Introduction

Water, particularly inland water bodies, plays a crucial role in various sectors such as agriculture, urban planning, industry, aquaculture, recreation, wildlife, and ecological health (Welcomme, 2011; Wang et al., 2022). With the rapid growth of the global population, the increased demand for water access and use has caused increased pollution and degraded water quality (United Nations, 2022). Water quality impairments can occur due to elevated nutrient levels, resulting from the discharge of pollutants in residential and industrial areas (Osibanjo et al., 2013; Karakoc et al., 2003), excessive use of fertilizers and pesticides (Muscutt et al., 1993; Dosskey, 2001), changes in land use/land cover (Rostom et al., 2017; El Saadi et al., 2014; Mbuh, 2019), and erosion during storm events (Grissinger & McDowell, 1970). Given the potential impact of poor water quality on both human health and marine ecosystems, the assessment of water quality holds utmost significance. Hence, it is essential to employ effective methods for measuring water quality to ensure sustainable management. Regular and extensive assessments are necessary for the adaptive management of water bodies, facilitating the development of long-term plans and the implementation of new policies and regulations based on the intricate relationships between human practices and water quality within lakes and reservoirs (United Nations Environment Programme, 2021).

Water quality parameters encompass the physical, chemical, and biological characteristics of a water body (Gholizadeh et al., 2016; Wen & Yang, 2011). Traditionally, these parameters are measured from small volumes of water using in situ sensors or grab samples that are subsequently analyzed in a laboratory. However, it is important to note that water quality exhibits spatial-temporal variability across the water surface (Sagan et al., 2020). Spatial variability refers to variations in parameter values or concentrations across different locations, while temporal variations represent changes caused by seasonal fluctuations in natural processes, such as temperature, precipitation, and hydrological conditions (Zhang et al., 2008). Consequently, attempting to discern water quality using limited samples poses challenges, necessitating thorough sampling and measurement. The use of traditional techniques in this regard can be expensive, labor-intensive, and time-consuming, and collecting a sufficient number of samples to adequately represent an entire water body is often impractical. These limitations impede the attainment of continuous and synoptic water quality assessments (Giardino et al., 2001; Skarbøvik & Roseth, 2014; Nas et al., 2009).

Remotely sensed water quality assessment presents a promising approach to overcome the limitations of traditional methods and enables more cost-effective, efficient, and larger-scale monitoring and assessments. In recent years, unmanned aerial vehicles (UAVs) have gained popularity for remote water quality assessment due to their ability to provide very high spatial and temporal resolution data. This capability is particularly valuable for monitoring small inland water bodies such as lakes, streams, rivers, wetlands, or reservoirs (Zhang et al., 2021; Yang et al., 2022; Zhang et al., 2023; Guimarães et al., 2019; Olivetti et al., 2020; Castro et al., 2020; Cui et al., 2022; Zang et al., 2012; Zeng et al., 2017; Keith et al., 2014; Murugan et al., 2016; Moses et al., 2015; Tan et al., 2011; Wu et al., 2014; Govender et al., 2007; Isgró et al., 2022; Su et al., 2015).

While several reviews have explored water quality assessment with a focus on utilizing multi-sourcing and multi-sensor satellite remote sensing data (Gholizadeh et al., 2016; Yang et al., 2022; Palmer et al., 2015; Yan et al., 2015), there is a literature gap regarding the joint review of UAV and satellite remote sensing of water quality in small inland water bodies. A comprehensive review is needed to support managers in evaluating trade-offs and selecting the most appropriate measurement technique. As a result, this study aims to bridge this gap in the literature by reviewing the current state-of-the-art water quality monitoring using both satellite and UAV remote sensing data for small inland water bodies. The review encompasses an analysis of the challenges associated with monitoring water quality in such water bodies using satellite remote sensing data, as well as the potential of UAVs to overcome these challenges and provide reliable data.

Remote sensing for water quality assessments

Remote sensing data plays a vital role in studying the Earth’s surface, examining changes, identifying problems, and formulating solutions for maintaining or enhancing the environment. Advancements in technology have contributed to the widespread use of remote sensing data in water quality assessments, yielding promising and satisfactory results (Zhang et al., 2023; Guimarães et al., 2019; Olivetti et al., 2020). This approach provides data across various spatial and temporal scales in a cost-effective and time-efficient manner (Mbuh, 2019). Moreover, the automatic and continuous acquisition of data allows for the timely identification of surface changes on Earth (Chebud et al., 2012). To date, numerous studies have leveraged remote sensing data for diverse applications, including water volume calculations (Lu et al., 2013), water resource management (Giardino et al., 2010), groundwater mapping (Elbeih, 2015), water bodies identification (Sun et al., 2012), water storage and level measurements (Frappart, 2005), and water quality monitoring in both small and large water bodies (Bresciani et al., 2017; Castro et al., 2020).

