Characterisation of First Flush for Rainwater Harvesting Purposes in Buildings

Abstract

The objective of this research was to assess the first flush of rainwater harvested from a fibre–cement roof in southern Brazil. Runoff samples were collected for quantifying pH, total suspended solids, turbidity, conductivity, apparent colour, total coliforms, and Escherichia coli. Statistical methods were employed to describe the data, establish correlations between variables, and assess if the antecedent dry weather periods and rainfall intensity affected water quality. The qualitative characterisation of the first flush was performed using principal component analysis and simple regression analyses. The results show that rainwater runoff can be highly contaminated. Hypothesis tests showed that initial rainfall intensity and antecedent dry weather periods affect the quality of the first flush. Principal component analysis suggested that the most significant variables to characterise the first flush were turbidity and apparent colour. Using first-flush diverters in rainwater harvesting systems does not ensure E. coli removal, but it may reduce the risk of users’ contamination. Practical implications include discussions on the suggested first flush and the consequential impact on the quantity and quality of rainwater harvested. Future studies may consider using the method used in this research to develop guidelines based on more samples across the country. As novelty, one includes a statistically robust qualitative study in a region that lacks research on the quantification and quality of first flush. Such assessment helps to build up Brazilian data for a better understanding of first flush management in rainwater harvesting.

1. Introduction

The relevance of studying the first flush associated with rainwater harvesting systems (RWHSs) is primarily related to water quality assessment. However, quantitative analyses of the volume of initial runoff to be diverted may also be important, especially in locations experiencing droughts [1]. According to the United Nations [2], climate change is expected to increase seasonal water variability, potentially leading to greater water scarcity where water stress is already being faced or even causing water shortages in previously unaffected areas. These changes highlight the importance of reducing losses, whether from conventional water supply systems or RWHSs.

The first flush corresponds to the initial amount of rainwater runoff in which the highest concentration of pollutants is found [3,4,5,6,7]. Although not all RWHSs have a device for the first flush diversion [8,9], the installation of such a mechanism is recommended, reducing the pollutant load in the harvested rainwater and providing a safer system for users [10,11,12,13].

Some definitions with a more quantitative approach regarding the first flush have already been proposed in the literature. Bertrand-Krajewski et al. [14], for example, proposed the existence of a first flush when at least 80% of the mass of pollutants on the runoff surface was transported during the first 30% of precipitation volume. According to Deletic [15], the first flush corresponds to the total percentage of the pollutant load of a rainfall event that is carried with the first 20% of the runoff. Geiger [16] states that a first flush effect exists when the percentage of the cumulative mass of pollutants is greater than the percentage of the corresponding cumulative runoff volume in an M(V) curve. Bach et al. [17] proposed a new methodology, also applied by Todeschini et al. [18], in which one should find the average concentrations of pollutants in the catchment area for a certain increment in runoff volume using pollutographs of rainfall events. Finally, non-parametric tests are used to define a characteristic photograph, and then it is possible to quantify the first flush.

Different studies, according to factors such as location and/or analysis methodology, point to different first flush diversion volumes. Amin et al. [19] considered the first flush as the first millimetre of runoff. Shaheed and Mohtar [20] defined the first flush as the first three millimetres of runoff. Such discrepancies are justified once many factors may affect the rainwater quality, such as the antecedent dry weather period, the material of the catchment surface, characteristics of the rainfall event, air quality, and the surrounding region of the building [21,22,23]. The Brazilian standard—NBR 15527 [24]—recommends, in the absence of data regarding the appropriate volume of first flush to be diverted, discarding the initial 2.0 mm of rainfall. Kus et al. [25] demonstrate that diverting the first two millimetres of runoff provides water with quality indicators that meet most of the requirements indicated by the Australian guidelines for drinking water quality. Martinson and Thomas [26] propose that diverting each millimetre of initial runoff from roofs reduces the contaminant load in rainwater by 50%.

In general, studies that evaluate the first flush quality in the context of RWHSs are conducted by collecting samples of runoff from real or pilot roofs. For catchment and storage purposes, arrangements are used. They consist of pipes and devices capable of storing the initial runoff in the arrangement order of the collector bottles [26,27,28,29]. This type of catchment system allows for assessing the quality of different runoff volumes throughout the rainfall event.

A decrease in pollutant concentration during roof runoff has commonly been observed [23,30,31]. Furthermore, there are other findings that should be highlighted: Sanjeeva and Puttaswamaiah [32] noted that rainfall events with acidic pH tend to cause the leaching of metals from roofs, thereby increasing metal concentration in the first flush; air quality parameters may be strongly correlated with runoff water quality parameters [33]; the roof material may affect the first flush quality [34,35]; the roof slope affects particle wash-off so that increasing the slope also increases the first flush effect [36]. Further research is needed to assess the influence of rainfall intensity, rainfall height, and antecedent dry weather period on the quality of the first flush. There is a lack of consensus among researchers regarding the potential effects of such parameters on the quality of the initial amount of the roof runoff [29,34,37,38].

With regard to more recent studies, Livhuwani et al. [39], for example, concluded that the type of roof does not influence the quality of the rainwater obtained, but that the first flush should be discarded because of the greater presence of Escherichia coli. Barriga et al. [40] found differences between roof types, demonstrating the variability in research into the influence of catchment parameters on quality. In the case of this study, all the samples, after treatment and considering the first flush, met the specified non-potable uses. The study by Lay et al. [41] should also be mentioned, as it is one of the most comprehensive to have recently evaluated the effect of the first flush. As a conclusion, the authors found that by removing 1.2 mm of runoff, up to 75% of the total suspended solids were removed, depending on which roofing material was used. In contrast, the study by Charlebois et al. [11] concluded that the ‘2 mm removal rule’, similar to that presented by the Brazilian standard—NBR 15527 [24]—is not sufficient to remove organic matter and dirt from the roof. The authors, therefore, recommended that the first flush be designed according to the rainfall characteristics and the building’s surroundings. Finally, the review paper by Gupta et al. [42] also evaluated recent studies, indicating that the volume of rainwater which needs to be discarded as the first flush remains unanswered, and, therefore, future research needs to address it. Rainwater treatment with regular monitoring is important for the control and assessment of rainwater harvesting systems.

