Effect of solar radiation on natural organic matter composition in surface waters and resulting impacts on drinking water treatment

ABSTRACT Solar radiation experiments showed a shift in the composition of natural organic matter (NOM). Due to irradiation, the concentration of high molecular weight (HMW) molecules decreased, and that of the low molecular weight (LMW) fraction increased. Microbiological analyses showed that biodegradation was neglectable. To assess the consequences for water treatment processes, coagulation jar tests were performed by comparing the removal effectivity for NOM fractions from irradiated and unirradiated raw water. The degree of dissolved organic carbon (DOC) removal by coagulation was lower for irradiated waters. As primarily HMW organic compounds are removed by coagulation, the decrease in coagulation performance is attributed to the increase in the LMW concentration due to photochemical reactions induced by solar radiation. Flocs were about 15% larger for irradiated water. Possibilities to adapt water treatment to respond to changes in DOC composition and concentration are outlined. Ozonation–biofiltration is judged as the most promising treatment process to cope with climate change-related challenges in drinking water treatment. GRAPHICAL ABSTRACT


Introduction
Since the beginning of the 1990s, increasing concentrations of natural organic matter (NOM) in surface waters have been noted [1][2][3][4][5]. As numerous studies have demonstrated, these increasing concentrations of organic substances are attributed, inter alia, to climate change. In the literature, it is suggested that increases in NOM concentrations in surface waters are mainly caused by the interaction of global warming and changes in runoff. The latter is due to the increasing frequency and strength of floods, heavy rain and storm events, or frequent changes between dry and wet periods. The consequence is an increase in erosion and leaching from catchment soils, resulting in an input of dissolved NOM into surface waters, as shown by Delpla et al. [6], Weyhenmeyer and Karlsson [7], Monteith et al. [8], Clark et al. [9], Evans et al. [10], Eikebrokk et al. [11], Chow et al. [12] and Hagedorn et al. [13]. It is assumed that global warming results in extended growth seasons, leading to increases in biomass production, which, in turn, can cause increased plant decomposition, hence greater production of NOM [5,14]. Besides the changes in runoff and erosion that affect soil leaching and with this the flux of organic material, special attention is paid to the leaching from peatlands as a large store of carbon [2,[15][16][17][18]. It has been concluded that rising temperatures stimulate the export of NOM and, hence, dissolved organic carbon (DOC). Furthermore, it has been found that droughts can cause increases in DOC production from peat [19]. Consequently, peatlands may act as an important source of DOC, and changes in biomass productivity and leaching can entail increases in the DOC concentration in surface waters connected to peatlands and their drainage system. In addition, other studies have demonstrated that rising trends in DOC concentration can be attributed to changes in deposition chemistry and/or catchment acid sensitivity [8,15,[20][21][22][23][24][25].
These changes may also be attributed to alterations associated with the so-called priming effect, that is, the impact of labile organic matter on the mineralisation and mobilisation of recalcitrant organic matter, as proposed by Guenet et al. [26]. Generally, processes in soils and surface waters are more complex, and climate change-induced alterations may directly and indirectly affect surface water quality in various ways. The interrelations between temperature, solar radiation, ground frost, snowmelt, ice covering, thermal mixing, biomass productivity and the activity of micro-and macro-organisms can furthermore be responsible for the observed increases in the DOC concentration in surface waters [27][28][29].
With respect to climate change, increases in global radiation are also likely to occur [30,31]. Since humic substances have high absorption capacities for radiant energy within the spectral ultraviolet (UV) range of light [32], photochemical reactions are expected in surface waters, possibly causing decreases in the DOC concentration [33][34][35][36][37] or at least decreases in the average molecular weight of dissolved organic compounds [32,[38][39][40][41][42][43]. This effect is limited to the upper layers of surface water [44] but may propagate into deeper layers due to mixing effects [3], or it might be affected by vertical motion [40]. Since global warming can cause increased stratification of lakes and reservoirs, mixing will be affected, resulting in an enhanced UVinduced bleaching of DOC in the upper layers during prolonged summer stratification, as proposed by Chen et al. [45], Del Vecchio and Blough [46] and Vodacek et al. [47].
Increasing contents of NOM and changes in their composition can result in adverse effects on water treatment [48,49], especially with respect to coagulation and flocculation processes [11,50,51]. For example, it is expected that floc formation will change as a consequence of lower molecular weight of organic compounds after irradiation. Sharp et al. [52] showed that a deterioration in treatment performance is caused by a change in the charge density of the NOM. Raeke et al. [53] linked the climate change-induced mobilisation of dissolved organic matter in catchments and its removal in drinking water treatment to its molecular characteristics. They found a preferential mobilisation of more oxygenated and unsaturated molecules of higher molecular weight at elevated discharge from a forested catchment, which could be efficiently removed through coagulation in drinking water treatment. However, in grassland and agriculture-dominated catchment areas, a higher proportion of sulphur-and nitrogen-containing molecules was detected, which, in turn, could be less efficiently coagulated. Ritson et al. [19] found that DOC produced from peatland after droughts and that had furthermore been exposed to oxygen was harder to be removed by conventional coagulation/flocculation treatment processes. Consequently, it is of interest to determine whether and to what extent solar UV radiation will contribute to the transformation of high molecular weight (HMW) compounds into smaller molecules. Furthermore, it is important to understand how changes in the composition of NOM affect their removability by coagulation during water treatment. Based on the results, conclusions regarding the optimisation of drinking water treatment of climate change-affected surface waters can be drawn.
The changes in raw water quality that cause decreases in treatment performance will not only have considerable impacts on the drinking water treatment itself but also on its distribution. When treatment effectivity decreases, more disinfectants will be required, more disinfection by-products will be formed and bacterial regrowth may increasingly become a problem [48,52,54]. However, opposite effects could also be observed, as indicated by Valencia et al. [55], who found decreasing formation potentials for trihalomethanes (THMs) and adsorbable organic halogens as a consequence of the UV-induced degradation of the larger molecules of DOC towards smaller molecular sizes. In the end, to guarantee high water quality, water treatment must be adapted in any case.
Consequently, the objectives of this research were to understand the impact of solar radiation on the composition of NOM in surface water for drinking water production, to determine the impact on the removability of NOM by coagulation, and to conclude on the effective adaptation strategies for water treatment in order to respond to the expected challenges.