Remote sensing has been proven to be a valuable and effective method for capturing water quality parameters (Saberioon et al., 2020; Hellweger et al., 2004; Yulong et al., 2022; Bonansea et al., 2018; Su & Lo, 2022). Various water quality components are developed and used to determine the presence and level of contamination using remote sensing, thus asserting water quality. These components are measured based on their reflectance signals, known as water-leaving radiance, captured by the sensors. Figure 1 displays the interaction of light between water quality components and sensors. The reflectance, absorbance, and scattering spectral characteristics of these components, called inherent optical properties (IOP) (IOCCG, 2006), make it possible to detect and monitor water pollutants using spectral reflectance signatures (Rostom et al., 2017; Wen & Yang, 2011). Optically active components can be directly measured using the relationship between inherent optical properties (IOP) and remote sensing reflection (Kirk, 1994; Kutser, 2004; Matthews, 2011). In contrast, optically inactive components do not exhibit optical activity and do not interact with light in the same way as optically active parameters. Due to the low optical properties and low signal-to-noise ratio of optical inactive components (Gholizadeh et al., 2016; Yang et al., 2022), several studies have primarily focused on optically active components. Similarly, the optically active properties are often used as proxies to estimate optically inactive properties (Yang et al., 2022). Table 1 presents commonly measured water quality parameters using remote sensing techniques.

Fig. 1
figure 1

Interactions of light between water quality components and sensors (Dörnhöfer & Oppelt, 2016)

Full size image
Table 1 Commonly measured water quality parameters
Full size table

Platforms and sensor types

Water quality components can be assessed using multispectral and hyperspectral optical sensors deployed on both airborne and satellite platforms (Fig. 2). Each of these platforms and sensor types is described next in the context of inland water quality monitoring.

Satellite platform

Satellite platforms have transformed water quality assessments by providing a broad-scale and comprehensive view of water bodies on the Earth’s surface. These platforms have the advantage of enabling continuous monitoring of large areas without causing any disturbance to the aquatic environment. This feature makes satellite-based platforms an invaluable tool for environmental assessment and management (Harvey et al., 2015; Topp et al., 2020).

UAVs platform

Unmanned aerial vehicles/systems (UAVs/UASs), commonly known as drones, have emerged as an effective remote sensing platform for environment monitoring (Cheng et al., 2020; Hashemi-Beni & Gebrehiwot, 2021; Gebrehiwot & Hashemi-Beni, 2021). UAVs offer several advantages over satellite remote sensing, such as the ability to fly at lower altitudes (Olivetti et al., 2020; Castro et al., 2020), providing higher spatial resolution, and offering flexible deployment options at a reasonable cost (Isgró et al., 2022).

Multispectral sensors

Multispectral sensors capture data within a limited number of bands, typically ranging from 5 to 10 bands with broad spectral resolution. These bands cover wavelengths from 430 to 1500 nm, encompassing the visible, near-infrared, and short-wave infrared regions of the electromagnetic spectrum. Multispectral sensors have been extensively employed for the detection and monitoring of water body dynamics and quality (Castro et al., 2020; Hellweger et al., 2004; Bonansea et al., 2018; He et al., 2008; Soriano-González et al., 2022; Knight & Voth, 2012; Pahlevan et al., 2020).

Hyperspectral sensors

Hyperspectral sensors capture several narrow and continuous spectral bands across the entire spectrum, from visible to thermal infrared regions. These sensors can record hundreds of spectral bands in a single acquisition (Govender et al., 2007), providing detailed spectral information for each pixel in an image. As a result, hyperspectral remote sensing data offers exceptional differentiation of water quality parameters based on their narrow-band spectral response. Its capability to capture fine-grained spectral information and identify specific water quality parameters has made hyperspectral data particularly valuable for monitoring and characterizing water quality parameters (Zeng et al., 2017; Palmer et al., 2015; Giardino et al., 2007; Östlund et al., 2001; Banerjee & Shanmugam, 2021).

In certain investigations, researchers have employed a combination of multispectral and hyperspectral sensors to obtain comprehensive data sets, resulting in cost-effective solutions (Yang et al., 2022; Topp et al., 2020; Giardino et al., 2010). Similarly, in ongoing efforts to maximize the periodicity of data acquisition and address data gaps in small inland water bodies, several studies have applied a combination of satellite and UAV platforms. These studies conduct separate analyses and develop predictive models (Castro et al., 2020). Additionally, other studies focus on integrating and fusing the datasets to construct predictive models (Rahul et al., 2023). This approach aims to enhance the accuracy of inland water quality monitoring by leveraging the unique strengths of both satellite and UAV technologies. Tables 2 and 3 provide an overview of commonly used satellite and UAV sensors, including their spectral and spatial resolutions.

Fig. 2
figure 2

Remote sensing platforms operating altitudes and spatial resolution

Full size image

Data pre- and postprocessing

Preprocessing

The preprocessing of remote sensing data is a crucial step in obtaining accurate information and improving the retrieval of water quality parameters. This process involves multiple steps, including geometric correction (Nas et al., 2009), radiometric correction, and primarily atmospheric correction. Each of these steps plays a significant role in ensuring the quality and accuracy of the data. Geometric correction involves aligning and georeferencing the remote sensing data to ensure precise spatial alignment. This step is essential for accurate analysis and comparison of different datasets. Radiometric correction is the process of calibrating the sensor data to convert digital numbers into physical units of radiance or reflectance. By removing sensor-specific effects and calibrating the data, radiometric correction allows for quantitative analyses and comparisons of images acquired at different times or by different sensors.