Considering the importance of characterising the initial runoff from catchment areas of RWHSs, in addition to the gaps found in the literature on this topic, this paper aims to assess the first flush from a fibre–cement roof located in Florianópolis. Tests for the runoff qualitative characterisation and statistical analyses of the data obtained were performed. The following quality parameters were investigated: pH, apparent colour, turbidity, total suspended solids, conductivity, total coliforms, and Escherichia coli (E. coli). Statistical analyses were performed to assess potential differences in the quality of rainwater based on the antecedent dry weather period and rainfall intensity. Lastly, a principal component analysis was carried out to identify the parameters that most influence the characteristics of the first flush. Therefore, one hopes to improve the literature on first flush by increasing the existing case studies on first flush quality parameters, and help to shape a better understanding of the necessary volume to be diverted.

2. Materials and Methods

In order to qualitatively characterise the first flush, an experiment was conducted to collect and analyse runoff samples from a roof at the Federal University of Santa Catarina in Florianópolis, southern Brazil. The innovation brought about by this research consists of a pioneering study in southern Brazil, evaluating different characteristics that can influence water quality from the removal of the first flush. Various studies have pointed to the inadequacy of the normative value of 2 mm as a suggested first flush, and this study stands as a reflection from a Brazilian point of view.

2.1. Study Location Climate

Florianópolis is located in the state of Santa Catarina and has a humid subtropical climate, with hot summers and cold winters [43]. Rainfall is distributed throughout the year without a well-defined dry season [44]. In winter, the rainfall volumes are the lowest, while summer tends to be the rainiest season. The combination of intense heat and high humidity leads to tropical convection, resulting in the formation of clouds and heavy rainfall [45].

2.2. Roof Characteristics and Experimental Site

The surrounding characteristics of the roof selected represent an unfavourable scenario for collecting samples of good-quality rainwater. The experimental site is close to trees (390 m away from an ecological park) and near two heavily trafficked roads. Other criteria were observed for the selection: the location is sparsely populated (reducing the possibility of interference in the experiment), and the roof is unpainted and made of fibre–cement sheets, a material widely used in buildings in Brazil.

Water samples were collected from a fraction of the building’s roof (horizontal projection area of 5 m²). The selected roof is over 35 years old (installed in 1985), contains asbestos in its composition, and has a slope of 13.5°. Currently, the use of asbestos in Brazil is prohibited by law, but older buildings still have this mineral, especially on roofs. Figure 1 shows the roof location and its surroundings.

2.3. Experimental Arrangement

The experimental arrangement consisted of a PVC gutter connected to a downspout, which had a flexible grate installed to prevent system blockage. The downspout was connected to vertical piping, with the lower end attached to a funnel. The funnel was supported by a PVC pipe—tilted at 5°, as recommended by Yaziz et al. [31]—which directed the runoff to containers and facilitated their sequential filling. Five glass bottles (with 4.70 L of capacity) were used as collectors. Each bottle had an expanded polystyrene ball inserted to block the entry of water after the bottle was completely filled. Butyl rubber fragments were used to ensure sealing at the bottle connections. The pipes used in the arrangement met the sizing requirements described in the Brazilian standard for rainwater systems [46].

Considering the losses on the roof, each bottle collected approximately 1 mm of rainfall. Some runoff accumulated in the region near the bottle opening surface (an approximate volume of 22 mL, or 0.5% of the volume corresponding to the bottle’s capacity). Since such accumulations represent a very small portion of the runoff, they were disregarded. Figure 2 shows the arrangement for collecting the first flush and a schematic drawing of the region where rainwater accumulation occurs.

2.4. Rainfall Intensities and Antecedent Dry Weather Period

Groups considering different ranges of rainfall intensity and antecedent dry weather periods were defined using statistical analysis of rainfall data from the UFSC meteorological station. The definition was created using the percentiles of the distributions. Strong intensity events were those whose intensities were above or equal to the 66.66% percentile. Moderate intensity events presented intensities within the percentile between 33.33% and 66.66%, weak intensity events presented intensities within the percentile less than or equal to 33.33%. Two groups of antecedent dry weather periods were considered: events with a long antecedent dry weather period and events with a short antecedent dry weather period. Events with a long antecedent dry weather period were those whose period of dry weather was equal to or greater than the integer value immediately above the mean. Events with a short antecedent dry weather period were those with a period smaller than the integer value immediately above the mean. Therefore, runoff samples were collected in six situations: during long and short antecedent dry weather periods between rainfall events and during high-, medium-, and low-intensity events. With sufficient rainfall volume in each event to fill all bottles, six samples were collected per event. Five samples were stored in the bottles, and one sample was directly collected from the sky into a glass container.

Near the experiment site, a tipping bucket rain gauge was installed, and in conjunction with a microcontroller and a data transmission module, it recorded the date and time whenever the rainfall depth reached 0.25 mm. Finally, the rainfall intensity corresponding to the first 6.25 mm of rainfall was calculated for each event, according to Equations (1) and (2).

$$\mathrm{I}={\displaystyle \frac{{\sum }_{\mathrm{i}}^{\mathrm{n}}{\mathrm{I}}_{\mathrm{i}}}{\mathrm{n}}}$$

(1)

$${\mathrm{I}}_{\mathrm{i}}={\displaystyle \frac{0.25}{{∆\mathrm{T}}_{\mathrm{i}}}}$$

(2)

where I is the mean intensity associated with the first flush of the rainfall event (mm/s); i is the number of the signal sent by the rain gauge during the precipitation event; n represents the total number of signals sent by the rain gauge up to a rainfall height of 6.25 mm; I_i is the rainfall intensity referring to signal i (mm/s); ∆T_i is the time interval required to complete the volume of one of the tipping bucket compartments (s).

