Rainfall data and rainwater samples were gathered between October 2023 and May 2024. The roof (clay tiles) of the N12 building at Universiti Teknologi Malaysia (UTM) in Johor Bahru, Malaysia, was used to collect roof-harvested rainwater (Figure 2). For rainfall amount measurement, rainwater was collected using a polypropylene basin (inner diameter of 22.2 cm and catchment area of 387.1 cm²) and measured using a measuring cylinder to the nearest 5 ml. Serving as a simplified rain gauge in this experiment, the basin must be kept at least 1.35 m above ground to reduce wind turbulence near the ground and away from nearby tall structures or trees to ensure accurate measurement [13]. The rainfall was calculated by dividing the volume of rainwater by the basin area. Each rainfall was tabulated and supplemented with the statistics of the nearest station (JPS Taman Aman) available from the website of Jabatan Pengairan dan Saliran Malaysia [14].

Figure 2.

Rainwater harvesting system in the N12 building of UTM

Because TDS measurement via the electrolytic conductivity method is inexpensive, simple, and repeatable, it was chosen for this investigation as a function of overall rainwater quality rather than the standard laboratory analysis of the contaminants found in rainwater [15]. Because pure water is very low in conductivity, a high value of TDS derived from a high value of conductivity means a high concentration of dissolved charged species in water. A portable TDS meter and a portable pH meter were used to measure the TDS and pH of rainwater, respectively. The Macherey-Nagel PF-12Plus photometer was used to measure colour and turbidity. All measurements were done at 25 ± 1 ℃.

Care must be taken during photometric measurements of the purge water sample, as small air bubbles are formed on the wall of the glass cuvette (Figure 3). These bubbles can affect the readings of the photometer and should be removed by gently swirling the cuvette. Interestingly, no air bubbles formed for thawed ice at room temperature. In addition, two water samples (roof-harvested rainwater and tap water) were sent to an external accredited laboratory for further water quality analysis in order to identify the parameters which were not within the national drinking water standard.

Figure 3.

Air bubbles formed on the wall of the glass cuvette

Considering a mass balance on a roof used to harvest rainwater, as the number of antecedent dry days (ADD) increases (i.e., in the longer period of dry days), the amount of contaminant deposited on the roof should increase, assuming the rate of deposition is constant. When a rainfall event does occur, higher rainfall causes a larger volume of rainwater to dissolve a finite amount of contaminant deposited on the roof. Therefore, the hypothesis is that higher rainfall produces rainwater with a lower amount of contaminants and lower TDS. Shorter ADD prior to the rainfall event should also produce rainwater with lower TDS due to shorter time available to deposit contaminant on the roof prior. Thus, Pearson correlation coefficient (r) derived from a simple linear regression (SLR) model calculated using the CORREL function in Microsoft Excel, was used to determine the strength of the relationship between rainfall amount, ADD and rainwater TDS [16]. If the value of r is greater than 0.7, a strong linear relationship exists, whereas if r is less than 0.3, no or a very weak linear relationship exists. The significance of the correlation was calculated using a t-test, shown in eq. (1) where r is the Pearson correlation coefficient and N is the sample size.

In this paper, the p-value (T.DIST.2T function in Microsoft Excel) of less than 0.05 was used as a basis to reject the null hypothesis. Distributions of the data were visualised and analysed using box plots. Skewness and excess kurtosis were calculated using the SKEW and KURT functions, respectively, in Microsoft Excel to verify the normality of the distributions. The correlations were re-examined using Spearman's rank correlation coefficient (r_s) for non-normal data. The value of r_s was calculated using the same method as r described earlier, after converting the lists of data into lists of ranks using the RANK.AVg function in Microsoft Excel. By assuming that the amount of pollutant that may be dissolved by rainwater during a storm event relies on the duration of rainfall and the initial concentration of pollutant available, it is possible to approximate the quality of rainwater runoff [17] numerically. Sea salts (Na⁺, Cl⁻, Mg²⁺ and K⁺) accounted for the majority of the dissolved substances in rain [18], and the concentration of these ions drops as one moves farther inland from the shore. However, the majority of Ca²⁺, NH₄⁺, SO₄²⁻, HCO₃⁻ and NO₃⁻ ions originated from terrestrial sources such as dust and human-caused emissions. This observation suggests that the ionic composition of rainwater depends on the geographical location of the rainfall event.

PFC experiments in this work were carried out using a commercial ice maker, the HICON HZB-60FAB, with a standard vapour-compression refrigeration system with R290 refrigerant (propane) to lower the temperature of the cooled surface made from stainless steel to a temperature below the water freezing point. During the PFC experiment, the feed solution was introduced using an on-board pump and circulated using a different pump. Figure 4 shows a schematic diagram of the operation of the ice maker.

Figure 4.

