Y.; Han, H.J.; Baek, Y.W.; Eom, I.C.; Kim, H.M.; Kim, P.; Choi, K. Aquatic Toxicity Assessment of Phosphate Compounds. Environ. Health Toxicol. 2013, 28, e2013002. [Google Scholar] [CrossRef] [Scilit]
- Ittekkot, V.; Humborg, C.; Schäfer, P. Hydrological Alterations and Marine Biogeochemistry: A Silicate Issue? BioScience 2000, 50, 776–782. [Google Scholar] [CrossRef] [Scilit]
- Van Dokkum, H.P.; Hulskotte, J.H.J.; Kramer, K.J.M.; Wilmot, J. Emission, Fate and Effects of Soluble Silicates (Waterglass) in the Aquatic Environment. Environ. Sci. Technol. 2004, 38, 515–521. [Google Scholar] [CrossRef] [Scilit]
- Burnett, W.C.; Bokuniewicz, H.; Huettel, M.; Moore, W.S.; Taniguchi, M. Groundwater and Pore Water Inputs to the Coastal Zone. Biogeochemistry 2003, 66, 3–33. [Google Scholar] [CrossRef] [Scilit]
- Burnett, W.C.; Aggarwal, P.K.; Aureli, A.; Bokuniewicz, H.; Cable, J.E.; Charette, M.A.; Kontar, E.; Krupa, S.; Kulkarni, K.M.; Loveless, A.; et al. Quantifying Submarine Groundwater Discharge in the Coastal Zone via Multiple Methods. Sci. Total Environ. 2006, 367, 498–543. [Google Scholar] [CrossRef] [Scilit]
- Wang, X.; Cuthbertson, A.; Gualtieri, C.; Shao, D. A Review on Mariculture Effluent: Characterization and Management Tools. Water 2020, 12, 2991. [Google Scholar] [CrossRef] [Scilit]
- Karakassis, I.; Pitta, P.; Krom, M.D. Contribution of Fish Farming to the Nutrient Loading of the Mediterranean. Sci. Mar. 2005, 69, 313–321. [Google Scholar] [CrossRef] [Scilit]
- Mascarenhas, V.; Keck, T. Marine Optics and Ocean Color Remote Sensing. In YOUMARES 8—Oceans Across Boundaries: Learning from Each Other. Proceedings of the 2017 Conference for YOUngMARineRESearchers in Kiel, Germany; Jungblut, S., Liebich, V., Bode, M., Eds.; Springer: Cham, Switzerland, 2018; pp. 41–54. [Google Scholar] [CrossRef] [Scilit]
- Zaneveld, R.; Barnard, A.; Lee, Z. Why are Inherent Optical Properties Needed in Ocean-Colour Remote Sensing? In Remote Sensing of Inherent Optical Properties: Fundamentals, Tests of Algorithms, and Applications; Lee, Z.P., Ed.; Reports of the International Ocean-Colour Coordinating Group; IOCCG: Dartmouth, MA, Canada, 2006; Volume 5, pp. 1–10. [Google Scholar]
- Jaywant, S.A.; Arif, K.M. Remote Sensing Techniques for Water Quality Monitoring: A Review. Sensors 2024, 24, 8041. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Sathyendranath, S. General Information. In Remote Sensing of Ocean Colour in Coastal, and Other Optically-Complex, Waters; Sathyendranath, S., Ed.; Reports of the International Ocean-Colour Coordinating Group; IOCCG: Dartmouth, MA, Canada, 2000; Volume 3, pp. 5–21. [Google Scholar]
- Binding, C.; Stumpf, R.P.; Schaeffer, B.A.; Tyler, A.; Hunter, P. Introducing to Deriving Water Quality Measures from Satellites. In Earth Observations in Support of Global Water Quality Monitoring; Greb, S., Dekker, A., Binding, C., Eds.; Reports of the International Ocean-Colour Coordinating Group; IOCCG: Dartmouth, MA, Canada, 2018; Volume 17, pp. 15–27. [Google Scholar]
- Sagan, V.; Peterson, K.T.; Maimaitijiang, M.; Sidike, P.; Sloan, J.; Greeling, B.A.; Maalouf, S.; Adams, C. Monitoring Inland Water Quality Using Remote Sensing: Potential and Limitations of Spectral Indices, Bio-Optical Simulations, Machine Learning, and Cloud Computing. Earth-Sci. Rev. 2020, 205, 103187. [Google Scholar] [CrossRef] [Scilit]
- Zhu, B.; Bai, Y.; Zhang, Z.; He, X.; Wang, Z.; Zhang, S.; Dai, Q. Satellite Remote Sensing of Water Quality Variation in a Semi-Enclosed Bay (Yueqing Bay) under Strong Anthropogenic Impact. Remote Sens. 2022, 14, 550. [Google Scholar] [CrossRef] [Scilit]
- Earth Science Data Systems. Remote Sensing. Available online: https://www.earthdata.nasa.gov/learn/earth-observation-data-basics/remote-sensing (accessed on 28 January 2025).
