Real-Time Monitoring of Seawater Quality Parameters in Ayia Napa, Cyprus

Abstract

Real-time monitoring systems are crucial for the comprehensive management of operations and processes, as well as for assessing the impacts of coastal infrastructures on the marine environment. These systems not only support environmental protection and data-driven decision-making but also enable the early detection of adverse events and the issuance of timely warnings for prompt responses. Although water quality is a critical parameter in this monitoring framework, there are currently limited permanent systems in place dedicated to maintaining these objectives. Even fewer systems leverage their data for research purposes, leading to a gap in the literature regarding effective processing approaches for real-time water quality data. In this context, this study presents a real-time water quality monitoring system integrated into a broader in-field laboratory installed at a coastal area off the coast of Ayia Napa, Cyprus, as well as an initial measured data set of different qualitative quantities. It proposes a holistic approach for post-processing real-time seawater quality data, employing both time and frequency domain analyses, alongside filtering techniques. The study discusses the advantages of each method and emphasizes the importance of their combined use. Utilizing data collected from a three-month operational period, the study assesses the current state of marine seawater quality and examines both temporal and cyclic variations in various seawater quality parameters. The findings reveal that the examined seawater parameters are within reasonable values, indicating that the construction and operation of a nearby marina and the necessary infrastructures (e.g., breakwater) did not affect the seawater quality in the area. Additionally, the study identifies pronounced daily cyclic responses in different seawater quality parameters, including temperature, density, pH, dissolved oxygen, and turbidity. Finally, notable correlations are observed between temperature and dissolved oxygen, temperature and conductivity, oxidation–reduction potential (ORP) and salinity, ORP and dissolved oxygen, and ORP and TDS.

1. Introduction

Marinas, ports, harbor and jetty constructions, land reclamation, beach nourishment, and offshore energy production farms can all negatively impact the coastal environment [1]. Such impacts can include water quality deterioration [2,3,4,5,6], harm to marine organisms [2], excessive algae photosynthesis [7], and sediment erosion or deposition that alters the coastal morphology [2,5,8]. Threats can also be posed by post-construction operations. Such threats can include urban and industrial discharges [6], maritime transport, loading and unloading operations [4,9] and permanent anchorages [9]. Water quality can be affected by the implications of human activities, such as climate change and global warming [10]. Additional changes to coastal water quality can occur due to physical phenomena such as freshwater discharge, land drainage and rainfall [6,7,11], as well as wind and waves [9]. Given these potential threats, monitoring the marine environment, particularly water quality, is essential for maintaining a sustainable and safe ecosystem under various pressures. The protection of human and environmental health is the main driving force behind the development and utilization of suitable monitoring equipment [10,12].

Real-time or near real-time monitoring of water quality has gained significant attention in recent years, serving various sectors, including environmental protection, regulatory authorities, harbor and port agencies, healthcare, aquaculture, and agriculture. Real-time water quality monitoring offers important advantages over the conventional methods of assessing water quality. The conventional methods comprise manual sample collection and laboratory analyses, requiring several days for their completion and high labor costs [11,13]. They do not allow for early detection of water quality degradation, unlike real-time monitoring, which can provide early warnings for prompt action [11,14,15]. Additionally, the manual sampling–transportation–lab process required by the conventional methods can yield unreliable results due to the time lag between sampling and analysis [10], as well as possible human errors [16]. Real-time monitoring equipment overcomes these issues, providing in situ data, minimizing the need for human intervention, reducing the operational costs, and increasing the sample rates [17,18,19]. This equipment can provide data in extreme marine environments and weather conditions, which would not be accessible for conventional sampling [17,18,19,20]. It can also offer both temporal and spatial perspectives of water quality parameters that cannot be achieved through the conventional methods [7,21]. Finally, real-time monitoring can enrich international large-scale networks for acquiring and sharing environmental data [22]. Such networks are commonly referred to as the Internet of Things (IoT), and the vast amount of data they gather are typically processed using machine learning (ML) approaches [1,4,5]. In recent years, the IoT has gained increasing popularity, further driving the adoption of real-time monitoring systems [20], including applications in water quality monitoring [23].

As mentioned by Karydis and Kitsiou [12], real-time data can be used for policy making and management practices for protecting the marine environment. According to Borja et al. [6], management practices can benefit from real-time data by facilitating the understanding of how water quality deteriorates due to human activities. Also, these data can provide critical information on the timing of interventions and generally enhance the efficacy of such actions. Real-time in situ monitoring of water quality has been adopted in a wide spectrum of operations. These include monitoring drinking water [13,24], aquaculture [25,26], and groundwater [27]. Real-time in situ monitoring has found important applications regarding surface water as well. Meyers et al. [21] proposed a vessel-based monitoring system to identify fresh and saline water conditions within canals of cities by exploiting conductivity measurements. Glasgow et al. [14] and Koričan et al. [15] mentioned the use of web-based real-time monitoring for early warning and prompt notification of the need for rapid responses to algal blooms in estuaries. Ramírez et al. [11] discussed the use of a sonde equipped with various sensors that provide real-time water quality data to identify potential contaminants in rivers. Additionally, real-time monitoring demonstrates promising applications in coastal areas. Duque et al. [27] monitored salinity and electrical conductivity to identify the location of the freshwater–saltwater interface at submarine discharge conditions. Similarly, Yu et al. [28] exploited measurements of salinity and dissolved oxygen to monitor the impact of the submarine groundwater discharge to the seawater in a bay. Other studies [17,18,19] have implemented real-time monitoring systems in coastal areas, designed for prolonged operation, to enhance the understanding of how coastal waters respond to anthropogenic impacts and natural phenomena.

It is unsurprising that Jacob et al. [5] identified monitoring of various environmental parameters as the predominant measure adopted by the existing studies for managing coastal environmental conditions. However, most of the studies mentioned earlier employed real-time monitoring only for short durations, with very few utilizing permanent systems for water quality monitoring. Given the dynamic nature of marine environments, permanent systems are essential for effective long-term monitoring of seawater quality. Additionally, the relevant data gathered from the existing networks are usually restricted to private use, while available data for research purposes remain scarce. This can partly explain the lack of effective methods for interpreting water quality data. The literature lacks studies that examine the periodic behavior of water quality parameters, which is crucial for understanding their daily and seasonal variations and their overall response to environmental factors. Furthermore, to the best of the authors’ knowledge, there are currently very limited smart systems designed specifically to holistically manage the operations, processes, and their associated impacts in coastal infrastructures.

Such systems exist in two distinct coastal areas of Cyprus, where in-field laboratories were developed by the East Med Energy Research for Growth and Education Centre of Excellence (EMERGE CoE) in recent years, as described by Onoufriou et al. [29]. This study presents a real-time water quality monitoring system deployed as part of a broader network of instruments installed in one of the abovementioned areas, a newly built marine infrastructure, specifically a marina in Ayia Napa, Cyprus. The aim is to design, establish, and test a new system that will provide initial monitored data for seawater quality, thus serving as the basis for developing a database over a period of time and contributing to further research work and the extension of such systems for wider national benefits and environmental protection of the coastal areas in Cyprus. The discussion includes the equipment utilized, its deployment, the challenges encountered, and the significance of the monitored water quality parameters. A holistic real-time data post-processing approach is proposed, involving both time and frequency domain analyses. The study presents the advantages and disadvantages of each approach, highlighting the need for their combination. This combined approach enables a thorough examination of the temporal variations and cyclic responses of water quality parameters, such as daily fluctuations, while also investigating the correlations between different parameters. The data collected from three months of continuous operation indicate a negligible impact of marine construction and operations on the water outside the marine infrastructure. Additionally, a strong daily cyclic response was observed for many of the water quality parameters, along with identified correlations between the parameters.