Raw water
All experiments were performed using raw water from a dammed drinking water reservoir in Saxony (Germany) with a catchment area being strongly influenced by peatland. The raw water was characterised by a high concentration of dissolved organic matter (DOC = 11-15 mg L -1 ; UV 254 = 54-58 m -1 ), a considerable brown colour (absorbance at 436 nm = VIS 436 = 3-4 m -1 ), a low pH of 5.5-5.8 (at 15°C) and a low conductivity of about 53 µS cm -1 (at 23°C), as well as low turbidity of 0.8-1.4 FNU. A UV 254 to DOC ratio (SUVA) of about 4 L mg -1 m -1 indicates a high content of humic substances [56,57] and that, consequently, coagulation will mainly be influenced by the organic compounds [58]. The DOC of the raw water was fractionated by liquid chromatography organic carbon detection (LC-OCD) analysis [59,60] as consisting of approximately 8% hydrophobic and 92% hydrophilic organics. Further fractionation of the latter revealed approximately 5% biopolymers, 67% humic substances, 12% building blocks and 8% low molecular weight (LMW) compounds (acids and neutrals).

Analytical methods
DOC and the concentrations of DOC fractions were measured using an 8th generation LC-OCD analyser (DOC Labor, Karlsruhe, Germany), which combines high-sensitivity size-exclusion chromatography with high-sensitivity organic carbon detection [59,60]. The accompanying software package ChromCALC was used for the evaluation and interpretation of the chromatograms. The analyser partitions DOC into hydrophobic organic carbon and chromatographable hydrophilic organic carbon (CDOC). CDOC is further fractionated into biopolymers (polysaccharides, amino sugars and proteins; >20,000 g mol -1 ), humic substances (350-10,000 g mol -1 ), building blocks (hydrolysis products of humic substances; 300-500 g mol -1 ), LMW acids (<350 g mol -1 ) and neutrals and amphiphilics (including aldehydes, ketones, alcohols and amino acids; <350 g mol -1 ) [59,61]. The measurement range was 10-5000 ppb. For the analysis, samples were diluted three times using deionised ultrapure water, except for samples of coagulated water, which were analysed without dilution.
The UV absorbance at a wavelength of 254 nm (UV 254 ) was measured after filtration of the sample through 0.4µm polycarbonate filters in a 1-cm quartz cell using a multi-cell spectrophotometer (DR 5000, Hach Lange GmbH, Germany). The same instrument and cuvette were used in scanning mode to determine the absorptivity for all wavelengths between 290 and 740 nm.
To determine total and viable bacterial cell concentrations, quantitative flow cytometry (FCM) using an Accuri C6 flow cytometer (BD Bioscience, Heidelberg, Germany) together with fluorescence staining of microbial cells with the nucleic acid stain SYBR Green I (1:100 dilution in DMSO) was applied according to the method described by Hammes et al. [62]. Staining was performed prior to FCM analysis in the dark for at least 13 min at 38°C. As described by Hammes et al. [63], SYBR Green I and propidium iodide (6 μM) were used in combination to differentiate between living and dead cells.