2.5. Sample Collection and Transportation

The materials for the first flush collection were prepared the day before the rainfall events: the expanded polystyrene spheres and the butyl rubber seal were rinsed; the first flush collection bottles were rinsed and sterilised using boiling water. After cleaning, the five bottles were arranged at the experiment site, attached to the pipeline, with the openings’ outer side surfaces covered by the sealing rubber. In order to collect the first flush samples, two types of containers were used: polypropylene bottles, with a capacity of 1000 mL for conducting physicochemical tests, and borosilicate glass bottles, for quantification of total coliforms and E. coli. The polypropylene bottles were first rinsed with potable water and then sanitised with distilled water. The borosilicate glass bottles were autoclaved at the Integrated Environment Laboratory (LIMA) of the Department of Sanitary and Environmental Engineering at UFSC for one hour at a pressure of 1.2 kgf/cm² and a temperature of 120 °C.

A sterilised glass container was also prepared to capture rainwater directly from the sky. Samples were manually collected on the terrace of the Civil Engineering Department at UFSC. The samples were collected immediately after the end of the rainfall event. Firstly, each collection bottle was shaken to homogenise the first flush. Then, a portion of the first flush was transferred to the autoclaved bottle, and another portion was transferred to the polypropylene bottle. Subsequently, a portion of the first flush collected (in the common glass bottle) was discarded, and the bottle was shaken again to homogenise the remaining runoff.

New portions of the first flush were transferred to each of the sample collectors. This procedure of transferring portions of homogenised runoff to collectors, followed by runoff discarding, was repeated until the capacity of the sample collectors was reached. The sample collection was carried out as described for all five bottles in the arrangement, with volumes corresponding to each runoff portion (1, 2, 3, 4, and 5 mm). This procedure was performed to ensure that the collected samples faithfully represent the quality of the captured first flush. The sample collectors were sealed and refrigerated until it was time to conduct the laboratory tests [47]. The samples were placed in thermal boxes and transported to the laboratory. All analyses were performed within 24 h after collection.

2.6. Laboratory Tests

Important water quality parameters for non-potable purposes were tested, namely, pH, apparent colour, turbidity, total suspended solids, conductivity, total coliforms, and E. coli. Before each test, the sample bottles were shaken to homogenise any pollutants present in the first flush. All equipment used in the analyses was previously calibrated.

For pH measurement, a pH meter was used. The reading was obtained once no further changes in the results were observed. Calibration was performed by the sensor immersion in buffer solutions with a pH of 4, 7, and 10 for recognition and calibration of the standard results.

Technical standards for water quality often provide threshold limits for parameters that characterise water appearance, such as turbidity, apparent colour, and total suspended solids [24,48,49]. Considering rainwater usage for non-potable purposes, the evaluation of such parameters is relevant as it can prevent wastage. For instance, if the water to be used in a toilet is highly turbid or yellowish, the user may repeatedly flush the toilet, thinking that a single flush was insufficient to transport the excreta.

The apparent colour test was performed using a spectrophotometer—calibrated by a specialised technician—at 455 nm using the platinum–cobalt method. The turbidimeter used in turbidity tests was calibrated using formazine standard solutions with different concentrations. The equipment measured turbidity by the nephelometric method. Conductivity was measured using a manually calibrated conductivity meter with the sensor inserted in a standard solution of 1413 µS/cm (±0.5% at 25 °C). Characteristics of the equipment used during the tests are presented in

Table 1. Characteristics of equipment used during laboratory tests.

Equipment	Brand and Model	Manufacturer	Measurement Ranges	Resolution	Accuracy	NBR Range for Non-Potable Uses 1
Portable pH meter	Kasvi K39-0014PA	Kasvi (Brazil)	0.0 to 14.0	0.1	±0.1	pH limits from 6.0 to 9.0
Benchtop turbidity meter	Tecnopon TB1000	MS Tecnopon (Brazil)	0 to 1 NTU	0.01 NTU	<0.05	Turbidity limits from 0 to 5 UT 2
			0 to 10 NTU	0.01 NTU
			0 to 100 NTU	0.1 NTU
			0 to 1000 NTU	1 NTU
Portable conductivity meter	Orion Model 115	Thermo Scientific (USA)	0.0 to 199.9 µS	0.1 µS	±0.5% of complete-scale reading into each band	Conductivity limits from 0 to 3200 µS/cm
			200 to 1999 µS
			2.00 to 19.99 mS
			20.0 to 199.9 mS
Spectrophotometer	Hach DR3900	Hach (USA)	320 nm to 1100 nm	1 nm	±1.5	Not included

Notes: ¹ Range of standard NBR 16783 (ABNT, 2019) [50]. Similar recommendations are shown in the standard NBR 15527 [24]. ² UT is the generic Unit of Turbidity, while NTU stands for Nephelometric Turbidity Unit.

.

The quantification of total suspended solids was performed using 0.6 µm porosity borosilicate glass fibre filter membranes, which were free from binding agents. They were previously heated in a muffle at 550 °C for approximately one hour. Then, rainwater samples were filtered using the pre-prepared membranes and a vacuum filtration system. The total suspended solids concentration was calculated based on the difference between the masses of the membranes before and after filtration.

The assessment of microbiological parameters was performed using samples collected in borosilicate glass collectors and analysed using Colilert tests from Idexx Laboratories (USA). The samples were dissolved in the test reagent, and the resulting solution was poured into a quanti-tray. Then, the quanti-tray was sealed and incubated at 35 °C (±0.5 °C) for 24 h. After incubation, the number of yellow and fluorescent cells was counted to estimate the total coliforms and E. coli colonies, respectively. Microbiological quantification was performed using the most probable number table for the 97-cell quanti-tray.

It is essential to highlight that rainwater and non-potable water quality requirements vary in each country, and many standards address this topic. In Brazil, two standards are applicable in this matter: NBR 16783 [50] and NBR 15527 [24]. The former is more comprehensive and covers requirements for non-potable use in buildings, addressing different water sources. The latter focuses exclusively on water collected from roofs and is more specific and similar to the one presented in this study. In general, NBR 16783 requires the following parameters: pH, E. coli, turbidity, biochemical oxygen demand, free residual chlorine, total dissolved solids, and electrical conductivity. NBR 15527 only requires pH, turbidity, and E. coli, and the water can be used for water cooling systems, toilet flushing, washing, firefighting, ornamental use, and irrigation.