2. Equipment

2.1. Multiparameter Monitoring Instrument: Features and Capabilities

Following a comprehensive evaluation of the requirements for seawater quality monitoring and an extensive review of market alternatives, the Aqua TROLL 600 by In-Situ Europe Ltd. (Worcestershire, UK) has been selected as the optimal choice [30]. It is a lightweight instrument, weighing only 1.45 kg, and has a cylindrical shape with a diameter of 47 mm and a length of 60.2 cm (Figure 1). Its compact design makes it easy to transport and install. Aqua TROLL 600 is a multiparameter sonde designed to accommodate four interchangeable sensors, in addition to the built-in pressure sensor. The four sensors can be chosen from a range of options, including temperature, conductivity, pH and oxidation–reduction potential (ORP), rugged dissolved oxygen (RDO), turbidity, chlorophyll a, phycocyanin, phycoerythrin, and more. The sonde can be equipped with an antifouling wiper, protecting the sensors from marine growth and mitigating its impact on the measurements. The wiper, along with the instrument’s battery life of over nine months, make the instrument well suited for extended in-field monitoring.

A great value of the instrument is its suitability for real-time monitoring as it enables real-time data transmission at the same frequency as the sampling one, which can be controlled by the user. Real-time transmission is achieved through cabled connection between the sonde and a computer. This method is preferred over wireless transmission, which, although possible, involves a more complex process of sending data from the instrument to local servers via the cloud. In contrast, a cabled connection allows for direct data transmission to an on-site industrial computer, enabling real-time data processing that is crucial for the in-field laboratory’s purposes.

2.2. Monitored Parameters

The employed sonde is equipped with turbidity, pH and ORP, RDO, and conductivity and temperature attachable sensors. All parameters monitored by the above sensors, as well as the corresponding units, are provided in Table 1. Various previous studies have monitored [6,13,21,25,26] and identified [7,10,11,12,16] these parameters as essential for assessing water quality. The importance of each monitored parameter for seawater quality is described in the subsequent sections (see Section 2.2.1, Section 2.2.2, Section 2.2.3, Section 2.2.4, Section 2.2.5, Section 2.2.6, Section 2.2.7 and Section 2.2.8), which justify their selection.

It is noted that salinity is derived from conductivity and temperature measurements and is not measured directly. In particular, it is derived by following the 2520B method of the Standard Methods for the Examination of Water and Wastewater of the American Public Health Association [31]. Specific conductivity (SC) is derived from actual conductivity (AC) and temperature (T) based on Equation (1), following the 2510B method of the same standards. Total dissolved solid (TDS) values are calculated from SC using Equation (2), where, in the absence of lab data, a conversion factor of 0.65 is adopted, as suggested by the 1030E method of the standards. An analysis of the importance of some of the monitored parameters follows.

$$\mathrm{S}\mathrm{C}={\displaystyle \frac{\mathrm{A}\mathrm{C}}{(1+0.0191\left(\mathrm{T}-25\right))}}$$

(1)

$$\mathrm{T}\mathrm{D}\mathrm{S}=0.65·\mathrm{S}\mathrm{C}$$

(2)

2.2.1. pH

pH is one of the main parameters for assessing water quality [7]. It is a measure of acidity or alkalinity of water, taking values on a scale of 0–14. pH of seawater is typically larger than 7, making it slightly alkaline due to the presence of dissolved basic minerals and the natural buffering effect of carbonates and bicarbonates [7,32]. Typical values of seawater pH range between 7.5 and 8.5 depending on the local conditions [7,13,16,26,32]. Although this parameter was found to have a moderate response to pressures, it can exhibit fluctuations. Human activities such as sewage overflows or runoff can cause significant short-term fluctuations in pH or long-term impacts that are extremely harmful to the surrounding environment. Seawater pH can decrease by acid precipitation, accidental spills or chemicals [11], as well as by the increase in organic loads [6] and the presence of industrial discharge [7,33]. pH can also decrease due to a reduction in salinity resulting from heavy rainfall events [6], which is more pertinent to ponds and lakes rather than seawater. The decrease in this parameter, known as water acidification, can reduce survival and growth of aquatic plants and animals [7,11,26,33], rendering the effective control of this parameter crucial. pH significantly affects other water quality parameters, including conductivity, hardness, and suspended solids [16].

Table 1. Monitored parameters and their units.

Sensor	Parameter	Unit
pH/ORP	pH	pH
	pH (mV)	mV
	ORP	mV
RDO	Dissolved Oxygen Concentration	mg/L
	Dissolved Oxygen Saturation	%
	Oxygen Partial Pressure (OPP)	torr
Conductivity/Temperature	Temperature	°C
	Actual Conductivity	µS/cm
	Specific Conductivity	µS/cm
	Salinity	PSU (Practical Salinity Units)
	Resistivity	ohm-cm
	Water Density	g/cm3
	Total Dissolved Solids	ppt (parts per thousand)
Turbidity	Turbidity	NTU (Nephelometric Turbidity Units)

Table 1. Monitored parameters and their units.

Sensor	Parameter	Unit
pH/ORP	pH	pH
	pH (mV)	mV
	ORP	mV
RDO	Dissolved Oxygen Concentration	mg/L
	Dissolved Oxygen Saturation	%
	Oxygen Partial Pressure (OPP)	torr
Conductivity/Temperature	Temperature	°C
	Actual Conductivity	µS/cm
	Specific Conductivity	µS/cm
	Salinity	PSU (Practical Salinity Units)
	Resistivity	ohm-cm
	Water Density	g/cm3
	Total Dissolved Solids	ppt (parts per thousand)
Turbidity	Turbidity	NTU (Nephelometric Turbidity Units)

2.2.2. Oxidation–Reduction Potential (ORP)

ORP indicates microbial activity and the activity of nitrogen cycle [26]. It is a measure of the overall oxidation of seawater and the ability of seawater to decompose waste products, including contaminants, and dead organic matter [11]. Large ORP values, i.e., larger than 200 mV, indicate seawater that can effectively cleanse itself, and therefore high-quality water. Conversely, low ORP values, i.e., lower than 50 mV, can lead to accumulation of organic matter, toxic substances, and poor water quality [26]. Measuring ORP is particularly important near the seabed, where organic matter predominantly accumulates. There, decomposition processes are highly accelerated compared to shallower waters, resulting in a significant depletion of available oxygen and consequently reducing ORP levels [11]. For this reason, Tsai et al. [26] suggests ORP monitoring at the water above seabed for effectively predicting changes in water quality.

2.2.3. Dissolved Oxygen

Similar to pH, dissolved oxygen (DO) is a vital indicator for water quality and marine health [7,11,16,26,33]. It strongly affects aquatic life as it is crucial for most of the chemical and biological reactions in the sea [7], while the harmful toxicity of CO₂ is mitigated [33]. DO is measured either as a concentration in mg/L or as a saturation percentage, which indicates the level of DO relative to the maximum capacity for full saturation. DO concentration is proportional to oxygen partial pressure according Henry’s law [32].

DO concentration larger than 5 mg/l is required for the survival of most aquatics [26], with high levels of DO indicating healthy waters [11,25]. Under these conditions, aquatics and aerobic bacteria can grow [26]. The amount of DO in water is influenced by temperature, salinity, and the rate at which oxygen is produced and consumed [34]. DO levels in seawater can decrease due to concentration of organic loads from waste discharge and nutrients [7], as well as from sewage discharges and rotting plants [11]. Also, it can decrease due to conditions that promote algae photosynthesis and high rates of respiration [7]. A decrease in DO can occur in warm weather because the oxygen-holding capacity of water decreases as temperature increases [33]. Due to DO’s dependence on temperature and pressure, its concentration can exhibit spatial variation in the sea [26]. It can also vary with changes in salinity, turbulence, atmospheric deposition, and biological oxygen demand [7,16]. Unlike pH, DO is a sensitive parameter of water quality as changes in its concentration can occur within some hours from the time pressure conditions are applied on the sea environment [7].

2.2.4. Conductivity

Seawater conductivity is a measure of the ability of the water to conduct electricity [7,16], and it is the reciprocal of resistivity. It provides an indicator about water’s ionic content, which in turn influences parameters such as hardness, alkalinity, and certain dissolved solids [16]. Although not the most widely used indicator [16], high conductivity can indicate low water quality, characterized by high concentration of inorganic salts and impurities [11,26]. Conductivity measurements were used by Duque et al. [27] to determine the interface between freshwater and saline water in a coastal area affected by submarine groundwater discharge conditions.