Solar radiation simulator and experimental setup
To determine whether and to what extent solar radiation induces the transformation of DOC, 2.5-L raw water samples were subjected to simulated solar radiation for 72 h. For the experiments, 2.5-L raw water samples in beakers with a 3-L volume, approximately 150 mm inner diameter and 210 mm height were placed in a cooling bath with dimensions of 30 cm × 60 cm × 30 cm (H × L × W), which was coated inside with black foil to prevent light reflection. In the beakers, the height of the 2.5-L water volume was 155 mm. Two or three beakers were irradiated simultaneously.
The custom-made solar radiation simulator (supplied by UV-EL, Dresden, Germany) was comprised of a 1000 W Hg medium-pressure lamp (UV 1150S, Hönle UV Technology, Gräfelfing, Germany) and an emitter (UV-F900, Hartmann, Ober-Mörlen, Germany). The emitter was equipped with a UV-C filter to eliminate wavelengths below 290 nm since solar radiation of shorter wavelengths does not reach the Earth's surface.
The irradiator was fixed 30 cm above the surface of the raw water samples. The whole setup was placed inside a laboratory fume hood whilst air replacement was switched on to avoid overheating. During the experiments, the samples were continuously stirred by magnetic stirrers at 200 rpm to keep the water volumes properly mixed. The temperature of the raw water samples was kept constant at 20°C and the pH was monitored since possibly induced radical reactions are pH dependent. A photograph of the setup is supplied in the supplementary material ( Figure S1).
The power of the lamp was dimmed at approximately 2/3 of full power and fixed for all experiments using a potentiometer. The total intensity of the light of the full spectrum from 290 to 740 nm hitting the water surface in the three beakers was determined to be 565 J m -2 s -1 using a pyranometer (CM6B, Kipp & Zonen B. V., The Netherlands) that had been placed below the lamp system and at the same height as the water level in the three beakers.
To assess the impact of solar radiation on the basis of reliable data, all experiments were performed in triplicate and simultaneously with and without simulated solar radiation. In this way, a direct comparison of all indicators and parameters between irradiated and unirradiated water samples was possible and thus enabled the identification and assessment of the most relevant processes and effects.

Coagulation jar tests
Coagulation jar test experiments were performed according to the Deutscher Verein des Gas-und Wasserfachs (DVGW) [64], with volumes of 1.8 L in 2 L beakers supplied with baffles. Ferric chloride was used as a coagulant. The temperature was kept at 20°C. Considering recommendations by Dennet et al. [65] and Vilgé-Ritter et al. [66] and preliminary experiments, a coagulation pH of 5 was chosen for maximum exploitation of the coagulant. A dosage of 14 mg Fe L -1 or higher had been determined in preliminary experiments to yield maximum DOC removal, and thus the dosage was set to this minimum value. The coagulation pH was adjusted with sodium hydroxide. For instantaneous dispersion of the coagulant and destabilisation of the particles, an Ultra Turrax Disperser (T 25 digital, IKA, Staufen, Germany) was used for the fast-stirring phase (G > 1000 s -1 ; t = 30 s). Mixing in the following slow-stirring phase (G = 40 s -1 ; t = 20 min) was performed using a Heidolph RZR 2041 single mixer.

Floc size measurements
The commercially available dynamic extinction probe (DEP) Aello 1400 from Semitec, Dresden, Germany, was used to monitor changes in floc size during coagulation. The measurement is based on mean transmission and its standard deviation as well as the fluctuation of these values. Since these key performance indicators are not only affected by different numbers and sizes of aggregates but also by aggregate overlapping and border zone effects, data processing includes empirical correction functions [67]. More details are given in the work by Slavik et al. [68].
With the DEP, the average particle size can be measured in situ within the range of 80 to 250 nm. Depending on floc size and extinction path, the particle volume concentration can also be determined within a range of 0.01−20%. As no pumping of water samples was involved, floc size could be determined without any risk for floc breakage due to shear forces.
However, the probe was calibrated with particles of known characteristics. Since the experiments were carried out with natural raw water containing natural organic matter with specific characteristics, the characteristics of the resulting flocs are certainly not equal to those of the particles used for calibration. Consequently, an exact floc size cannot be given, but rather an apparent floc size is provided. Moreover, floc sizes can, after all, be compared in relation to each other. This means that a relative floc size in relation to the average maximum floc size at a steady state can be determined and used for discussion and interpretation of the results. The moving average of five data points was used for the interpretation of the floc size data.

