Online UV-Vis spectrophotometers can be effective and practical for measuring water quality parameters continuously and without the need for physical filtration using software particle compensation techniques. The water industry has deployed more online instruments to monitor water quality from catchment to tap for online and in-situ measurements as well as the treatment process control. However, the reputation of lacking reliability of the measurements is the general restriction of these instruments to expand to a wider range of water quality management applications. This section discusses those issues and limitations.

With the advancement of photodetectors’ development, there is an increasing variety of online UV-Vis spectrophotometers. Various UV-Vis sensors developed from different detection technologies and instrument designs are available for water quality monitoring and process control. The principle of UV-Vis spectrophotometry is based on the substance molecules in the water that can absorb UV-Vis light of a specific wavelength and the correlation between absorption spectrum and the concentration of the substance ^[8]. These sensing devices generally do not require sample filtrations (software particle compensation), are reagent free and allow fast measurements of water quality. They also have low maintenance requirements for parameters such as UV₂₅₄ and the spectral absorption coefficient (SAC254); in addition, they are equipped with automatic ultrasonic cleaning systems without the need for manual cleaning; however, in some situations, particularly for turbid source waters, regular/on-demand manual cleanings of the measurement ports are still required to ensure reliable measurements. Commonly used commercially available online UV-Vis instruments are summarised in Table 1. There are mainly two categories of these online instruments: single wavelength (SW) and spectrum (full or partial).

Online SW UV-Vis instruments determine concentrations of a particular parameter in water based on the absorbance of a selected single wavelength ^[23]. The SW UV-Vis instruments, also called UV sensors, are usually manufactured with a specific wavelength to measure UV₂₅₄ or nitrogen as nitrate and nitrite. The most common SW instruments are UV₂₅₄ sensors which measure the absorbance at 254 nm with the absorbance at 550 nm for particle compensation. UV₂₅₄ sensors can generate a surrogate parameter—SAC254 to determine dissolved organics and provide measurements of correlated parameters such as DOC and COD ^[23]. These surrogate parameters determined by the sensors are generated based on the correlations (often linear) of UV₂₅₄ and the parameters, which is the wavelength of 254 nm absorbed by organic matter in the water. This concept is used by some commercial instruments including the HACH UV probe ^[24], Burkert SAC254 sensor ^[25] and YSI UV-Vis sensor; they use a single wavelength (absorbance at 550 nm) to compensate for the particle effect. These instruments are often SW instruments that use absorbance at 254 nm to determine the concentrations of a particular parameter, such as DOC.

In comparison to the SW UV sensors, UV-Vis spectral or full-spectrum sensors record the absorbances of a certain band of wavelengths or full spectra. These sensors produce fingerprints of spectra which are then used to determine and calculate concentrations of water quality parameters based on the instrument’s built-in algorithms. The spectral sensors can provide measurements for various parameters such as UV₂₅₄, colour, DOC and turbidity using the algorithms. These instruments are generally factory calibrated for the particular water quality parameters using their proprietary algorithms but with site-specific re-calibration options. The instrument built-in proprietary algorithms are first used to remove the particle’s effect on the measurements of the water to replace the physical filtration step ^[26]. Some instruments, such as the IQ Sensor NET, provide surrogate parameters for DOC and COD, but the results are often not comparable to standard laboratory methods without specific calibration. This is because the correlations between the surrogate and standard analytical methods depend on the compositions of the water ^[18]. Re-calibration is often needed if significant changes happen in the compositions of water ^[18][27], which is different to the specific water type used in the original algorithm development ^[26].

Generally, when comparing the performance of full-spectrum and SW sensors, SW sensors can provide measurements and trends of the parameters varied during certain periods but may not compensate for the particle effect accurately, particularly when comparing the results with the standard laboratory procedures and measurements. The SW sensors may only provide a rough surrogate measurement of organic content and the total nitrogen content of nitrate and nitrite, and they do not have accurate particle compensation for most surface waters. In comparison, the spectral sensors provide better particle compensation and can be calibrated to specific locations with higher accuracy; they are better for precise applications, such as real-time water monitoring and treatment processes. In addition, the calibrations of online sensors are normally based on the grab sample collected at the same time of the measurement compared against laboratory measurement of that same water sample; these calibration procedures are more susceptible to the errors of grab samples.