Atmospheric correction is a critical preprocessing method that corrects the influence of atmospheric components on the remote sensing data. Atmospheric effects, such as aerosols, water vapor, and atmospheric path radiance, can significantly impact the accuracy of the data, especially in aquatic environments. The successful application of remote sensing algorithms for water quality assessment relies on employing appropriate atmospheric correction methods to accurately retrieve the remote sensing reflectance (Hussein & Assaf, 2020; Hu et al., 2004; Moses et al., 2017). This correction is particularly important in mitigating the atmospheric effects on water bodies, which typically exhibit low reflectance values (Martins et al., 2017; Pahlevan et al., 2021). The specific preprocessing methods required may vary depending on the type of remote sensing data, platforms, and water type (Pahlevan et al., 2021).

Satellite remote sensing

Satellite data often requires more rigorous atmospheric correction than UAV data due to the higher altitude and greater atmospheric attenuation. Several atmospheric correction algorithms have been designed for aquatic environments to obtain reliable water-leaving radiance estimates from satellite measurements. In brief, these algorithms can be categorized into two groups: image-based models and radiative transfer codes models (RTCs) (Hadjimitsis et al., 2004). The first group, image-based atmospheric correction techniques, utilizes the information contained within the satellite image itself to estimate and correct atmospheric effects. These methods employ look-up tables to simulate the interaction of radiation with the atmosphere based on the estimated aerosol optical thickness (Mobley et al., 2016). Some commonly used algorithms in this category include atmospheric correction for L8 OLI and Sentinel 3 Ocean and Land Colour Instrument (ACOLITE) (Saberioon et al., 2020; Pahlevan et al., 2020; Ansper & Alikas, 2018; Rodrigues et al., 2017; Page et al., 2019), Case 2 Regional Coast Colour processor (C2RCC) (Isgró et al., 2022; Ansper & Alikas, 2018; Shen et al., 2020), Improved Contrast between Ocean and Land (ICOL) (Bresciani et al., 2017; Harvey et al., 2015; Philipson et al., 2016), POLYnomial-based algorithm applied to MERIS (POLYMER) (Pahlevan et al., 2020; Ansper & Alikas, 2018; Shen et al., 2020), and SeaWiFS Data Analysis System (SeaDAS) (Pahlevan et al., 2020; Banerjee & Shanmugam, 2021; Shen et al., 2020; Chavula et al., 2009; Pahlevan et al., 2021; Dev et al., 2022). These algorithms are relatively simple to implement and require minimal ancillary data, making them accessible to a wide range of users and applications. However, they may encounter limitations in complex atmospheric conditions or when dealing with specific optical properties of water bodies (Rodrigues et al., 2017; Ansper & Alikas, 2018).

Table 2 Commonly used satellite-based sensor for water quality assessment
Full size table
Table 3 Commonly used UAV-based sensors for water quality assessment
Full size table

The second group, radiative transfer code (RTC) models, simulates the interaction of solar radiation with the atmosphere and the underlying surface. These models take into account atmospheric constituents, such as gases, aerosols, clouds, and profiles (e.g., temperature, pressure, humidity), as well as surface properties (e.g., reflectance, emissivity) and sensor characteristics to quantify the atmospheric effects on satellite measurements (Vermote et al., 1997). The most commonly used RTC models for inland and coastal waters are quasi-analytical algorithm (QAA) (Li et al., 2016; Ogashawara et al., 2022) and generalized IOP algorithm (GIOP) (Werdell et al., 2013; Shi & Wang, 2019). RTC-based algorithms provide physically meaningful surface reflectance values, allowing for quantitative analysis and accurate comparison of remote sensing data. However, they require accurate estimation of atmospheric parameters and input data, and uncertainties in these parameters can introduce errors in the correction process (Kutser, 2012).

In addition to water-specific atmospheric correction models, several researchers have also utilized atmospheric correction algorithms designed for land environments. The dark object subtraction (DOS) algorithm is widely used in water quality monitoring (Bonansea et al., 2015; Matthews et al., 2010; Markogianni et al., 2020; Zhou et al., 2008; Carvalho et al., 2022). It assumes that certain dark targets in an image, such as deep-water bodies or dark pixels, exhibit no atmospheric signal and can serve as references for atmospheric correction. Other atmospheric correction algorithms, such as Sen2cor atmospheric correction procedure for MSI imagery (Yang et al., 2022; Toming et al., 2016; Ansper & Alikas, 2018; Kutser et al., 2016; Grendaitė et al., 2018), atmospheric and topographic correction method (ATCOR) (Bresciani et al., 2019; Rodrigues et al., 2017; Kutser et al., 2016; Chebud et al., 2012), Land Surface Reflectance Code (LaSRC) (Peterson et al., 2020; Rubin et al., 2021), Second Simulation of a Satellite Signal in the Solar Spectrum (6 S) (Bresciani et al., 2017; Bonansea et al., 2015; Matthews et al., 2010; Ma & Dai, 2007; Oyama et al., 2009; Flores-Anderson et al., 2020; Shen et al., 2020), and Fast Line-of-Sight Atmospheric Analysis of Hypercubes (FLAASH) (Yang et al., 2022; Rodrigues et al., 2017; Kutser, 2012; Ha et al., 2017; Watanabe et al., 2015; Tebbs et al., 2013; Abdelmalik, 2018; Kutser et al., 2005) have been applied for different types of satellite imagery.