Conductivity measured directly by a sensor is referred to as actual conductivity and corresponds to the site’s water temperature conditions. Actual conductivity is usually converted to specific conductivity to describe the water conductivity at a standard temperature of 25 °C. In addition to its dependence on temperature, conductivity can vary with pH, dissolved oxygen, turbidity, and total dissolved solids [16].

2.2.5. Salinity

Salinity reflects the concentration of dissolved salts and ions in water. This parameter is crucial for the growth and reproduction of marine life [7] as excessive salinity levels can pose a threat to aquatic organisms. Salinity increase can lead to aggregation of suspended particles, affecting the water quality in coastal areas [35]. Moreover, high salinity can accelerate corrosion of steel components of marine infrastructure [11,36]. It is well known that this parameter is inherently associated with conductivity, allowing it to be calculated from conductivity measurements [7,37]. Similar to conductivity, salinity monitoring data can be used to identify the interface between freshwater and saline water in submarine groundwater conditions. Salinity data have also been used to comprehend the mixing conditions between bay seawater and outer seawater [27].

2.2.6. Temperature

Temperature is one of the most vital water quality parameters for all aquatic life, impacting reproduction and growth [16,26,33]. Given that each species requires specific temperature, temperature fluctuations can diminish their disease resistance and increase their susceptibility to infections, thereby hindering their survival [26]. It influences gas transfer rates and the amount of dissolved oxygen. Higher temperatures can lead to decreased dissolved oxygen levels, posing health risks to aquatic life. Additionally, as mentioned earlier, water temperature has a significant impact on water conductivity, while it is less correlated with pH [16].

2.2.7. Turbidity

Turbidity is a measure of water transparency, clarity, and suspended particles, including soil, plant waste, and other organisms [11,13]. Low turbidity levels indicate clear water and health conditions for aquatic life. Conversely, prolonged high levels can reduce the survival of aquatic organisms and inhibit photosynthesis in submerged aquatic plants [11]. Excessive levels can also pose health risks to humans, potentially exposing them to various illnesses [13]. Turbidity is found to be highly sensitive to various factors, such as rainfall, land drainage, and freshwater discharge [6]. Additionally, in marine and coastal environments, it can increase by the effects of industrial waste, construction, soil erosion due to runoff, and excessive phytoplankton growth [7]. According to Borja et al. [6], monitored turbidity data are particularly useful for detecting wastewater discharge due to their aforementioned sensitivity to this factor. It is associated with water conductivity, hardness, and total dissolved solids [16].

2.2.8. Total Dissolved Solids (TDS)

TDS refer to the concentration of the remains of inorganic and organic solids in seawater. They are closely associated with turbidity, conductivity, hardness, and total suspended solids. Moreover, an increase in TDS can result in higher salinity levels [16]. Increased TDS have comparable effects to turbidity. They reduce water clarity, elevate water temperature, lower oxygen levels due to reduced photosynthesis, and facilitate the binding of solids to toxic compounds and heavy metals [33].

2.3. Previous In-Field Monitoring Using the Employed Equipment

The equipment adopted for the purposes of this study, Aqua Troll 600 multiparameter sonde, has been utilized in previous research studies to monitor water quality parameters in coastal areas [17,18,19,27,28] and rivers [11,21]. The monitoring systems referred to were either stationary, mobile, or installed for a short duration of time before removal. These studies have employed the equipment for monitoring durations ranging from some hours to several days.

Ushlkov et al. [38] showed that this equipment has the potential to be used for long-term real-time monitoring and be an integral part of an IoT system of coastal environments. By leveraging IoT services, the collected data can be transmitted to the internet. The authors also developed software and hardware for processing the monitored data, facilitating the effective protection of the marine environment. Ramirez et al. [11] proposed a monitoring system featuring the Aqua TROLL 600 as its main component for long-term water quality monitoring in a river, with data transmitted over a 9 km optical fiber connection. The system was tested in a laboratory using various liquid solutions before being experimentally tested on site at two locations. The development of this network aimed at identifying potential contaminants in river water through continuous real-time monitoring, along with associated data processing and analysis.

Duque et al. [27] employed 30 Aqua TROLL 600 units uniformly distributed over a coastal area of approximately 150 m² to identify the interface between freshwater and saltwater through conductivity measurements. The monitoring campaign involved several two-hour intervals of monitoring before the monitoring system was withdrawn from the site. Salinity was inferred from conductivity data, enabling the identification of freshwater and saltwater of interface areas. Similarly, Yu et al. [28] used the same equipment to measure temperature and salinity at various locations within a bay, but not in real time. Analyzing the distribution of these parameters in the study area, they found that salinity was higher at the mouth of the bay and lower at the top, while the opposite trend was observed for temperature. Additionally, salinity measurements were used to assess the mixing conditions between the bay seawater and the outer seawater.

Unlike the above stationary systems, Snow et al. [18] and Bennett et al. [17,19] proposed a monitoring system for measuring water quality at various depths and positions in a coastal area. The design comprised an inclined cable-mounted Aqua TROLL 600 sonde, with the cable fixed on the deck on the shore side and anchored at the seabed on the sea side. The sonde can move along the inclined cable, allowing the system to gather high-frequency (i.e., 1 per minute) water quality data from various positions. However, real-time monitoring is not feasible as data can be retrieved when instrument is removed from the sea. Meyers et al. [21] designed another mobile water quality monitoring system that included a vessel-mounted Aqua TROLL 600 sonde. The vessel continuously navigates through an urban canal network, capturing the spatial variations in the monitored water quality parameters. The data are transmitted in real time through a cloud-based IoT system and stored in a database.

3. Deployment

3.1. Site

The equipment is installed in a coastal area South–East of Cyprus, near a recently built marina (Ayia Napa Marina). The Ayia Napa Marina provides an ideal testbed for innovative digitization solutions, given the substantial coastal interventions required for this project and the high density of new marine infrastructure, operations, and processes in a relatively confined area. Primarily serving the tourism sector, the area supports dense onshore and offshore activities, which can pose significant environmental risks. Any pollution in this region could significantly affect the local ecosystem and potentially lead to a decline in tourism, thereby impacting the country’s economy. To address these challenges, an in-field laboratory has been established, equipped with accelerometers, a weather station, a wave reader, and a water quality monitoring system [29]. The water quality monitoring system enables the early detection of extreme or unexpected events, such as oil spills, and provides early warnings to ensure prompt responses and environmental safety.

3.2. Installation and Connectivity

The equipment is part of a broader in-field laboratory developed in recent years [29] by the EMERGE CoE in Ayia Napa Marina, designed to function as a real-time data management system for monitoring environmental and structural quantities. Specifically, the instrument for water quality monitoring is positioned near the seabed outside the marina at a horizontal distance of 25 m from the round head of the outer breakwater, as shown in Figure 2. The instrument is connected via a cable to an industrial computer, which is permanently installed at the breakwater at the location indicated in the same figure.

The equipment was installed under the guidance of the EMERGE research team in collaboration with a specialist technical team from Ayia Napa Marina. The operation engaged engineers, technicians, divers, and boats, all crucial in supporting and coordinating the safe installation of the equipment. The anchoring system comprises a reinforced concrete block weighing approximately 45 kg and a 30 cm mooring line that connects the instrument to the anchor, as depicted in Figure 3a. The instrument is positioned to maintain vertical orientation, achieved by connecting the back of the instrument with a float, as demonstrated in Figure 3b. This setup ensures that its sensors face the seabed while the cable faces the surface, preventing entanglement between the mooring line and the cable. The cable runs along the seabed, where it is anchored at multiple points. As it approaches the breakwater, it is elevated and crosses over it to connect with the industrial computer installed in that location. At the breakwater, the cable passes through a flexible plastic tube to protect it from frictional forces induced by the contact with the breakwater’s tetrapods (Figure 3c).