Schematic diagram of the operation of the ice maker

The circulating solution cooled down gradually when flowing down on the cooled surface, eventually reaching the freezing point of water. At that point, ice started to form on the cooled surface while the circulation pump continued running until an internal timer stopped the pump when a cycle was completed. The cooled surface was heated by hot refrigerant to partially thaw the ice and detach it from the cooled surface. The ice cubes were then directed to an ice tank via the ice diverter. Once the cooled surface was free of ice, the circulation pump resumed circulation to initiate the next cycle.

Water supercooling in PFC is undesired since it will cause fractal or treelike ice dendrites to grow rapidly and contaminants to be trapped and included in the growing ice crystal lattice, lowering the quality of the generated ice. Nonetheless, certain plant and animal species depend on it to keep deadly ice crystals from forming and rupturing their cells so that they can survive the subfreezing temperatures during winter. In the context of this research, supercooling of water is defined as water being in the liquid phase below its usual freezing point, even if a more stable state, such as solid ice, may exist [10]. Lining the cooled surface with pure water to form ice [19] before feed solution is introduced to the surface [20], the use of ice-nucleating protein [21], and holes on the cooled surface [22] are some of the strategies that have been suggested to avoid supercooling in PFC. To the authors’ knowledge, no technique for preventing supercooling in falling-film PFC equipment has been put forth to date [23]. This is most likely because the equipment circulating flow is inherently more turbulent compared to other types of PFC setups, thus preventing supercooling.

To monitor temperature change during the experiment, the sump of the circulating fluid housed a calibrated temperature sensor (ISOLAB), which was monitored every 30 seconds. A graph of temperature against time was plotted throughout the experiment. Supercooling was determined to have occurred if the graph showed a characteristic dip below the freezing point of water. Distilled water and roof-harvested rainwater were used as the feed solution to replicate the experiment. After each experimental run, a dilute detergent solution was used to clean the equipment to prevent microbial growth.

If the roof-harvested rainwater is clean enough and the target use of the rainwater is not in critical applications, it may not require any water treatment prior to using it. In reality, roof-harvested rainwater is usually treated minimally, even for non-potable uses, due to the variability of rainwater quality. The use of green roofs combined with artificially constructed wetlands with aquatic plants was proposed to treat roof-harvested rainwater for non-potable applications [24].

Figure 5.

Stainless steel cooled surface

On the other hand, additional purification steps are necessary if the roof-harvested rainwater is intended for drinking. To investigate the effects of circulation time, every batch of rainwater was divided into two portions to be circulated for 20 minutes and 26 minutes, with five cycles in each set.

The stainless steel cooled surface (Figure 5) can produce 44 ice cubes every cycle. Purified rainwater, feed rainwater and purge (the concentrated, unfrozen fraction of the rainwater at the end of the experiment) were assessed for TDS, pH, colour, turbidity, and volume. The formed ice was thawed using a hotplate magnetic stirrer (Ingenieurbüro CAT M 6).

While longer circulation time produces bigger ice per cycle, the increase in ice size does not increase proportionally to the increase in time. Therefore, it is hypothesised that overall ice output per unit time (yield) is expected to be lower at longer circulation time. Rainwater quality is not expected to affect yield. Previous PFC studies relied on a few common measures to quantify the separation efficiency of PFC, such as effective partition coefficient (K), removal efficiency (RE), and concentration factor (CF). Using PFC for a dilute solution like rainwater, the concentration of ice does not vary as much as the purge does with different feed concentrations. Therefore, K and RE are expected to be very similar for different initial rainwater TDS, unlike CF. Longer circulation time produces higher purge TDS due to more ice formation, concentrating TDS in the purge, resulting in higher CF. The formulae for these three are defined in the next section. Since each batch of rainwater sample was experimented on twice under the different effects of circulation time (20 min and 26 min), the experimental data were analysed using a paired-sample t-test with Analysis ToolPak (t-test: Paired Two Sample for Means) in Microsoft Excel, while the SLR model was used to analyse the effects of feed rainwater quality. Yield, partition coefficient K, removal efficiency RE, concentration factor CF and energy consumption were quantified with equations (2) to (5).

Partial rainfall data are shown in Table 1 and displayed as box plots in Figure 6. The ADD for the majority of rainfall occurrences was relatively brief, with a mean of 2.4 days and a median of 1 day. Given that Malaysia is a tropical nation, this is hardly unexpected. The rainfall amount was 17.4 mm on the median and 23.6 mm on the mean. With a median and mean of 14.5 ppm and 16.2 ppm, respectively, the TDS of each rainwater sample were below 50 ppm, which was the municipal water TDS.

Figure 6.