UAV remote sensing

One significant advantage of UAV remote sensing is that the imagery captured by UAVs is less affected by atmospheric conditions compared to satellite imagery (Castro et al., 2020; Zeng et al., 2017; Del Pozo et al., 2014). Since UAVs fly at lower altitudes, the atmospheric effects on the imagery are minimized, but rather, the acquired imagery may suffer from more distortions and perspective effects (Zang et al., 2012). Because of this, in UAV remote sensing, more attention is typically given to geometric correction to rectify these distortions and align the UAV imagery accurately (Su et al., 2015). Additionally, other preprocessing steps, such as georeferencing, reflectance correction, and ortho-mosaicking, are necessary for UAV imagery (Isgró et al., 2022; Castro et al., 2020; Guimarães et al., 2017; Cheng et al., 2020; Olivetti et al., 2020; Zhang et al., 2021, 2020).

Data analysis

Coastal and inland water bodies are often characterized as Case II and exhibit complex optical properties due to the presence and varying concentrations of organic matter, suspended particles, and dissolved substances (Morel & Prieur, 1977; Doerffer et al., 1999). For example, chlorophyll-a, a pigment utilized in photosynthesis by plants and algae, shows high reflectance in the green (reflectance peak at 550 nm), red-edge (670–675 nm), and near-infrared (reflectance peak at 700 nm) spectral bands, as well as high absorption in the blue (450–475 nm) spectral range (Gholizadeh et al., 2016; Castro et al., 2020; Avdan et al., 2019; Hussein & Assaf, 2020; Zhang et al., 2022). On the other hand, suspended particles, which consist of sediment, organic matter, and plankton (containing algal cells) suspended in the water column, are characterized by high reflectance throughout the spectral range, while colored dissolved organic matter (CDOM) absorption increases with decreasing wavelength, from blue to some portion of the green. Figure 3 shows the spectral characteristics of water with high turbidity, chlorophyll-a, and colored dissolved organic matter. Complex waters can contain all three constituents, introducing complexity in the relationship between the sensor radiance and the water quality parameters (Sudheer et al., 2006) and posing challenges for remote sensing-based water quality retrieval (Maciel et al., 2021). To overcome these challenges, specific retrieval algorithms that account for the unique spectral responses of the constituents are needed (Yang et al., 2022; Harvey et al., 2015). Indeed, numerous models, such as those implemented in the NASA SeaDAS software (https://seadas.gsfc.nasa.gov), have been developed on global and larger regional scales. However, these models may not be suitable for most inland water bodies since they were originally designed for Case I water bodies, such as oceans, characterized by a dominant presence of chlorophyll-a. These models often rely on simpler algorithms that may not adequately capture the complexity of inland water systems (Qin et al., 2007). As a result, developing models for inland water bodies necessitates a high degree of customization, tailoring them to the specific characteristics of the inland water, including its unique properties and water type. Unlike open oceans, inland water ecosystems exhibit diverse dynamics, requiring more meticulous and context-specific approaches to achieve accurate predictions of water quality. Developing and utilizing these algorithms are essential for effective water quality monitoring and management in complex aquatic environments.

Fig. 3
figure 3

Spectral illustration of water with high turbidity, chlorophyll-a (Chl-a), and colored dissolved organic matter (CDOM) (Pahlevan et al., 2021). OWT stands for optical water types, a classification based on water’s reflectance and absorption. OWT1 and OWT2 represent clear waters, OWT3 shows high chlorophyll concentrations, OWT4, OWT5, and OWT6 indicate high concentrations of different phytoplankton blooms and high turbidity, and OWT7 represents waters with high sediment concentrations

Full size image
Fig. 4
figure 4

Workflow for water quality parameters prediction using remote sensing data

Full size image

To address the difficulties associated with monitoring Case II waters, various approaches have been developed and used for the accurate retrieval of water constituent concentrations from remote sensing data. Figure 4 shows a framework acquisition and analysis of water quality parameters using remote sensing data. In this review, these approaches are categorized as follows: empirical methods, semi-analytical methods, and artificial intelligence (AI) approaches.

Empirical approach

The empirical approach establishes statistical relationships between remote sensing observations and water quality parameters. This approach can involve single band analysis (El Saadi et al., 2014; Brezonik et al., 2009; Tarrant et al., 2010; Simis et al., 2005), band combinations including ratio (He et al., 2008; Kutser et al., 2005), or band indices (Castro et al., 2020). The relationships can be expressed as linear functions (Toming et al., 2016), power functions (Ha et al., 2017), or polynomial functions (Flores-Anderson et al., 2020), depending on the regression analysis and the dynamic range of the calibration data used (Yang et al., 2011). The empirical approach is simple, flexible, and relatively easy to implement (Matthews et al., 2010). However, the physical interpretability of models is limited, as they do not explicitly incorporate underlying physical processes, making it difficult to derive mechanistic insights. Transferability and generalization of empirical models can also be challenging when applied to different regions or time periods due to variations in water body composition and optical properties (Bukata, 2005; Oyama et al., 2009; Fukushima et al., 2016). Likewise, empirical models heavily rely on the availability of accurate and representative ground truth data, which can be limited or expensive to acquire.