- EARTHDATA. Algorithm Publication Tool. Remote Sensing Reflectance. Available online: https://www.earthdata.nasa.gov/apt/documents/rrs/v1.1#doc-header (accessed on 28 January 2025).
- Werdell, P.J.; McClain, C.R. Satellite Remote Sensing: Ocean Color. In Encyclopedia of Ocean Sciences, 3rd ed.; Elsevier: Amsterdam, The Netherlands, 2019; Volume 5, pp. 443–454. [Google Scholar] [CrossRef] [Scilit]
- Earth Science Data Systems. Synthetic Aperture Radar (SAR). Available online: https://www.earthdata.nasa.gov/learn/earth-observation-data-basics/sar (accessed on 2 February 2025).
- Gentemann, C.L.; Wentz, F.J.; Brewer, M.; Hilburn, K.; Smith, D. Passive Microwave Remote Sensing of the Ocean: An Overview. In Oceanography from Space; Barale, V., Gower, J.F.R., Alberotanza, L., Eds.; Springer: Hague, The Netherlands, 2010; pp. 13–33. [Google Scholar]
- Asiyabi, R.M.; Ghorbanian, A.; Tameh, S.N.; Amani, M.; Jin, S.; Mohammadzadeh, A. Synthetic Aperture Radar (SAR) for Ocean: A Review. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2023, 16, 9106–9138. [Google Scholar] [CrossRef] [Scilit]
- Mohseni, F.; Saba, F.; Mirmazloumi, S.M.; Amani, M.; Mokhtarzade, M.; Jamali, S.; Mahdavi, S. Ocean Water Quality Monitoring Using Remote Sensing Techniques: A Review. Mar. Environ. Res. 2022, 180, 105701. [Google Scholar] [CrossRef] [Scilit]
- Topp, S.N.; Pavelsky, T.M.; Jensen, D.; Simard, M.; Ross, M.R.V. Research Trends in the Use of Remote Sensing for Inland Water Quality Science: Moving towards Multidisciplinary Applications. Water 2020, 12, 169. [Google Scholar] [CrossRef] [Scilit]
- Janiesch, C.; Zschech, P.; Heinrich, K. Machine Learning and Deep Learning. Electron. Mark. 2021, 31, 685–695. [Google Scholar] [CrossRef] [Scilit]
- Sarker, I.H. AI-Based Modeling: Techniques, Applications and Research Issues Towards Automation, Intelligent and Smart Systems. SN Comput. Sci. 2022, 3, 158. [Google Scholar] [CrossRef] [Scilit]
- Alaskar, S. Proceedings of Integrated Intelligence Enable Networks and Computing; Singh Mer, K.K., Semwal, V.B., Bijalwan, V., Crespo, R.G., Eds.; Springer: Singapore, 2021. [Google Scholar] [CrossRef] [Scilit]
- Nguyen, G.; Dlugolinsky, S.; Bobák, M.; Tran, V.; LópezGarcía, Á.; Heredia, I.; Malík, P.; Hluchý, L. Machine Learning and Deep Learning Frameworks and Libraries for Large-Scale Data Mining: A Survey. Artif. Intell. Rev. 2019, 52, 77–124. [Google Scholar] [CrossRef] [Scilit]
- He, W.; Chen, S.; Liu, X.; Chen, J. Water Quality Monitoring in a Slightly-Polluted Inland Water Body Through Remote Sensing—Case study of the Guanting Reservoir in Beijing, China. Front. Environ. Sci. Eng. China 2008, 2, 163–171. [Google Scholar] [CrossRef] [Scilit]
- Dong, G.; Hu, Z.; Liu, X.; Fu, Y.; Zhang, W. Spatio-Temporal Variation of Total Nitrogen and Ammonia Nitrogen in the Water Source of the Middle Route of the South-to-North Water Diversion Project. Water 2020, 12, 2615. [Google Scholar] [CrossRef] [Scilit]
- Theologou, I.; Patelaki, M.; Karantzalos, K. Can Single Empirical Algorithms Accurately Predict Inland Shallow Water Quality Status from High Resolution, Multi-Sensor, Multi-Temporal Satellite Data? Int. Arch. Photogramm. Remote Sens. Spt. Inf. Sci. 2015, 40, 1511. [Google Scholar] [CrossRef] [Scilit]
