Overview of the Application of Remote Sensing in Effective Monitoring of Water Quality Parameters

Abstract

This study provides an overview of the techniques, shortcomings, and strengths of remote sensing (RS) applications in the effective retrieval and monitoring of water quality parameters (WQPs) such as chlorophyll-a concentration, turbidity, total suspended solids, colored dissolved organic matter, total dissolved solids among others. To be effectively retrieved by RS, these WQPs are categorized as optically active or inactive based on their influence on the optical characteristics measured by RS sensors. RS applications offer the opportunity for decisionmakers to quantify and monitor WQPs on a spatiotemporal scale effectively. The use of RS for water quality monitoring has been explored in many studies using empirical, analytical, semi-empirical, and machine-learning algorithms. RS spectral signatures have been applied for the estimation of WQPs using two categories of RS, namely, microwave and optical sensors. Optical RS, which has been heavily applied in the estimation of WQPs, is further grouped as spaceborne and airborne sensors based on the platform they are on board. The choice of a particular sensor to be used in any RS application depends on various factors including cost, and spatial, spectral, and temporal resolutions of the images. Some of the known satellite sensors used in the literature and reviewed in this paper include the Multispectral Instrument aboard Sentinel-2A/B, Moderate Resolution Imaging Spectroradiometer, Landsat Thematic Mapper, Enhanced Thematic Mapper, and Operational Land Imager.

1. Introduction

Water quality parameters (WQPs) have traditionally been analyzed and monitored through in situ sampling and laboratory testing; however, these approaches are expensive, labor-intensive, time-consuming, and not suitable for large-scale analysis. Sample collections are therefore limited over spatial and temporal scales [1,2,3,4,5]. The use of these conventional methods in water quality monitoring has been taxing due to limited resources in terms of capital and labor available particularly in developing countries. Conventional methods have limited capability to capture the horizontal heterogeneity in the water ecosystem on a large scale and extent. Monitoring, management, and forecasting of the entire water bodies may be inaccessible due to the topographic terrain of the water body. Furthermore, the accuracy and precision of sampled in situ data can be questionable owing to both errors that may arise from field sampling and laboratory analysis [6,7,8,9]. These limitations have put constraints on efforts to monitor the ecological suitability of water bodies over the past decades. Remote sensing (RS) application alternatives have been used in water quality monitoring at local, regional, and global scales in the last six decades and can overcome the limitations associated with conventional methods [10,11,12].

The paper provides a compilation of efforts utilized in the monitoring of WQPs using RS, satellite sensors, and wave bands as were available through 2022 for several WQPs. This review also evaluates the strengths and limitations that could lead to errors, including image processing errors, pixel resampling problems, and atmospheric interference of the sensors and platforms, which are necessary for continued research in the field of RS in water quality monitoring.

RS applications for water quality monitoring offer synoptic spatial and temporal coverage of the water body that is unattainable by conventional methods, making it ideal and suited for the cost-effective assessment of water quality [8,13]. Most of the satellite images are made available at no cost to users [3]. RS applications, unlike in situ field measurement, allow for analyses of WQPs on a large scale, and hence, are not limited to a single location in space and time. These applications aid to fill the global gap in the spatiotemporal data. They also provide present-time and comparable accurate results at a very fast pace [2,14,15,16,17,18]. Water resources researchers, decision, and policymakers can rely on remotely sensed data to reinforce their abilities in the effective monitoring of the fast-depleting water resources [19]. RS coupled with strategic in situ measurement and monitoring is a useful tool in aiding to predict, mitigate, and possibly prevent future water quality issues [20].

RS sensors capture spectral signatures of water constituents, which makes them an effective tool for the estimation of different WQPs including TSS and chlorophyll-a concentration [1]. Several studies have been carried out that have utilized RS applications in the qualitative and quantitative determinations of WQPs. Qualitative determinations include the identification of the presence, absence, or approximation of WQPs. When RS is combined with calibration curves from ground truth observations, quantitative estimates can be obtained, which are of much greater value to water scientists, engineers, planners, regulators, and policymakers.

Different RS sensors have been reportedly used in the estimation of WQPs, including Landsat 5, 7, and 8 Operational Land Imager (OLI) sensors, Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) images, Sentinel-2A/B Multispectral Instruments (MSI), Pleiades-1A, Moderate Resolution Imaging Spectroradiometer (MODIS), and Medium-Resolution Imaging Spectrometer (MERIS); these sensors have been used extensively to effectively quantify various parameters, including but not limited to, chlorophyll-a concentration, turbidity, TSS, total phosphate, COD concentration, algae content, total dissolved organic matter, colored dissolved organic matter (CDOM), dissolved organic carbon (DOC), Secchi disk depth (SDD), pH, and total dissolved solids (TDS), in inland and coastal water bodies [8,15,17,21,22,23,24,25,26,27,28,29,30,31,32]. WQPs such as TSS and CDOM have been effectively determined using RS techniques because they are spectrally active in the visible and IR portions of the electromagnetic spectrum.

RS has been utilized as an efficient tool for supporting the implementation of the European Union Water Framework Directive (EU WFD) as it provides needed tools for the derivation of WQPs including CDOM, total suspended matter (TSM), chlorophyll-a, the spectral attenuation coefficient, and phycocyanin [33,34]. Methods used to retrieve WQPs are dependent on the optical properties of the inland water bodies since the measurement parameters are mostly physical, chemical, or biological [1,35]. WQPs are categorized as either optically active or weak variables. Optically active parameters are constituents of the water that interact with light and subsequently change the energy spectrum of reflected solar radiation from water bodies. These parameters can be measured with RS. Parameters such as total suspended solids (TSS), suspended materials/solids (SM), chlorophyll-a, and turbidity constitute optically active parameters, while total nitrogen (TN), ammonia nitrogen (NH3-N), pH, dissolved oxygen, micro-pollutants, and dissolved phosphorus (DP) are either optically weak with a low signal/noise ratio or are optically inactive in the visible and near IR. These weak parameters are estimated through inference by association with active optical parameters, and proxy assimilation.

Optically active parameters are, however, complex if a reflected signal contains information from more than one constituent and is spatially and temporally variable. The retrieval of these parameters is therefore applicable using specific sensors and in particular environments [1,10,16,36,37].

One of the important parameters for reflecting water quality is water color. The water color is determined by the absorption and scattering of water components due to the interactions between sunlight and various materials in the water. The interaction between the concentration of optically active substances in the water and the color of the water makes it a crucial criterion for affecting water quality [1]. For example, dissolved organic matter absorbs light in both the visible and ultraviolet range of the electromagnetic spectrum, thereby affecting the volume of reflectance at a shorter wavelength [38].

2. Water Quality Monitoring

This section provides an overview of the literature on the efforts carried out on water degradation and WQPs.

2.1. Water Resources Degradation

Water resources cover about 75% of the earth’s surface, However, only 3% is portable. Approximately, 25% of the world’s population does not have access to a satisfactory quantity and or quality of freshwater [39]. Water bodies are crucial to the existence of humans, aquatic life, and the whole ecosystem. Water bodies including inland waters are the main source of water for drinking, power generation, industrial cooling, supporting biodiversity, ecosystem services, recreation, transportation routes, waste disposal, agriculture production, energy production, regional planning, and fish farming [11,18,40,41,42,43,44,45]. Inland water bodies are at the receiving end of the effect of climate change, development, urbanization, and contamination caused by rapid and uncontrolled environmental changes including nutrient pollution, drought, and changes in land use and land cover, which results in negative impacts such as the proliferation of toxic blue-green algae, extreme turbidity, accelerated eutrophication among others, which impacts the sustainability of the water resources [46,47,48,49]. Eutrophication in water bodies is a condition that leads to significant changes in the algal nutrient loads, which then leads to an imbalance in the water ecosystem with a bloom of algal and intense oxygen consumption at the bottom of the lakes posing a threat to the water ecosystem. This situation has the potential to alter the pristine ecological conditions of water bodies [12,42,50]. Anthropogenic activities such as irrigation, the use of fertilizers, grazing, indiscriminate discharge of untreated sewage from wastewater treatments, and economic development around watersheds, particularly in industrial areas, are continually causing the impairment and degradation of freshwater systems such as lakes and rivers [17,45,51,52,53,54]. The degradation of water bodies including river catchments is one of the important issues of environmental concern affecting the quality of water in tropical countries [55]. Studies have shown that rivers and streams flowing through the Great Plains of Central North America are now endangered and deteriorated and are no longer a sustainer of biodiversity [56]. In some watersheds, the deterioration of water bodies continues to worsen due to urbanization and climate change.

Climate change is said to be a worsening factor that poses an unending stressor on the ecological status of the water systems including lakes. The quantification of the exact role played by climate change in the ecology of the water bodies is however problematic due to the importance played by nutrients compared with slow changes occurring in the physical characteristics of the water column [50]. Climate change is known to result in corresponding changes in streamflow and the temperature of water bodies. An increase in water temperatures alters the biochemical reaction rates, causing thermal stratification in freshwater bodies. Water temperatures increase and can cause lake mixing reduction, which combined with increased oxygen consumption of biota can reduces the quality of water [57]. Climate change is also a cause of salinity intrusion, which can have a deleterious effects on agriculture production, economics, and the environment [58]. Predicting ecological and hydrological responses of water systems to climate change and their future variability is a challenge for scientists across the globe [59].

Water bodies are under intense pressure due to environmental and ecological changes caused by rapid population growth and development. A national survey on the quality of water bodies in China reported the quality of 22.1% of the country’s inland water bodies as bad. The severity of the pollution of water resources has resulted in a shortage of freshwater [1]. Water pollution is now one of the most serious issues posing a threat to the water environment, human health, and the water as a resource, which needs to be addressed with the needed urgency [5]. Pollution of water bodies was, however, appreciated as far back in human history. The Romans in 312 B.C. [60] noted that the river Tiber was too polluted to be used as drinking water. The English started considering river pollution as environmental pollution as far back as the 13th century when the laws prohibited the washing of charcoals in the River Thames in Middlesex for the fear of causing downstream pollution. By the late 19th and early 20th centuries, many of the major rivers in Europe including the Elbe, Danube, and Rhine (used as a source of drinking water by the Dutch) experienced marked pollution problems [60]. Known water quality-related issues include high organic suspended solids, high biological oxygen demand, and high ammonia levels. These led to a reduction in plant and fish species, invertebrates, and the subsequent total disappearance of fish [60].

Impairment of water resources brings about economic losses. Changes in the quality of water in rivers and lakes cause changes in the quality of recreational and commercial activities. There is a decline in boating, fishing, and swimming in these water bodies when they are impaired, which brings about significant economic losses. A study performed using biophysical modeling establishes that Virginia households may need to pay as much as USD 184 million yearly to ensure there is an improvement in the quality of their water [57].

Drinking impaired water results in waterborne diseases. Most of the reported waterborne outbreaks in North America are linked to contaminated rural drinking water systems exposed to microbial agents such as viruses, parasites, and bacteria. Water in these areas is exposed to these microbial agents due to the high agricultural activities (i.e., livestock production), and inadequate advanced treatment for human, animal, and wastewater, leading to an increase in human exposure to water contaminants [61]. An estimated 50 and 120 million cases of acute respiratory and gastrointestinal diseases associated with swimming in contaminated beach waters, respectively, are reported worldwide [62].

These issues stemming from exposure to contaminated water show the importance of managing and protecting water resources, especially concerning water quality. Protecting, restoring, and managing water resources is, however, a complex process that requires a systemic framework to deal with the issues. The process involved in water management often faces political and social contestation as opposed to regular problem-solving approaches, which are often undisputed and straightforward [63]. For example, the EU WFD, which came into action in December 2020 for all European Union member countries, ordered good ecological and chemical status in terms of pollutant levels by the year 2027 for all water bodies larger than 50 hectares in Europe. A good ecological status is described as a condition in which alterations in hydro-morphological conditions cause no more than a slight change in the abundance and composition of essential biological indicators including phytoplankton, macrophytes, and fish fauna among others [33,59].