Wavelength range relevant for DOC transformation
For the light of a certain wavelength, the energy carried by a photon is absorbed if at least one chemical bond is elevated to a higher energy level. Once the energy is transferred, this can be measured as absorbance. Figure 1 shows the spectral absorption coefficient (SAC λ ), which is the normalised absorbance per length of the light path, before and after 72 h of irradiation of the raw water in the setup described above as a function of the wavelength. It can be seen clearly that the absorption of light is very small at wavelengths above 450 nm.
The relative change in the spectral absorption coefficient during the 72 h of irradiation that is plotted on the secondary axis is negative for wavelengths below 450 nm and equals zero at 450 nm. This confirms that the few organic substances in the raw water that absorb light at wavelengths above 450 nm, or their respective chemical bonds, are not consumed during the 72 h of irradiation. Rather, they seem to be produced in very small amounts, as indicated by the 1% increase in the specific absorption coefficient for wavelengths above 450 nm. Vice versa, chemical bond absorbing light at wavelengths below 440 nm are consumed, which is indicated by the considerable decrease in the SAC λ . Chemical bonds belonging to DOC and absorbing in these wavelength ranges are typically aromatic groups with various substitutions and are primarily associated with humic substances [69].
According to the Planck-Einstein relation, the following is given: where c vac is the vacuum speed of light (2.26·10 8 m s -1 in water) and h is Planck's constant. The energy, E, of a photon is inversely proportional to the wavelength, l.
Using this equation, the photon energy was calculated as 310 kJ mol -1 at 290 nm and as 196 kJ mol -1 at 450 nm. These values are in the same magnitude as for bond energy in organic compounds, which are 413 kJ mol -1 for C-H bonds, 358 kJ mol -1 for C-O bonds, 347 kJ mol -1 for C-C bonds and 256 kJ mol -1 for the first bond of a C=C double bond [70]. This confirms that visible light at wavelengths above 450 nm does not contribute to a conversion of organic matter and is consistent with conclusions by Frimmel [32], whilst light at wavelengths below 450 nm contributes to a breakage of primarily the first of C=C double bonds, which are typical for HMW organic matter. Consequently, for comparison of the spectral irradiance of the solar radiation simulator setup with that of natural sunlight (or with other setups and results), only the wavelength ranges from 290 to 450 nm must be considered.

Verification of the solar radiation simulator spectrum
To ensure that the spectrum of the solar radiation simulator was very similar to the spectrum of solar light, the spectral irradiances at the full spectrum of wavelengths from 290 to 738 nm were compared. The spectral irradiance of sunlight in Central Europe in June was calculated according to Frank and Klöpffer [71]. Details are given in the supplementary material S2 a). For the simulator, the spectral irradiance had been determined by the manufacturer of the system (UV-EL, Dresden, Germany) at full power by radiometry.
The wavelength range relevant for photochemical NOM transformation had been found between 290 and 450 nm (see section 'Wavelength range relevant for DOC transformation'). For comparison of the solar spectrum and the solar radiation simulator spectrum, both irradiance spectra were normalised by the respective maximum spectral irradiance intensity in the range from 290 to 450 nm. The resulting normalised irradiance spectra are shown in Figure 2. In the relevant wavelength range, the normalised spectral irradiance of the simulator was very similar to that of solar irradiation that reaches the Earth's surface. However, spectral irradiances were higher in the range of 350-400 nm and lower between 430 and 450 nm.

Comparison of relevant irradiance from the solar irradiation simulator with the irradiation from sunlight
For the wavelength range from 290 to 450 nm, irradiance from natural sunlight in Central Europe in June was calculated as 68.4 W m -2 , and for the full spectrum from 290 to 750 nm it was calculated as 306.7 W m -2 . Details are given in the supplementary material S2 b). Thus, the portion of irradiance contributed within the wavelength range effective for photochemical conversion of organic matter is 22.3% of the total irradiance from sunlight.
The spectral irradiance emitted directly from the solar light simulator was measured using a pyranometer. From these data, it was calculated that the portion of irradiance contributed within the wavelength range from 290 nm to 450 nm was 229.8 W m -2 or 40.7% of the total irradiance (i.e. 565 W m -2 from the full wavelength range of 290-750 nm). Thus, the irradiance from the simulator was 3.36 times higher than for natural sunlight in June (229.8 W m -2 versus 68.4 W m -2 ), or, considering that in nature most of the irradiance is delivered over 8 h per day, 24 h of solar simulator irradiation are equivalent to approximately 10 days in nature. This means that the duration of an experiment of 72 h corresponds to 30 days in nature in Central Europe.