Semi-analytical approach

The semi-analytical approach combines physical principles, such as radiative transfer models and bio-optical algorithms, along with empirical relationships to retrieve water quality information from remote sensing data. This approach provides a more mechanistic understanding of the underlying processes (Lee et al., 2015). Retrieving inherent optical properties allows for improved atmospheric correction, improving water quality parameter estimation accuracy. The semi-analytical approach can capture complex bio-optical relationships, making it applicable to various aquatic environments (Mishra et al., 2014). However, it requires accurate and representative input parameters, including accurate atmospheric correction and knowledge of inherent optical properties (Matthews et al., 2010). Representing the complexity of bio-optical processes and the variability of environmental conditions also poses challenges in modeling (Yang et al., 2011). Calibration and validation can be difficult, particularly when ground truth data are limited or hard to obtain (Yang et al., 2011).

Artificial intelligence techniques

The emergence of AI techniques has revolutionized water quality monitoring from remote sensing data. These data-driven techniques can handle large volumes of data and extract complex patterns without prior knowledge that may not be evident to traditional approaches (Chang & Vannah, 2013; Keller et al., 2018), and capture nonlinear relationships between remote sensing variables and water quality parameters (Rubin et al., 2021; Sudheer et al., 2006; Maciel et al., 2021). They can adapt to changing environmental conditions and integrate multi-source and multi-sensor data (Chang & Vannah, 2013), resulting in improved estimation accuracy (Sudheer et al., 2006). AI models enable automation and efficiency of water quality monitoring processes. Support vector regression (SVR), artificial neural network (ANN), and extreme gradient boosting (XGBoost) are commonly used AI models in water quality retrieval, and they have shown great success and relatively satisfactory results in many recent studies (Hafeez et al., 2019; Xiao et al., 2022; Arias-Rodriguez et al., 2021; Tian et al., 2023; Yan et al., 2023). Figure 5 illustrates a neural network structure for water quality retrieval. Although most studies using AI approaches have reported significant results, there are limitations to consider. Large and representative training datasets are required to effectively train AI models. Insufficient or biased training data may lead to poor generalization and inaccurate predictions. The black-box nature of some AI algorithms limits their interpretability, hindering the understanding of underlying processes (Petch et al., 2022). Overfitting can occur if models are overly complex or trained on limited datasets, compromising their performance on new data. Moreover, implementing AI approaches requires computational resources and expertise for model development, implementation, and maintenance.

Fig. 5
figure 5

Illustration of neural network for chlorophyll-a retrieval (Pahlevan et al., 2020)

Full size image

Satellite remote sensing

Multispectral remote sensing

A number of multispectral satellite sensors have been used for quantitative monitoring of water quality parameters. All the images are free of charge, publicly available, and provided in single and multi-day aggregate products in a variety of spatial and spectral resolutions.

Among the several multispectral satellite sensors specifically designed for ocean color measurements, Sentinel-3 Ocean and Land Color Instrument (OLCI), Medium Resolution Imaging Spectrometer (MERIS), Coastal Zone Color Scanner (CZCS), Moderate Resolution Imaging Spectroradiometer (MODIS), and Sea-viewing Wide Field-of-view Sensor (SeaWiFS) have been widely used in inland water quality monitoring. These sensors offer advantages such as repeat cycles (e.g., 1–2 days), narrow spectral bands (ranging from 400 to 1100 nm), and high radiometric resolution. However, their spatial resolution (respectively of the order of 300 m to 1.1 km) is limited to large-scale water monitoring, such as an open ocean, where chlorophyll-a is the major optically active constituent. Inland water bodies with complex optical characteristics may experience reduced estimation precision, which can hinder the monitoring of water quality changes, particularly at specific sites of interest (Topp et al., 2020). Some of the models for water quality parameters are listed in Table 4.

Table 4 Retrieval models using low spatial resolution satellite multispectral data
Full size table

There are several satellites that have sufficient spatial resolution for use in inland water quality monitoring studies. These include the Landsat program (including Landsat’s Thematic Mapper (TM), Enhanced Thematic Mapper Plus (ETM+) and Operational Land Imager (OLI)), Sentinel-2, Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), WorldView, and Satellite Pour I’Observation de la Terre (SPOT). Despite these sensors being primarily designed for land-based remote sensing applications and having less frequent satellite revisit time (3–16 days vs. 1 daily revisit for ocean color sensors), several studies have estimated and proposed reliable algorithms between the remote sensing data and water quality parameters in inland water. Examples of parameters studied include chlorophyll-a (Chebud et al., 2012; Sudheer et al., 2006; Cao et al., 2020; Lu et al., 2021), water clarity (Shen et al., 2020; Maciel et al., 2021), and total suspended sediments (Sudheer et al., 2006). The patterns observed in these satellite sensors demonstrate more detail than those of the water-based sensors due to the resolution of 10–30 m as opposed to 300 m respectively. Some of the models for water quality parameters using land-based sensors are presented in Table 5.