Following the installation of the equipment, continuous real-time monitoring of the water quality parameters listed in Table 1 is conducted. These data are initially stored temporarily on the industrial computer before being transmitted via the cloud to the university’s local servers. The next stage in data transmission involves storing the data in a dedicated database designed to organize all information gathered from the two in-field laboratories in a user-friendly interface [29].

3.3. Deployment Considerations

The deployment of the water quality real-time monitoring system was carefully designed to ensure reliable data collection, real-time data processing, and durability in the harsh marine environment. One of the primary challenges in marine networks is data transmission. Underwater wireless signal transmission poses significant difficulties and is rarely used for such deployments [23]. An alternative is to use a wired connection between the underwater instrument and a buoy equipped with a cellular telemetry device for transmitting data to a cloud-based database. This deployment can work effectively in cases where a physical computer is not accessible in situ. However, cellular data transmission is challenged by connectivity issues and electricity power demands [10]. Additionally, it does not allow real-time data processing that is necessary for the purpose of developing the in-field laboratories in Cyprus, described by Onoufriou et al. [29].

For real-time data processing, this study suggests using a direct cable connection between the instrument and the computer, provided that physical access to a computer is available. As mentioned by Jahanbakht and Lanzo [23], cable connection is suitable for data and energy transmission, while optical fibers should be preferred for image and video transmission, which is beyond the scope of the proposed monitoring system. As noted by the same authors, cable connections have some significant drawbacks to be considered in the deployment design. These include high maintenance costs, challenging installation and retrieval processes, limited flexibility in spatial monitoring, lack of security against vandalism and random events, and an increased probability of failures due to loads induced by waves and currents. The described system is installed and will be maintained by the Ayia Napa marina stakeholders, who possess extensive related expertise. Provisions have been made for regular inspections and maintenance of the mooring and anchoring system, including biofouling removal. The deployment method, as outlined earlier, facilitates straightforward retrieval and repositioning of the instrument, thereby simplifying calibration. Additionally, comprehensive maintenance and management plans were implemented to ensure the long-term durability and effectiveness of the in-field laboratories under dynamic marine conditions.

3.4. Sampling Frequency

The availability of real-time monitoring equipment has significantly increased the sampling frequency in comparison with conventional laboratory-based water quality measurements. Typically, these instruments allow users to configure the sampling rate, with previous studies reporting monitoring intervals as short as 1 min [17]. The optimum selected sampling frequency is the smallest one that enables the estimation of mean values for the monitored parameters with a certain confidence level [20] while also capturing extreme values. A high sampling frequency can result in a large database and increased operational costs. A sampling frequency that is too low might obscure the maximum and minimum values, and also affect the accuracy of other statistical values such as the mean values. This is particularly relevant for dynamic environment impacted by transient events and anthropogenic pressures [39]. Such environments include marinas, which can be affected by boats spills, industrial discharges, or construction, which can promote rapid alterations to the water quality.

As demonstrated by Vilmin et al. [39], the selection of sampling frequency depends on various project-specific parameters. These include the sampling location, the sensitivity of the monitored parameters to environmental factors and anthropogenic pressures, the period of the study, and the monitoring goals. Low sampling frequency may be adequate for projects focused on examining the long-term variations regarding certain parameters through extended monitoring periods. However, the purpose of the current project is to verify the reliability of the monitoring system and investigate the variations in the parameters within a short pilot period of time (i.e., three months). Therefore, a higher sampling rate is required. Moreover, a higher sampling frequency is required due to the fact that one of the aims of the broader monitoring system is to enable early warning in case of pollutants in the area around the marina. Considering that a spill of one ton of oil can spread across a 50 m radius within 10 min [40], it is crucial to take measurements at frequent intervals. This allows authorities to promptly respond and take necessary actions to mitigate such incidents. A pertinent early warning system proposed by Koričan et al. [15] collected data every three minutes, which were transmitted to the platform for analysis every 15 min.

For the purposes of the current study, the adopted multiparameter monitoring instrument was configured to measure seawater quality parameters at 30 min intervals, with the measurements transmitted to the industrial computer immediately after each measurement. This interval is considered sufficiently low to capture transient variations in the monitored parameters listed in Table 1 with adequate resolution. It is low enough to enable early warnings for spills or other events while avoiding excessive data accumulation. It is noted that the sampling frequency was temporarily set to one sample every 10 min during the instrument’s assessment in the laboratory. This adjustment enabled a more detailed examination of the temporal variations in the monitored parameters.

4. Data Quality Assessment and Processing

The instrument outputs data in meaningful units, organized in parallel columns, all corresponding to a common time and date column. Data interpretation includes analyses in the time and frequency domains.

Time domain analysis involves presenting time histories of the monitored variables, with the x-axis being discretized in week units for better visualization of the results. These figures are used for identifying unreliable measurements that can be incorporated in real-time in situ data. Such measurements result from various error sources, such as sensor malfunction, calibration drift, biofouling, contamination or encrustation of the sensors, signal loss, data corruption logging, or transmission. To ensure data quality, the collected data were thoroughly assessed before further analysis. Time series were examined for unreasonable trends or deviations from values expected based on the literature. The expected values were also compared with mean and extreme values, output from statistical analyses that were performed. In particular, the reliability of the measurements was ensured by comparing the above with measurements in both tap water (see Section 5) and seawater (see Section 6). Tap water measurements were conducted in a controlled laboratory environment, ensuring that factors such as biofouling, contamination, or sensor encrustation were not present, allowing the focus to remain on other potential sources of error. The aforementioned sources of error are also unlikely to have impacted the seawater measurements during the three-month pilot operation, as no marine growth was observed on the instrument upon retrieval.

Time series are also used to identify the variability in monitored water quality parameters over time. Additionally, by leveraging the common date time stamp across different datasets, two variables are presented together to identify potential correlations. This investigation is further supported by the use of Pearson correlation and linear regression analysis, comprising statistical tools for assessing the linear relationship between two variables. Finally, time series can be interpreted to estimate probabilities of occurrence, which can be useful for infrastructures design.

Although time series can indicate time variability, they do not provide evidence of cyclic patterns in the monitored data, such as daily, seasonal, or annual cycles. The period of occurrence of these cycles can be determined through analysis in the frequency domain, typically facilitated by the use of the Fourier transform. Fourier transform is widely used for interpreting signals with cyclic characteristics, such as those from earthquakes or dynamic motions on structures and soil [41,42], while it has also been adopted for interpreting real-time water quality data [43]. The present study computes the discrete Fourier transform of the recorded time series to produce Fourier spectra, providing insights into the significance of certain frequencies (or, equivalently, periods). The ordinates of Fourier spectra refer to the Fourier amplitude (FA), while the abscissas denote periods (P) instead of the more conventional frequency to better visualize recurrence times. Any peak in the spectra signifies a strong contribution from the corresponding period to the signal, indicating a prominent cyclic pattern with the specific period.

Frequency domain analyses are also employed to identify relationships at specific periods or frequencies that cannot be identified by time domain analyses. They also have the potential to identify correlations that could not be found from time domain analysis when data are contaminated with high levels of noise. Aiming to provide a more comprehensive interpretation of water quality data, this study conducts a complementary investigation of the linear relationship between two signals in the frequency domain. This can be achieved through the use of cross-power spectral density (CPSD), phase difference, and coherence between two monitored time series of different water quality parameters. These have been widely used in earthquake engineering, and their mathematical derivations can be found in [44].

The CPSD between two signals is a complex number comprising both magnitude and phase. Its magnitude expresses the correlation between the signals at certain periods (or, equivalently, frequencies), with peaks indicating strong correlations. The phase of a CPSD indicates the phase difference between the two signals at each period, ranging from −π to +π radians. A phase of zero indicates that the signals are in phase, while values near −π or +π suggest that the signals are out of phase. Coherence is a measure of the degree of linear correlation between two signals in the frequency domain, with values ranging from 0 to 1. Higher coherence values indicate stronger linear correlation, while values close to zero suggest no relationship, a nonlinear relationship, or significant noise interference. Two signals are linearly related at periods or frequencies where the CPSD exhibits peak magnitudes, coherence values are near unity, and phase differences are close to zero, −π or +π. This approach enables the assessment of the correlation of water quality parameters in periods with low noise levels while also identifying periods where high noise is prominent.