2.7. Analysis of Experimental Data

Firstly, the experimental data were summarised using tables, indicating means, standard deviations, maximums and minimums of rainfall intensity, antecedent dry weather periods, rainfall duration, and the water quality indicators analysed. Then, statistical tests were applied to investigate the behaviour of the runoff water quality parameters, as well as the influence of the variables, such as antecedent dry weather periods between rainfall events and rainfall intensity, on the first flush data.

2.7.1. Water Quality Parameters

The behaviour of the quality parameters during the runoff was explored using scatter plots, including all first flush samples. The analysis allowed for examining the possibility of approximating the indicators’ behaviour to an exponential decay function of pollutant concentration with the increase in the first flush volume, as shown by Doyle [30]. Through the analysis of the behaviours obtained from the scatter plots, simple regression analyses were performed, testing the exponential model for each water quality parameter. Regression analyses were carried out with the data normalised in relation to the data obtained for the first millimetre collected. All regression analyses were conducted using the R programming language. The models were obtained through the least squares method.

2.7.2. Parameters Related to Rainfall

Hypothesis tests were carried out to assess the influence of initial rainfall intensity and the antecedent dry weather period between rainfall events on the quality of rainwater, with a significance level of 5%. Firstly, the Shapiro–Wilk normality test was applied to the water quality parameters. It assumes, by the null hypothesis, that the sample is derived from a normally distributed population. In the case of non-rejection of the null hypothesis, that is, when the p-value of the test is higher than the significance level, the analysis proceeded with the parametric test, either One-Way ANOVA or Student’s t-test. The t-test was applied to determine statistically significant differences in runoff water quality between long and short antecedent dry weather periods between rainfall events. The ANOVA test was used to investigate the potential existence of significant differences in the runoff water quality among different categories of rainfall intensity.

Non-parametric tests were applied when the null hypothesis of the normality test was rejected. The Kruskal–Wallis test assessed the potential differences in pollutant concentrations in the runoff between rainfall events of weak, moderate, and heavy intensities. The Mann–Whitney U test evaluated the possible existence of significant differences in pollutant concentrations of the runoff considering two antecedent dry weather period categories between rainfall events (short and long). When significant differences were identified for data from different rainfall intensity categories, Dunn’s post hoc test was applied to identify which groups were statistically different.

2.7.3. Linear Regression

In order to identify the parameters that most influence the characteristics of roof runoff, principal component analysis (PCA) was conducted, a commonly employed procedure in studies investigating the first flush phenomenon [29,51,52,53,54,55]. The following variables were included in the analysis: collected rainfall depth, rainfall intensity, antecedent dry weather period, rainfall volume, rainfall duration, faecal coliforms, E. coli, turbidity, apparent colour, pH, conductivity, and total suspended solids. All variables included in the analysis were standardised.

Firstly, the Pearson correlation matrix was determined. Subsequently, pairs of variables with Pearson correlation coefficients higher than 0.70 were identified, and linear regression analyses were conducted between them. Only the analyses with higher coefficients of determination were presented, and the overall significance of the models was assessed.

Afterwards, the factors were defined, with each factor corresponding to a grouping of variables. Since twelve variables were included in the analysis, the maximum number of factors to be obtained is twelve. The determination of each factor depends on calculating its scores, which were determined after obtaining the eigenvalues and eigenvectors of the Pearson correlation matrix.

According to Favero and Belfiore [56], determining the appropriate number of factors to represent the original variables is the researcher’s responsibility. Thus, in our analysis, factors with the three highest eigenvalues were considered. In order to assess which factors have the strongest correlation with the main variables, factor loadings were defined, i.e., Pearson correlations were performed between the original variables and each of the factors considered. Additionally, the overall adequacy of the principal component analysis was assessed using the Kaiser–Meyer–Olkin (KMO) statistic and Barlett’s test of sphericity, as recommended by Favero and Belfiore [56].

3. Results

3.1. Categories of Parameters Associated with Rainfall Events

The rainfall data analysis resulted in the categorisation of events into strong, moderate, and weak intensity. Two groups of antecedent dry weather periods were established: long periods and short dry weather periods. The numerical range considered for each group is shown in

Table 2. Definition of categories of rainfall intensity and antecedent dry weather period.

Variable	Category	Range Definition	Range
Rainfall intensity	Intense	Intensities higher than or equal to the percentile 66.66%	Intensity ≥ 13.7 mm/h
	Moderate	Intensities higher than the percentile 33.33% and less than the percentile 66.66%	2.7 mm/h < Intensity < 13.7 mm/h
	Weak	Intensities less than or equal to the 33.33% percentile	Intensity ≤ 2.7 mm/h
Antecedent dry weather period	Long	Period higher than or equal to the integer value immediately above average	Dry weather period ≥ 4 days
	Short	Period less than the integer value immediately above average	Dry weather period < 4 days

. It is important to remember that rainfall intensity corresponds to the average intensity measured for every 0.25 mm of rain during the first 6.25 mm. If less than 6.25 mm had been measured, total rainfall in the event was considered.

3.2. Description of Environmental Variables and Water Quality Parameters

During the study, 58 samples were collected from eleven rainfall events. The collection of samples was carried out between August 2020 and February 2021. The information regarding the number of collections conducted per rainfall intensity category and antecedent dry weather period between rainfall events, as well as the amount of first flush collected in each event, is presented in

Table 3. Rainfall events analysed and the amount of the first flush collected.

Antecedent Dry Weather Period	Rainfall Intensity	Date of the Rainfall Event	First Flush Collected (mm)
Long	Weak	27–28 September 2020	5
		12–13 August 2020	5
	Intense	13–14 December 2020	5
		24 February 2021	5
Short	Weak	1 September 2020	4
		19–20 January 2021	5
	Moderate	25–26 October 2020	4
		16 November 2020	5
		30 November 2020	5
	Intense	7 October 2020	1
		1 February 2021	5

. Three events had less than 5 mm of the first flush collected because they had a small rainfall height in the period.

The mean duration of the rainfall events was 8.56 h, with the longest event lasting 31 h and the shortest event lasting just under one hour (0.88 h). The mean total accumulated rainfall was 16.90 mm, ranging from 4.5 mm to 57.25 mm. The mean initial rainfall intensity of the sampled events was 10.25 mm/h, ranging from 0.62 mm/h to 35.04 mm/h, and the mean antecedent dry weather period was 4.39 days (minimum of 0.33 days and maximum of 14.00 days).