Table 5 Retrieval models using medium to high spatial resolution satellite multispectral data
Full size table

Although multispectral satellites provide data in different spectral bands, the number of bands and their specific wavelengths may not be optimized for water quality monitoring (Fu et al., 2018). Some water quality parameters, such as chlorophyll-a concentration, as shown in Tables 4 and 5, are typically measured using visible NIR bands, and their spectral ranges may overlap with other optically active water quality parameters. This overlap can lead to challenges in accurately estimating these parameters using multispectral satellite data. Certain water quality parameters, such as suspended particle matter and turbidity, have shown higher correlations with specific spectral combinations (Avdan et al., 2019), emphasizing the need for more specific spectral bands or higher spectral resolution than what is available in multispectral satellite data.

Researchers (Gholizadeh et al., 2016; Kutser et al., 2016; Dekker et al., 2001; Ha et al., 2017) have highlighted that the distribution of spectral bands in most multispectral satellite sensors, such as Landsat-7 ETM+, Landsat 8 OLI, Sentinel-2, ASTER, and SPOT HRV, further complicates the extraction of exact reflectance and absorbance peak points for each water quality parameter. The spectral band positioning of Landsat-7 ETM + and Landsat 8 OLI with respect to spectral characteristics of water quality parameters are presented in Fig. 6. This limitation can indirectly result in the overlapping of reflectance signature positions and omission of reflectance peak for certain parameters, leading to errors in their estimation.

Fig. 6
figure 6

Example of chlorophyll-a reflection in relation to band positions of Landsat-7 ETM+ (left) (Gholizadeh et al., 2016) and Landsat 8 OLI (right) (Ha et al., 2017). The different colors in the right image represent chlorophyll-a reflectance in nine lakes

Full size image

Hyperspectral remote sensing

The fine spectral resolution of hyperspectral satellite sensors allows for more precise and accurate identification and quantification of specific absorption and reflectance of certain water quality parameters, thus reducing the issue of parameter overlapping that commonly occurs in multispectral data. Additionally, it facilitates the development of advanced retrieval algorithms specifically tailored to estimating water quality parameters. These algorithms can exploit detailed spectral information to derive more precise relationships between spectral signatures and the corresponding water quality measurements. This leads to improved accuracy in estimating parameters such as temperature, pH, turbidity, oxidation-reduction potential (ORP), specific conductivity, high dissolved oxygen (HDO), crude oil contamination (Rostom et al., 2017), chlorophyll-a concentration (Murugan et al., 2016), total suspended solids (Tan et al., 2011; Wu et al., 2014), and colored dissolved organic matter (CDOM) (Moses et al., 2015). Hyperspectral satellite sensors that have been used in water quality studies include Hyperion on NASA EO-1 and PRecursore IperSpettrale della Missione Applicativa (PRISMA), and some retrieval models using hyperspectral satellite data are presented in Table 6.

Table 6 Retrieval models using hyperspectral satellite data
Full size table

Although hyperspectral imaging offers high spectral resolution, allowing for precise identification and measurement of water quality parameters, data processing can be complex. Zhong et al. (2021) noted that one challenge is the spectral combination problem, where multiple parameters can contribute to a single spectral signature. This makes it difficult to isolate and quantify individual parameters accurately. Another limitation of hyperspectral satellite data is its limited availability, especially for specific regions or sensors (Yang et al., 2022). This can restrict the temporal resolution of water quality monitoring in small water bodies. Similar to multispectral satellite sensors, noise and atmospheric influences and coarse spatial resolution can further complicate the analysis of hyperspectral data. These challenges can be addressed through the use of advanced data processing techniques and algorithms, as well as through the development of new and improved remote sensing technologies.

Limitations of satellite remote sensing

The use of satellite remote sensing in monitoring water quality parameters in inland water bodies has been tested and applied in several individual studies. However, developing operational and widely applicable monitoring tools is difficult due to several drawbacks.

The correction of atmospheric effects in satellite images is vital as most reflectance comes from the atmosphere (Hussein & Assaf, 2020; Moses et al., 2017). Failure to do so may lead to inaccuracies in water quality measurements. Atmospheric corrections, on the other hand, may significantly impact the satellite product (Harvey et al., 2015) and lead to uncertainty (El Saadi et al., 2014; Bresciani et al., 2017; Castro et al., 2020; Toming et al., 2016). This uncertainty is attributed to different factors, such as variations in atmospheric conditions (Werdell et al., 2010), the complexity of inland water bodies (IOCCG, 2018; Shen et al., 2020), and errors in atmospheric correction algorithms (Kutser, 2012). It has been observed that using top-of-atmosphere (TOA) measurements instead of atmospherically corrected data or bottom-of-atmosphere (BOA) data can yield better results (Toming et al., 2016; Kutser, 2012; Grendaitė et al., 2018; Tebbs et al., 2013).