Finally, frequency domain analyses are used for data noise cancellation. Field measurements are frequently affected by high-frequency noise due to the sensitivity of the sensors. The method used in this study for noise removal is detailed in Section 6.2.

5. Lab Testing

Prior to deploying the water quality monitoring instrument in the sea, as described in Section 3, it underwent calibration and testing in a laboratory setting. The calibration process adhered to the manufacturer’s guidelines, with each sensor being individually calibrated using designated solutions. This solution-based calibration involved submerging the sensors into specific solutions provided by the manufacturer. The instrument then performed an automatic calibration procedure, facilitated by a smartphone application, also supplied by the manufacturer, which connects the instrument to the phone for exchanging information and commands.

Following calibration, the instrument was submerged in tap water in the laboratory under controlled conditions. It remained in this setting for three weeks, monitoring the parameters listed in Table 1.

Table 2. Statistical values of tap water quality parameters monitored in the lab, superimposed by parametric values aligned with Directive (EU) 2020/2184 [45] and reference values provided by the State General Laboratory of the Republic of Cyprus [46].

Parameter	Mean	Min	Max	Parametric Value Aligned with EU Directive	Reference Values
pH (pH)	8.90	8.62	9.50	≥6.5 and ≤9.5	8.5–8.7
DO Conc. (mg/L)	6.16	5.67	7.16	-	7.8–8.2 *
Specific Cond. (µS/cm)	633.15	586.13	687.06	≤2500	760–900
TDS (ppt)	0.41	0.38	0.45	-	0.3–0.4
Turbidity (NTU)	0.0044	0	1.3	Acceptable to consumers and no abnormal change	0.3 *

* values measured at water treatment plants, which might change before consumption.

presents statistical values (mean, min, and max) of pH, DO concentration, specific conductivity, TDS, and turbidity. The Directive (EU) 2020/2184 on the quality of water intended for human consumption [45] specifies acceptable limits for pH, conductivity, and turbidity. The recorded values for these parameters comply with the specified limits, as indicated in the table. Additionally, all recorded parameters closely align with measurements taken by the State General Laboratory of the Republic of Cyprus [46]. With the exception of turbidity, all parameters included in Table 2 exhibited limited fluctuation.

The variability in the measurements over time is better illustrated through the time series presented in Figure 4. Although very subtle, pH, actual conductivity, salinity, temperature, and TDS showed increasing trends over time, whereas ORP and density exhibited decreasing trends. These trends can be attributed to impacts from environmental factors. One significant factor is the daily environmental temperature variation, the impact of which is evidenced by the pronounced cycles observed in water temperature. Similar albeit weaker cycles can be observed in all other parameters depicted in Figure 4, except for turbidity.

The period of occurrence of these cycles can be determined through Figure 5, which presents the Fourier spectra of the time series of variables in Figure 4, which have already been discussed. Figure 5 demonstrates that all variables exhibited daily variations, as indicated by the peaks at periods of twenty-four hours. As expected, the daily cycles of the monitored data were very pronounced on the water temperature, as indicated by the prominent peak observed in the corresponding Fourier spectra. Pronounced daily cycles were also observed in density, pH, actual conductivity, and turbidity. It is noted that an efficient way to interpret water quality data should encompass both the analysis of time series and the corresponding Fourier spectra. Relying solely on time series analysis may lead to overlooking the predominant periods of the signal. Conversely, interpreting only the Fourier spectra may result in a misunderstanding of the parameter’s variation over time. An example of the latter is illustrated in Figure 5h, which gives the impression that density varies significantly on a daily basis, whereas the variation is minimal, as shown in Figure 4h.

Overall, lab testing produced data that closely matched reference values found in the literature. Additionally, the time variations in the monitored parameters were within the expected ranges. These observations verify the reliable operation of the instrument and the credibility of the data obtained.

6. In-Field Monitoring

The pilot monitoring system operated for three months, monitoring the seawater quality in a coastal area southeast of Cyprus near a newly built marine infrastructure.

Table 3. Statistical values of seawater parameters monitored in a coastal area of Cyprus compared with reference values for nearby coastal regions and in general coastal waters.

	Measured			Reference Values
Parameter	Mean ± SD	Min	Max	For the Coastal Area of Cyprus [47,48] ** Mean ± SD (min–max)	For Coastal Waters in General
pH (pH)	8.25 ± 0.05	8.17	8.36	7.9 ± 0.5 (6.6–8.8)	7.9–9.0 [26]
ORP (mV)	210.33 ± 36.46	128.13	274.22	-	100–500 [49,50]
DO conc. (mg/L)	5.13 ± 0.22	2.89	5.74	7.8 ± 0.4 (1.6–8.6)	1.5–8.5 [50] but typically 7–8
DO sat. (%)	80.40 ± 4.49	40.24	92.54	-	-
Actual Cond. (mS/cm) *	41.74 ± 17.98	10.92	76.11	56.3 ± 4.8 (0.8–67.2)	30–60 [51]
Specific Cond. (mS/cm) *	36.42 ± 15.26	9.77	66.44	-	-
Salinity (PSU) *	23.78 ± 10.75	5.58	46.11	37.7 ± 2.9 (0.86–44.1)	31–36 [28]
Temperature (°C)	32.19 ± 1.34	29.50	34.59	16 (winter), 26 (summer) (15–36) ***	-
Turbidity (NTU)	38.54 ± 223.21	0.00	4410.20	-	0.1–100 [52]
TDS (ppt) *	23.67 ± 9.92	6.35	43.19	-	19–40 [53,54]
Density (g/cm3) *	1.01 ± 0.008	1.00	1.03	-	1.0

* The statistical values of the measured data are calculated using data recorded during the first seven weeks of operation. ** Values are inferred from a statistical analysis conducted by Hadjisolomou et al. [47,48] by exploiting data monitored by the Department of Fisheries and Marine Research of the Cyprus Republic. *** Values refer to surface water.

presents the statistical mean, minimum, maximum, and standard deviation (SD) values of the monitored parameters, along with reference values for the coastal waters of Cyprus and coastal waters in general. For most of the parameters, the recorded values fall within the range of the reference values, indicating that the construction and operation of the marina have not adversely affected these specific water quality parameters. A discrepancy between the recorded and reference values is observed for temperature, where the sensor provided unexpectedly high readings for seawater at a depth of 10 m during winter. This suggests a malfunction of the temperature sensor, underscoring the need for its maintenance. Additionally, the dissolved oxygen concentration levels appeared to be low. This can be attributed to the measurements being taken near the seabed, where the concentration is typically lower compared to near-surface water [11].

6.1. Time and Frequency Domain Data Interpretation

For some of the parameters presented in Table 3 (i.e., ORP, actual conductivity, salinity, turbidity, and TDS), the minimum and maximum values deviated significantly from the mean. However, interpreting these statistical values alone cannot ascertain whether the extreme values result from an isolated event (e.g., spurious spikes) or are part of a sustained trend. A better visualization of the time variations in the water quality parameters can be provided by interpreting the recorded time series. Figure 6 presents the time series of selected monitored variables. The figure illustrates relatively stable values for pH, ORP, dissolved oxygen, and density. The water temperature initially increased during the first two weeks before declining as a result of the decrease in environmental temperature. The conductivity, salinity, and TDS values showed unexpectedly rapid decreases after seven weeks of operation. Since all these parameters were measured from the same sensor (see Table 1), it can be concluded that the sensor began malfunctioning in the eighth week, underscoring the need for calibration.