Table 4. Statistical description of water quality parameters.

Parameters	Rainwater Collected Directly from the Sky			First Flush
	Mean	Standard Deviation	Range	Mean	Standard Deviation	Range
Total Suspended Solids (mg/L)	5.21	3.47	0.30 to 9.33	9.37	5.39	1.00 to 21.67
pH	6.6	0.3	6.1 to 7.0	8.2	0.3	7.3 to 8.8
Turbidity (NTU)	2.34	3.58	0.14 to 11.73	3.1	4.2	0.21 to 16.40
Conductivity (µS)	36.8	36.4	6.6 to 113.5	86.3	50.8	17.1 to 278.0
Apparent colour (PtCo)	17	15	0 to 50	70	58	24 to 291
Total coliforms (MPN/100 mL)	31.4	45.3	2.0 to 137.6	1954.3	706.4	461.1 to >2419.6
E. coli (MPN/100 mL)	0.9	1.4	<1 to 3.1	375.8	692.8	<1 to >2419.6

shows the statistical description of the water quality parameters for the samples collected directly from the sky and for the first flush samples.

The water collected directly from the sky has lower pollutant concentrations when compared to the first flush. The samples from direct collection showed slightly acidic pH, but close to 7.0, and relatively low concentrations of E. coli and turbidity, generally complying with the limits established in NBR 15527 [24] for non-potable uses. Total coliforms were detected in all collected rainwater samples. Among the samples collected directly from the sky, only one showed a total coliform concentration of colonies exceeding 50 MPN/100 mL.

Regarding the presence of total coliforms in rainwater collected directly from the sky, the detection was observed in 100% of the samples, with a mean most probable number (MPN) of 31.4 colonies per 100 mL of sample. Only one rainwater sample had a total coliform concentration exceeding 50 MPN/100 mL. The mean concentration of E. coli in the samples which were not intercepted by the roof was low, and in five events, no colonies were detected.

The pH of first flush samples collected from the fibre–cement roof is higher than the pH of directly collected rainwater, which is consistent with findings from other studies [22,32,57,58]. However, the pH obtained for all samples complies with the recommendations established in the Brazilian standard for non-potable rainwater quality [24]. Analyses of samples from fibre–cement roofs in Poland also had an alkaline pH [59], which is consistent with the findings of this study. Due to the age of the fibre–cement roofing, it is presumed that leaching of cementitious components occurs, neutralising the acidity of the runoff water and making it more alkaline.

The concentration of total suspended solids in the runoff samples was relatively low when compared to results of other studies that also analysed the quality of runoff from fibre–cement roofs [57,60]. However, according to the Brazilian Water Agency [61], for Class 1 reuse water (intended for use in toilet flushing; floor, clothes, vehicle washing; or for ornamental purposes), the concentration of total suspended solids should not exceed 5 mg/L. Thus, at least a portion of the runoff collected would not be suitable for direct use.

The mean conductivity of the runoff samples was higher than the mean obtained in the study conducted by Chizoruo and Onyekachi [57]—65.68 µS—and lower than the mean reported in the work of Polkowska [59], corresponding to 171.27 µS. In this experiment, most of the samples showed conductivity ranging from zero to 139.5 µS. Electrical conductivity is indicative of dissolved ions in water. Considering that the asbestos–cement roofing used in this experiment has been exposed to weathering for a long period, and among the roofing components, the cementitious matrix has the lowest resistance [62], it is, therefore, the first to undergo disintegration. Consequently, ion dissolution is presumed, such as calcium (derived from the dissolution of calcium hydroxide), as well as other ions like chlorine, nitrate, sulfate, sodium, potassium, and magnesium [59].

Regarding colour, the mean obtained is seven times higher than the limit indicated for reuse water [61]. However, Table 4 shows that not even the minimum apparent colour obtained meets the limit defined by the Brazilian Water Agency. In general, the samples had apparent colour ranging from zero to 100 PtCo. The mean turbidity of the initial runoff was below the limit established in NBR 15527 [24]; nevertheless, some samples had higher turbidity concentrations, with the highest concentration being 16.40 NTU. Although some turbidity concentrations exceeded the limit, the majority of samples (slightly over 70%) had concentrations within the range from 0 to 2.75 NTU. Furthermore, the mean turbidity found in the samples of this study was lower than those obtained in studies conducted in Colombia and Nigeria [57,60].

Regarding the total coliforms in the first flush, the majority of runoff samples exceeded the upper limit of the test (2419.6 MPN/100 mL). However, concerning the E. coli concentration, most of the runoff had concentrations within the range from zero to 483.7 MPN/100 mL. Although the mean concentration of such microorganisms in the first flush exceeded the limit established in NBR 15527 [24], slightly over 65% of the samples had concentrations below the standard’s limit. Mendez et al. [63] also found high concentrations (with a mean between 1000 and 2000 CFU/100 mL and a maximum concentration exceeding 3000 CFU/100 mL) for the total coliform concentration in runoff from concrete tile roofs. However, those authors observed lower concentrations of faecal coliforms when compared to results found in this study (with an average below 25 CFU/100 mL and a maximum concentration below 50 CFU/100 mL). According to the World Health Organization [64], E. coli should not be detected in drinking water.

3.3. Behaviour of Water Quality Parameters as a Function of Runoff

The behaviour of first flush quality parameters was assessed using normalised data (based on concentrations obtained in the first millimetre). Figure 3 shows scatter plots for the parameters pH, total suspended solids, total coliforms, and E. coli, as well as their corresponding coefficients of determination, regression equations, and fitted curves for the exponential model (points with the same colour represent data from the same sampled event). It is essential to note that different types of curves were tested, and those with the best R² are presented in the figures. However, the method does not aim to predict dependent variable values from independent variables. This correlation was conducted solely to observe trends and discuss possible conclusions and practical perspectives.

The microbiological parameters, pH, and total suspended solids did not have a clear trend during the first flush collection. The low coefficients of the determination indicate that the exponential decrease obtained for such variables does not fit the sampled data well. Therefore, predicting pH, total suspended solids, total coliforms, or E. coli solely based on the deviation of the initial rainfall runoff is not possible. The remaining water quality parameters showed more defined trends in behaviour, as shown in Figure 4.