- Liu, D.; Song, L.; Ji, D.; Hui, E. Nutrient Estimation by HJ-1 Satellite Imagery of Xiangxi Bay, Three Gorges Reservoir, China. Environ. Earth Sci. 2016, 75, 633. [Google Scholar] [CrossRef] [Scilit]
- Markogianni, V.; Kalivas, D.; Petropoulos, G.; Dimitriou, E. Analysis on the Feasibility of Landsat 8 Imagery for Water Quality Parameters Assessment in an Oligotrophic Mediterranean Lake. Int. J. Geol. Environ. Eng. 2017, 11, 906–914. [Google Scholar]
- Markogianni, V.; Kalvas, D.; Petropoulos, G.P.; Dimitriou, E. An Appraisal of the Potential of Landsat 8 in Estimating Chlorophyll-a, Ammonium Concentrations and Other Water Quality Indicators. Remote Sens. 2018, 10, 1018. [Google Scholar] [CrossRef] [Scilit]
- Al-Shaibah, B.; Liu, X.; Zhang, J.; Tong, Z.; Zhang, M.; El-Zeiny, A.; Faichia, C.; Hussain, M.; Tayyab, M. Modeling Water Quality Parameters Using Landsat Multispectral Images: A Case Study of Erlong Lake, Northeast China. Remote Sens. 2021, 13, 1603. [Google Scholar] [CrossRef] [Scilit]
- Khattab, M.F.O.; Merkel, B.J. Application of Landsat 5 and Landsat 7 Images Data for Water Quality Mapping in Mosul Dam Lake, Northern Iraq. Arab. J. Geosci. 2014, 7, 3557–3573. [Google Scholar] [CrossRef] [Scilit]
- Bilge, F.; Yazici, B.; Dogeroglu, T.; Ayday, C. Statistical Evaluation of Remotely Sensed Data for Water Quality Monitoring. Int. J. Remote Sens. 2003, 24, 5317–5326. [Google Scholar] [CrossRef] [Scilit]
- Singh, A.; Jakubowski, A.R.; Chidister, I.; Townsend, P.A. A MODIS Approach to Predicting Stream Water Quality in Wisconsin. Remote Sens. Environ. 2013, 128, 74–86. [Google Scholar] [CrossRef] [Scilit]
- Elhag, M.; Gitas, I.; Othman, A.; Bahrawi, J.; Gikas, P. Assessment of Water Quality Parameters Using Temporal Remote Sensing Spectral Reflectance in Arid Environments, Saudi Arabia. Water 2019, 11, 556. [Google Scholar] [CrossRef] [Scilit]
- Karaoui, I.; Arioua, A.; Boudhar, A.; Hssaisoune, M.; Mouatassime, S.E.; Ouhamchich, K.A.; Elhamdouni, D. Evaluating the Potential of Sentinel-2 Satellite Images for Water Quality Characterization of Artificial Reservoirs: The Bin El Ouidane Reservoir Case Study (Morocco). Meteorol. Hydrol. Water Manag. 2019, 7, 31–39. [Google Scholar] [CrossRef] [Scilit]
- Padilla-Mendoza, C.; Torres-Bejarano, F.; Campo-Daza, G.; González-Márquez, L.C. Potential of Sentinel Images to Evaluate Physicochemical Parameters Concentrations in Water Bodies-Application in a Wetlands System in Northern Colombia. Water 2023, 15, 789. [Google Scholar] [CrossRef] [Scilit]
- El Saadi, A.M.; Yousry, M.M.; Jahin, H.S. Statistical Estimation of Rosetta Branch Water Quality Using Multi-Spectral Data. Water Sci. 2014, 28, 18–30. [Google Scholar] [CrossRef] [Scilit]
- Mushtaq, F.; Nee Lala, M.G. Remote Estimation of Water Quality Parameters of Himalayan Lake (Kashmir) Using Landsat 8 OLI Imagery. Geocarto Int. 2017, 32, 274–285. [Google Scholar] [CrossRef] [Scilit]
- Abdelmalik, K.W. Role of Statistical Remote Sensing for Inland Water Quality Parameters Prediction. Egypt. J. Remote Sens. Space Sci. 2018, 21, 193–200. [Google Scholar] [CrossRef] [Scilit]
- Dewidar, K.; Khedr, A.A. Remote Sensing of Water Quality for Burullus Lake, Egypt. Geocarto Int. 2005, 20, 43–49. [Google Scholar] [CrossRef] [Scilit]