In addition, the use of land-based atmospheric correction algorithms on water surfaces may lead to significant errors (Toming et al., 2016; Grendaitė et al., 2018; Maciel et al., 2021) due to the distinct optical properties and atmospheric effects specific to water (Bonansea et al., 2015; Liu et al., 2016). Several studies have compared the performance of land-based atmospheric correction algorithms with specialized algorithms developed for water quality studies (Ansper & Alikas, 2018; Rodrigues et al., 2017). For example, Wang et al. (2019) recommend the use of water-specific atmospheric correction algorithms, such as the EXP atmospheric algorithm (based on exponential extrapolation) that is integrated into the ACOLITE algorithm, to improve the accuracy of satellite-based water quality monitoring. In parallel, a study by Pahlevan et al. (2021) that evaluated the performance of different water-based atmospheric correction methods for Landsat-8 and Sentinel-2 over various water bodies, including lakes, rivers, and coastal waters, showed a performance difference of water-specific algorithms in different optical characteristics of inland waters. Therefore, specialized water-specific atmospheric correction algorithms should be used to ensure accurate monitoring of water parameters. The choice of the atmospheric correction method should consider the specific characteristics of the water body and the goals of the analysis (Tyler et al., 2006; Pahlevan et al., 2021). Similarly, validation of the atmospheric correction outputs against ground-based measurements is necessary to assess accuracy and quality.

The impact of cloud coverage on satellite images results in data gaps (Bonansea et al., 2015) and limits the availability of suitable images for analysis (Cheng et al., 2020; Olivetti et al., 2020), particularly during winter and rainy seasons. Small water bodies are especially susceptible to cloud coverage, making it challenging to obtain consistent and representative satellite imagery. The frequent occurrence of clouds and haze can further complicate the study of small areas and hinder the monitoring of water quality parameters (Olivetti et al., 2020).

Satellite remote sensing data with both high spatial and spectral resolution, suitable for water quality studies, is currently limited (Palmer et al., 2015). Many multispectral sensors mounted on satellites show potential for evaluating water quality parameters, such as Sentinel-2 and Landsat series data, which have comparable high spatial and low spectral resolution (Brezonik et al., 2009). However, their spatial resolution is generally not high enough to study small inland water bodies (McCullough et al., 2012). Moreover, sensors built solely for aquatic remote sensing, like MODIS and MERIS, have valuable narrow wavelength bands for global-scale water quality assessments (Hellweger et al., 2004; Sayers et al., 2015), but they also have limitations in studying small inland water bodies (Keith et al., 2014; Tyler et al., 2006). The lack of high spatial and spectral resolution data poses a challenge in accurately monitoring water quality parameters.

The temporal resolution of satellite remote sensing data is another significant challenge in water quality monitoring. Obtaining high temporal resolution satellite imagery for water quality studies is often not readily available. There is usually a temporal gap between satellite imagery and in situ measurements, which could be a few days to weeks, making it difficult to develop accurate regression models between water quality parameters and satellite data. For instance, Harvey et al. (2015) used MERIS images with a 0–3-day gap, while Kutser et al. (2016) obtained Sentinel-2 and Landsat-8 images with a 3-day gap. He et al. (2008) used the closest Landsat-5 images with a 9-day gap, and in some cases, it was challenging to find imagery that approximated the date of field measurement (Toming et al., 2016). Bonansea et al. (2015) noted that the maximum time lag between in situ measurements and satellite overpass should be no more than 1 day to ensure effective matching. Larger time lags may lead to discrepancies between the field measurements and satellite data, resulting in less accurate estimation of water quality parameters (Kabiri, 2023). Furthermore, it is difficult to observe rapid changes in the trophic state of water bodies, such as algal blooms, with satellite imagery due to its lower temporal resolution (Gholizadeh et al., 2016; Su et al., 2015; McCullough et al., 2012; Sayers et al., 2015; Kloiber et al., n.d.).

The limited availability of satellite imagery with high spectral, spatial, and temporal resolution has been a significant challenge in studying water quality parameters in small water bodies. To address this issue, some researchers have proposed data fusion of multiple satellite images to combine the advantages of the high spatial, spectral, and temporal resolution (Castro et al., 2020; Lai et al., 2021; Fu et al., 2018). However, this approach can be computationally demanding (Yulong et al., 2022). In studies conducted by Hellweger et al. (2004) and Niroumand-Jadidi et al. (2020), it was concluded that achieving satellite images with all three resolutions is nearly impossible due to signal-to-noise conditions. As a result, unmanned aerial vehicles (UAVs) have emerged as a promising alternative for conducting measurements instead of relying solely on satellite-based observation (Brezonik et al., 2009). UAVs have the potential to provide high-resolution images with improved spatial and temporal resolution based on the quality of their onboard sensors and camera capabilities. This allows researchers to overcome some of the limitations posed by satellite imagery. UAVs also have the ability to capture data with greater detail, enabling the detection of water quality parameters such as harmful algal blooms and pollutants at small scales.

Unmanned aerial vehicle (UAV) remote sensing

Unmanned aerial vehicles (UAVs) have emerged as a promising tool for water quality monitoring, providing high-resolution images with a spatial resolution of up to a few centimeters (Olivetti et al., 2020). This enhanced resolution enables the detailed and accurate mapping of water quality parameters, allowing for the detection of subtle changes in water quality that might be missed. Furthermore, the versatility of UAVs in capturing images at different times of the day facilitates the monitoring of diurnal cycles and can provide information on daily fluctuations in temperature, dissolved oxygen, and pH (Castro et al., 2020). One of the key advantages of UAVs is their ability to bridge the gap between in situ sampling and satellite sensors; UAVs offer a unique way to obtain water quality data at the local scale while also integrating with regional and global data sets obtained from satellites (Isgró et al., 2022). These reasons make UAVs an attractive alternative for water quality monitoring, particularly for small waterbodies that are not well served by satellite data. In recent years, several studies have explored the use of UAV remote sensing data for water quality monitoring (Zhang et al., 2023; Guimarães et al., 2019; Olivetti et al., 2020; Castro et al., 2020; Cui et al., 2022).