Effect of solar radiation on DOC composition
The difference in DOC, UV 254 and VIS 436 due to simulated solar radiation is shown in Figure 3 by comparing radiation experiments with experiments performed in the dark. There was a significant decrease in UV 254 at a 95% confidence level, whereas the change in DOC indicated by comparing the averages is not significant, as can be seen from the overlap of the error bars. For VIS 436 , no significant difference can be proven. It appears from the significant decrease in UV 254 that solar-radiation-induced photochemical reactions caused a considerable change in DOC composition. The small decrease in total DOC suggests degradation to a certain degree. Analyses of the concentration of microorganisms should give a clue as to whether the decrease in DOC concentration was due to biological degradation or caused by photochemical decomposition to carbon dioxide (CO 2 ).
The results of the analyses performed to determine the concentration of microorganisms are illustrated in Figure 4. Total cell counts (total counts) and viable cell counts (live counts) are given for the beginning of the experiments and after 24, 48 and 72 h. During the experiment, there was a significant increase in both total cell count and viable cell count. This gives an indication of biological activity, that is, biological degradation of DOC is likely to occur. It is striking, however, that the increase in viable cell count was significantly lower when the water was irradiated. Consequently, simulated solar radiation seems to suppress bacterial growth even when the range of wavelengths is outside the range of disinfecting effects. Due to the considerable difference in the number of viable microorganisms between irradiated and unirradiated water samples, biological degradation apparently seems to play a subordinate role in the decrease in DOC.
The results of the LC-OCD analyses are presented with their respective confidence intervals in Figure 5 and Table S1 in the supplementary material and prove the effect of solar radiation on DOC composition. There was no change in DOC and DOC composition after 72 h in the dark, that is, without irradiation. In contrast, the concentrations of the DOC, CDOC and humic substances decreased significantly, and that of the neutrals increased to a certain degree when the water was irradiated for 72 h. The photochemical decomposition of DOC is also evident by the LC-OCD chromatograms of the raw water and the irradiated water, as shown in Figure S2 in the supplementary material. There, the significant decrease in the HMW fraction and the increase in the LMW fraction are apparent.
The change in the concentration of carbon in humics resulted in a decrease in the concentration of HMW molecules, whilst the concentration of LMW fractions of bulk DOC increased. Balancing the changes gives a decrease of 3 mg L -1 of carbon in humics, but the increase in building blocks and neutrals was less than 1 mg L -1 . The difference of 2 mg L -1 corresponds with the decrease in total DOC, which proves the carbon mass balance.
Since there was no change in DOC without radiation, it is assumed that the decrease in total DOC is due to a photochemical decomposition of CO 2 or due to the formation of particulate organic carbon. This conclusion is  supported by the fact that the concentration of viable microorganisms was considerably lower in the irradiated water samples, as shown in Figure 4. To summarise, simulated solar radiation results in the decomposition of humic substances into smaller molecules, that is, building blocks and neutrals, and to a certain degree to CO 2 . This corresponds with results from the literature where photochemical mineralisation of DOM was observed. Based on their investigation at Lake Skjervatjern in western Norway, Salonen and Vähätalo [37] assumed an epilimnetic light mineralisation of DOM of 25 mg C m -3 d -1 in this region. This is 0.75 mg C L -1 in 30 days, which is lower than the 2 mg C L -1 observed in this study. The lower mineralisation rate in Lake Skjervatjern can probably be attributed to the more northerly position of the study area. Amon and Benner [36] determined an unusually high photochemical DOC consumption of up to 0.05 mg C L -1 h -1 in investigations using water from the Rio Negro. Assuming 10 h of sunlight per day, this rate results in a mineralisation of 15 mg C L -1 in 30 days, which is more than seven times the mineralisation found in our experiments. Once again, the southern location and the associated intensity of solar radiation can be the cause of this effect. In bottle experiments, Molot and Dillon [35] found that up to 50% of DOC can be transferred to inorganic C over 6-11 days of solar radiation. Their experiments were performed using water from headwater streams of Dickie Lake, Ontario, which are characterised by a high DOC concentration of about 18 mgL -1 . Unfortunately, the measurement of inorganic C is not described there in a comprehensible manner. Brinkmann et al. [33] observed a decrease in DOC of up to 1 mg L -1 in 24 h, which is about 3 mg L -1 in 72 h, due to simulated solar irradiance in the summer in Germany and assumed that this was primarily caused by mineralisation because the formation of bigger entities or small particles by photocoagulation could not be detected.
These effects are also identifiable in Figure 1, where the absorption spectrum of the irradiated water shows a significant decrease in the UVA and UVB range, that is, between 290 and 400 nm. This means that dissolved organic compounds, which are stimulated at lower wavelengths in particular, were removed and converted to a certain degree to LMW compounds, which usually exhibit lower absorbance.

DOC fraction removal
To assess the impact of solar-radiation-induced changes in DOC composition on its removability by coagulation, coagulation jar tests were performed. The jar tests were performed with unirradiated and irradiated raw waters to enable a direct comparison. The results presented in Figure 6 refer to the jar tests performed with a coagulant dosage of 14 mg Fe 3+ L -1 at pH = 5. It is known that in coagulation processes removal efficiency is higher for larger molecules such as biopolymers and humic substances [54,65,72]. Therefore, it can be expected that LMW compounds are removed to a lesser extent.
This aspect could be confirmed by the coagulation jar test results shown in Figure 6. The removal efficiency of biopolymers and humics was around 90%, whereas the degree of removal of building blocks and LMW acids and neutrals was less than 60%. It can furthermore be seen in Figure 6 that the removal efficiency of the LMW compounds was even worse when the raw water was irradiated, especially that of the LMW acids and neutrals. This means that the degree of DOC removal by coagulation decreased due to photo-oxidation of humic substances. Moreover, it can be concluded that less of those compounds can be removed from solarradiation-affected water by coagulation, which shows a high potential for biological regrowth [73] and THM formation [74]. This aspect is relevant for the quality of the finished drinking water.
The decrease in removal effectivity due to changes in DOC composition in waters affected by solar radiation can also be demonstrated by comparing the residual concentration of DOC fractions after coagulation. As shown in Figure 7, the residual concentration of the LMW fractions (building blocks, LMW acids and neutrals) was significantly higher when coagulation was performed with irradiated raw water.
The difference in the DOC fractions after coagulation between unirradiated and irradiated water is also noticeable by the LC-OCD chromatograms included in the supplementary material ( Figure S3). The curve of the irradiated water shows higher peaks in the range of LMW compounds, which correspond to the higher concentrations shown in Figure 7.