- Dewidar, K.; Khedr, A. Water Quality Assessment with Simultaneous Landsat-5 TM at Manzala Lagoon, Egypt. Hydrobiologia 2001, 457, 49–58. [Google Scholar] [CrossRef] [Scilit]
- Wang, X.J.; Ma, T. Application of Remote Sensing Techniques in Monitoring and Assessing the Water Quality of Taihu Lake. Bull. Environ. Contam. Toxicol. 2001, 67, 863–870. [Google Scholar] [CrossRef] [PubMed]
- Qiu, Y.; Zhang, H.E.; Tong, X.; Zhang, Y.; Zhao, J. Monitoring the water quality of water resources reservation area in upper region of Huangpu River using remote sensing. In Proceedings of the 2006 IEEE International Symposium on Geoscience and Remote Sensing, Denver, CO, USA, 31 July–4 August 2006; pp. 1074–1077. [Google Scholar] [CrossRef] [Scilit]
- Mallick, J.; Hasan, M.A.; Alashker, Y.; Ahmed, M. Bathymetric and Geochemical Analysis of Lake Al-Saad, Abha, Kingdom of Saudi Arabia Using Geoinformatics Technology. J. Geogr. Inf. Syst. 2014, 6, 440–452. [Google Scholar] [CrossRef]
- Somvanshi, S.; Kunwar, P.; Singh, N.B.; Shukla, S.P.; Pathak, V. Integrated Remote Sensing and GIS Approach for Water Quality Analysis of Gomti River, Uttar Pradesh. Int.J. Environ. Sci. 2012, 3, 62. [Google Scholar]
- Khalil, M.T.; Ahmed, M.H.; Kafrawy, S.B.E.; Emam, W.W.M.; El-Halim, A.; Saad, A.; Ahmed, M.H.M. Integrated Field Study, Remote Sensing and GIS Approach for Assessing and Monitoring Some Chemical Water Quality Parameters in Bardawil Lagoon, Egypt. Int. J. Innov. Res. Sci. Eng. Technol. 2016, 5, 14656–14669. [Google Scholar] [CrossRef]
- Gonzalez-Marquez, L.C.; Torres-Bejarano, F.M.; Torregroza-Espinosa, A.C.; Hansen-Rodríguez, I.V.; Rodríguez-Gallegos, H.B. Use of LANDSAT 8 Images for Depth and Water Quality Assessment of El Guajaro Reservoir, Colombia. J. S. Am. Earth Sci. 2018, 82, 231–238. [Google Scholar] [CrossRef] [Scilit]
- Japitana, M.V.; Burce, M.E.C. A Satellite-Based Remote Sensing Technique for Surface Water Quality Estimation. Eng. Technol. Appl. Sci. Res. 2019, 9, 3965–3970. [Google Scholar] [CrossRef] [Scilit]
- Pizani, F.M.C.; Maillard, P.; Ferreira, A.F.F.; De Amorim, C.C. Estimation of Water Quality in a Reservoir from Sentinel-2 MSI and Landsat-8 OLI Sensors. ISPRS Ann. Photogramm. Remote Sens. Spat. Inf. Sci. 2020, 3, 401–408. [Google Scholar] [CrossRef] [Scilit]
- Ahmed, W.; Mohammed, S.; El-Shazly, A.; Morsy, S. Tigris River Water Surface Quality Monitoring Using Remote Sensing Data and GIS Techniques. Egypt. J. Remote Sens. Space Sci. 2023, 26, 816–825. [Google Scholar] [CrossRef] [Scilit]
- Zhang, Y.; Jing, W.; Deng, Y.; Zhou, W.; Yang, J.; Li, Y.; Cai, Y.; Hu, Y.; Peng, X.; Lan, W.; et al. Water Quality Parameters Retrieval of Coastal Mariculture Ponds Based on UAV Multispectral Remote Sensing. Front. Environ. Sci. 2023, 11, 1079397. [Google Scholar] [CrossRef] [Scilit]
- Satapathy, D.R.; Vijay, R.; Kamble, S.R.; Sohony, R.A. Remote Sensing of Turbidity and Phosphate in Creeks and Coast of Mumbai: An Effect of Organic Matter. Trans. GIS 2010, 14, 811–832. [Google Scholar] [CrossRef] [Scilit]
- Xu, Y.; Zhang, Y.; Zhang, D. Retrieval of Dissolved Inorganic Nitrogen from Multi-Temporal MODIS Data in Haizhou Bay. Mar. Geod. 2010, 33, 1–15. [Google Scholar] [CrossRef] [Scilit]
- Yu, X.; Yi, H.; Liu, X.; Wang, Y.; Liu, X.; Zhang, H. Remote-Sensing Estimation of Dissolved Inorganic Nitrogen Concentration in the Bohai Sea Using Band Combinations Derived from MODIS Data. Int. J. Remote Sens. 2016, 37, 327–340. [Google Scholar] [CrossRef] [Scilit]