Box plots of rainfall data (ADD, rainfall and TDS)

All three distributions of ADD, rainfall and TDS showed positive skewness (skewness > 0) and leptokurtic distribution (excess kurtosis > 0). Graphically, the box plots show that the value of the mean (cross) was greater than the median (horizontal line inside the box). As a normal distribution has skewness and excess kurtosis of 0 [25], all three distributions were not normally distributed, with ADD and TDS showing the most and the least deviation, respectively.

Table 2 shows the summary of rainfall data analysis. TDS was found to be not correlated (r = 0.076 and r_s = 0.001) with ADD (Figure 7) and weakly negatively correlated (r = −0.320, p = 0.013 and r_s = −0.386, p = 0.002) for Pearson r and Spearman r_s, respectively, with rainfall amount (Figure 8). The relationship between TDS and the rainfall amount was considered significant as the p-values were less than 0.05.

Overall, the relationships were weak, and other factors such as the location and nature of roof surfaces and storage tanks, length and intensity of the rainfall events, concentration of greenhouse gases and air pollutants in the atmosphere may affect the quality of the roof-harvested rainwater to various degrees. The list of these factors was not exhaustive and beyond the scope of this study.

Figure 7.

Graph of TDS of rainwater against ADD

Figure 8.

Graph of TDS of rainwater against rainfall amount

Table 3 shows the partial results of the water quality analysis of rainwater and tap water in comparison with the national drinking water standard [26]. From the results, the quality of roof-harvested rainwater and tap water is within the water standard, except for free residual chlorine in rainwater and fluoride in tap water. Tap water generally has a higher concentration of dissolved ions, notably chloride, sodium and sulphate, compared to rainwater (Figure 9). Bacterial growth is absent in tap water due to the presence of residual chlorine. No toxic heavy metal was detected in either of the water samples.

Figure 9.

Graph of ionic concentration in rainwater and tap water samples

Provided the roof was in good condition, zinc ion level from roof-harvested rainwater collected from galvanised metal roofs in Austin, United States, was not higher than that of other types of roofs [27], despite the fact that metallic roofs tend to leach metallic ions into roof-harvested rainwater because rainwater is somewhat acidic. Heavy metals such as Cr, Ni, Cu, Pb, Mn, Co, Zn, and Cd ions were detected in roof-harvested rainwater samples collected in Yatta, Palestine, with Pb, Cr, Ni and Zn ions in some samples exceeding WHO limits [28]. Roof-harvested rainwater from concrete and ceramic tile roofs generally has better overall water quality compared to asphalt and galvanised metal roofs [29] in Shanghai, China. Rainfall amount and rainfall intensity have a negative correlation with various rainwater contaminant levels, −0.44 to −0.31 and −0.36 to −0.21, respectively. Simply put, rainwater quality improved with increasing rainfall amount and intensity. Interestingly, there was no significant correlation between ADD and rainwater quality.

Solid, clear ice crystallised gradually underneath the falling film of circulating solution as soon as a few minutes after the start of the experiment, as the temperature of the stainless steel cooled surface (Figure 10) reached below the freezing point of water.

Figure 10.

Initial crystallisation of ice

Figure 11.

Graph of temperature against time during freezing of distilled water

As seen in Figure 11, the temperature of the circulating solution decreased linearly with time. It never fell below the freezing point before ice formed, indicating that neither distilled water nor roof-harvested rainwater experienced supercooling during cooling and ice formation. If supercooling had occurred during the experiment, a characteristic dip (shown in dotted line) would appear in the actual temperature graph. Supercooling might not happen in a falling-film PFC configuration, in contrast to other setups, most likely because of the way the fluid was circulated in the former.

In a falling-film equipment, the circulating fluid became turbulent and aerated when it was forced through the distributor above the cooled surface and flowed in a thin film of a few millimetres, hence the name falling-film. Air bubbles entrapped in the falling film seemed to prevent supercooling by acting as ice-nucleating sites [30]. As opposed to the falling-film configuration, the traditional PFC system that uses stirrers was unable to produce enough agitation to encourage heterogeneous nucleation. More PFC studies should be done with falling-film equipment to replicate these observations.

While supercooling has been widely cited as the leading cause of impure ice, several other theories offer alternative explanations for how impurities get included in ice. In the v/k criterion theory, where v is the maximum allowable growth rate of ice and k is the mass transfer coefficient, pure ice crystals can form provided the value of v is not exceeded [31]. The value of v depends on the concentration of impurities in the bulk fluid and the mass transfer coefficient k. This theory also explains that forced convection can improve crystal purity as the value of k increases with stronger circulation of the bulk fluid. However, this theory is limited due to its temperature independence and does not take into account the effects of heat transfer. Various derivations and forms of the theory can be found in [32]. On the other hand, in the gradient criterion theory, pure ice crystals can form as long as constitutional supercooling does not occur by having the real temperature gradient higher than or equal to the equilibrium temperature gradient [31]. The real temperature gradient is a function of heat transfer coefficient, heat conductivity and temperature of bulk fluid, while the equilibrium temperature gradient is a function of linear gradient in the phase diagram, crystal growth rate and diffusion coefficient.