As reported earlier, some advanced full-spectrum online UV-Vis spectrophotometers can determine a range of water quality parameters, including UV₂₅₄, colour, DOC, turbidity and nitrate. The parameters can be computed based on the applications via the instrument’s built-in proprietary algorithms. According to the manufacturer literature, these algorithms were developed based on chemometrics techniques, such as partial least squares (PLS) and multiple linear regression, to establish the relationship between UV-Vis spectra and laboratory measurements of water samples ^[7]. The algorithms were created by hundreds of global datasets containing both UV-Vis spectra and reference laboratory data obtained from a wide range of water quality ^[7]. Instruments have default configurations to apply the generic calibration for particle compensation to the raw spectral data using the built-in algorithms.

UV₂₅₄ measures the amount of light absorbed by conjugated organic compounds, which have been widely used as a rapid water quality measurement technique to control water treatment processes ^[35]. UV₂₅₄ from the online UV-Vis instruments based on built-in algorithms generally performs well for the treated water as less interference exists ^[26]. However, performing site-specific calibrations may be needed if it is used for source water with a complex matrix. It is more difficult to judge the accuracy of the colour measurement than UV₂₅₄ using the online UV-Vis instruments, as the standard laboratory colour measurement method relies on SW measurement. Besides, the wavelength selected to measure colour may be different based on regions and water sources; for instance, 456 nm is used by Australia and the USA, whereas 410 nm is used by Russia to measure colour in natural water ^[36]. To measure the colour of water, the online instrument needs to be set up according to the required wavelength. DOC is used to monitor water quality from catchment to tap water, and measurement is usually carried out using the laboratory-based standard method. Online UV-Vis instruments with built-in algorithms can be used as alternative measurements; however, they are often reported as water-specific and lacking accuracy and thus need additional calibrations against different water matrices ^[37]. Turbidity determined by the online UV-Vis spectrophotometers with the generic built-in algorithms is comparable to the turbidity results analysed in the laboratory ^[38][39]. Nitrates generated from the online instruments with the generic built-in algorithms are generally not satisfactory ^[38], but another group reported that the results are comparable with a laboratory analysis ^[39]; this could be related to the specific algorithm used and the monitoring application.

The measurements of water quality parameters using the online UV-Vis spectrophotometers are often source-water dependent. Thus, additional site-specific calibrations are needed to improve the accuracy of measurements ^{[7][18][40][41][42]} which is an instrument function provided to enter the laboratory-determined value of the collected grab sample measured at the same time by the online instrument. The site-specific calibration process involves modifying the slope and intercept of the built-in regression function using laboratory data from the reference grab samples ^[18][40]. To achieve the best calibration results, grab samples are needed to be representative and cover the whole measurement range of the water. Measurement accuracies can be enhanced with the use of an increased number of grab samples for the calibrations of the online UV-Vis spectrophotometers.

Applications of online UV-Vis instruments can reveal that the significant fluctuations of water quality could affect the accuracy of the measurements, and long-term monitoring required regular calibration to compensate for the variation of the particle character issue ^[16][26][40]. A site-specific calibration was conducted for a submersible UV-Vis instrument to monitor water quality in a forested catchment, and comparable results were achieved ^[41]. In contrast, a site-specific calibration was performed for a UV-Vis submersible instrument to measure the water quality of stream water, but concentrations were overestimated because of inaccurate particle compensation ^[42]. Therefore, accurate site-specific calibration of the UV-Vis instrument is crucial to obtain measurements for water quality monitoring.

Calibrations of online UV-Vis instruments should be performed as needed for situations such as an instrument being installed for the first time, changing the monitoring locations of the instrument, or the instrument being inaccurate to the reference water samples. Routine calibration may not be necessary for monitoring less variable source water or stable water quality, such as treated water, but routine verification of the measurements using lab references is recommended to ensure the accuracies of the instruments ^[37]. Site-specific calibrations have been approved to achieve the desirable measurement outcomes and can adequately account for the differences in large water quality changes or between different types of water ^{[13][15][43][44][45]}.

When there are difficulties to obtain accurate continuous measurements using online UV-Vis spectrophotometers with the instrument built-in algorithms or when the instruments do not come with built-in algorithms, alternative particle compensation (calibration) techniques can be developed by end-users. To better utilise the online instruments for water quality monitoring and water treatment process control, as well as lower the maintenance costs, more importantly gaining in-house experience and knowledge of the instruments, researchers and water utilities have been seeking techniques for particle compensation alternative to the built-in particle compensation methods. The particle compensation techniques based on the UV-Vis spectra can be categorised into direct subtraction compensation and chemometric modelling. Table 2 summarises particle compensation techniques from the literature for online water quality monitoring using the UV-Vis instruments.