Floc sizes
Based on floc size measurements, the impact of solar radiation on coagulation performance was evaluated by comparing floc formation of irradiated and unirradiated raw water used in coagulation experiments. Floc size data show that the final floc size reached in coagulation experiments differs between the irradiated and unirradiated raw waters. According to the apparent   floc size (shown in Figure S4 in the supplementary material), coagulation of irradiated water results in somewhat bigger flocs, whereas flocs formed in unirradiated raw water are smaller. The specific extent of the difference is illustrated in Figure 8. Flocs formed during coagulation of irradiated water are 10-15% larger. There is also a difference in the rate of floc formation. The final floc size is attained sooner when coagulation is performed with irradiated water.
The larger flocs formed in solar simulator irradiated water coincide with the lower removal effectivity after solar radiation. Solar radiation resulted in a decrease in the HMW humic substances concentration, and an increase in the LMW substances concentration. The former can be removed more effectively by coagulation at low pH since they adsorb well on metal hydroxide surfaces [75], whilst the LMW substances adsorb less. Flocs with adsorbed NOM are considered as weak and fragile [76]. Jarvis et al. [77] found that alum and ferric flocs with adsorbed NOM were smaller than flocs formed in the absence of NOM. They concluded that the adsorption of NOM decreased the number of bridging bonds that otherwise hold the metal hydroxide flocs together. In conclusion, solar radiation resulted in lower concentrations of organic matter that adsorbs to metal hydroxides and disturbs bindings inside the hydroxide flocs. Consequently, flocs after solar radiation can better resist shear forces and grow to a larger size.
The observed difference in floc size may be relevant for the performance of the subsequent solid-liquid separation process because of the impact of floc size on settling behaviour as well as on the effectiveness and performance of fixed-bed filtration. Whilst solar radiation decreases NOM removal effectivity in coagulation, it increases floc sizes and thus sedimentation rates of flocs after coagulation, and their removability in deep bed filters.

General remarks
Due to climate change, water suppliers that produce drinking water from surface water will increasingly be faced with increases in NOM concentrations. However, increasing solar radiation will, as demonstrated by the results of this study, partly bring about the conversion of HMW organic matter to smaller compounds. Consequently, besides increasing DOC concentrations, shifts in the molecular weight distribution of NOM are to be expected. Such changes will bring about a need to adapt water treatment by applying processes that are robust to those changes in raw water quality, that is, to rising concentrations as well as changes in the molecular weight distribution of DOC. In the following, different processes are evaluated for their potential to cope with such challenges.

Coagulation and flocculation
For coagulation processes, Ritson et al. [78] recommend switching from aluminium to ferric coagulants and/or the incorporation of coagulant aids to improve the DOC removal processes. This recommendation is based on findings by Matilainen et al. [79]. The authors showed that 84% of the NOM was removed by coagulation with ferric sulphate and only 75% of the NOM was removed with aluminium sulphate as the coagulant. The authors showed that this mainly resulted from a 28% more efficient removal of the NOM fraction ranging from 3000 to 4000 g mol -1 and a 20% more efficient removal of NOM compounds having a molar mass between 1000 and 3000 g mol -1 when using an iron-based coagulant. Explanations for the observed effects can be found in the literature from the 1980s and 1990s. For example, Crozes et al. [80] pointed to the importance of active charge density of the coagulant, the floc surface area available for adsorption, the nature of the bonds between the organics and the metal hydroxide floc and the associated mechanism of coprecipitation by adsorption of NOM on the metal hydroxide. The authors stated that at similar coagulant dosages ferric chloride presents roughly two times more active positive charges than hydrated aluminium sulphate. Moreover, a ferric chloride solution is more acidic than an aluminium solution, and the alkalinity consumed during the formation of metal hydroxides is two times higher for ferric chloride than aluminium. As a result, the pH will be lower (if not adjusted) with ferric chloride, causing increased protonation of humic substances and increased positive charge of the coagulating species, which, in turn, favours adsorption of organics onto the metal hydroxides. Because of the higher concentration of active metal in an iron-based solution and the higher molecular weight of iron, ferric chloride is likely to produce 2.8 times more metal hydroxide by weight than a similar dosage of aluminium. Consequently, the surface area for NOM uptake is larger for ferric chloride and provides more active adsorption sites.
With respect to the recommendation of incorporating coagulant aids, Ritson et al. [78] referred to Lindqvist et al. [81], without further explanation. According to Bratby [82], the mechanisms associated with polymers added in coagulation and flocculation steps are complex and cannot solely be attributed to one particular phenomenon. The improvement in DOC removal by adding a polymer as a coagulation or flocculation aid is possible, but very individual and requires a case-by-case analysis.