2.2. Water Quality Parameters

Water quality is an overall indicator of the condition and general well-being of the water body, being comprised of its physical, chemical, biological, and radiological properties [1]. To protect and restore the quality of water resources, monitoring efforts and management strategies are key and need to be successful to achieve desirable results of a healthy water body in the ecosystem economy [64,65]. We cannot overemphasize the knowledge of the WQPs and their characterization in the monitoring efforts. WQPs are generally categorized as physical, chemical, and biological parameters. Physical water indicators include pH, dissolved oxygen (DO), total suspended solids (TSS), electrical conductivity (EC), salinity or total dissolved solids (TDS), nutrients such as nitrogen and phosphorus, and aesthetics (i.e., odor, color, and taints) Chemical parameters, however, include, but are not limited to, cations such as calcium (Ca²⁺), potassium (Na⁺), and magnesium (Mg²⁺) as well as anions such as nitrate (NO³⁻), sulfate (SO₄²⁻), and chloride (Cl⁻); meanwhile, biological parameters include algae and bacteria such as Escherichia coli (E. coli) and total coliforms. Categorization of these parameters was performed through correlation analysis [18,66]. Other contaminants in water include volatile organic contaminants (VOCs) that are detrimental to public health and are of environmental concern due to their usage by industry-inclined countries. Some of the most organic contaminants include aromatic hydrocarbons (AHs) and chlorinated hydrocarbons (CHCs) [67]. VOCs such as tetrachloroethylene and trichloroethylene are found in groundwater while polynuclear aromatic hydrocarbons (PAHs) generated during the combustion of organic materials are found in surface water [68].

Measured WQPs are used to evaluate the degree of organic contamination of water bodies including chemical oxygen demand (COD), biochemical oxygen demand (BOD), and total organic carbon (TOC). The COD and BOD are two indicators that have been accepted as surrogate indicators of organic contaminants in water bodies [64]. TOC is made from a fraction of particulate organic carbon (POC) and dissolved organic carbon (DOC) and has been used in studies as an alternative test procedure to rapidly assess the quality of untreated and treated water. DOC in natural water is mostly composed of hydrophilic and hydrophobic [69,70]. DOC has also been considered a surrogate parameter in the quantification of dissolved organic matter (DOM) in water [71]. DOC and CDOM are key water quality indicators to examine the biogeochemical state of coastal waters [72]. Most organic matter in the water column of water systems is usually in dissolved form. DOM in water bodies such as lakes is made of a large degree of colored humic substances that find their way into water bodies from the terrestrial environments of the water bodies [73]. Quantifying CDOM in water bodies is essential to know its absolute concentration of DOM [74]. Soluble COD and DOC are parameters used in industrial or municipal plants to optimize carbon degradation processes. They are for persistent compounds in wastewater effluents that are not eliminated in the biological stage of the wastewater treatment plants (WWTPs) [75]. DOC has been linked to the spectral shape of water spectral reflectance and, in turn, used to capture mash DOC export in a study to examine the spatial features and temporal changes in DOC over a range of conditions using Landsat 8 OLI and Sentinel-2 images [76].

The detection of organic pollutants such as COD has been performed through instrumentalmethods, which include dichromate-spectrophotometry and dichromate-titration. These methods are however faced with some deficiencies including the longer amount of time needed for the determination of COD and reagents, the high cost associated with them, the need for online digestion of water samples at high pressure and temperature, and the production of secondary pollutions. Other known methods of COD measurement include hydroxyl radical oxidation-electrochemical analysis and UV absorption spectrum analysis. While UV absorption spectrum analysis does not need reagents, it has poor stability and universality owing to its absorbance and poor linear correlation with COD concentration. The hydroxyl radical oxidation-electrochemistry analysis is however able to meet the need for the rapid determination of COD at normal conditions of temperature and pressure and has no secondary pollution [77].

A known phenomenon that also affects the quality of water is lake turnover. Lake turnover has also been linked to water quality issues. Failure for lakes to turnover causes a condition known as anoxic hypolimnion, which is not suitable for the life of aquatic organisms. Water quality monitoring is therefore key to determining if there are any changes in lake mixing frequency because disturbance in the mixing capabilities of the lake reduces its quality for human and aquatic life [78]. The turnover is expected to occur twice annually in dimictic temperate lakes that freeze in wintertime, usually in fall and spring. Oxygen distribution is one of the factors that affect a lake’s turnover. Oxygen is supplied to the lake’s bacteria from the phytoplankton, which undergoes photosynthesis in the upper water layers due to the presence of light on the water surface since sunlight reduces with increased water depth. In the winter, when there is little or no sunlight, bacteria consume oxygen in the sediments at the bottom of the lake, which releases nutrients from the decomposition of debris. Temperature changes and wind forces during the spring and fall are the driving forces for the water layers to mix. This phenomenon allows for the circulation of oxygen from the upper water layer known as the epilimnion and the bottom water layer called the hypolimnion. When the lake is unable to mix, an anoxic condition is created in the hypolimnion with fewer nutrients in the epilimnion impacting the productivity of phytoplankton [78].

Water bodies including lakes and rivers can also be impaired by pollutants emanating from highway stormwater runoff, industrial waste discharges, and agricultural activities, as well as municipal and residential discharges. With increasing urbanization and climate change, impairments to some water bodies continue to worsen, exposing more than half of the world’s water bodies to a myriad of anthropogenic activities that ultimately destroy ecological environments and impair public health [79,80,81]. Most of the stormwater discharges into freshwater bodies are largely untreated and about 80% of the wastewater discharged from municipalities and industries released worldwide ends up in freshwater bodies [66]. The environmental damage caused by these pollutants is signifcant and measures need to be taken to control the rate of pollution.

Studies have looked at effective ways of reducing the amount of WQPs. One such study is the use of chitosan and the effectiveness of chitosan as a bio-coagulant in shrimp washing wastewater on three WQPs, namely, color, turbidity, and TDS. Chitosan is an eco-friendly, non-toxic, reactive natural bio-coagulant, and renewable polysaccharide obtained from the chitin deacetylation used for water and wastewater purification. Chitosan solution effectiveness was measured by computing the percentage decrease in the three WQPs. Analysis showed about a 54% reduction in TDS as opposed to an 84% reduction in color and an 83% reduction in turbidity [82]. A similar study by [83] on the effectiveness of chitosan in the removal of turbidity and bacteria from turbid waters showed a removal efficiency of >74% for turbidity at a pH of 7.0–7.5. Additionally, chitosan significantly reduced the required dosage of the primary coagulant required by more than 50%. Chitosan was also found to have reduced bacteria at 2–4 log units (i.e., 99 to 99.99%) within the first 2 h of treatments. Efficiency in the removal of Escherichia coli (E. coli) was found to be better than for S. faecalis. Bacteria reduction of 2–4 long units, i.e., 99–99.99%, was obtained within the first 1 to 2 h of treatment. Overall results indicate that E. coli was removed better than Enterococcus faecalis (S. faecalis).

Monitoring of water quality is carried out through conventional methods such as field measurement and laboratory analysis or RS applications. The following paragraphs summarize the water quality monitoring efforts as performed by past researchers.

Researchers in [84] investigated the use of satellite imagery for water quality studies in the New York Harbor in the US. This study compared in situ data collected from routine sampling with images from Landsat 5 Thematic Mapper (TM) and MODIS sensors and found a strong correlation between Landsat 5 TM red of bandwidth 0.63–0.69 µm reflectance and turbidity (R² of 85% for a sample size of 21) and chlorophyll-a concentration (R² of 78% for a sample size of 16). Thus, turbidity increments due to runoff and spring tide resuspension in Newark Bay were identified and mapped based on the reflectance in the red band of Landsat 5 TM images. There was no correlation between the Terra MODIS images and the in situ chlorophyll-a concentration observation. The MODIS images were not directly useful for determining the concentration of chlorophyll-a in the study area.

Researchers in [85] performed studies on water quality monitoring in support of the EU WFD in the Gulf of Finland using RS techniques. The study aimed to extract the spatial distribution of water quality with a focus on nitrate concentration in support of the EU WFD. Results show that the value of nitrate concentration was below 25 mg/L, which is an indication of significantly less affected quality when compared to most of the countries in Europe. Nitrates are important species in water bodies for several years because they are associated with Phosphorus (P), which is a cause of eutrophication in water bodies [50]. Chlorophyll-a concentration is another indicator of algal biomass leading to eutrophication that may impact the ecological functions of water ecosystems [86].

Researchers in [35] assessed water turbidity in the Alqueva reservoir located in the south of Portugal using MERIS images. The study proposed a linear empirical algorithm of the green and blue spectral bands with a correlation coefficient (R) of 0.96 to estimate surface water turbidity.

Researchers in [87] performed a study using UV–visible RS reflectance to assess the spectral slopes of the absorption coefficient of colored dissolved and detrital material (CDM). In the said study, unique features of hyperspectral RS reflectance in the UV–visible wavelength were observed in the various water bodies in a wide range of coastal and open ocean environments. It was established that ocean color in the UV domain was sensitive to the variation in the CDM spectral slope. It was also established that adding UV wavelengths to the ocean color measurements will improve the retrieval of the CDM spectral slope. Further analysis of the results supports the concept for UV wavelengths to be included in the next generation of satellite ocean color sensors.

Researchers in [88] applied unpolarized Raman spectra for the determination of salinity and temperature from a natural water sample in Australia. The results demonstrate that Raman spectra of both temperature and salinity can be measured with RMSEs below 0.2 °C for temperature and 0.6 Practical Salinity Units (PSU) for salinity. The Raman spectra showed potential in determining both temperature and salinity at the same time using a numerical model based on multiple linear regressions (MLR). Results showed that both salinity and temperature have a similar influence on the spectra signatures. An increase in both parameters causes a corresponding decrease in the signal intensity.

Researchers in [66] assessed the impact of effluents on the quality of a river body and quantified the various water quality indicators such as physical, bacteriological, and chemical parameters. Knowing the concentration of these parameters aids in assessing the quality of water. While the bacteriological indicators include total coliform and E. coli, the physical indicator includes TDS, TSS, EC, pH, Total hardness (TH), DO, Biochemical oxygen demand (BOD), and chemical oxygen demand (COD), as well as the chemical parameters of various cations and anions.

Researchers in [89] developed a numerical model to estimate the turbidity for the Tennessee River and its tributaries in southeast Tennessee using Landsat 8 OLI imageries and near real-time in situ measurements. The study establishes a nonlinear regression-based numerical model using the red band (band 4) surface reflectance values of the Landsat 8 OLI sensors with bands of 0.64–0.67 µm wavelength in the estimation of turbidity in the study area. The turbidity estimation provided high accuracy with R² of 97% and RMSE of 1.41 Nephelometric Turbidity Unit (NTU). Images acquired on various dates were used to test the predictive performance of the model, which generated a promising R² of 0.94.RS methods have also been used to successfully estimate several other parameters, including suspended solids, Secchi disk, total phosphate, nitrogen, nitrate-nitrogen, dissolved inorganic nitrogen (DIN), and phytoplankton [17,21,24,45,81,90,91].

2.3. Remote Sensing

This subsection describes remote sensing systems, the principles used, and their applications in water quality monitoring.

2.3.1. Remote Sensing Systems

RS systems have been used to acquire images with solar reflectance spectral bands which are utilized for several studies including water quality, snow mapping, and plant water content based on light scattering properties of the remote sensors. RS systems used are categorized as microwave and optical systems and further divided into active and passive sensors described in the subsequent paragraphs [92,93,94,95,96].

Microwave and optical RS technologies are two technologies used in various studies including water quality monitoring. Optical RS sensors collect data in the visible, near-infrared, and shortwave infrared regions of the electromagnetic spectrum, but microwave sensors use a longer wavelength (cm to m), which makes them able to penetrate through cloud cover, dust, haze, and all kinds of rainfall except the heaviest rains. The longer wavelengths of microwave sensors make them unsusceptible to atmospheric scattering, which impacts shorter optical wavelengths and makes it possible for them to detect microwave energy under all weather and environmental conditions at any time [97,98,99].

The following subsection describes microwave and optical remote-sensing technologies.