The models found for turbidity and conductivity explain 33% and 40% of the variance in the data, respectively, based on the runoff rainfall depth. Using the regression equation obtained for turbidity, on average, to reduce the turbidity of the runoff rainfall by 50%, approximately 2.5 mm of initial runoff would need to be diverted, unlike the result found by Martinson and Thomas [65]. The model obtained for apparent colour was the best fit for the sample data, explaining approximately 64% of the data variance. Considering the equations obtained, diverting the initial 2.5 mm of runoff would result in a reduction of 43.7% for both conductivity and apparent colour. Taking into account the deviation of the first millimetre of runoff, the reductions in turbidity and conductivity concentration would be approximately 23.7% and 20.5%, respectively.

3.4. Influence of Rainfall Intensity and Antecedent Dry Weather Period

The Shapiro–Wilk normality test and Levene’s test indicated that only the data distribution of total suspended solids concentration is approximately normal, with homogeneity of variances. For this reason, the influence of rainfall intensity and antecedent dry weather period on total suspended solids concentration was analysed using parametric tests. The data distribution for the other water quality parameters did not approach a normal curve. Therefore, the influence analysis of the independent variables on the other parameters was based on non-parametric tests.

Table 5. Assessment of the influence of the antecedent dry weather period on the first flush quality parameters.

Parameters	Test Performed	Statistical Measure	p-Value
Total suspended solids	t-test	−2.17	<0.05
pH	Mann–Whitney	448	<0.05
Apparent colour	Mann–Whitney	136	<0.05
Turbidity	Mann–Whitney	72	<0.05
Conductivity	Mann–Whitney	249	0.53
Total coliforms	Mann–Whitney	120	<0.05
E. coli	Mann–Whitney	400.5	<0.05

shows the results of the tests between the first flush quality parameters and the antecedent dry weather period. The results demonstrate that there is an effect of the antecedent dry weather period on nearly all investigated water quality parameters, except for conductivity (p-value = 0.53).

Table 6. Influence of rainfall intensity on the water quality parameters of the first flush.

Parameters	Test Performed	Statistical Measure	p-Value
Total suspended solids	One-way ANOVA	0.64	0.53
pH	Kruskal–Wallis	0.99	0.61
Apparent colour	Kruskal–Wallis	2.63	0.27
Turbidity	Kruskal–Wallis	8.99	<0.05
Conductivity	Kruskal–Wallis	9.42	<0.05
Total coliforms	Kruskal–Wallis	5.25	0.07
E. coli	Kruskal–Wallis	3.26	0.20

shows the results regarding the possible influence of rainfall intensity on the quality parameters of runoff water. Most of the parameters analysed are not affected due to rainfall intensity (p-value > 0.05). However, the Kruskal–Wallis test showed a statistically significant difference among the groups of rainfall intensity for turbidity and conductivity (p-value < 0.05).

Table 7. Description of not normally distributed water quality parameters, separated by groups of antecedent dry weather periods.

Parameters	Groups of Antecedent Dry Weather Period	Median	Interquartile Range
pH	Short	8.3	0.2
	Long	8.1	0.3
Apparent colour	Short	40	29
	Long	64	82
Turbidity	Short	0.68	1.02
	Long	3.84	8.20
Conductivity	Short	78.4	77.5
	Long	68.4	86.2
Total coliforms	Short	1733.0	1553.0
	Long	2419.6	0.0
E. coli	Short	151.0	286.0
	Long	4.6	543.0

shows the statistical description of the pH, apparent colour, turbidity, conductivity, total coliforms, and E. coli data, separated by groups of antecedent dry weather periods. The median is the measure of central tendency, and the interquartile range is the measure of data dispersion. The mean for the concentration of total suspended solids in samples collected during a short dry weather period between rainfall events was statistically lower (8.10 mg/L, with a standard deviation of 4.97 mg/L) than the mean concentration of total suspended solids in samples collected during a long dry weather period (12.10 mg/L, with a standard deviation of 5.51 mg/L). Overall, the medians of the water quality parameters for the first flush were significantly higher for long antecedent dry weather periods. Regarding conductivity, it is not possible to assert the existence of significant differences between the medians in events with long and short antecedent dry weather periods.

However, the median for the pH is significantly higher in events with short antecedent dry weather periods. This may occur because, in long dry weather periods, the probability of higher concentrations of pollutants that acidify rainfall water is increased. Regarding the median obtained for E. coli concentrations, the Mann–Whitney test showed that the median in events with short antecedent dry weather periods is statistically higher than the median for long dry weather periods, which diverges from results commonly found in the literature. A discrepant result regarding faecal coliform concentrations was also found in the study conducted by Friedler et al. [34]. They observed a negative correlation between the antecedent dry weather period and faecal coliform concentration.

Table 8. Description of water quality parameters without normal distribution, separated by rainfall intensity groups.

Parameters	Groups of Rainfall Intensity	Median	Interquartile Range
pH	Intense	8.2	0.2
	Moderate	8.2	0.2
	Weak	8.2	0.6
Apparent colour	Intense	58	26
	Moderate	40	25
	Weak	41	92
Turbidity	Intense	1.97	2.64
	Moderate	0.68	0.78
	Weak	1.15	5.87
Conductivity	Intense	45.6	25.8
	Moderate	79.4	40.3
	Weak	124.0	89.8
Total coliforms	Intense	2419.6	608.0
	Moderate	1733.0	1508.0
	Weak	2419.6	0.0
E. coli	Intense	22.8	139.0
	Moderate	142.0	1348.0
	Weak	148.0	276.0

shows the statistical description of the water quality parameters analysed (except for the concentration of total suspended solids), separated by groups of rainfall intensity. According to the one-way ANOVA test, the means of the total suspended solids concentrations for the different rainfall intensity groups (9.38 mg/L for strong intensity, 10.30 mg/L for weak intensity, and 8.00 mg/L for moderate intensity) do not differ statistically from each other. Similarly, the Kruskal–Wallis test showed no significant differences in the medians of the different rainfall intensity groups for apparent colour, total coliforms, and E. coli.