Ion exchange
To remove NOM or to improve the treatment performance of coagulation processes for NOM removal, the magnetic ion exchange (MIEX ® ) process is recommended as an alternative treatment or pre-treatment process [83][84][85]. However, a change in DOC composition towards smaller molecules with higher hydrophilicity will also have an adverse effect on the treatment performance of the MIEX ® process. This can be attributed to the fact that photo-oxidised waters with a lower SUVA and smaller molecular sizes are dominated by hydrophilic acids and neutral organic compounds. The hydrophilic acids tend to occupy more sites on the anion exchange resin per milligram of DOC removed, whereas hydrophilic neutrals are not removed by ion exchange at all [83]. Consequently, the capacity of the ion exchangers to remove DOC from waters with increased LMW concentrations, due to solar radiation, will be lower. A possibility to improve the overall DOC removal of waters dominated by molecules of smaller size might be the application of ion exchange processes as a post-treatment step. Thus, hydrophilic acids remaining after coagulation and filtration could still be removed to a certain degree. However, whether and to what extent this is a viable alternative has to be determined individually. With respect to changes in total raw water DOC, the benefit of a MIEX® process as a pre-treatment to microfiltration or conventional coagulation treatment was demonstrated by Drikas et al. [86]. Their investigations revealed more consistent water quality when MIEX® was incorporated into the treatment processes, which resulted in better performance of the following microfiltration or conventional coagulation steps.

Membrane filtration
To enhance NOM removal in water treatment, membrane filtration processes can be applied [87]. Since LMW organic compounds are not removed in ultrafiltration (UF) membranes, only nanofiltration (NF) and reverse osmosis (RO) applications make sense in order to adapt water treatment to raw water quality changes associated with shifts in molecular weight distribution of DOC. For NF applications, attention should be paid to the fact that at high NOM concentrations, the effect of concentration polarisation will be more pronounced, resulting in enhanced rejection, as shown by Jarusutthirak et al. [88]. Basically, the biggest drawback of membrane applications in water treatment is membrane fouling, which causes a considerable decline in productivity over time [89]. Fouling means an accumulation of colloidal matter, organic and inorganic compounds and microorganisms on membrane surfaces and within membrane pores, which often results in an irreversible loss of permeate flux. In particular, the hydrophilic fraction of NOM causes irreversible fouling, as described by Yamamura et al. [90]. Kennedy et al. [91] showed that the reversibility of fouling due to the hydrophilic fraction was very poor. Therefore, it must be expected that NF and RO filtration will experience enhanced irreversible fouling when the molecular weight distribution of NOM shifts towards LMW substances due to solar radiation.
A common way to enhance NOM removal and to decrease membrane fouling is to combine membrane processes with other unit processes, such as coagulation, adsorption, MIEX ® and ozonation [87]. This, however, involves substantial effort and expenses. Strategies to control membrane fouling in reverse osmosis applications without expensive and extensive pre-treatment were considered by Cornelissen et al. [92]. However, besides in-process adaptations of flux operation, cross-flow velocity and hydraulic cleaning, the authors considered UF pre-treated RO systems that are also associated with a considerable effort. Consequently, it will not be possible to adapt drinking water treatment by implementing membrane processes without extra and significant expenditure. Coupling conventional filtration (coagulation-sedimentation-sand filtration) with NF, as suggested by Chang et al. [93], seems to be the most reasonable and economical approach in this context to enhance water treatment in the light of climate-change-associated challenges. However, pretreatment by coagulation and flocculation will not help to mitigate enhanced irreversible fouling due to higher concentrations of the hydrophilic fraction of NOM. But ozonation-biofiltration will remove both HMW and LMW substances and will contribute to the mitigation of fouling [94].