Microwave Remote Sensing

Microwaves are the portion of the electromagnetic spectrum with a wavelength between 1000 µm and 1 m that fall within regions of electromagnetic radiation that are almost unimpeded by the atmosphere. Microwave RS is categorized as active, also known as a non-optical sensor, and passive remote, also called an optical sensor. Optical sensors depend on the energy of the sun, unlike non-optical sensors, which produce their energy. Although most studies have measured WQPs using optical RS, there is an opportunity to measure these parameters from the microwave region of the electromagnetic spectrum [92,97,98,100]. Passive microwave sensors detect the emitted energy within their field of view. Active microwave sensors are those sensors that provide their source of radiation to illuminate their target. They are grouped as imaging and non-imaging with a common form of imaging active microwave sensors being RADAR; non-imaging active microwave sensors include scatterometers and altimeters [98].

Active microwave sensors send pulses of electromagnetic radiation, particularly microwave radiation. The sensor then records the energy that is scattered back or reflected. Properties of the reflected energy received at the antenna of the sensor depend on its distance from the antenna, target properties, and the wavelength of the signal received. Some examples of active sensors include Radio Detection and Ranging (for example Synthetic Aperture Radar (SAR)) and Light Detection and Ranging (LIDAR). These sensors are used in managing disasters, precision, agriculture, and exploring crime [99,101,102,103]. The SAR is a well-developed non-optical microwave sensor that operates by recording the amount of energy reflected after interacting with the Earth. Microwave sensors have been used to effectively monitor and solve environmental issues in different applications. They have been used in the study of crops, forest cover, snow ice, soil moisture, hydrological processes such as floods and rainfall measurements, water surface pollution, the productivity of natural and agricultural ecosystems, and surface temperature detection of anomalies among others [92,93,98,100]. An example of a rainfall sensor is the Tropical Rainfall Measuring Mission (TRMM). These sensors have moderate spatial resolution compared to visible and infrared satellite sensors and present an effective means of measuring rainfall, unlike conventional methods. Conventional methods of measuring rainfall such as a network of rain gauges suffer major limitations due to inappropriate spatial coverage for the capturing of spatial variations in the rainfall [100,101]. The energy recorded by the passive microwave RS can be emitted, reflected, or transmitted from the surface. Passive microwave sensors are therefore mostly categorized by low spatial resolution [98].

Passive microwave radiometers and active SAR are commonly used for the RS of soil moisture and snow water equivalent (SWE). Studies have explored the usefulness of microwave signals for the determination of the surface soil layer. Microwave bands of wavelength ranging from 0.3 to 30 cm are noted to be effective in soil moisture measurement. The microwave sensor however has a major drawback in soil estimation, which is the poor surface penetration of its signals through forest canopies and vegetation covering the soil surface [101,104]. There are known satellite microwaves including the Soil Moisture and Ocean Salinity (SMOS) by the ESA and Soil Moisture Active Passive by the National Aeronautics and Space Administration (NASA) that have been explored in studies [104].

Optical Remote Sensing

Optical RS sensors are mainly passive sensors that make use of the sun’s energy. The USGS Landsat, NASA’S MODIS, and the ESA’s Sentinel-2 among many others are some optical remote sensors that have been used by several scientists including water researchers for monitoring and managing water resources [99]. This research focuses on passive RS sensors. Passive sensors can be either airborne or spaceborne sensors based on the platforms launched. Images from these sensors can be multispectral or hyperspectral based on the spectral and spatial resolutions as described in the subsequent subsection.

Spectral Imaging

Spectral imaging is the acquisition of several digital images at a different and well-defined number of optical wavelengths [105]. The passive optical remote sensors are categorized as multispectral or hyperspectral images based on the spectral resolution of the sensors used [103].

Multi and hyperspectral RS images have been used in several studies including water and land observation monitoring. Hyperspectral and multispectral satellite missions offer the opportunity for the detection of species-specific spectral features. This functionality of the sensors was postulated as far back as the 1990s and is a reality today [106]. Multispectral systems collect data in 3–10 spectral bands in a single observation from the visible and the near-infrared range of the electromagnetic spectrum. The spectral bands of multispectral bands range from 0.4 to 0.7 µm for red-green-blue, and infrared wavelengths within the range of 0.7–10 µm, or more for near, middle, and far infrared [5,8]. The use of multispectral images is, however, restrictive because the spectral resolution of the images influences the quality and quantity of the information they can provide [107].

Several multispectral sensors have been successfully used in the estimation of several WQPs. These include Landsat sensors, which have been used for the estimation of TSM, COD, and TP [5,8]. There have, however, been advances in sensor development through the development of hyperspectral sensor technologies to overcome the limitations of multispectral sensor systems over the past two decades. Hyperspectral systems can collect and store several hundred spectral bands in a single acquisition producing much more detailed spectral data and high spectral resolutions in the third dimension. They have been widely applied in medical imaging, processing, food safety testing, and agricultural yield [103,108,109]. Hyperspectral RS applications offer an effective mechanism for frequent, synoptic water quality monitoring over a large spatial extent [110]. They have been used in several studies including studies to distinguish between cyanobacterial and algal blooms in inland waters. Several hyperspectral sensors have been developed and have been applied in several studies. Some of the hyperspectral satellite missions include NASA’s Plankton, Aerosol, Cloud, and ocean Ecosystem (PACE), Germany’s Environmental Mapping and Analysis Program (Nmap), ESA’s Fluorescence Explorer (FLEX), Italy’s PRecursore IperSpettrale della Missione Applicativa (PRISMA), and India’s Hyperspectral Imaging Satellite (HySIS) [106]. Other miniature hyperspectral sensors used for image acquisitions and monitoring missions include Norway’s HySpex VNIR, US’ Micro- and Nano-Hyperspec, and Germany’s Fire- flEY, which can be installed on manned or unmanned airborne platforms such as Unmanned Aerial Vehicle (UAVs), airplanes, or helicopters [103]. There are different hyperspectral cameras for image acquisition. These include pushbroom, whiskbroom, and snapshot cameras. The principle used by each hyperspectral sensor is a function of its ability to obtain the whole picture referred to as a snapshot at one time, one point of the picture known as whiskbroom, or one line of the picture called pushbroom [103].

Hyperspectral sensors collect 200 or more bands, enabling the construction of a continuous reflectance spectrum for all the pixels in the scene. Hyperspectral RS has been applied in a variety of water resource studies including flood detection, water quality monitoring, wetland mapping, estimation of bathymetry, land-use and vegetation classification, and evapotranspiration among others [108,111]. Despite the emergence of hyperspectral images, multispectral sensors have been the preferred sensors for the mapping and monitoring of WQPs including salinity and TSS due to the low cost of image acquisition [107].

Airborne and Spaceborne Sensors

Sensors used in RS applications are grouped into two main characterizations based on the platforms in which they are installed. These categories are space and airborne sensors. Spaceborne sensors are those carried by satellites or spacecraft to areas outside the Earth’s atmosphere. Examples of spaceborne sensors include Landsat satellite, Advanced Spaceborne Thermal Emission and Reflection Radiation (ASTER), Moderate Resolution Imaging Spectroradiometer (MODIS) Sensor, GeoEye, and IKONOS among others. Airborne sensors are mounted on platforms flown within the Earth’s atmosphere. These platforms include boats, helicopters, aircraft, or balloons. Both airborne and spaceborne sensors have been used in past studies to examine a wide range of WQPs since the early 1970s [112]. Airborne sensors are hyperspectral passive sensors that measure the reflectance of solar radiation throughout the visible, near-infrared, and mid-infrared, and thermal regions of the electromagnetic spectrum. Images acquired are numerous but narrower. Spaceborne are also passive sensors that cover a broader geographic area compared to the coverage by airborne sensors. Spaceborne sensors make use of the constant orbital patterns of the satellite they are launched on. Images acquired by spaceborne sensors can either be multispectral or hyperspectral. The choice of a particular sensor to be used in a study depends on spectral properties and associated strengths and limitations [8].

The strengths and limitations of airborne and spaceborne sensors used in water quality monitoring are outlined [8].

The strengths of airborne RS sensors include:

• Spectral and spatial resolutions: airborne sensors have higher spectral and spatial resolutions making them appropriate for the measurement of WQPs. Hyperspectral airborne sensors produce images that allow for more detailed spatial and spectral models for the accurate classification of an image.
• Flexibility in terms of configuration: airborne sensors have higher flexibility in terms of their configuration in terms of spectral range, number bands, and wavelength. The time of survey for any project work is not fixed. There is no repeat cycle as in the case of spaceborne sensors.
• Larger geographic areas: airborne sensors can operate at higher altitudes covering a larger geographic area, and hence are suitable for regional water quality monitoring.
• Monitoring of small water bodies: airborne sensors can assess small water bodies including rivers, estuaries, ponds, and tributaries.

Airborne sensors however present the following limitations:

• Greater planning: there is a need for greater planning ahead before the airborne survey. Considerations need to be made for external factors including air traffic, weather, flight line orientations, and solar radiation.
• Cost: airborne surveys are more cost-intensive compared with spaceborne sensors. Sensitive detectors, large data storage capacity, and fast computers are needed for the image processing of airborne sensors.
• Altitude: airborne sensors usually cover smaller areas as compared to spaceborne sensors due to the lower altitude of image acquisition.

The strengths of spaceborne RS sensors include:

• Cost: spaceborne sensors produce images at typically zero to little cost and can be used for large-scale monitoring of water quality.
• Less complexity with image processing: compared to airborne sensors, processing the spaceborne image is less complex.
• Multi-temporal studies: spaceborne sensors have a revisit frequency ranging from daily to monthly making them useful for multi-temporal scale monitoring of the trends and variations in WQPs.
• Coverage of large geographic areas: spaceborne sensors can cover large geographic areas making them suitable for moderate, regional, and global water quality monitoring.

Limitations in the use of spaceborne sensors include:

• Cloud constraints: spaceborne sensors are faced with enormous limitations when a project requires a cloud-free image.
• Overestimation of water parameters: empirical and semi-empirical approaches in analyzing multispectral images by spaceborne sensors may lead to overestimations at areas where there is a contribution of reflectance from the bottom to the water leaving reflectance.
• High-resolution images acquired are costly. The limitation with coverage: there is a limitation with coverage of the electromagnetic spectrum by spaceborne sensors. Some bands such as middle infrared and thermal bands may not be covered, which may impact the accuracy of the estimation of WQPs.

Airborne sensors provide either hyperspectral or multispectral images depending on the sensor used in the image acquisition. Some of the airborne sensors used in RS applications are summarized as follows [8,16].

• Airborne Visible Infrared Imaging Spectrometer (AVIRIS): This sensor is manufactured by NASA Jet Propulsion Lab with a hyperspectral image with 224 bands. This image has a 17 m resolution with a spectral range of 0.40–2.50 µm. The sensor utilizes the whiskbroom scan system.
• Hyperspectral Digital Imagery Collection Experiment (HYDICE). This sensor is manufactured by the Naval Research Lab with a pushbroom scan system. This is also a hyperspectral image with 210 bands with a spectral range of 0.40–2.50 µm and a spatial resolution of 0.8 to 4 m.
• Airborne Prism Experiment (APEX). This sensor is manufactured by VITO and produces up to 300 hyperspectral bands with a pushbroom scan system. The spatial resolution of images produced ranges from 2 to 5 m and the spectral range of 0.38–2.50 µm.
• Compact Airborne Spectrographic Imager (CASI-1500). This sensor is manufactured by ITRES Research Limited and produces up to 228 hyperspectral bands of spectral range 0.40–1.00 µm and a spatial resolution of 0.5–3 m with a pushbroom scan system.
• Multispectral Infrared and Visible Imaging Spectrometer (MIVIS): This sensor is manufactured by Daedalus Enterprise Inc., USA, and produces multispectral images of visible, near-infrared, mid-infrared, and thermal bands. The sensor operates on the whiskbroom scan systems with resolutions ranging from 3 to 8 m depending on the altitude.
• Airborne Imaging Spectrometer (AISA). This sensor is manufactured by Spectral Imaging and produces hyperspectral bands of up to 288 bands with a spatial resolution of 0.43 to 0.9 µm. Images have a spatial resolution of 1 m. The sensor operates on the pushbroom scan system.
• Digital Airborne Imaging Spectrometer (DAIS 7915). This sensor is manufactured by GER Corporation and produces hyperspectral bands with a spatial resolution of 0.43 to 12.30 µm. Images have a spatial resolution of 3–20 m depending on the altitude. The sensor operates on the whiskbroom scan system.