Although the Kruskal–Wallis test showed differences among one or more medians of each rainfall intensity group for turbidity and conductivity, it does not indicate which specific groups have differences. In this context, the Dunn post hoc test was conducted for pairwise comparisons between the groups. The Dunn test showed that there is a significant difference between the medians of the strong and moderate groups (p-value < 0.05) for turbidity. Therefore, it indicates that turbidity tends to be higher in events of strong intensity and lower in events of moderate intensity. The Dunn test for conductivity showed that it tends to be lower in events of strong intensity compared to events of moderate and weak intensity.

3.5. Principal Component Analysis

Principal component analysis (PCA) was performed to examine potential correlations among variables and to identify which variables, among those measured, have the most influence on characterising a rainfall event.

Table 9. Pearson’s correlation matrix.

Variables	Rainfall Duration	Rainfall Height	Initial Rainfall Intensity	Antecedent Dry Weather Period	First Flush Collected	Total Suspended Solids	pH	Turbidity	Conductivity	Apparent Colour	Total Coliforms	E. coli
Rainfall duration	1
Rainfall height	0.68	1
Initial rainfall intensity	−0.58	−0.11	1
Antecedent dry weather period	0.07	−0.17	0.03	1
First flush collected	0.00	0.07	0.02	0.03	1
Total suspended solids	0.12	0.09	0.03	0.05	−0.22	1
pH	−0.10	0.25	0.33	−0.71	−0.09	−0.41	1
Turbidity	0.12	−0.14	−0.10	0.90	−0.13	0.49	−0.76	1
Conductivity	−0.01	−0.40	−0.37	−0.10	−0.58	−0.12	−0.10	0.04	1
Apparent colour	0.16	−0.09	−0.16	0.78	−0.33	0.56	−0.78	0.89	0.24	1
Total coliforms	0.01	−0.40	−0.23	0.39	−0.40	0.09	−0.29	0.38	0.45	0.35	1
E. coli	0.38	0.11	−0.29	0.71	0.02	0.43	−0.82	0.79	0.05	0.84	0.22	1

shows the Pearson correlation matrix obtained.

The correlation matrix shows strong correlations between antecedent dry weather periods and several water quality parameters, including pH (negative correlation with a correlation coefficient of −0.71), turbidity (0.90), apparent colour (0.78), and E. coli (0.71). Correlations were also found between pH and three other quality parameters: turbidity (correlation coefficient of −0.76), apparent colour (correlation coefficient of −0.78), and E. coli (correlation coefficient of −0.82). Strong correlations were observed between turbidity and apparent colour, as well as between turbidity and E. coli and between apparent colour and E. coli.

Figure 5 shows the eigenvalues for each component obtained. The variance of each component is represented by its corresponding eigenvalue. When summing the eigenvalues, the importance of each component in explaining the data variance may be determined. Among the twelve components obtained, the first component can explain 40.14% of the data variance. The second and third components explain 18.59% and 15.86% of the data variance, respectively. Therefore, considering the first three components, it is possible to explain 74.59% of the data variance. Each component is formed by a linear combination of the twelve variables included in the analysis.

The scores and factor loadings between the variables and the three extracted components are shown in

Table 10. Scores for the first three components (factors) and factor loadings.

Variables	Component Scores			Factor Loadings Between Variables and Components
	F1	F2	F3	F1	F2	F3
Rainfall duration	0.048	0.262	0.372	0.233	0.585	0.707
Rainfall height	−0.028	0.354	0.196	−0.137	0.789	0.374
Initial rainfall intensity	−0.057	−0.037	−0.402	−0.275	−0.083	−0.764
Antecedent dry weather period	0.179	0.035	−0.178	0.864	0.077	−0.338
First flush collected	−0.041	0.246	−0.200	−0.198	0.548	−0.38
Total suspended solids	0.121	0.081	−0.089	0.581	0.182	−0.170
pH	−0.180	−0.023	0.053	−0.866	−0.052	0.102
Turbidity	0.193	0.014	−0.096	0.929	0.031	−0.183
Conductivity	0.039	−0.320	0.285	0.189	−0.714	0.543
Apparent colour	0.197	−0.017	−0.006	0.947	−0.038	−0.012
Total coliforms	0.098	−0.255	0.117	0.472	−0.569	0.222
E. coli	0.183	0.128	0.032	0.882	0.285	0.060

. The factor loadings represent the Pearson correlation coefficients obtained by correlating the original variables with each of the factors.

The variables that most contributed to the formation of the first component were the ones that characterise the visual aspect of water (turbidity and apparent colour), followed by one of the microbiological parameters analysed (E. coli), antecedent dry weather period, and pH. For the second component generation, the variables that contributed the most were the environmental ones, which show the characteristics of the rainfall events (rainfall height, duration of the rainfall event), collected rainfall height, as well as two other quality parameters: conductivity and total coliform concentration. The variables that contribute the most to the formation of the third component are the initial rainfall intensity, rainfall duration, and conductivity.

The Kaiser–Meyer–Olkin (KMO) statistic and Bartlett’s test of sphericity were used to assess the overall adequacy of the factor extraction. The result of the KMO statistic, which may range from 0 to 1, represents the proportion of variance shared among the variables considered [56]. Therefore, results close to zero indicate that the factor extraction is inadequate. For the principal component analysis conducted, the KMO statistic yielded a result of 0.68, indicating reasonable overall adequacy. The analysis performed using Bartlett’s test of sphericity demonstrated adequacy in the factor extraction, as the correlation matrix obtained differed from the identity matrix (χ² = 399.40 and p-value < 0.05).

3.6. Assessment of Correlated Water Quality Parameters

Since no trend was observed in the behaviour of E. coli concentration with runoff, and strong correlations were found between such parameter and turbidity and apparent colour, linear regression analyses were performed to further investigate the relationships between such correlated parameters. Figure 6 shows the fitted curves between ‘E. coli’ and ‘turbidity’, ‘E. coli’ and ‘apparent colour’, and ‘turbidity’ and ‘apparent colour’.

Table 11. Statistics from regression analyses for E. coli, turbidity, and apparent colour.