- Kim, Y.H.; Son, S.; Kim, H.C.; Kim, B.; Park, Y.G.; Nam, J.; Ryu, J. Application of Satellite Remote Sensing in Monitoring Dissolved Oxygen Variabilities: A Case Study for Coastal Waters in Korea. Environ. Int. 2020, 134, 105301. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Bourouhou, I.; Salmoun, F. Sea Water Quality Monitoring Using Remote Sensing Techniques: A Case Study in Tangier-Ksar Sghir Coastline. Environ. Monit. Assess. 2021, 193, 557. [Google Scholar] [CrossRef] [Scilit]
- Dong, L.; Wang, D.; Song, L.; Gong, F.; Chen, S.; Huang, J.; He, X. Monitoring Dissolved Oxygen Concentrations in the Coastal Waters of Zhejiang Using Landsat-8/9 Imagery. Remote Sens. 2024, 16, 1951. [Google Scholar] [CrossRef] [Scilit]
- Abualhin, K.; Abushaban, S. Predictive models of non-optically active coastal water quality parameters by remote sensing imagery. Marit. Technol. Res. 2024, 7, 274944. [Google Scholar] [CrossRef] [Scilit]
- Wang, X.; Ma, L.; Wang, X. Apply Semi-Supervised Support Vector Regression for Remote Sensing Water Quality Retrieving. In Proceedings of the IEEE International Geoscience and Remote Sensing Symposium, Honolulu, HI, USA, 25–30 July 2010; pp. 2757–2760. [Google Scholar] [CrossRef] [Scilit]
- Wang, X.; Fu, L.; He, C. Applying Support Vector Regression to Water Quality Modelling by Remote Sensing Data. Int. J. Remote Sens. 2011, 32, 8615–8627. [Google Scholar] [CrossRef] [Scilit]
- Tian, S.; Guo, H.; Xu, W.; Zhu, X.; Wang, B.; Zeng, Q.; Mai, Y.; Huang, J.J. Remote Sensing Retrieval of Inland Water Quality Parameters Using Sentinel-2 and Multiple Machine Learning Algorithms. Environ. Sci. Pollut. Res. 2023, 30, 18617–18630. [Google Scholar] [CrossRef] [Scilit]
- Song, W.; Yinglan, A.; Wang, Y.; Fang, Q.; Tang, R. Study on Remote Sensing Inversion and Temporal-Spatial Variation of Hulun Lake Water Quality Based on Machine Learning. J. Contam. Hydrol. 2024, 260, 104282. [Google Scholar] [CrossRef] [Scilit]
- Liu, Y.; Wang, Z.; Yue, H. Water Quality Parameters Retrieval and Nutrient Status Evaluation Based on Machine Learning Methods and Sentinel-2 Imagery: A Case Study of the Hongjiannao Lake. Environ. Monit. Assess. 2025, 197, 556. [Google Scholar] [CrossRef] [Scilit]
- Chen, B.; Mu, X.; Chen, P.; Wang, B.; Choi, J.; Park, H.; Xu, S.; Wu, Y.; Yang, H. Machine Learning-Based Inversion of Water Quality Parameters in Typical Reach of the Urban River by UAV Multispectral Data. Ecol. Indic. 2021, 133, 108434. [Google Scholar] [CrossRef] [Scilit]
- Lo, Y.; Fu, L.; Lu, T.; Huang, H.; Kong, L.; Xu, Y.; Zhang, C. Medium-Sized Lake Water Quality Parameters Retrieval Using Multispectral UAV Image and Machine Learning Algorithms: A Case Study of the Yuandang Lake, China. Drones 2023, 7, 244. [Google Scholar] [CrossRef] [Scilit]
- Zhao, Y.; Yu, T.; Hu, B.; Zhang, Z.; Liu, Y.; Liu, X.; Liu, H.; Liu, J.; Wang, X.; Song, S. Retrieval of Water Quality Parameters Based on Near-Surface Remote Sensing and Machine Learning Algorithm. Remote Sens. 2022, 14, 5305. [Google Scholar] [CrossRef] [Scilit]