A variety of spaceborne or satellite sensors are used in RS applications. Images from sensors on US-government-funded satellite platforms are generally available at no cost to the user. Each sensor has different spatial, spectral, radiometric, and temporal resolutions. The spatial resolution defines the surface area, and it is represented by the pixel or cells. The spectral resolution measures the sensor’s ability to resolve features in the electromagnetic spectrum. The radiometric resolution measures the sensor’s potential to record varying brightness levels, while the temporal resolution measures the amount of time, usually in days, it takes satellite sensors to revisit a particular point on the surface of the earth [113].

The following paragraphs highlight some widely used remote sensors used in water quality monitoring.

Some of the widely used sensors are summarized below.

• Landsat satellite images: Landsat programs are joint efforts of the USGS and NASA for Earth Observation and have been in existence since 1972. Landsat images have been used by stakeholders in many applications such as land use planning, natural resources management, public safety, climate research, natural disaster management, home security, and agriculture, among others [114]. The Landsat 8 Operational Land Imager (OLI), Landsat 7 Enhanced Thematic Mapper (ETM+), and Landsat 5 TM have all been used in water quality monitoring efforts. The Landsat 9 sensor has a 14-bit quantization, which can differentiate 16,384 shades, i.e., a brightness uncertainty of ±0.006%) at a given wavelength. The Landsat 8 OLI sensor captures data over a 12-bit instrument with improved precision in radiometry. The sensor captures images with the overall improvement in the signal-to-noise ratio, which translates to 4096 grey levels (i.e., a brightness uncertainty of ±0.024%). Landsat 1 to 7 sensors capture data with 256 grey levels (i.e., a brightness uncertainty of ±0.4%) over an 8-bit dynamic range. The Landsat 8 OLI data with 12-bit are scaled to 16-bit and made available in the form of level-1 data products. These are scaled to 55,000 grey levels from the 65,536 grey levels, which can be subsequently rescaled with radiometric coefficients, which come with the product metadata file (MTL file). The rescaling is performed to the Top of the Atmosphere (TOA) reflectance and or radiance [115,116].
  • Landsat 9 OLI/TIRS: The Landsat 9 sensor is the latest series of Earth-observing satellites launched on 27 September 2021, and its data are publicly available (Yang et al., 2022). The sensor was launched from Vandenberg Air Force Base, California, USA, onboard a United Launched Alliance Atlas V 401 rocket and it is an improved replica of the Landsat 8 sensor. It carries the Operational Land Imager 2 (OLI-2), built by Ball Aerospace & Technologies Cooperation, Boulder, CO, USA, and the Thermal Infrared Sensor 2 (TIRS-2), built at the NASA Goddard Space Flight Center, Greenbelt, MA, USA. The Landsat 9 OLI-2 provides images consistent with Landsat 8 spectral, spatial, geometric, and radiometric qualities. It has nine spectral bands over a 185 km swath of 30 m resolutions for all bands except for the panchromatic band, which is 15 m, at a maximum ground sampling distance (GSD) both in and cross tracks. Landsat 9 offers a 16-day revisit earth coverage and an 8-day offset with Landsat 8. It acquires more than 700 scenes per day [117]. The sensor was launched to continue with the collection, distribution, and archival of multispectral imagery offering users the synoptic, global, and repetitive coverage of the Earth’s surface for the detection of natural and human-induced changes on a spatiotemporal scale [117].
  • Landsat 8 OLI/TIRS: The Landsat 8 carries the OLI and Thermal Infrared Sensor (TIRS) sensors. It was previously called the Landsat Data Continuity Mission (LDCM) and was launched on an Atlas-V rocket from Vandenberg Air Force Base, California, USA, on 11 February 2013. Landsat 8 OLI has nine spectral bands. Studies have used some of these bands and their combinations for water quality estimations. All the bands have a spatial resolution of 30 m, except band 8 with a finer spatial resolution of 15 m. A 30 m resolution means each pixel of the image provides an average reflectance value recorded of an area of 900 m² (30 m by 30 m) [89]. Landsat 8 OLI is available by the United States Geological Survey (USGS) on 16 days of repeat time with an equatorial crossing time of 10:00 am ± 15 min mean local time [116,118]. The Landsat 8 OLI sensor acquires around 740 scenes in a day on the Worldwide Reference System-2 (WRS-2) path/row system. It has a swath overlap that varies from 7% at the equator to about 85% at extreme latitudes. Each scene size is 114 mi × 112 mi [116].
  • Landsat 7 Enhanced Thematic Mapper (ETM+): The Landsat 7 was launched from Vandenberg Air Force Base, California, USA, on 15 April 1999, with a repeat cycle of 16 days. The Landsat 7 carries an improved version of the Thematic Mapper I instrument, which was onboard Landsat 4 and 5. ETM+ sensor Landsat 7 ETM+ data collected since 2003 have gaps caused by the Scan Line Corrector (SLC) failure. Each scene of this image is 185 km wide. Landsat 7 images are delivered in 8 bits with 256 grey levels [119]. Landsat 7 ETM+ is also available by the USGS on 16 days repeat time with an equatorial crossing time of 10:00 am ± 15 mean local time [120,121]. All the bands have a 30 m resolution, except band 6 and band 8 with 60 m and 15 m resolutions, respectively [119].
  • Landsat 5 TM: These data sets consist of seven bands with 30 m spatial resolutions for all bands except band 6, which has 120 m resolutions and, in turn, is resampled to 30 m. This sensor was launched on March 1, 1984, from Vandenberg Air Force Base in California, USA, with its data products quantized to eight bits. It carries both the Multispectral Scanner (MSS) and the TM instruments. The sensor delivered data for close to 29 years and was retired on 5 June 2013 [121,122,123]. Landsat 5 TM images are delivered in 8 bits with 256 grey levels [124].
• RapidEye images: The RapidEye sensor was launched on 29 August 2008, with a 6.5 m spatial resolution. This sensor produces images used in a variety of applications, including agriculture, engineering, construction, mining, and cartography. The RapidEye sensor has five (5) spectral bands and a spacecraft lifetime of 7 years. It has a 5.5-day revisit time off-nadir and at nadir, respectively, and an approximate equator crossing time of 11:00 am local time. It has a swath width of 77 km, a camera dynamic range of 12 bits, and an image capture capacity of 5 million km²/day. The RapidEye constellation was retired on 31 March 2020.
• Advanced Spaceborne Thermal Emission and Reflection Radiation (ASTER): The National Aeronautics and Space Administration (NASA) launched the ASTER sensor on the Terra satellite in December 1999. It has an equatorial crossing time at 10:30 a.m. local time with 16 days of repeat time. It has a resolution of 15, 30, and 90 m depending on the bands with quantization levels of 8 bits for the 15 and 30 m resolution bands and 12 bits for the 90 m resolution. ASTER is a collaboration between NASA, Japan’s Economy Ministry, Trade and Industry, and Japan Space System. ASTER data are used in the mapping of land surface temperature, elevation, and reflectance Additionally, they are used in applied geology, soils, hydrology, ecosystems dynamics, and land cover change studies [125]. ASTER is made of different subsystems, namely, the Visible and Near-infrared (VNIR), SWIR, and Thermal Infrared (TIR). Each of these subsystems collects data in a separate set of wavebands with each set having its spatial resolution. The sensor has fourteen different bands with varying spatial resolutions for each subsystem.
• Moderate Resolution Imaging Spectroradiometer (MODIS): The first MODIS instrument was launched aboard NASA Terra in December 1999, with the second instrument launched in May 2002 aboard the NASA Aqua platform. These data have been available since February 2000 and June 2002, respectively, for the Terra and Aqua platforms. The Terra and Aqua satellite platforms have local equatorial crossing times at 10:30 a.m. and 1:30 p.m., respectively [126,127]. The MODIS sensor has a repeat cycle of 16 days and 12-bit quantization. MODIS data consist of 36 spectral bands with varying wavelengths and spatial resolutions. The MODIS sensor has 250 m spatial resolution for bands 1 and 2, 500 m for bands 3–7, and 1 km for bands 8–36. The wavelength of bands 1 to 19 ranges from 405 to 2155 nm, with bands 20–36 ranging from 3.66 to 14.28 µm. MODIS images are utilized in studying and understanding environmental phenomena, and dynamic variations in inland, ocean, and the lower atmosphere [127]. Although the MODIS data with medium resolution (i.e., 250 and 500 m resolutions) were originally designed for land studies and cloud monitoring and have lower sensitivities than MODIS ocean bands. The moderate resolution MODIS was compared with Landsat 7 ETM+ and Coastal Zone Color Scanner (CZCS), and SeaWiFS in a study and were found to provide sufficient sensitivity for water application data. The MODIS Ocean band (i.e., 1 km resolution) was found to be 3 to 6 times more sensitive than the SeaWiFS bands, which makes it possible for them to detect subtle ocean features. Compared to Landsat 7 ETM+ bands, the moderate resolution bands of MODIS (250 and 500 m) are found to be more sensitive with nearly twice the sensitivity of CZCS blue-green bands. MODIS moderate resolution bands are hence expected to be as useful as CZCS for coastal ocean studies [128].
• Sentinel-2 MSI: Several group efforts led by the Airbus Defense resulted in the design and building of the Sentinel-2 sensor. The Sentinel-2 satellites are a constellation of the European Space Agency (ESA) and were launched on multispectral scanners [129]. The Sentinel-2 sensor comprises two identical satellites, namely, Sentinel-2A and Sentinel-2B, in the same sun-synchronous orbit. These satellites are separated at an angle of 180° from each other with a mean orbital altitude of 786 km. The crossing time of the sentinel sensor at the descending node is 10:30 am Mean Local Solar Time (MLST). The sensors were launched on a Vega rocket from Kourou in French Guiana on 23 June 2015, and 7 March 2017, respectively, for the Sentinel 2A and 2B [130,131]. Sentinel-2 images are used in the study of water quality, impervious surface mapping, monitoring of land ecosystems and land-use land cover (LULC), forest management, agriculture, and disaster mapping [129,131,132,133]. The sensor has 10 days of revisit time with one satellite and 5 days of revisit time with the two satellites, at the equator under cloud-free conditions [134]. The sensor has a swath width of 290 km. The Sentinel-2 has a 12-bit radiometric resolution and spatial resolution ranging from 10 m for bands 2, 3, 4, and 8; 20 m for bands 5, 6, 7, 8a, 11, and 12; to 60 m for bands 1, 9, and 10, with wavelengths varying from 442.7 to 2202.4 nm for band 1 to 11, respectively [113].

Some of the sensors used in the estimation of WQPs are summarized in

Table 1. Optical Satellite Sensors used in WQP Estimations and their specifications.

Satellite Sensor	Year Launched	Spatial Resolution (m)	Temporal Resolution (Days)	Spectral Resolution
				Number of Bands	Wavelength (µm)
Landsat-1,2,3 MSS	1972	80	18	4	0.50–1.10
Landsat-4,5 TM	1982	30–120	16	7	0.50–2.35
Landsat-7 ETM+	1993	15–60	16	8	0.45–0.90
Landsat-8 OLI/TIRS	2013	15–100	16	11	0.43–12.51
Landsat-9 OLI/TIRS	2021	15–100	16	11	0.43–12.51
NOAA-20 VIIRS	2011	375–750	16	22	0.412–12.01
Sentinel-2 A/B MSI	2015	10–60	5	12	0.44–2.19
RapidEye	2008	5	5.5	5	0.44–0.85
SPOT-5 HRG	2002	2.5–20	2–3	5	0.48–1.75
GOCI	2010	500	1	8	0.41–0.87
ALOS AVNIR-2	2006	2.5–10	2	5	0.42–0.89
ERS-2 ATSR-2	1995	1000	3–6	7	0.56–12.00
GeoEye IKONOS	1999	3.2–0.82	~3	5	0.45–0.93
GeoEye Geoeye-1	2010	0.41–1.65	< 3	5	0.45–0.92
GOES Imager	1975	1000–4000	-	16	0.45–13.60
Japan Earth Resources Satellite Optical Sensor	1992	18.3–24.2	3	8	0.520–2.40
MERIS	2002	300–1200	1	15	0.39–1.04
Terra ASTER	1999	15–90	16	14	0.52–11.65
Terra MODIS	1999	250–1000	1–2	36	0.41–2.16
VIIRS	2011	375–750	0.5	22	0.50–12.01
HICO	2009	100	10	128	0.35–1.08
EO-1 Hyperion	2000	30	16	242	0.35–2.57
NOAA-16 AVHRR	2000	1100–4000	9	6	0.65–1.23
NOAA Polar Orbiting Environmental Satellites AVHRR	2009, 2012	1100–4000	1	6	0.580–12.5
EO-1 ALI	2000	10–30	16	10	0.43–2.35
Digital Globe WorldView-1	2007	0.5	1.7	1	-
Digital Globe WorldView-2	2009	1.85–0.46	1.1	9	0.40–0.92
NIMBUS-7 CZCS	1978	825	6	6	0.433–12.50
ENVISAT MERIS	2002	300–1200	35	15	0.390–1.04
Ziyuan 3 Multispectral Camera	2012	5.8	4–5	4	0.450–0.89
QuickBird Multispectral and Panchromatic Sensors	2001	10–30	16	-	0.450–0.90
NOAA WorldView-3	2014	0.31–0.31	1–4.5	17	0.433–12.50
CARTOSAT	2005	2.5	5	1	0.50–0.85
Digital Globe Quick bird	2001	2.62–0.65	2.5	5	0.65–1.23
OrbView-2 SeaWiFS	1997	1130	16	8	0.41–0.87

for optical sensors and 2 for microwave sensors. Table 1 and

Table 2. Microwave Satellite Sensors used in WQP Estimations and their specifications.