Adsorption, ozonation and biofiltration
The removal of NOM by adsorption on activated carbon and oxidation/biofiltration processes has already been described by Ødegaard et al. [95]. These processes are particularly interesting for the treatment of waters with a high content of LMW and hydrophilic fractions since these compounds are in principle less adsorbable [96] but better biodegradable [97,98] compared to bigger and therefore more hydrophobic molecules. With respect to absorbability, two opposite effects must be taken into consideration. Although smaller and more hydrophilic molecules are less adsorbable than bigger and more hydrophobic ones, these molecules can enter smaller pores that are not available for the bigger compounds. Consequently, a larger number of adsorption sites are available for the LMW compounds, resulting in higher overall removal of this DOC fraction [99].
The purpose of an oxidation step, ozonation in most cases, prior to biofiltration is to produce LMW and hydrophilic compounds to increase biodegradability of the water [98,100]. Oxidation during treatment means to continue the photo-oxidation processes that have taken place in the water resource. When granular activated carbon is used as a filter material in the ozonation/biofiltration process, adsorption and biofiltration can appropriately be combined. Then, the combined effects of decreased molecular size and improved availability for biodegradation are of special importance. The total DOC removal by activated carbon (bio)filtration will be higher when there is an oxidation step in front of it (e.g. [101,102]). This is primarily due to the improvements in biodegradation supported by the increased access to inner pores of smaller molecules. Applications of such process combinations can not only be found in drinking water treatment but also for continued wastewater treatment to remove micropollutants. Similar to photo-oxidised raw water, secondary effluents of wastewater treatment plants contain both a high concentration and a high proportion of LMW organics [103], which is why biofiltration can be operated at maximum efficiency.
Besides ozonation, advanced oxidation processes (AOPs) can be used to degrade NOM for subsequent biodegradation since complete oxidation, that is, mineralisation, is unlikely to be achieved. A comprehensive overview of AOPs for the removal of NOM within the scope of drinking water production is given by Matilainen and Sillanpää [48] and Sillanpää et al. [104](2018). The authors present an overview of recent research and development on the application of AOPs. Although Sillanpää et al. [104] identified a serious lack of investigations focusing on the costs of AOP applications, the effort connected with the use of an AOP in practise seems to be higher compared to ozonation. Thus, ozonation is rather a standard when oxidation is applied in drinking water treatment and for the treatment of secondary wastewaters [105][106][107][108]. Compared to RO-based treatments, ozonation/biofiltration requires up to 80% less in capital costs and costs 90% less with respect to operation and maintenance [109].
Due to the level of development as a state-of-the-art application, the high effectiveness in removing NOM, the robustness towards changes in raw water quality, the cost advantage towards membrane applications and AOP processes, and the fact that nearly no residuals arise, ozonation/biofiltration seems to be the most promising treatment to cope with climate changerelated challenges in drinking water treatment. Even disinfection by-product formation is manageable when biofiltration is optimally operated [110]. Modelling the ozonation-biofiltration treatment process allows for continuous optimisation and adaptation to changes in raw water quality and respective process control. This will become crucial in the future due to the increasingly intense, rapid and more frequently occurring changes in raw water quality.

Summary and conclusions
Raw water with a high DOC concentration of about 10 mg L -1 and a UV 254 of 38 m -1 was subjected to solar spectrum irradiation under controlled conditions in the lab using a solar spectrum simulator.
. Out of the full wavelength range from 290 to 750 nm, it was shown that only wavelengths from 290 to 450 nm are relevant for the transformation of organic matter. . Photon energies in the relevant wavelength range are of the magnitude of bond energies of the first bond in C=C double bonds. This points to the breakage of C=C double bonds due to solar radiation. . Experiments with 72 h solar simulator irradiation (equivalent to 30 days in nature) resulted in partial decomposition of humic substances and building blocks to smaller molecules such as LMW acids and neutrals, as demonstrated using LC-OCD analytics. . The changes in DOC composition due to solar radiation resulted in decreases in the degree of DOC removal in coagulation jar tests. Since HMW organic compounds in particular are removed by coagulation, the decrease in coagulation performance could be attributed to a higher proportion of LMW as a result of photochemical reactions due to solar radiation. . Larger flocs are formed in coagulation of water containing NOM subjected to solar radiation, compared to non-irradiated water. This is explained by the fact there is less adsorption of the smaller LMW NOM molecules compared to HMW the metal hydroxide flocs. Therefore, as HMW NOM is partly destroyed and LMW NOM is generated due to solar radiation, the forces that hold the flocs together are disturbed less for solar radiated water.
. The options to adapt drinking water treatment to changes in the concentration and composition of DOC, as observed and expected due to climate change, were reviewed and discussed. Treatment processes will need to be adjusted not only to increase NOM concentrations. Even more, they will need to be adjusted to a shift from HMW to LMW substances as well. . For coagulation and flocculation processes, a switch from aluminium to ferric coagulants and/or the incorporation of coagulant aids could result in improved DOC removal. However, it is not possible to cope with increasing concentrations of LMW by coagulation/flocculation. . Ion exchange does not appear to be suitable as an easily adjustable treatment or pre-treatment process since increasing LMW concentrations have adverse effects on treatment performance. . With respect to membrane filtration, only NF and RO applications make sense for adapting water treatment to raw water quality changes associated with shifts in molecular weight distribution of DOC from HMW to LMW. However, due to the drawback related to membrane fouling and increasing NOM concentrations, these processes are associated with both high investments and high operation costs. . The process combination of ozonation and biofiltration with granular activated carbon as filter medium is judged as the most promising treatment to cope with climate change-related challenges in drinking water treatment. However, continuous process optimisation with regard to changes in NOM concentration and composition, is recommended.