- Tian, D.; Zhao, X.; Gao, L.; Liang, Z.; Yang, Z.; Zhang, P.; Wu, Q.; Ren, K.; Li, R.; Yang, C.; et al. Estimation of Water Quality Variables Based on Machine Learning Model and Cluster Analysis-Based Empirical Model Using Multi-Source Remote Sensing Data in Inland Reservoirs, South China. Environ. Pollut. 2024, 342, 123104. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Niu, C.; Tan, K.; Jia, X.; Wang, X. Deep Learning Based Regression for Optically Inactive Inland Water Quality Parameter Estimation Using Airborne Hyperspectral Imagery. Environ. Pollut. 2021, 286, 117534. [Google Scholar] [CrossRef] [Scilit]
- Wang, Z.; Wei, L.; He, C.; Lu, Q. Ammonia Nitrogen Monitoring of Urban Rivers with UAV-Borne Hyperspectral Remote Sensing Imagery. In Proceedings of the 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS, Brussels, Belgium, 11–16 July 2021; pp. 3713–3716. [Google Scholar] [CrossRef] [Scilit]
- Fu, B.; Lao, Z.; Liang, Y.; Sun, J.; He, X.; Deng, T.; He, W.; Fan, D.; Gao, E.; Hou, Q. Evaluating Optically and Non-Optically Active Water Quality and Its Response Relationship to Hydro-Meteorology Using Multi-Source Data in Poyang Lake, China. Ecol. Indic. 2022, 145, 109675. [Google Scholar] [CrossRef] [Scilit]
- Toming, K.; Liu, H.; Soomets, T.; Uuemaa, E.; Nõges, T.; Kutser, T. Estimation of the Biogeochemical and Physical Properties of Lakes Based on Remote Sensing and Artificial Intelligence Applications. Remote Sens. 2024, 16, 464. [Google Scholar] [CrossRef] [Scilit]
- Cao, Q.; Yu, G.; Sun, S.; Dou, Y.; Li, H.; Qiao, Z. Monitoring Water Quality of the Haihe River Based on Ground-Based Hyperspectral Remote Sensing. Water 2022, 14, 22. [Google Scholar] [CrossRef] [Scilit]
- Li, N.; Ning, Z.; Chen, M.; Wu, D.; Hao, C.; Zhang, D.; Bai, R.; Liu, H.; Chen, X.; Li, W.; et al. Satellite and Machine Learning Monitoring of Optically Inactive Water Quality Variability in a Tropical River. Remote Sens. 2022, 14, 5466. [Google Scholar] [CrossRef] [Scilit]
- Amanollahi, J.; Kaboodvandpour, S.; Majidi, H. Evaluating the Accuracy of ANN and LR Models to Estimate the Water Quality in Zarivar International Wetland, Iran. Nat. Hazards 2017, 85, 1511–1527. [Google Scholar] [CrossRef] [Scilit]
- Hassan, G.; Goher, M.E.; Shaheen, M.; Taie, A.S.A. Hybrid Predictive Model for Water Quality Monitoring Based on Sentinel-2A L1C Data. IEEE Access 2021, 9, 65730–65749. [Google Scholar] [CrossRef] [Scilit]
- Krishnaraj, A.; Honnasiddaiah, R. Remote Sensing and Machine Learning Based Framework for the Assessment of Spatio-Temporal Water Quality in the Middle Ganga Basin. Environ. Sci. Pollut. Res. 2022, 29, 64939–64958. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Yang, W.; Fu, B.; Li, S.; Lao, Z.; Deng, T.; He, W.; He, H.; Chen, Z. Monitoring Multi-Water Quality of Internationally Important Karst Wetland through Deep Learning, Multi-Sensor and Multi-Platform Remote Sensing Images: A Case Study of Guilin, China. Ecol. Indic. 2023, 154, 110755. [Google Scholar] [CrossRef] [Scilit]
- Vatitsi, K.; Siachalou, S.; Latinopoulos, D.; Kagalou, I.; Akratos, C.S.; Mallinis, G. Monitoring Water Quality Parameters in Small Rivers Using SuperDove Imagery. Water 2024, 16, 758. [Google Scholar] [CrossRef] [Scilit]
- Li, H.; Wang, N.; Du, Z.; Huang, D.; Shi, M.; Zhong, Z.; Yuan, D. Multi-Parameter Water Quality Inversion in Heterogeneous Inland Waters Using UAV-Based Hyperspectral Data and Deep Learning Methods. Remote Sens. 2025, 17, 2191. [Google Scholar] [CrossRef] [Scilit]