Satellite Sensor	Year Launched/ Deployed	Spatial Resolution (km)	Temporal Resolution (Days)	Spectral Resolution
				Number of Bands	Wavelength (m)
SMOS MIRAS	2009	3.5–50	3	1	0.212
SAC-D Aquarius	2011	100	7	1	0.212
GCOM-W1 Advanced Microwave Scanning Radiometer 2 (AMSR-2)	2012	5–10	2	7	0.003–0.043
Geophysical Satellite Radar Altimeter (GRA)	1985	5	17	1	0.022
European Remote Sensing Satellite (ERS-1/2) Radar Altimeter	1991	16–20	35		0.022
ENVISAT Radar Altimeter	2002	20	30		0.022–0.094
ERS-2 SAR	1995	≤0.03	35	1	0.057
Cryosat-2 SAR/Interferometric Radar Altimeter-2 (SIRAL-2)	2010	15	369	1	0.022
Sentinel 3 A/B SAR Radar Altimeter (SRAL)	2016	(0.30 × 1.64)–(1.64 × 1.64)	27	2	0.022–0.055
Japan Earth Resources Satellite L-band SAR	1992	0.018	44	1	0.235
SAtélite Argentino de Observación COn Microondas (SAOCOM-1A) L band SAR	2018	0.01–0.1	16	1	0.235
MetOp-A/B Advanced SCATterometer (ASCAT)	2007	25	1–2	1	0.057
RADARSAT-2 SAR C-band	2007	0.003–0.1	24	1	0.055
ERS-1/2 SAR	1991	50	2–7	1	0.057
QuickSCAT Ku-band scatterometer	1999	25	1–2	1	0.022
Nimbus-5, 6 ESMR	1972	2.5	1	1	0.016
Priroda-MIR IKAR	1996	1.5–7.5	1.7	5	0.003–0.040
TRMM TMI	1997	4.4	0.5	5	0.003–0.028
Aqua AMSR-E	2002	5–10	16	6	0.003–0.043
DMSP-F16 SSMIS	2003	0.013–0.069	1	4	0.004–0.015
GPM Core GMI	2014	3	2	2	0.002–0.03
Coriolis WindSat	2003	25	8	5	0.008–0.044
(Airborne) Electronically Scanning Thinned-Array (ESTAR)	1990	100	-	-	0.212
(Airborne) Scanning Low-Frequency Microwave Radiometer (SLFMR)	1999	0.5–1	-	-	0.212
Airborne Salinity, Temperature, and Roughness Remote Scanner (STARRS)	2001	1	-	-	Up to 0.212
SEASAT SMMR	1978	22–100	-	-	0.008–0.045
(Airborne) Passive Active L- and S-band Sensor	1999	0.350–1	-	-	0.212
(Airborne) Two-Dimensional Electronically Scanning Thinned-Array Radiometer	2003	0.800	-	-	0.212

present their year of launch, and the spatial, temporal, and spectral resolution to aid researchers in their decision on what sensor to use in their studies [5,6,8,16,40,43,135,136,137]. Microwave sensors have been used in the measurement of WQPs such as sea surface temperature (SST) and sea surface salinity (SSS). These parameters are important in determining the density of seawater, which is a critical indicator driving the currents of the ocean. Ocean circulation is a critical phenomenon to analyze global water balance, evaporation rates, and productivity forecast models [16].

2.3.2. Characterization of Water Quality Parameters

WQPs are grouped as optically active and non-optically active parameters for RS studies. Optical active water parameters are those parameters that are likely to impact the optical characteristics measured by RS sensors while non-optically active parameters are those parameters that are less likely to influence the optical characteristics measured by the RS sensors [138]. Examples of optically active parameters include suspended particulate matter (SPM) or sediment concentrations, CDOM, turbidity, and chlorophyll-a concentrations. Non-optically active parameters on the other hand include TP, TN, COD, DO, and TDS [138,139]. Studies have also proposed two classifications of water bodies based on reflectance ratios at 443 and 550 nm namely Case I and II waters [47]. Case I water is water with reflectance ratios at 443 and 550 nm greater than one, while Case II waters is water with reflectance ratios of at 443 and 550 nm less than one. The optical properties of Case I waters are mainly influenced by phytoplankton and their covarying compounds including CDOM and detritus. Case II waters, on the other hand, are optically deep and complex and have their non-optically active parameters determined by constituents such as mineral particles. The concentrations of these constituents do not covary with phytoplankton [47,140].

2.3.3. Remote Sensing Principles in Water Quality Monitoring

Water color and turbidity are two main optical parameters that affect the energy levels detected by RS applications. Water color increases the absorption of light in the water thereby causing a reduction in the remotely sensed signal [141]. Changes in water color can be detected by a camera or multispectral scanner. The presumption that the appearance of water is quantitatively related to water quality may not be justified, as many variables have the potential to affect the energy levels measured by satellite instruments. The variables include the time of year, sun elevation angle, aerosol and trace gas concentrations in the atmosphere, water-vapor content of the atmosphere, specular reflection of the skylight from the water surface, the roughness of the water surface, the presence of film, foam, debris, or floating plants on the water surface, water color, water turbidity, multiple reflections and scattering of solar energy in water, submerged or emergent vegetation, depth of water, and reflectance of bottom sediments [141]. Turbidity is another optically active constituent that affects the light spectrum. The effect of turbidity results in the scattering and absorption of light, preventing them from being transmitted. Turbidity has the potential to induce changes in a reservoir’s vertical thermal structure. The suitability of water color and turbidity estimates using satellites can be assessed through the evaluation of the influence of hydrological and atmospheric variables. While turbidity results from colloidal or suspended substances, the color of the water is a function of water appearance resulting from dissolved substances [35,141]. The differences in water color and turbidity affect signals in the visible and very ultraviolet and infrared wavelengths. The energy flux reaching the sensor is decreased by an increase in water color as more of the sun’s energy is absorbed in the water. On the other hand, an increase in turbidity results in an increase in the energy flux reaching the sensor. This is because more solar energy is backscattered or reflected by the particles that produce turbidity [141].

The principle of RS used in the measurement of water color or turbidity is represented by Equation (1) as given by [141];

$$I_{o }=I_{SR }+I_{A }+I_{B }$$

(1)

where I_o is the solar energy that reaches the surface of the water, I_SR is the solar flux that is spectrally reflected at the water surface, I_A is the flux absorbed by the water, and I_B is the flux backscattered to the water surface and thereby available for remote detection. The I_SR is equal at all wavelengths, while the I_A and I_B produce distinctive spectral signatures depending on the water constituents. Scattering in clear water is caused by molecules and is strongly dependent on the wavelength of the light. Most of the solar energy in natural waters is in a forward direction and subsequently absorbed by the water [141]. An estimated 2% of the light flux is backscattered in clear, infinitely deep water [141]. Fifty percent of the signal for the blue range of the spectrum comes from a depth of fewer than 15 m, with most of the signals coming from the red light at a depth of less than close to 1.1 m [141]. The portion of the backscattered energy that returns to the water surface is the signal utilized for RS. To be detected by RS, pollutants must affect turbidity, temperature, color, or wave emissivity. Dissolved constituents that add no color to water do not affect the absorption and scattering of water as measured by a multispectral scanner or spectrophotometer [141]. Dissolved colored substances increase light absorption in water and cause a reduction in the remote signal but they do not affect the light scatter. The spectral distribution of backscattered energy is however affected by additional absorption. Suspended materials on the other hand increase the backscattering of light and hence may increase the remote signal [141]. The interaction of light with these suspended materials produces a signal reflectance that is detectable from a distance. These signals can therefore be used to estimate the presence and the quantity of TSS.TSS concentration can be determined using the spectral backscattering coeefieicnt of particles as a proxy. The backscatterings of in situ measurements have been found to show susbstantial variability based on particulate loadings. The spectral properties of the water are categorized as the Inherent Optical Properties (IOPs) and the Apparent Optical Properties (AOPs). The IOPs, which describe the spectral light absorption and backscattering, are key components in linking reflectance measurements to the concentration of water parameters such as TSS. The absorption coefficient of particles in optically complex waters such as inland, coastal, and estuarine is broken into components caused by phytoplankton and those caused by non-algal particles. While the IOPs can be quantified in the laboratory, the AOPs, which include radiance and downwelling plane irradiance, are only measured using the IOPs and depend on the IOPs and the geometric structure of the radiance distribution including the wind speed, surface water structure, and atmospheric conditions [137,142,143].

2.3.4. RS Retrieval of Water Quality Parameters

RS applications in water quality monitoring of optical active parameters involve the determination of WQPs by relating them empirically, semi-empirically, analytically, or through artificial intelligence to remotely sensed reflectance [5,18,144]. The techniques used in RS are a function of the changes in the spectral signature backscattered from water. The spectral signature is related by measuring changes in empirical or analytical models to WQPs. The optimal wavelength used in WQP measurement is a function of the constituent being measured and the sensor characteristics [145]. Researchers in [5,144] outlined the different approaches by which WQPs are extracted from remotely sensed spectral data. These include:

• Empirical method: this approach utilizes statistical relationships derived between measured RS spectral values and measured water quality. The relationship is established using regression techniques. Estimations performed by empirical models need in situ data as many parameters are likely to change between RS missions. The empirical methods are simple and easier approaches to the retrieval of water quality.
• Analytical method: the IOPs and AOPs are used to model the reflectance. Physical relationships are then derived between the WQP, the underwater light field, and the remotely sensed radiance. This method involves the use of bio-optical and transmission models to simulate the propagation of light in water bodies and the relationship between the WQP and reflection. Parameters such as TDS are, however, determined due to their association with other colored WQPs. Salinity is only determined in microwave bands.
• The semi-empirical methods: this approach is the combination of the empirical and analytical methods for the retrieval of WQPs. In this method, the spectral characteristics of the parameters are known. Here, the appropriate combination of wavebands is used as correlates. The spectral radiance is recalculated to above the surface irradiance reflectance, and then through regression techniques related to the WQP.
• Artificial Intelligence (AI) methods: this is an implicit algorithm approach that differs from the three other approaches outlined i–iii. The complications from various water surfaces, WQP combinations, and sediment deposits will mean the need to use implicit algorithms for the retrieval of WQPs. AI applications capture both linear and nonlinear relationships compared with conventional statistical approaches. Studies have applied various AI applications including the neural network (NN), which is non-linear, as compared to a linear MLR model and SVM in water quality retrieval, and produced satisfactory results. ML models such as the ANN outperforms regression models such as MLR models [48,146].