Additionally, time histories can be interpreted to estimate probabilities of occurrence, which can be useful for infrastructure design. Although three months of operation are not sufficient to establish reliable design parameters,

Table 4. Values of water quality variables corresponding to the 10th, 20th, 50th, and 80th percentiles of probability.

	pH	ORP	DO (Sat.)	AC *	SC *	Salinity *	Temp.	Turbidity	TDS *	Density *
P10	8.188	161.04	69.71	47,120.56	41,153.48	26.84	30.02	0.00	26.75	1.015
P20	8.201	163.89	74.68	50,121.00	43,793.11	28.77	30.54	0.04	28.47	1.016
P50	8.259	205.05	80.72	55,547.79	48,202.94	32.04	31.68	7.39	31.33	1.018
P80	8.314	246.49	83.34	61,997.15	52,618.17	35.37	33.60	25.39	34.20	1.020

* The correlations for these parameters were carried out using data recorded during the first seven weeks of operation. DO (Sat.): saturated dissolved oxygen; AC: actual conductivity; SC: specific conductivity; TDS: total dissolved solids.

provides indicative values for the monitored water quality parameters that correspond to the 10th, 20th, 50th, and 80th percentiles of probability (P10, P20, P50, and P80). A value at a given percentile is greater than the corresponding percentage (i.e., 10%, 20%, 50%, or 80%) of all the data. The table verifies the small variability in pH, ORP, temperature, and density, contrasted with a broader distribution observed in the remaining parameters.

As previously mentioned, time series are not particularly useful for studying cyclic events. In the context of water quality parameters, these events include daily, seasonal, and annual variations in temperature and other parameters associated with temperature. Given that the monitored duration for the present study was three months, daily variation is the only apparent cyclic event that can be studied. As discussed in Section 4, cyclic events can be more effectively interpreted in the frequency domain through Fourier transform. Figure 7 presents the Fourier spectra for the water quality parameters shown in Figure 6. These spectra were inferred from the time histories that were considered to be reliable based on the previous discussion. They focus on the periodic responses within a timeframe of 48 h. In the spectra of pH, ORP, dissolved oxygen, temperature, and turbidity, a pronounced peak in amplitude is observed at a period of approximately twenty-four hours. This indicates that these parameters predominantly vary on a daily basis, with temperature showing the most significant variation. Additionally, the pH and dissolved oxygen measurements show a cyclic variation within a twelve-hour period. This variation is weaker than the daily one, as indicated by the smaller peak amplitude at a period of twelve hours in the corresponding spectra (Figure 7a,c).

6.2. Data Noise Cancellation

In-field measurements are usually contaminated with high-frequency noise, mainly due to the sensitivity of the sensors. Jahanbakht and Lanzo [23] outlined noise cancellation methods used for water quality monitoring data. The most widely used method is the lowpass filter, which attenuates the contribution of frequencies above a specified cutoff frequency in the signal. The selection of this cutoff frequency is crucial; it should be sufficiently low to filter out high-frequency content while ensuring that frequencies of interest are retained. Filters can be classified as either causal or acausal. Causal filters introduce phase shifts to the signal, whereas acausal filters achieve zero phase shift [55]. Alternatively, the moving average approach can be used to smooth out high-frequency peaks. The time window over which the average values are calculated should be carefully selected. It should be wide enough to enclose high-frequency noise, but it should not be wider than the time periods of interest. This study employs both an acausal lowpass filter and time average methods.

The water quality parameter that exhibited excessive high-frequency noise is turbidity (Figure 6h). Turbidity data undergo filtration using a fourth-order lowpass Butterworth filter with a cutoff frequency set at the reciprocal of half a day (2/day = 2.31 × 10⁻⁵ Hz, equivalent to a twelve-hour period). Additionally, a moving average approach is employed, utilizing a sliding window spanning twelve hours. The cutoff frequency and window width are selected with the consideration that the cyclic variation in turbidity within twelve hours is not critical for this study’s objectives. These values enable filtering out high-frequency variations in turbidity within this time window while emphasizing the daily cyclic variations, which are of primary interest.

Figure 8 depicts the outcomes of applying filtering and moving average methods, illustrating the impact of noise cancellation on (a,b) time histories and (c) Fourier amplitude spectra. Both methods effectively reduce the high-frequency peaks, aligning the results more closely with the values reported in the literature [52]. The suppression of high-frequency content (short periods) is evident from the small amplitudes observed in the Fourier spectra at short periods, as depicted in Figure 8c. The same figure also illustrates that the rigorous filtering ensures that the signal content at periods larger than the cutoff (twelve hours) remains unaffected, demonstrating its effectiveness in noise cancellation. In contrast, the moving average approach impacts periods longer than the cutoff, thereby rendering this approach less suitable. Another drawback of the moving average approach is its potential to eliminate single-time extreme events that may be crucial for early detection and early response purposes. In contrast, rigorous lowpass filtering selectively removes cyclic content with frequencies above the cutoff, preserving single-time events.

6.3. Water Quality Parameter Correlations

The linear dependence between any two water quality parameters is evaluated using the Pearson correlation coefficient. This coefficient ranges from −1 to 1, where values closer to −1 or 1 indicate a stronger linear relationship, either negative (inverse) or positive (direct). The resulting correlation is considered to be significant when the associated probability p-value is smaller than the predetermined significance level, typically assumed to be 5%. In this case the null-hypothesis, which states that there is no relationship between the two parameters, is rejected.

Table 5. Correlation coefficients between parameters, with black numbers indicating very strong correlations.

	pH	ORP	DO (Sat.)	AC *	SC *	Salinity *	Temp.	Turbidity	TDS *	Density *
pH	1	0.070	0.307	−0.664	−0.589	−0.593	−0.403	0.089	−0.589	−0.517
ORP	0.070	1	0.590	−0.796	−0.804	−0.809	0.583	−0.013	−0.804	−0.802
DO (Sat.)	0.307	0.590	1	0.649	0.687	0.692	0.605	−0.013	0.687	0.707
AC *	−0.664	−0.796	0.649	1.000	0.999	0.999	0.869	−0.158	0.989	0.955
SC *	−0.589	−0.804	0.687	0.999	1.000	1.000	0.854	−0.142	1.000	0.986
Salinity *	−0.593	−0.809	0.692	0.999	1.000	1.000	0.850	−0.142	1.000	0.986
Temp.	−0.403	0.583	0.605	0.869	0.854	0.850	1.000	−0.095	0.854	0.036 **
Turbidity	0.089	−0.013	−0.013	−0.158	−0.142	−0.142	−0.095	1	−0.142	−0.126
TDSs *	−0.589	−0.804	0.687	0.989	1.000	1.000	0.854	−0.142	1	0.986
Density *	−0.517	−0.802	0.707	0.955	0.986	0.986	0.036 **	−0.126	0.986	1

* The correlations for these parameters were carried out using data recorded during the first seven weeks of operation. ** The correlation is not significant as the probability value is larger than the significance level (5%). DO (Sat.): saturated dissolved oxygen; AC: actual conductivity; SC: specific conductivity; TDS: total dissolved solids.

presents the resulting correlation coefficients between all the pairs of the monitored parameters. Except for the temperature–density pair, all the correlations are statistically significant, with p-values smaller than the predefined significance level of 5%. The p-values are provided in Appendix A. Table 5 shows a very strong linear correlation between salinity, TDS, conductivity, and temperature. However, the correlations among these parameters, except for the conductivity–temperature pair, may be considered biased. This bias arises because salinity and TDS were not directly measured in situ; they were derived from the conductivity and temperature measurements (see Section 2). Nevertheless, the strong correlation between these variables verifies the applicability of the method. Another outcome of the Pearson correlation analysis is that ORP shows a strong linear relationship with conductivity, salinity, TDS, and density, whereas density is linearly correlated with conductivity, salinity, temperature, and TDS. pH shows a moderate correlation with conductivity, salinity, temperature, and TDS, while temperature is moderately correlated with pH, ORP, and DO. Similarly, DO exhibits a moderate correlation with ORP, conductivity, salinity, temperature, and TDS. Finally, turbidity was found to be very weakly correlated with any other water quality parameter.