Direct subtraction compensation is based on the absorbance of wavelength characterised by the particles in the water ^[7][58]. Wavelengths, including 275 nm, 350 nm, 545 nm, 546 nm and 550 nm, have been utilised to characterise the particles in the water and to remove the particle effect from the UV-Vis measurements ^[46][47][48]. The absorbance at 546 nm was used to remove the particle effect on the COD in river water ^[47]. The wavelength at 545 nm was used to reduce the particle influence on the UV for surface water ^[48]. The wavelength at 550 nm is commonly selected for SW particle compensation for individual water quality parameters and has often been used for the measurements of UV₂₅₄ ^[26]. UV at 350 nm has been used to compensate for the online measurements of COD using UV spectrophotometry to detect groundwater quality to remove the influence from particles ^[46]. In the same study, the absorbance at 275 nm was also used to compensate for nitrate at 220 nm. Figure 1 shows an example of using an SW particle compensation method to remove the particle’s influence on a raw spectrum for surface water ^[26].

The multi-wavelength particle compensation technique, also called chemometric modelling, includes a selection of chemometrics such as multiple linear regression (MLR), multiple stepwise regression (MSR), support vector machine (SVM), support vector regression (SVR), multiplicative scatter correction (MSC), principal-component analysis, and PLS. The multi-wavelength particle compensation technique is based on relationships between the raw spectra and laboratory reference values of the water quality parameters. The MLR determines the linear relationship between a dependent variable (the laboratory values) and independent variables (wavelengths of the raw spectra) which can directly define the coefficient of each variable. MLR was employed to remove the particle effect on the UV-Vis spectra of brackish water for the rapid measurement of water quality parameters ^[55]. Multiple linear regression was also used to quantify the DOC content in the stream water ^[17] and TOC in the drinking water, seawater and river water ^[21]. Multiple stepwise regression was used to compensate for the particle effect on the DOC measurements for surface water ^[56]. SVM is a machine learning algorithm that can be used for classification, regression and outlier detection. SVM was used to determine the concentration of dissolved nutrients in surface water using the full spectral wavelengths and laboratory values and demonstrated the effectiveness of the approach ^[57]. SVR is similar machine learning method as SVM, but it works with continuous values instead of classification, as in SVM. SVR was used to predict the combined nitrate and nitrite concentration for treated water samples using spectral features, and the predicted values were matched with the standard laboratory values ^[59]. MSC is a normalization technique to correct the particle’s effect on spectra by changing the scale and the offset based on the reference spectrum, which is the mean of the spectra. The MSC method was applied to compensate for the particle’s effect on the COD in lake water ^[19], UV₂₅₄ in reservoir water ^[26] simulated surface water ^[50] and DOC in drinking water production ^[51].

PLS constructs components by projecting the predictor variables to a new space. Then, the linear regression models were built between new predictors and responses. PLS can be used to extract important information from a large data matrix ^[60]. PLS regression is a commonly used method to remove the particle’s effect on the measurements of water quality parameters based on the multiple wavelength spectra. It has been used to remove the particle’s effect on the water quality multi-parameter, such as COD in artificial seawater ^[61], suspended solid in brackish water ^[55], COD in lake water ^[62], COD and TOC in seawater ^[53] and stream water ^[17], nitrate in water ^[52], nitrate and nitrite in seawater ^[63], DOC in surface water ^[56] and drinking water ^[51] and ozone in drinking water ^[54].

The use of a subtraction method for particle compensation generally works well for low- and medium-turbid source water and treated water, but it may lead to less accurate measurements in some cases, such as highly turbid water ^[49][56]. The accuracies of the measurements such as DOC in the surface waters can be improved by using the multi-wavelength particle compensation methods. Water matrix-specific particle compensation is often recommended for water quality monitoring. There are some benefits of using alternative particle compensation methods to the built-in algorithms. Firstly, it creates simplicity and flexibility of custom-made particle compensation methods for water quality measurements of the particulate water matrix, as the details of the built-in algorithms for the commercial online instruments are often not accessible to the users. Secondly, the use of alternative particle compensation methods can lower the calibration costs of the instruments. Moreover, cost-effective UV-Vis instruments with a single wavelength or short wavelength band could be employed in the field to monitor water quality instead of using full-spectrum UV-Vis instruments.