- Sharaf El Din, E.; Zhang, Y.; Suliman, A. Mapping Concentrations of Surface Water Quality Parameters Using a Novel Remote Sensing and Artificial Intelligence Framework. Int. J. Remote Sens. 2017, 38, 1023–1042. [Google Scholar] [CrossRef] [Scilit]
- Peterson, K.T.; Sagan, V.; Sloan, J.J. Deep Learning-Based Water Quality Estimation and Anomaly Detection Using Landsat-8/Sentinel-2 Virtual Constellation and Cloud Computing. GIsci. Remote Sens. 2020, 57, 510–525. [Google Scholar] [CrossRef] [Scilit]
- Wang, L.; Yue, X.; Wang, H.; Ling, K.; Liu, Y.; Wang, J.; Hong, J.; Pen, W.; Song, H. Dynamic Inversion of Inland Aquaculture Water Quality Based on UAVs-WSN Spectral Analysis. Remote Sens. 2020, 12, 402. [Google Scholar] [CrossRef] [Scilit]
- Ahmed, M.; Mumtaz, R.; Anwar, Z.; Shaukat, A.; Arif, O.; Shafait, F. A Multi–Step Approach for Optically Active and Inactive Water Quality Parameter Estimation Using Deep Learning and Remote Sensing. Water 2022, 14, 2112. [Google Scholar] [CrossRef] [Scilit]
- Wang, D.; Cui, Q.; Gong, F.; Wang, L.; He, X.; Bai, Y. Satellite Retrieval of Surface Water Nutrients in the Coastal Regions of the East China Sea. Remote Sens. 2018, 10, 1896. [Google Scholar] [CrossRef] [Scilit]
- Huang, J.; Wang, D.; Pan, S.; Li, H.; Gong, F.; Hu, H.; He, X.; Bai, Y.; Zheng, Z. A New High-Resolution Remote Sensing Monitoring Method for Nutrients in Coastal Waters. IEEE Trans. Geosci. Remote Sens. 2023, 61, 4206315. [Google Scholar] [CrossRef] [Scilit]
- Zhu, L.; Cui, T.; Runa, A.; Pan, X.; Zhao, W.; Xiang, J.; Cao, M. Robust Remote Sensing Retrieval of Key Eutrophication Indicators in Coastal Waters Based on Explainable Machine Learning. ISPRS J. Photogramm. Remote Sens. 2024, 211, 262–280. [Google Scholar] [CrossRef] [Scilit]
- Wu, S.; Qi, J.; Yan, Z.; Lyu, F.; Lin, T.; Wang, Y.; Du, Z. Spatiotemporal Assessments of Nutrients and Water Quality in Coastal Areas Using Remote Sensing and a Spatiotemporal Deep Learning Model. Int. J. Appl. Earth Obs. Geoinf 2022, 112, 102897. [Google Scholar] [CrossRef] [Scilit]
- Chatziantoniou, A.; Spondylidis, S.C.; Stavrakidis-Zachou, O.; Papandroulakis, N.; Topouzelis, K. Dissolved Oxygen Estimation in Aquaculture Sites Using Remote Sensing and Machine Learning. Remote Sens. Appl. Soc. Environ. 2022, 28, 100865. [Google Scholar]
- Zhu, X.; Guo, H.; Huang, J.J.; Tian, S.; Xu, W.; Mai, Y. An Ensemble Machine Learning Model for Water Quality Estimation in Coastal Area Based on Remote Sensing Imagery. J. Environ. Manag. 2022, 323, 116187. [Google Scholar] [CrossRef] [Scilit]
- Ouma, Y.O.; Waga, J.; Okech, M.; Lavisa, O.; Mbuthia, D. Estimation of Reservoir Bio-Optical Water Quality Parameters Using Smartphone Sensor Apps and Landsat ETM+: Review and Comparative Experimental Results. J. Sens. 2018, 2018, 1–32. [Google Scholar] [CrossRef] [Scilit]
- Dimoudi, A.; Domenikiotis, C.; Klaoudatos, D.; Neofitou, N. A Methodological Approach for Retrieval of Ammonium in Coastal Aquaculture Based on Remotely Sensed Imagery and Machine Learning Algorithms. Aquac. Int. 2025, 33, 459. [Google Scholar] [CrossRef] [Scilit]
- NASA. Landsat Science. Landsat 4. Available online: https://landsat.gsfc.nasa.gov/satellites/landsat-4/ (accessed on 16 February 2025).