2.3.5. Detection of Optically Active Parameters

Optically active parameters such as CDOM and TSS that are important for the structure and function of the ecosystem of water bodies, absorb light in the ultraviolet and visible wavelength range and influence the optical properties enabling it to be sensed from satellite observations. CDOM plays an important role in regulating light attenuation in water bodies. CDOM concentrations are conventionally estimated using the absorption coefficient. The equation for estimating absorption coefficient is presented in Equation (2) below [73,74,147];

$$a_{CDOM,m }\left(\mathsf{\lambda }\right)=2.303\times \mathrm{A} \left(\mathsf{\lambda }\right)/\mathit{l}$$

(2)

where the value 2.303 is the conversion between the base 10 and the natural logarithm, the subscript ‘m’ represents measured value, aCDOM,m(λ) and A(λ) represents the measured absorption coefficient (m⁻¹) and absorbance wavelength λ, and 𝓁 is the path length in m [147]. This equation can be narrowed to specific wavelengths of light in the near-UV and visible spectral regions (shown in Equation (3)) as the absorption coefficient of CDOM generally decreases nearly exponentially with increasing wavelengths in those regions of the electromagnetic spectrum [147].

$$a_{CDOM }\left(\mathsf{\lambda }\right)=a_{CDOM,m }\left({\mathsf{\lambda }}_0\right) \exp [-\mathrm{S}\left(\mathsf{\lambda }-{\mathsf{\lambda }}_0\right]$$

(3)

where aCDOM(λ) and aCDOM(λ₀) represents the absorption coefficients at the observation wavelength λ and reference wavelength λ₀, respectively. S represents the spectral slope measured in (nm⁻¹). The aCDOM(λ₀) is utilized in the characterization of CDOM concentration in a specific waterbody [147].

Optical water parameters are also used to monitor several other processes aside from water quality. The high amount of CDOM and sediment loads has the potential to reduce water clarity and block radiation needed for submerged aquatic vegetation growth. Sediments, CDOM, and nutrient inputs affect the optical characteristics of estuarine waters making a color in both in situ and remote sensed a good indicator for monitoring phenomena including run-off processes [148].

The ratio between the green and red bands of satellite images is useful in the extraction of the CDOM because the reflectance is low or almost zero in the blue part of the electromagnetic spectrum due to the high absorbance by CDOM with or without interference from SPM or phytoplankton [74].

Researchers in [149] demonstrated the capability of using CDOM algorithm retrieval to monitor DOC fluxes in the Yenisei River, which is the largest Arctic River in terms of annual runoff of 630 km³ and basin area of 2.58 million km². The researchers used Landsat 8 OLI and SPOT5 images in their study. It was established that the use of shorter/longer wavelength combinations can retrieve CDOM in the arctic river. The mean concentration of DOC measurement was 8.30 mg/L while the absorption coefficient of CDOM was 4.9 m⁻¹. The concentration of TSS was however found to vary from 2.63 to 19.90 mg/L. TSS was found to be highly correlated with DOC measurement with an R² of 74%. The best-performing model was found to be based on the green band and green-red band interaction. The model recorded an R² value of 76% and an RMSE of 1.21, which is an indication of high prediction. The authors recommend an extensive sampling protocol to limit the time gaps between the RS and in situ measurements. They also propose Sentinel-2-3/Landsat 8 synergies for the reproducibility of CDOM algorithm retrievals.

Studies have also found a good correlation between the SPM concentrations and water reflectance in the green and red regions of the electromagnetic spectrum for low to moderate turbid waters [150].

A study by [73] found that the 8-bit radiometric data of Landsat 7 were not adequate for the estimation of CDOM in lakes located in southern Finland when the absorption coefficient of CDOM at wavelength 420 nm was higher than 3 m⁻¹. However, the study found the Advanced Land Imager (ALI), which is a prototype of the next-generation Landsat, to be suitable for mapping CDOM in a wider range of concentrations. Chlorophyll-a concentrations serve as differential absorbers causing a decrement in the spectral response at the blue region (450–520 nm) of the electromagnetic spectrum with reflectance being greatest at the green and red wavelengths. Suspended solids are, however, found to be associated with an increase in reflected energy at longer wavelengths (630–690 nm) [7,121].

A study was performed by [4] to estimate optical water parameters using a guided approach. The optical WQPs considered included chlorophyll-a concentration, TSM, the absorption coefficient of CDOM, SDD from Ocean and Land Color Instrument (OLCI), and MSI data at Estonian and Finnish lakes and the Baltic Sea coastal area. The study used empirical algorithms. High correlations (R) ≥ 0.87, were established for in situ measured optical WQPs and the parameters predicted by the optical water-type guided approach. The presence of multiple constituents such as phytoplankton, CDOM, and non-algal particles (NAP) at different compositions in inland waters has however made the use of RS applications for the monitoring of inland water quality to be less successful as compared to open oceans [47]. RS techniques can monitor optically active parameters but have limited capacity to detect and monitor optically inactive parameters such as TN and TP [48].

2.3.6. Detection of Optically Inactive Parameters

Researchers in [48] utilized optically active parameters such as chlorophyll-a, Secchi disk depth (SDD), and TSS to determine optically inactive parameters such as TP and TN concentrations in the Geshlagh reservoir located in the west of Iran using Landsat 8 OLI images.The study established that band ratios (B3/B2) and bands 3 and 4 were the most effective for the determination of chlorophyll-a concentration and were subsequently used for the prediction of TP and TN concentrations in the reservoir. Previous research established a relationship between band combinations for TN and TP predictions based on their correlations with SDD, TSS, and chlorophyll-a concentrations [48].

Retrieving optically inactive parameters using satellite visible, NIR, or IR bands such as TN, TP, pH, and COD among others is possible because of their correlation with the optically active parameters [89,138]. Researchers in [138] found that bands 3, 4, and 5 of Sentinel-2 images to be the most influential bands for the detection of TN, TP, and COD. Estimation of optically non-active parameters including TDS assumes that there is a process that (usually) co-occurs with changes in the TDS or salinity that affects a WQP that is optically active. These processes may, however, not always co-occur in the same magnitude.

2.3.7. Remote Sensing Applications in Water Quality Monitoring

Water quality monitoring involves the systematic collection and analysis of water samples to provide knowledge and information about the water body to inform critical decisions [37]. Although RS has been used since the 1980s for the synoptic and multi-temporal studying of inland water quality [34,151], it was not until the early 21st century that CHRIS Compact High-Resolution Imaging Spectrometer (CHRIS), which is an experimental multi-angle sensor with a good spatial and spectral resolution, and MERIS, which has good spectral and temporal resolutions, started to acquire good resolutions. These images allow for the development of a reliable imagery acquisition system to be used as a water management tool [34]. The monitoring of small urban water bodies by RS applications can be challenging due to the coarse spatial resolutions of some sensors [138].

Most studies have applied sensors with shorter wavelengths in the study of many types of earth and water features including water monitoring. Some of the shorter wavelength bands include those in the visible and near-infrared. Visible radiation is that portion of the electromagnetic spectrum with a wavelength of between 0.4 to 0.7 µm, which the human eye can perceive. The near-infrared has a wavelength ranging from 0.7 to 1.0 µm [100]. This study focuses on sensors with shorter wavelengths such as visible and infrared wavelengths that have been used in the monitoring of water quality due to their high frequency of data acquisition with large-scale coverage [5].

RS techniques have been utilized in the monitoring and management of water quality. Several WQPs have been monitored with these techniques. Some of these parameters include TDS or salinity, SS, chlorophyll-a concentration, turbidity, surface water temperature, cations, and anions [8,17,42,86,152,153,154,155,156,157,158,159,160,161].

Surface WQPs are detected through different satellite reflectance values on the principle that every object has its reflectance value based on its physical properties. The use of satellite reflectance allows for a vast scope and several temporal monitoring of surface water parameters [162]. Satellite reflectance is the signal of the image on the water surface, which is a function of absorption and backscatter in the water body absorption/backscatter in atmosphere.

Studies have estimated WQPs using digital evaluations of optical satellite data at visible and near-infrared wavelengths [45]. Solar radiation can penetrate and interact with different components as it travels within the column of water in the visible and near-infrared wavelength ranges of the electromagnetic spectrum [163].

Researchers in [164] modeled the spatiotemporal changes of Lake Urmia, which, in September 2010, was the twentieth-largest lake and second-largest hypersaline lake in the world. The study was performed for the period 2000–2013 using RS images from Landsat 5 TM, 7 ETM+, and 8 OLI sensors. Their study considered various indices including the Normalized Difference Moisture Index (NDMI), Normalized Difference Water Index (NDWI), Modified NDWI (MNDWI), Automated Water Extraction Index (AWEI), Water Ratio Index (WRI), and Normalized Difference Vegetation Index (NDVI). The study found NDWI to be a better index and adopted it for spatiotemporal changes in the lake. A principal components analysis (PCA) of the multi-temporal NDWI was evaluated for the detection of surface water quality. Results show a decreasing trend in the lake surface area in the period under study. The PCA technique is useful for producing uncorrelated output bands, separating noise components, and further reducing the dimensionality of data.

Although most studies have measured WQPs using the visible and near-infrared wavelengths of the electromagnetic spectrum, there is an opportunity to measure this parameter from the microwave region of the electromagnetic spectrum. Unlike visible and near-infrared sensors, microwave sensors are less susceptible to atmospheric attenuations, making them suitable for the detection of biophysical properties including those of crops and soils, coastal areas, and extreme weather monitoring in any condition during the day or night [165,166]. Microwave sensors flown on aircraft can be used in estimating temperature distributions to about 1 °C. Similarly, salinity distribution can also be determined by microwave radiometers [167]. Microwave RS has also been explored for the estimation of several WQPs including suspended sediment concentration (SSC), SDD or Secchi disk transparency (SDT), turbidity, chlorophyll-a concentration, salinity, or TDS.

Researchers in [146] applied an empirical neural network in the estimation of WQPs such as SSC, SDD, turbidity, and chlorophyll-a concentration in the Gulf of Finland using combined optical (Landsat 5 TM) and microwave data (European RS (ERS)-2 SAR). Results indicate that the neural network was adequate in describing the nonlinear transfer function between the optical and microwave sensors and the water surface parameters. The optical and microwave sensors including the thermal band produced an R² of >91% for all the measured surface WQPs without the thermal bands, and the performance reduced slightly to an R² of >85% of the WQPs. Thermal bands measure the emitted thermal radiance of the water body. They do not measure the sunlight reflected.

Several studies have utilized different RS sensors and techniques for the retrieval of WQPs and have achieved varying degrees of accuracy. Some of the studies that have retrieved WQPs with RS data have been summarized in

Table 3. Model Retrieval of Optically Active WQPs widely measured using RS Applications.