Like the v/k criterion, these theories do not predict the separation efficiencies given process parameters, as they mainly serve to provide a theoretical framework of how impurities are incorporated during the crystallisation process. Lastly, in the inclusion migration theory, pure ice crystals can form as long as the crystal growth rate is lower than the migration rate of inclusion, which is a function of diffusion coefficient, concentration at the crystallising side of the inclusion and temperature gradient. More detailed explanations of the various theories can be found in [31]. The theories described here are not exhaustive and may be helpful in models to describe PFC under different circumstances.

Table 4 displays a summary of the experimental data. By limiting the incorporation of foreign ions during the growth of ice crystals, PFC significantly reduced the TDS of rainwater.

Purge or concentrate was found to have higher level of TDS and colour, suggesting that the coloured organic materials dissolved in the rainwater were distributed unevenly between solid ice phase and liquid purge phase, attaining separation of contaminants in rainwater according to mass balance, since contaminants must be concentrated in purge phase, even if no discernible change was seen in colour or turbidity between feed and purified rainwater. Since it was not possible to measure the purified rainwater TDS due to the TDS meter detection limit (< 1 ppm), the values of RE and K for both circulation times were given as ranges in Table 4.

Figure 12 shows that the yield of 20 min circulation time is 15.0% higher than 26 min (p < 0.001). Although a longer circulation time did increase the mass of ice formed, the increase in mass did not scale proportionally with the increase in time. As ice increased in thickness, the rate of heat transfer decreased, making any increase in circulation time a diminishing return in the increase in mass of ice. Therefore, the yield decreased for longer circulation time. Meanwhile, Figure 13 shows that the CF of 26 min circulation time is 15.0 % higher than 20 min (p = 0.001). This result means that longer circulation time can produce a purge with higher concentration but lower yield of ice. Since the ideal circulation time varies depending on the experimental setting and feed solution, it should be assessed case-by-case in pilot experiments before scaling up to industrial applications.

Figure 12.

Graph of yield against feed TDS

Feed rainwater TDS did not significantly influence yield, K, and RE. On the other hand, CF was found to be strongly negatively correlated (r = −0.766, p = 0.045) and (r = −0.756, p = 0.049) for 20 min and 26 min, respectively, with feed rainwater TDS (Figure 13). This implies that the TDS concentration effect of PFC weakens as feed TDS increases. No significant correlation was found between yield and feed rainwater TDS (Figure 12). In a sodium chloride solution ranging from 1.75 wt% to 3.5 wt%, K improved (i.e. approached zero) with a more dilute solution [33]. CF of calcium ions increased (1.6 to 2.1) with initial concentration [34] while the performance of PFC did not correlate with the pH of the feed solution. However, CF decreased with increasing concentration of cow cheese whey protein [35]. In glucose solution, when the circulation duration increased from 40 min to 80 min, both K and yield showed a declining trend [36].

Figure 13.

Graph of CF against feed TDS

Nonetheless, the yield of both sets of experiments (840 – 970 ml/h) was far lower than the rated capacity of 68 kg ice production per 24 hours or 2,830 ml/h. This result is due to the rated capacity of the equipment, which was measured by the manufacturer with water inlet temperature and ambient temperature of 20 ℃, whereas in this study, the water inlet temperature and ambient temperature were 25 ℃. The rate of ice production is dependent on two factors: the amount of heat to be rejected and the rate of heat rejection to the environment. The higher the water inlet temperature, the higher the amount of sensible heat to be rejected and the longer the time required to lower the water temperature to freezing point. Likewise, the higher the ambient temperature, the lower the rate of heat rejection and the less ice is produced per cycle. These two reasons combined resulted in the ice-making equipment yielding less ice compared to the rated capacity. This situation is actual for all refrigeration equipment as their performances are highly sensitive to operating temperatures.

Table 5 shows the comparison of energy consumption and footprint between PFC in the current study and a conventional method for producing pure water. The energy consumption of PFC is 280 kJ/kg water, which is only 10.4% of the energy consumption of a similarly-sized distillation water purifier. PFC energy consumption can be further reduced by scaling up capacity, improving heat recovery (cooling feed rainwater with thawing ice), and improving heat transfer efficiency. The equipment footprints for both methods are similar and require a space of less than 1 m² for 1 kg/h water production capacity.

The main reason the PFC freezing technique was chosen in this research to treat rainwater was that freezing water can produce high-purity ice crystals in a single process with low energy consumption. In contrast, other processes require several different processing steps in series to produce final pure water. Therefore, by furthering research into PFC to overcome its drawbacks, PFC can be used to purify rainwater into drinking water with less costly and complicated equipment.