Linear correlations between variables are better illustrated through linear regression analysis. Figure 9 illustrates the analysis results by plotting pairs of water quality parameters that exhibited moderate to strong correlations, as indicated in Table 5. The data points are superimposed by the associated predicted linear line, its equation, the coefficient of determination (R²), and the p-values. R² is a measure of the goodness of fit between the measured and predicted responses, with values closer to the unity indicating a better prediction. All the p-values are below the 5% significance level, indicating that all the observed correlations are statistically significant.

As previously mentioned, salinity and TDS were derived from the conductivity measurements. This explains the precise linear relationship between actual conductivity and salinity, as illustrated in Figure 9a. Similar relationships were observed between conductivity and TDS, as well as salinity and TDS, but are not presented herein for brevity. Figure 9b,c demonstrate that the increase in temperature leads to a pronounced increase in both salinity and conductivity. It also results in a decrease in dissolved oxygen (Figure 9d), aligning with previous knowledge [33]. Due to the established relationships between temperature, conductivity, and salinity, dissolved oxygen is also found to decrease with actual conductivity (Figure 9e) and salinity (Figure 9f), while a moderate decrease is observed with the increase in ORP (Figure 9g). Although further verification is required due to data scatter, an increase in ORP appears to be associated with a decrease in salinity and TDS levels (Figure 9h,i). Finally, no clear relationship between pH and salinity can be discerned from the monitored data, as indicated by the scatter plot in Figure 9j. Consequently, the expected increase in pH with rising salinity, as reported in the literature [6], cannot be verified.

A disadvantage of the Pearson correlation coefficients presented in Table 5 and the linear correlation illustrated in Figure 9 is their potential unreliability when the signals are contaminated with high-frequency noise. Additionally, these methods do not capture relationships at specific periods or frequencies, such as the daily correlation between two parameters. To address these issues and provide a more comprehensive interpretation of water quality data, the linear relationships between two signals were investigated in the frequency domain. This was achieved through the use of cross-power spectral density (CPSD), phase difference, and coherence between two monitored time series of different water quality parameters. The interpretation of this approach has been explained in Section 4 [44].

Figure 10a provides the CPSD, phase difference, and coherence between actual conductivity and salinity, which, as previously explained, are interconnected. The figure demonstrates the applicability of the method as it verifies that the two variables exhibit a linear relationship for the majority of the periods (i.e., >12 h). In these periods, the CPSD shows large magnitudes, the coherence values are very close to the unity, and the phase difference is zero. Figure 10b,c depict the same spectra computed for temperature and actual conductivity and temperature and salinity, respectively. Both figures indicate strong correlations (high CPSD amplitude) and nearly zero phase difference between the two examined pairs of water quality parameters during a twenty-four-hour period. However, the low coherence values (around 0.5) indicate a moderate degree of linear relationship. This observation between temperature and salinity supports the validity of the proposed method given that salinity was derived from a nonlinear relationship with temperature, as explained in Section 2.2.

Figure 11 presents the CPSD, phase difference, and coherence to evaluate the relationships between dissolved oxygen and temperature, as well as actual conductivity and ORP. Dissolved oxygen exhibits a linear in-phase correlation with temperature and actual conductivity at a period of twenty-four hours. This is evidenced by the peak in CPSD, the nearly zero phase difference, and high coherence at this period. As shown by Figure 11c, dissolved oxygen is also correlated with ORP, but the response of the two parameters is out of phase and their correlation is nonlinear, indicated by the phase difference in −π and low coherence.

Figure 12a,b illustrate that ORP is in-phase correlated with salinity and TDS at a period of twenty-four hours; however, the correlation can be characterized as moderately linear due to the low coherence levels. During the same period, Figure 12c demonstrates a strong linear in-phase correlation between pH and salinity. This observation contradicts the negative and weak relationship found between the two variables through the linear regression analysis (see Figure 9j). These contrasting observations suggest that, while the overall trend shows a decrease in pH with increasing salinity, both parameters exhibit similar daily variations. This discrepancy underscores the importance of utilizing both frequency domain correlation methods and traditional linear regression analysis to gain a comprehensive understanding of the relationships between water quality parameters.

Another notable observation from the frequency domain correlation analysis is a strong in-phase linear relationship observed at a period of around forty hours between the following variables: dissolved oxygen and actual conductivity (Figure 11b), ORP and salinity (Figure 12a), ORP and TDS (Figure 12b), and pH and salinity (Figure 12c). For some of the pairs, this correlation is even stronger than the correlation during the twenty-four-hour period. The physical explanation of these correlations requires further investigation, which is beyond the scope of this study.

7. Discussion

7.1. Practical Outcomes from the Pilot System

The three-month pilot operation of the real-time water quality monitoring system offers valuable practical insights into the functionality of such systems. This experience will inform the installation of a sustainable, permanent monitoring system in the examined coastal area, enhancing the existing in-field laboratory [29]. This study highlights key considerations for designing effective coastal real-time water quality monitoring systems to ensure their long-term viability. These considerations encompass installation and retrieval processes, connectivity and data transmission, and sampling frequency. Additionally, based on the authors’ experience, comprehensive maintenance and management plans are crucial to ensure the long-term durability and effectiveness of the in-field laboratories. These plans will facilitate the mitigation of the impact of the dynamic marine environment on the monitoring system, addressing issues such as marine growth on the mooring lines and the contamination or encrustation of the sensors.

7.2. Data Post-Processing

As noted by Manjakkal et al. [10], a fundamental decision in time-series modeling, including in the context of water quality, is whether to conduct the analysis in the time domain or the frequency domain. The current study proposes a data post-processing approach that includes investigation in both domains, as well as data filtering. Filtering is particularly important regarding noise-contaminated signals, such as turbidity measurements. The proposed filter is a lowpass Butterworth filter that can effectively reduce the contribution of those frequencies larger than a selected low cutoff frequency. Although this filter is widely used for processing monitored data [23] and is easy to apply, it does require some fundamental knowledge of data processing. The study also proposes an alternative and equally effective approach for filtering out unwanted frequency content (noise cancellation) by employing the moving average technique, which is well-known in the engineering community. In both approaches, careful consideration should be afforded to selecting the low cutoff frequency for the lowpass filtering technique and the moving window width (time duration) for the moving average method.

A drawback of the moving average technique is that it can impact the frequencies that were intended to remain unaffected by the noise-cancellation process. An additional limitation is that it can obscure the impact of extreme events regarding the monitored data. During an extreme event, such as an oil spill, unfiltered data would show a sharp change in some water quality parameters. However, a moving average with a broad time window will smooth the data over the selected period, attenuating the evidence of the extreme event. For this reason, the moving average is not recommended for marine real-time water quality monitoring systems that aim to provide early warnings and prompt notifications for rapid responses to extreme events.

For identifying extreme events and issuing early warnings, frequent inspection of time histories is essential. This practice can also enhance the understanding of water quality parameters’ variation with time, as well as aid in detecting malfunctioning sensors and instruments within a water quality monitoring network. The above cannot be achieved by relying solely on statistical data (mean, minimum, maximum, etc.) or Fourier spectra. Statistical values derived from long-term measurements can be particularly important for infrastructure designing purposes. Fourier spectra, on the other hand, are useful for identifying the periodic responses of the parameters, such as daily or seasonal cyclic variations. To identify daily variations, measurements over several days are required, while identifying seasonal variations necessitates continuous monitoring over several years. This highlights the importance of investing in water quality monitoring systems that are permanently installed in critical infrastructures subject to dynamic changes (e.g., oil spills), such as the system presented in this work. Also, the present work can be extended for the early detection of adverse events, such as oil spills.

In addition to investigating the response of each water quality measurement separately, this study also examined the statistically significant correlations between the monitored parameters at a 5% significance level. This was achieved through time domain analyses, utilizing Pearson correlation coefficient and linear regression analysis, as well as through frequency domain analyses, employing a combination of cross-power spectral density, phase difference, and coherence spectra. The two time domain approaches facilitate the overall assessment of the linear correlation between two variables, while the frequency domain approach characterizes the relationship across different periods (or frequencies) of the signals. The relationship can be linear, nonlinear, or there may be no relationship, while low coherence can indicate high noise levels, as described in Section 6.3. Frequency domain correlation analysis is particularly valuable for investigating the cyclic correlations of parameter responses over specific periods (e.g., within a day). It has the advantage over the above linear correlation approaches that it can handle noisy datasets, while it can provide information regarding the phase between two examined signals. Also, frequency domain correlations can be useful in identifying the causes of the cyclic responses of the values of water quality parameters. For instance, the pronounced in-phase correlations between temperature and conductivity and salinity and DO demonstrated in Figure 10b,c and Figure 11a, suggest that the daily fluctuations in temperature drive the variations in these other parameters.