WQPs	Sensor/Data	Model/Algorithms	R2	Study Area (Region/Country)	Reference
TSS/TSM/SSC/SS/SPM	MODIS	Empirical	0.89	Lake Pontchartrain, Mississippi River. Mississippi Sound (US)	[168]
TSS/TSM/SSC/SS/SPM	RapidEye, SPOT 6, Pleiades-1A	Empirical	0.65	Didipio catchment (Philippines)	[25]
TSS/TSM/SSC/SS/SPM	Landsat-4, 5 TM, Landsat 8 OLI	Semi-empirical	-	Pear River Estuary (China)	[169]
TSS/TSM/SSC/SS/SPM	MODIS	Neural Network	0.72	Bohai Sea, Yellow Sea, East China Sea (China)	[170]
TSS/TSM/SSC/SS/SPM	ALOS/AVNIR-2	Empirical	>0.70	Monobe River (Japan), Altamaha River (US), St. Marys River (US)	[171]
TSS/TSM/SSC/SS/SPM	Landsat 5 TM	Neural Network	> 0.90	Beaver Reservoir (US)	[172]
TSS/TSM/SSC/SS/SPM	MODIS	Empirical	0.83	Green Bay of Lake Michigan (US)	[173]
TSS/TSM/SSC/SS/SPM	Landsat 8 OLI	Empirical	-	Keenjhar Lake (Pakistan)	[174]
TSS/TSM/SSC/SS/SPM	Landsat 8 OLI	Empirical	>0.50	Nakdong River (South Korea)	[175]
TSS/TSM/SSC/SS/SPM	Indian Remote-Sensing Satellite (IRS-P6)	Empirical	0.94	Shitoukoumen Reservoir (China)	[176]
TSS/TSM/SSC/SS/SPM	CASI	Empirical	0.96	Tamar estuary (UK)	[177]
TSS/TSM/SSC/SS/SPM	MODIS	Empirical	0.90	Tampa Bay (USA)	[128]
TSS/TSM/SSC/SS/SPM	Sentinel-2A MSI	Semi-empirical and ML	0.80	Water Reservoirs (Czech Republic)	[178]
TSS/TSM/SSC/SS/SPM	Landsat 8 OLI	AI	>0.93	Saint John River (Canada and US)	[179]
TSS/TSM/SSC/SS/SPM	MERIS	-	-	Lake Maggiore (Italy)	[180]
TSS/TSM/SSC/SS/SPM	IRS LISS III	Empirical	>0.23	Gomti River (India)	[181]
TSS/TSM/SSC/SS/SPM	Hyperspectral Imager for the Coastal Ocean (HICOTM)	Semi-empirical	0.85	Northern Adriatic Sea	[182]
Chlorophyll-a	Landsat 5 TM	Neural Network	>0.50	Beaver Reservoir (US)	[172]
Chlorophyll-a	ALOS/AVNIR-2	Empirical	>0.70	Monobe River (Japan) Altamaha River (US), St. Marys River (US)	[171]
Chlorophyll-a	EO-1 Hyperion	Empirical	0.59	Lake Garda (Italy)	[183]
Chlorophyll-a	Landsat-5 TM, SPOT-Pan, IRS-1C/D LISS, Pan, Landsat-5 TM	Empirical	0.26	Küçükçekmece Lake (Turkey)	[184]
Chlorophyll-a	Landsat 8 OLI	Empirical	>0.50	Nakdong River (South Korea)	[175]
Chlorophyll-a	Landsat 8 OLI	Empirical	>0.34	Nakdong River (South Korea)	[175]
Chlorophyll-a	Landsat 7 ETM+	Empirical	>0.70	Rotorua Lakes (New Zealand)	[7]
Chlorophyll-a	MERIS, MODIS	Semi-empirical	>0.90	Lake Okoboji, Lakes, and Reservoirs in Nebraska, Lake Minnetonka, Choptank River (US)	[23]
Chlorophyll-a	MERIS, MODIS	Semi-empirical	0.89	Fremont State Lakes (US)	[185]
Chlorophyll-a	MERIS	Empirical	>0.90	Taganrog Bay, Azov Sea (Russia)	[186]
Chlorophyll-a	Landsat 5 TM		>0.60	Minnesota Lakes (US)	[187]
Chlorophyll-a	Landsat 5 TM		>0.80	Minnesota Lakes (US)	[187]
Chlorophyll-a	Landsat 5 TM	Neural network	>0.95	Kissimmee River basin (US)	[188]
Chlorophyll-a	Landsat 7 ETM+	Empirical and semi-analytical	0.68	Rotorua Lakes (New Zealand)	[152]
Chlorophyll-a	Sentinel-2A MSI	semi-empirical and ML	0.85	Water Reservoirs (Czech Republic)	[178]
Chlorophyll-a	PROBA-CHRIS	Empirical	0.89	Mazurian Lakes (Poland)	[189]
Chlorophyll-a	MERIS	-	-	Lake Maggiore (Italy)	[180]
Chlorophyll-a	HICOTM	Semi-empirical	0.71	Northern Adriatic Sea	[182]
Chlorophyll-a	MODIS	Empirical, Neural network,	>0.61	Chaohu Lake (China)	[190]
Chlorophyll-a	Airborne real-time cueing hyperspectral enhanced reconnaissance (ARCHER)	Analytical	-	Shenandoah River Basin (US)	[95]
CDOM	Airborne real-time cueing hyperspectral enhanced reconnaissance (ARCHER)	Analytical	-	Shenandoah River Basin (US)	[95]
CDOM	ALOS/AVNIR-2	Empirical	>0.70	Monobe River (Japan), Altamaha River (US), St. Marys River (US)	[171]
CDOM	Landsat 5 TM		>0.60	Minnesota Lakes (US)	[187]
CDOM	MERIS	-	-	Lake Maggiore (Italy)	[180]
CDOM	Ship—mounted spectroradiometer Hyperspectral data	Empirical, Analytical, Semi-Analytical	>0.65	Mississippi River and Atchafalaya River (US), The northern Gulf of Mexico	[191]
CDOM	Landsat 8 OLI	Empirical	>0.7	Saginaw and Kawkawlin Rivers (US)	[169]
CDOM	Landsat 8 OLI, RapidEye	Semi-Analytical	0.52	Lake Garda (Italy)	[151]
Turbidity	ARCHER	Analytical	-	Shenandoah River Basin (US)	[95]
Turbidity	Landsat-5 TM + SPOT-Pan, IRS-1C/D LISS + Pan, and Landsat-5 TM	Empirical	0.68	Küçükçekmece Lake (Turkey)	[184]
Turbidity	Landsat 7 ETM+	Empirical	>0.76	Rotorua Lakes (New Zealand)	[121]
Turbidity	Landsat 5 TM	Neural network	>0.95	Kissimmee River basin (US)	[188]
Turbidity	Landsat 8 OLI	AI	>0.99	Saint John River (Canada, and US)	[179]
Turbidity	Indian Remote Sensing (IRS) P6 LISS IV	Empirical	>0.11	Malad Creek (India)	[192]
Turbidity	MODIS	Empirical	0.87	Green Bay of Lake Michigan (US)	[173]
SDD/SDT	Landsat 7 ETM+	Empirical	>0.80	Rotorua Lakes (New Zealand)	[7]
SDD/SDT	Landsat 5 TM	Empirical	>0.7	Ebro river basin (Western Europe)	[193]
SDD/SDT	PROBA-CHRIS	Empirical	0.95	Mazurian Lakes (Poland)	[189]
SDD/SDT	MODIS	Empirical, Neural Network,	>0.45	Chaohu Lake (China)	[190]
Algal bloom	Landsat 1, 3 MSS, IRS LISS-III, IRS LISS-IV	Empirical	0.88	Sambhar Lake (India)	[194]
Algae bloom	Landsat 5 TM	Empirical, Theoretical algorithms	0.86	Guanting Reservoir (China)	[24]
EC	Worldview-2	Empirical	0.68	Lake Al-Saad (Saudi Arabia)	[195]

and

Table 4. Model Retrieval of Optically Inactive WQPs widely measured using Remote Sensing Applications.

WQPs	Sensor/Data	Model/Algorithms	R2	Study Area (Region/Country)	Reference
TP	Landsat-5 TM, SPOT-Pan, IRS-1C/D LISS, Pan, and Landsat-5 TM	Empirical	0.62	Küçükçekmece Lake (Turkey)	[184]
TP	Airborne imaging data (AISA)	Empirical	>0.60	Eagle Creek Reservoir, Geist Reservoir, Morse Reservoir (US)	[176]
TP	Landsat 8 OLI	Empirical	0.80	Xin’anjiang Reservoir (China)	[196]
TP	Landsat 5 TM	Neural network	>0.95	Kissimmee River basin (US)	[188]
TP	Landsat 5 TM	Empirical	0.77	Qiantang River (China)	[197]
TP	MODIS	Empirical, Neural network	>0.24	Chaohu Lake (China)	[190]
TP	Landsat 5 TM	Empirical and theoretical	0.38	Guanting Reservoir (China)	[24]
DP	Landsat 5 TM	Empirical and theoretical	0.91	Guanting Reservoir (China)	[24]
TN	Landsat-5 TM, SPOT-Pan, IRS-1C/D LISS, Pan, Landsat-5 TM	Empirical	0.70	Küçükçekmece Lake (Turkey)	[184]
TN	Landsat 8 OLI	Empirical	>0.20	Nakdong River (South Korea)	[175]
TN	Landsat 8 OLI	Empirical	0.84	Xin’anjiang Reservoir (China)	[196]
NH3-N	Landsat 5 TM	Empirical and theoretical	0.65	Guanting Reservoir (China)	[24]
NO3-N	Landsat 5 TM	Empirical and theoretical	0.85	Guanting Reservoir (China)	[24]
TN	MODIS	Empirical, Neural network	>0.50	Chaohu Lake (China)	[190]
TN	Landsat 5 TM	Empirical and theoretical	0.56	Guanting Reservoir (China)	[24]
TDS/DS	Sentinel-2 MSI	Empirical	-	Guartinaja, Momil wetlands (Columbia)	[139]
TDS/DS	Landsat 8 OLI	Empirical	>55	Shatt al-Arab River (Iraq)	[28]
TDS/DS	Landsat 8 OLI	Empirical	0.64	Tigris River (India)	[198]
TDS/DS	Landsat 5 TM	Empirical	>0.83	Tigris and Euphrates Rivers (Iraq)	[145]
TDS/DS	Landsat 8 OLI	Empirical	>0.95	Tubay River (Philippines)	[199]
TDS/DS	IRS LISS III	Empirical	>0.46	Gomti River (India)	[181]
COD	Landsat 8 OLI	Artificial Intelligence	>0.93	Saint John River (Canada and US)	[179]
COD	Landsat 5 TM	Empirical	>0.63	Shenzhen Reservoir (China)	[200]
COD	IRS LISS III	Empirical	>0.28	Gomti River (India)	[181]
COD	Landsat 5 TM, Landsat 7 ETM+	Empirical	>0.66	Dongting Lake (China)	[201]
BOD	Landsat 8 OLI	Artificial Intelligence	>0.92	Saint John River (Canada and US)	[179]
BOD	Landsat 5 TM	Empirical	>0.70	Shenzhen Reservoir (China)	[200]
BOD	IRS LISS III	Empirical	>0.48	Gomti River (India)	[181]
DO	Landsat 8 OLI	Artificial Intelligence	>0.93	Saint John River (Canada and US)	[179]
DO	IRS LISS III	Empirical	>0.56	Gomti River (India)	[181]
DO	Worldview-2	Empirical	0.67	Lake Al-Saad (Saudi Arabia)	[195]
TOC	Landsat 5 TM	Empirical	>0.82	Shenzhen Reservoir (China)	[200]
pH	IRS LISS III	Empirical	>0.70	Gomti River (India)	[181]

for optically active and optically non-active parameters.

Based on information obtained from the literature reviews, a generalized systematic overview to be utilized for any study involving the monitoring and management of water quality using RS applications is presented in Figure 1 and described in the following paragraphs.

The first approach to be employed in water quality assessment with RS involves conducting a literature review: this involves searching through literature databases on the background and the state-of-the-art work performed in the area and identifying gaps to be addressed. Some of the known databases include Web of Science™, Compendex™, Scopus™, and Google Scholar™ with keywords such as water quality, RS, and satellite sensors.

The second approach entails methodology development. This includes establishing WQPs to be measured and the identification of RS and ground truth data to be used. RS sensors include airborne and spaceborne. The sensor and image acquisition to be used depend on a lot of factors including the availability of resources and the desired need. Ground truth data sets include field measurements or laboratory analysis of the WQPs. The USGS collects several WQPs that are available for free on their website for download for any RS analysis. After obtaining RS data through drones and satellites, level 1 images need to be corrected by application of atmospheric corrections. Some of the known atmospheric algorithms used in literature include Dark Object Subtraction (DOS) and Atmospheric Correction based on Turbidity Classification (ACbTC). These algorithms have the potential to overcome the adjacency effects caused by dark waters and bright land in the NIR bands. Atmospheric corrections are essential prerequisites needed to be performed to obtain accurate inland water information. Studies have also found that combining different RS data using merging algorithms proved to be a plausible approach for the precise retrieval of WQPs [5,202,203]. After establishing a method to be used, the next step involves model development: models are developed typically by using band reflectance in the RS image and the ground truth data.

Several statistical models may be explored depending on the need of the researcher. Some of the models include empirical regression models, and machine learning techniques such as ANN, SVM, and PCA, among others. After the model is developed, models then need to be calibrated and validated. Developed models are trained and calibrated to ensure accuracy in the prediction of WQPs and evaluated using statistical metrics such as Nash–Sutcliffe Efficiency (NSE), Percent Bias (PBIAS), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Percent Mean Absolute Relative Error (PMARE) [204,205,206,207,208]. The evaluated models are then used to predict and map of WQPs: WQPs are predicted using the developed models and then mapped to assess the spatiotemporal extent of water degradation or pollution. Once the WQPs are mapped, the sources of pollution can then be analyzed; once the sources are known, proposals and recommendations can then be offered on the strategies to control the sources and the mitigating effects.