Multispectral remote sensing

Various multispectral remote sensing sensors such as Canon Powershot S110 and RedEdge Micasense have been used for water quality studies. In Table 7, the use of multi-temporal imagery from UAV remote sensing data shows good performance with \(R^2\) ranging from 0.84 to 0.94, and higher \(R^2\) values are observed for single images, ranging from 0.86 to 1.

Table 7 Retrieval models using UAV multispectral data
Full size table

Hyperspectral remote sensing

Similarly, hyperspectral UAV remote sensing has been used to monitor water quality parameters, as shown in Table 8. The results of these studies have demonstrated satisfactory performance with \(R^2\) ranging from 0.94 to 0.96 using multi-temporal imagery and 0.72 to 0.955 using single-date images. These findings suggest that UAV-based models can be used to expand water quality databases in both space and time.

Table 8 Retrieval models using UAV hyperspectral data
Full size table

Discussion

The studies reviewed in this paper demonstrate a strong agreement between remote sensing-driven values and field measurements, indicating the potential of remote sensing-based water quality algorithms in Case II water bodies. However, it is important to consider several factors that contribute to the consistency and robustness of these algorithms.

One significant challenge in developing robust water retrieval algorithms is the sample size. Inland water bodies exhibit spatial and temporal variability influenced by factors such as land use and point source pollution. Obtaining a small sample size can introduce potential biases in assessing water quality conditions. Several studies (mentioning some (Bresciani et al., 2017; Bonansea et al., 2018; Avdan et al., 2019; Lai et al., 2021; Harvey et al., 2015; Ma & Dai, 2007)) have relied on small sample sizes to build their remote sensing-based water quality algorithms, with only a few surface waters considered in each study. Limited observations make it difficult to capture the complex nature of inland water, resulting in reduced statistical power, which hampers drawing robust conclusions and making accurate predictions. Moreover, small sample sizes pose challenges in obtaining a sufficient number of ground truth measurements for validation purposes (Fu et al., 2018; Moses et al., 2009), which ultimately hinders effective management and mitigation strategies.

Validation plays a crucial role in establishing the accuracy and reliability of remote sensing water retrieval algorithms. It provides insights into how well algorithms capture specific optical properties and helps identify their strengths, weaknesses, and potential areas for improvement (Yen et al., 2015). However, several studies have neglected the validation step (El Saadi et al., 2014; Gons et al., 2008; Nas et al., 2010; Kloiber et al., n.d.), possibly due to limited sample sizes, as calibration and validation require a large dataset. This limitation can make it challenging to perform cross-comparisons between different sensors, platforms, and algorithms. Further, the absence of validation deprives researchers of the opportunity to identify and address deficiencies in sensors, platforms, or algorithms.

A commonly used approach in water quality assessment is the utilization of a single-date or single-image model, where a single snapshot of the water body is used. This approach can be practical when continuous monitoring or extensive temporal coverage is not required. The literature review indicates the good performance of these approaches under optimal conditions. However, water quality parameters often exhibit significant temporal variations influenced by seasonal cycles, diurnal patterns, and occasional events. By relying solely on a single-date approach, these variations are not captured, potentially leading to an incomplete understanding of the dynamics and trends of water quality over time. Furthermore, due to the fact that these models are sensitive to water-specific characteristics and the atmospheric conditions of the day, they can not necessarily be applied to other places or time frames (Rubin et al., 2021).

Conclusion

The use of remote sensing data offers advantages over conventional water quality assessment techniques, which often necessitate extensive fieldwork and costly laboratory analysis. Based on the comprehensive review conducted, it can be concluded that the advancements in remote sensing technology can support the monitoring, assessment, and estimation of various water quality parameters, including CDOM, Chl-a, TSS, TS, SD, and turbidity, among others. Several retrieval algorithms are frequently employed, such as empirical, semi-analytical, and artificial intelligence approaches.

The development of multispectral and hyperspectral satellite sensors has played a significant role in this regard, enabling high-resolution spatial and temporal observations of water bodies and offering valuable insights into variations in water quality. The integration of UAVs in remote sensing has significantly addressed limitations of low spatial and temporal resolution, atmospheric effects, and cloud conditions. UAV imagery provides high spatial resolution by operating at flexible and low-flight altitudes. This capability allows for the detection of short-term changes in small water bodies and facilitates the measurement and monitoring of water quality with the appropriate spatial and temporal resolution, even under challenging climatic conditions like clouds or haze.

Overall, this review provides a comprehensive overview of the current knowledge and applications of remote sensing in inland water quality assessment. Additionally, it offers invaluable insights into data preprocessing and analysis techniques and the limitations of remote sensing approaches in inland water quality assessment. These insights can guide researchers in selecting from alternative remote sensing approaches for water quality assessments in inland water bodies.