The Pearson correlation and regression analysis provided compatible correlations between the monitored parameters. However, it is advisable to present the examined parameters in a scatter plot format to more accurately assess the dispersion and thereby validate the correlation results. For instance, scatter plots were essential for evaluating the correlation between TDS and ORP, as well as between pH and salinity. Although the Pearson correlation coefficient in Table 5 indicates a strong negative linear correlation between these parameter pairs, a finding that is also supported by linear regression analysis (as shown in Figure 9i,j), the scatter plots presented in these figures raise questions about the validity of the correlations. In fact, a contrasting conclusion emerged from the frequency domain correlation analysis, which utilized the combination of CPSD, phase difference, and coherence spectra. Specifically, the parameter pairs TDS–ORP and pH–salinity were found to exhibit strong linear in-phase correlations during a twenty-four-hour period (see Figure 12b,c), which is the period that dominates the responses of all the parameters. This discrepancy suggests that although the two parameters are generally negatively correlated, they may exhibit positive correlation over certain periods (e.g., twenty-four hours) due to their similar response to a third factor (e.g., daily temperature fluctuations). The discrepancy also underscores the importance of investigating correlations in the frequency domain as a complement to time domain analysis.

7.3. Seawater Quality

For most water quality parameters, the monitored values align with those expected for the coastal waters of Cyprus, as documented in the literature [47,48]. The pH values fall within the range of 7 to 8.5, which is the range mentioned by Tavakoly Sany et al. [7] as suitable for marine life. The dissolved oxygen levels are around or exceed 5 mg/L, meeting the minimum requirement for marine life [26]. The ORP levels were observed to be approximately 200 mV or higher, indicating high-quality water with low organic matter content, and conditions conducive to the effective decomposition of organic matter by microorganisms, according to Tsai et al. [26]. The actual conductivity was found to be around 50 mS/cm during the period the sensor worked properly, complying with the mean value mentioned for seawater by Zheng et al. [51].

The turbidity values fluctuated widely during the monitoring period. As shown in Figure 8b, the turbidity values were exceptionally high between the ninth and tenth weeks, the origin of which requires further investigation. For the remaining dates, the turbidity values were within a reasonable range. Many measurements were below 1 NTU, indicative of calm weather, as noted by Macdonald et al. [52]. According to the same authors, more turbid water, i.e., values greater than 20 NTU, can be associated with windy weather. Verifying these claims would require wind data measurements from a weather station as well as wave measurements. Although such equipment is available at the site where the water quality monitoring system is installed [29], interpreting its records is beyond the scope of this study.

7.4. Time Variations and Correlations

Most of the monitored seawater parameters exhibited daily variations, with dissolved oxygen and temperature showing the most significant fluctuations (Figure 7c,g). Apparently, seawater temperature is affected by the daily variations in environmental temperature, while dissolved oxygen’s daily variation can be attributed to its dependence on water temperature. The daily variations in the two parameters were found to be linearly but negatively correlated (see Figure 11a). The same correlation was also found for the overall variations in the two parameters over three months of monitoring (Figure 9d), verifying that dissolved oxygen can decrease in warm weather as a consequence of a water temperature increase [33]. Additionally, this study verifies that water temperature is linearly correlated with conductivity, salinity, and TDS (refer to Table 5).

The overall correlation between dissolved oxygen and actual conductivity indicated that an increase in one parameter can lead to an increase in the other (Figure 9e), as also noted by [16]. However, the daily variations in these two parameters were found to be linearly and positively related (Figure 11b). This relationship should not be misconstrued as a direct interaction between the two variables; rather, the daily variations in both are driven by the daily fluctuations in water temperature. An additional observation is that dissolved oxygen was found to be relatively stable with respect to ORP, exhibiting only a very weak correlation, as observed in Figure 9g. On a daily basis, though, dissolved oxygen and ORP responses are out of phase; i.e., when one increases, the other decreases, with their relationship being nonlinear, as indicated by the low coherence values in Figure 11c. Similar to the previous discussion, the out-of-phase relationship is attributed to the fact that dissolved oxygen decreases while ORP increases with increasing temperature, which is the primary driving factor of the daily responses of the two parameters. Finally, ORP was found to have a moderately linear in-phase correlation with both salinity and TDS when considering their daily variations (Figure 12a,b). However, more data are required to assess their correlation over extended periods.

Drawing correlations with respect to pH is challenging because the pH levels remained relatively stable throughout the monitored period. Small daily variations were observed, which were found to be linearly correlated with the daily variations in salinity (see Figure 12c), consistent with the existing knowledge [6]. Although not presented herein for brevity, the same correlation was observed between pH and TDS, as well as pH and actual conductivity, aligning with the established literature [16] as well.

8. Conclusions

This study has presented a pilot three-month operation of a seawater quality system for real-time monitoring, integrated within a broader in-field laboratory situated in a coastal area of Cyprus. The successful pilot validated the system’s reliability and provided valuable insights for deploying similar systems. The study proposes a deployment plan to address the challenges identified through this experience, including those related to system installation and retrieval, data transmission, and maintenance. Addressing these challenges is essential for ensuring both the sustainability and operational safety of the system, and the reliability of the data collected.

A holistic data post-processing approach was proposed, incorporating both time and frequency domain analyses. Time domain analysis, which involves interpreting monitored time series and their statistical values, is particularly useful for identifying temporal variations and extreme values in water quality properties due to extreme events (e.g., wind or waves) that can be used in designing infrastructure. They are also crucial for identifying accidental events, such as oil spills, and issuing early warnings to enable prompt responses. However, time series can be contaminated by high-frequency noise, which complicates their interpretation. Noise can be mitigated by applying a lowpass filter, considering the careful selection of the low cutoff frequency, as suggested in this study. Additionally, time domain analysis alone is inadequate for investigating the cyclic patterns in the response of water quality parameters. As this study demonstrates, cyclicity can be effectively investigated using Fourier transform, specifically through Fourier amplitude spectra, which provide the magnitude of the contribution of each frequency (or period) to the monitored signal. Despite its utility, frequency domain analysis can obscure the actual physical magnitude of the variations in the examined water quality parameters over time. Considering the distinct advantages and limitations of time and frequency domain analyses, it is recommended to combine both methods for a more comprehensive interpretation of real-time water quality data.

Similarly, the present study recommends using both time and frequency domain analyses for correlating water quality parameters at a significance level of 5%. Time domain correlation analyses, particularly involving the Pearson correlation coefficient and linear regression analysis, are useful for identifying the overall trends between the examined parameters. On the other hand, frequency domain correlation analysis, which includes cross-power spectral density, phase difference, and coherence spectra, provides insights into the relationship and phase alignment of two parameters at specific periods. It is crucial to carefully interpret frequency domain correlations as the correlations observed at specific periods (e.g., daily cycles) may result from a common driving factor rather than a direct relationship between the two parameters.

The application of the proposed data processing approach, utilizing data recorded over three months, yielded water quality parameters that align with the expected values for the open sea conditions in the coastal waters of Cyprus. This finding suggests that the construction and operation of the marina have not adversely impacted the water quality in the area. Also, significant daily cyclic variations were observed in most of the examined parameters, most of them driven by their dependence on water temperature. Correlations between the parameters were identified, with some aligning with the existing knowledge, while others revealed new preliminary relationships that necessitate further investigation with extended data to confirm their validity. The investigation also identified that water quality parameters, such as turbidity, may correlate with the weather and wave conditions. This warrants further exploration using the relevant data that are currently available from broader in-field laboratory monitoring systems.