3. Strengths and Shortcomings of RS Applications

3.1. Strengths

RS techniques present the opportunity to manage and protect water resources. RS applications offer researchers and policymakers the opportunity to carry out synoptic, consistent, repetitive, detailed, and cost-effective assessments of water quality on a spatiotemporal scale compared with point-based in situ measurements [16,18]. Measurement of WQPs with RS satellite sensors can be performed at a great distance, i.e., several hundred to thousands of miles, which helps in the coverage of large areas on the ground and can observe water surfaces repeatedly [43].

The use of satellite images saves users a huge sum of money that would have been spent on field or laboratory work. The use of satellite images and improved statistical and mechanistic modeling of WQPs has been performed at local, watershed, and regional scales with a high degree of accuracy. These models also can effectively enhance the real-time detection of hydrological variability, which is key in providing early warning and rapid response to harmful algal bloom events [18].

Most satellite images are freely available, open-source data that can be accessed at no cost to the users. Different sensors offer different advantages compared to others in terms of their spatial and temporal resolutions. Comparing Landsat sensor data to sensors of higher resolution such as those aboard the SPOT, ASTER, or IKONOS satellite program still places the Landsat programs at an advantage owing to their more extensive temporal record dating back as far as the 1970s. Landsat satellites have been extensively used for water quality assessment due to their relatively low cost, and spatial and temporal resolutions. The utilization of Landsat image data has saved users a sum of about USD 3.45 billion as of 2017, with users in the United States accounting for about 59.7% [114]. The Landsat program serves as a central reference comparison focus for many different moderate and high-resolution optical satellite sensors operated by commercial and government institutions due to their associated benefits including the length of a record, availability of data, calibration standards, and global coverage. Landsat programs have influenced the development of other earth RS satellites by government agencies such as the ESA’s Copernicus Program Sentinel-2 satellite constellation and private agencies’ systems [114].

3.2. Limitations

Although RS satellite images offer a great tool for the monitoring of WQPs, they possess some underlying limitations that when properly addressed will improve their performance and provide many comparable results with conventional methods. The limitations of RS techniques are outlined in the following paragraphs.

Image acquisition errors may occur. The atmospheric path existing between the sensor and the water surface impacts remotely sensed reflectance, which, in turn, affects the data acquired. The interactions of the sensor’s incident energy with atmospheric conditions are significant enough to impact the incoming radiance, which has the potential to compromise the image. Cloud cover, fog, and haze will blur or obscure target water bodies. Hence, they impact images collected in the visible and near-infrared regions of the electromagnetic spectrum [5,6]. As a result, atmospheric corrections need to be carried out before analysis is performed. Atmospheric corrections are important for water surfaces where the reflected light fraction is very low (i.e., not more than 10% of the radiance measured by the sensor) [18].

Atmospheric interference: This has the potential to restrict the optical signals sensed from the surface of the water bodies. The maximum penetration of light in seawater is at a depth of around 55 m near the 475 nm wavelength. At this depth, a greater percentage of incident energy on the water surface is transmitted. With increased concentrations of sediments to about 400 ppm, light penetration reduces to about 60 cm, making only a progressively thinner layer of surface water detectable [16]. The interactions of incident energy and the atmospheric conditions are significant enough to quantitatively affect the incoming radiance and hence can comprise the results [18].

Image processing errors such as processes involved in the atmospheric corrections and pan-sharpening of satellite images may induce errors when using these images for the estimation of water quality. Satellite images with coarse spatial resolutions may possess issues related to the contamination of pixels with the land surface presenting an inaccurate spectral signature. Additionally, remote sensing image processing errors may result in an invalid detection of WQPs, bringing the accuracy of extracted WQPs into question [40]. For example, the densest areas of cyanobacteria blooms in the Baltic Sea are said to be usually masked out in chlorophyll product maps as processing errors or an atmospheric correction. This is because high reflectance in the near-infrared range of the electromagnetic spectrum is not anticipated by ocean color algorithms [209].

Studies have considered ways of improving the accuracy of their results obtained from high-resolution images through techniques such as pan-sharpening. Other studies resampled their pixel resolution such as the case of [139] where the 20 m resolution Sentinel 2 image was resampled to 10 m resolution. Pan-sharpening or downscaling, unlike conventional resampling, is a fusion technique that allows for the artificial enhancement of the spatial resolution of the image by fusing it with a higher-resolution image. Pan-sharpening introduces less spectral and spatial distortion and provides visually more coherent data compared with resampling of the pixel [210,211]. Conventional resampling of images selects the closest image pixel and preserves the input data, which has the potential to cause blockings in the pixel and location shifts [210]. The authors in the study [210] compared the results of a conventional resampling of a 30 m resolution Landsat 8 OLI to 20 m and Panchromatic-Assisted Downscaling (LPAD) or pan-sharpening of 30 m reflective wavelength bands to Sentinel-2 20 m resolution and found that the LPAD resulted in clearer object boundaries and finer spatial detail than the conventionally resampled data. Although these techniques have been applied to images, it is noted that the pan-sharpened approach alters the radiometric and spectral features of the satellite image and is only useful for image visualization and interpretation, and not for analytical purposes [211]. Researchers in [212] also noted that the sharpening procedure introduces artifacts, particularly around bright non-water pixels, and is therefore not recommended to resample high-resolution bands for noise reduction. The authors pointed out the spatial resolution of the image is its strength, and the native resolution band will produce adequate performance.

Another drawback in the use of RS techniques in the monitoring of water quality is the lack of in situ data in some areas, particularly in sub-Sahara Africa, for the calibration and validation of remotely sensed models due to associated on-site cost and high level technical expertise involved in conducting in situ water quality monitoring efforts [18]. Models developed using RS data need to be properly calibrated and validated using in situ data. The absence of in situ measurements for calibration and validation raises accuracy concerns [16,40].

Weak or inactive optical characteristics of some WQPs: Most of the WQPs detected directly by RS applications are optically active such as TSS, CDOM, and chlorophyll-a concentration. These parameters absorb light in the EM spectrum and influence water optical properties enabling these parameters to be sensed from satellite observations, unlike the optically inactive or weak optical parameters such as TDS, pH, TN, and NH₃-N, which have a low signal-to-noise ratio. The estimation of optically inactive parameters such as TDS is due to their statistical associations with other colored WQPs with which they may co-vary. Additionally, retrieval of most WQPs is performed through methods that require accurate parameterization, which may vary with the optical property of the water body [16,135].

Limitations of the sensor’s spatial, spectral, and temporal properties: Other limitations that arise with the use of RS have to do with spatial, temporal, and spectral resolutions of RS sensors. RS sensors have different spatial, spectral, and temporal resolutions, and depending on the need, it may be difficult to rely on a particular sensor to detect temporal trends, especially when they have a longer repeat cycle. An example is comparing the temporal resolutions of Sentinel-2 A/B MSI, which has a 5-day revisit time, to Landsat 8 OLI, which has a 16-day revisit time.

Airborne sensors compared with spaceborne sensors have finer spatial and temporal resolutions, but the data may, however, come at a higher cost to the user and may defeat the purpose of cost savings with the use of RS. Data from most sensors such as MODIS (250 m–1.1 km), MERIS (300 m), and SeaWiFS (1.1 km) are suitable for the effective retrieval of ocean WQPs but may not work efficiently for inland water bodies such as narrow river channels and lakes due to their coarse spatial resolutions and their low signal-to-noise ratios. Data from some sensors with high spatial resolution may not provide an effective time series of effective assessments of WQPs due to their low temporal resolutions. Those sensors with high temporal resolutions may also not support comprehensive and effective investigations owing to their coarse spatial resolutions [5].

4. Summary and Recommendations

In summary, traditional in situ methods of monitoring WQPs are not only cost-intensive but require constant laboratory and field efforts. The use of RS techniques and geospatial tools, however, provides a cost-effective approach for the assessment of WQPs on a wide spatial scale in a reliable manner in space and time. Both spaceborne and airborne RS sensors have the potential to reliably estimate WQPs. Retrieval of WQPs from RS images is performed through empirical, analytical, semi-empirical, and AI methods, as described in Section 2.3.5.

Multispectral sensors in water quality monitoring have been used widely due to the availability of their images across the globe. Their usage is said to be gaining ground in developing countries. The use of multispectral images such as MERIS in the study of small inland lakes may pose some accuracy issues due to their coarse resolution. Coarse-resolution images have only one pixel representing a large area, which is a potential source of error since each small area may have its distinct features. An example is a study to compare the Chinese high-resolution GF-1 Wide Field Imager (WFI) data with MODIS for the estimation of SPM. High consistency in the spatial distribution and concentration of SPM maps was seen between the GF-1 and Landsat 8 OLI data. More than 75% of the spatial variations were resolved by the GF-1 while only 40% were resolved by the MODIS band with 250 m resolution, indicating the limitations of using a coarse resolution image such as the MODIS [213].

The use of UAV airborne sensors is on the ascendency due to the rapid development of these sensors. Airborne-based spectrometers provide efficient and flexible solutions to underlying temporal, spectral, and spatial resolution issues associated with some satellite sensors.

Retrieval of WQPs using RS is based on the optical properties of the WQPs. Parameters such as TSS, CDOM, chlorophyll-a concentration, and turbidity can be directly determined using RS images since they are optically active. Retrieving optically weak or inactive parameters such as TDS, TN, TP, COD, and pH is possible because they have been found to correlate with optically active parameters.

Based on the literature review, the following recommendations are made for future considerations for the effective assessment of water quality.

Establishment of atmospheric correction model: When using level 1 raw satellite images, it is recommended that high-precision atmospheric correction algorithms such as DOS and ACbTC be employed to improve the accuracy of the retrieval models for water quality. UAV platforms should be used for capturing RS images without inferences from the atmosphere. When available, level 2 atmospherically corrected surface reflectance data sets should be used to minimize the potential errors associated with atmospheric corrections.

The Use of High-Resolution Images: Analyses of water quality should be performed using high-resolution satellite data. High-resolution images with shorter revisit intervals should be used for water quality assessment. Examples of high-resolution images include GeoEye IKONOS with a spatial resolution of 0.82–5 m and revisit time of 2–3 days; and Digital Globe WorldView-1 with a resolution of 0.5 m and a revisit interval of 1.7 days. Additionally, hyperspectral images should be utilized and explored for the estimation of WQPs, Examples include Airborne Visible Infrared Imaging Spectrometer (AVIRIS) by NASA’s Jet Propulsion Lab, Airborne Imaging Spectrometer Multispectral by Spectral Imaging, and Hyperspectral Digital Imagery Collection Experiment by the Naval Research Lab. Hyperspectral images can aid to eliminate both the potential limitations of low spatial and spectral resolutions and the limitations of discrete spectral signatures associated with other images. Hyperspectral sensors can simultaneously collect narrower and contiguous images with up to 200 spectral bands, which enable measurement and monitoring of WQPs.

Improvement in methods of analysis: Analysis and interpretation of WQPs using RS data should involve machine learning (ML) approaches instead of simple linear regressions as outlined for the retrieval of WQPs in Section 2.3.5. Although empirical methods of retrieval of WQPs are easy to implement and require less computation time and effort, they may not be effective in distinguishing some WQPs. ML models such as PCA, ANN, and SVM models, which have the potential to detect complicated nonlinear relationships through training and testing to improve the accuracy of WQP retrieved through RS, should be used.

Utilization of free and low-cost satellite images with fine spatial resolution: Although high resolution and hyperspectral images have high capabilities in discriminating WQPs, some of them come with a cost and may defeat the purpose of utilizing RS in water quality monitoring. Therefore, it is recommended that researchers should utilize Landsat images in the areas of water quality assessment. Landsat images have proven to be useful for small inland water systems including lakes, owing to their high spatial resolution (30 m) compared to coarse resolution images such as MODIS (250 m–1.1 km), MERIS (300 m), and SeaWiFS (1.1 km), which are more useful for coastal studies. Comparing Landsat programs to sensors of higher resolution such as those aboard the SPOT satellite program, ASTER, or IKONOS still places Landsat programs at an advantage owing to their more extensive long-term records dating back as far as the 1970s. The Landsat program is an open source with a highly accessible global spatial coverage data repository.