- NASA. Landsat Science. Thematic Mapper (TM). Available online: https://landsat.gsfc.nasa.gov/thematic-mapper/ (accessed on 16 February 2025).
- Sayler, K. Landsat 7 (L7) Data Users Handbook. Version 3.0. 2024. Available online: https://www.usgs.gov/media/files/landsat-7-data-users-handbook (accessed on 16 February 2025).
- GeoEye. IKONOS Imagery Products Guide. Version 1.5. 2006. Available online: https://earth.esa.int/eogateway/documents/20142/37627/IKONOS-Imagery-Product-Guide.pdf (accessed on 16 February 2025).
- LAADS DAAC. Terra & Aqua Moderate Resolution Imaging Spectroradiometer (MODIS). Available online: https://ladsweb.modaps.eosdis.nasa.gov/missions-and-measurements/modis/ (accessed on 16 February 2025).
- ESA. SPOT 5. Available online: https://earth.esa.int/eogateway/missions/spot-5 (accessed on 16 February 2025).
- ESA. IRS-R2 (ResourceSat-2). Available online: https://earth.esa.int/eogateway/missions/resourcesat-2 (accessed on 16 February 2025).
- Pérez-Planells, L.; Valor, E.; Coll, C.; Niclòs, R. Comparison and Evaluation of the TES and ANEM Algorithms for Land Surface Temperature and Emissivity Separation Over the Area of Valencia, Spain. Remote Sens. 2017, 9, 1251. [Google Scholar] [CrossRef] [Scilit]
- Wang, Q. Technical System Design and Construction of China’s HJ-1 Satellites. Int. J. Digit. Earth 2012, 5, 202–216. [Google Scholar] [CrossRef] [Scilit]
- ESA. WorldView-2. Available online: https://earth.esa.int/eogateway/missions/worldview-2 (accessed on 16 February 2025).
- Lamquin, N.; Mazeran, C.; Doxaran, D.; Ryu, J.H.; Park, Y.J. Assessment of GOCI Radiometric Products Using MERIS, MODIS and Field Measurements. Ocean Sci. J. 2012, 47, 287–311. [Google Scholar] [CrossRef] [Scilit]
- CEOS; ESA. ZY-3-01 Mission. Available online: https://database.eohandbook.com/database/missionsummary.aspx?missionID=732&utm_source=eoportal&utm_content=zy-3a (accessed on 16 February 2025).
- Ihlen, V. Landsat 8 (L8) Data Users Handbook. Version 5.0. 2019. Available online: https://www.usgs.gov/media/files/landsat-8-data-users-handbook (accessed on 16 February 2025).
- Sayler, K. Landsat 9 Data Users Handbook. Version 1.0. 2022. Available online: https://www.usgs.gov/media/files/landsat-9-data-users-handbook (accessed on 16 February 2025).
- ESA. S2 Mission. Available online: https://sentiwiki.copernicus.eu/web/s2-mission (accessed on 16 February 2025).
- LAADS DAAC. Copernicus Sentinel-3 Ocean and Land Colour Instrument (OLCI). Available online: https://ladsweb.modaps.eosdis.nasa.gov/missions-and-measurements/olci/ (accessed on 27 April 2025).
- ESA. PlanetScope. Available online: https://earth.esa.int/eogateway/missions/planetscope (accessed on 16 February 2025).
- Jian, W.; Yi, W.; Wenlong, W.; Lei, S.; Haiping, S. Transfer-Learning-Based Cloud Detection for Zhuhai-1 Satellite Hyperspectral Imagery. Front. Environ. Sci. 2022, 10, 1039249. [Google Scholar] [CrossRef] [Scilit]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2025 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).