Correlated Water Quality Parameters	Test Statistics	p-Value
	F	ANOVA
E. coli versus turbidity	63.93	1.00 × 10−10
E. coli versus apparent colour	115.79	4.79 × 10−15
Turbidity versus apparent colour	108.99	1.46 × 10−14

shows the resulting significance probabilities from the F-test. The lighter and darker points in Figure 6 represent data from events with short and long antecedent dry weather periods, respectively. The best fit was obtained for the regression model that correlates apparent colour with E. coli concentration (R² = 0.68).

The variance tests revealed that the angular coefficients are statistically significant in all three cases, and the models are statistically significant (p-value < 0.05). When analysing Figure 6, one can observe that samples collected during long antecedent dry weather periods showed not only relatively low E. coli concentrations but also higher concentrations exceeding the limit established in NBR 15527 [24]. There is a tendency for E. coli concentrations to be lower during short dry weather periods between rainfall events. Therefore, considering the antecedent dry weather period before rainfall in first flush deviations is vital for improving the harvested rainwater quality, as it helps prevent contamination of users.

Based on the data obtained during short dry weather periods between rainfall events, it was found that 39.3% of the samples exceeded the limit provided by NBR 15527 [24] for E. coli concentration. For long periods between rainfall events, 25.0% of the samples exceeded the test’s limit detection (concentrations above 2419.6 MPN/100 mL). The regression equation obtained from the correlation between turbidity and E. coli shows that turbidity concentrations greater than 2.32 NTU in roof runoff indicate E. coli concentrations higher than 200 MPN/100 mL.

4. Discussion

The usage of first flush devices in RWHSs can drastically improve the quality of harvested water [66]. The use of a first flush device in a rainwater harvesting system in a house in Florianópolis, for example, reduced the concentrations of turbidity and apparent colour in harvested rainwater [67]. It is not uncommon to find studies in the literature on RWHSs with first flush devices that divert a small portion of the initial runoff volume. In some systems, less than 1 mm of runoff is diverted [51,68]. This study, which did not include roof cleaning before rainfall, found that even when diverting the first five millimetres of runoff, there is still a possibility of rainwater contamination with E. coli concentrations exceeding the maximum recommended by NBR 15527 [24]. Therefore, in areas with heavy vehicle traffic, near trees or other potentially polluting sources, it may be necessary to divert a larger runoff volume or use additional techniques for runoff treatment.

Shubo et al. [28], on the other hand, conducted a study in an urbanised region of Rio de Janeiro, assessing the presence of harmful viruses in the first 10 mm of runoff from a 5 m² rooftop area. The authors detected high concentrations of such microorganisms and reported that even with a larger volume of runoff diverted, rainwater may not be suitable for use unless other treatment methods are applied. Given the literature’s findings and the results found in this experiment, it is recommended that further research be conducted to assess the bacteriological concentration of rainwater, especially E. coli, in subsequent runoff volumes, such as the next 5 mm, for instance.

4.1. Practical Implications

Diverting larger volumes of runoff would have two main implications: first, it would result in the loss of rainwater available for harvesting, reducing the performance of the RWHS, and second, it would require diversion devices to have sufficient storage capacity for the first flush. This means that buildings with extensive roof areas would need relatively large storage tanks for first-flush collection, which could increase the implementation costs of the system. An alternative to large-sized first flush diversion devices would be the use of smart devices, equipped with sensors and automatic valves that allow continuous initial runoff flow (without the need for storage) until the volume indicated for diversion has been discharged.

When considering the water quality of the first flush, several points should be taken into account. The assessment of parameters such as turbidity and apparent colour has an aesthetic appeal and aims to reduce the maintenance of the system. Thus, as global water scarcity increases and users become aware of the possibility of a reduction in the aesthetic quality of water from a harvesting system, there may be greater acceptability among the population regarding rainwater aesthetics. In this context, water quality standards for non-potable purposes may be more flexible concerning such parameters. However, parameters indicative of pathogenic microorganisms, such as E. coli concentration, are more complex to be relaxed due to the risks they pose to users.

This topic introduces the context of quality management in rainwater harvesting. In many cases, even with the use of first flush, disinfection and even filtration are used to improve water quality, reducing the number of pathogens and improving aesthetic aspects that influence the use of non-potable water. In this way, one of the possible practical implications of using larger first flushes could be the lower amount of chemicals needed for treatment. In any case, this assumption needs to be tested, and this topic begins with understanding the qualitative profile of rainwater with different first flush volumes.

Given the current and future negative scenario of water distribution on Earth, the development and publication of a global international standard becomes urgent. The specification of maximum values for water quality parameter concentrations considering different types of water usage and building occupancy is suggested. For example, it may not be necessary to require the absence of E. coli in harvested rainwater used only for flushing toilets in buildings predominantly occupied by adults. The points mentioned here are not intended to discourage the use of harvesting systems but rather to foster the emergence of new proposals and research to expand the use of such systems, making them more accessible, safe, and with improved performance.

4.2. Limitations and Future Research

Some limitations are present in the article, and it is necessary to describe them to encourage new studies on the subject. First, the relatively short period of data collection is a significant limitation of the study. Future research, with a more extended period of data and sample collection, is encouraged, and it could provide more information regarding the concentration of Escherichia coli in the runoff samples collected, for example. It also may provide more data on the differences between rainfall intensities and antecedent dry weather periods.

Another limitation is the single location of the study so that the environmental variables of geographical characterisation could not be isolated. Further studies could compare results between urban and non-urban regions, near trees and in different cities in Brazil—and abroad—to understand the possible impact of these variables in the context of first flush. For example, it is expected that areas near high-traffic zones and those with a higher level of pollution will yield worse qualitative results, leading to different conclusions about the first flush. The same can be said about the type of roof, which varies between fibre–cement, concrete, and ceramic roofs, potentially producing differences of interest to practitioners in terms of practical considerations related to the first flush.

Regarding the statistical methods employed, uncertainties are associated with the data used and the prospect of univariate regression. This choice was made to use a simple method capable of observing trends from the variation in one of the parameters assessed. Other alternative methods, such as the use of multivariate regressions, the use of artificial intelligence, and machine learning methods, can be incorporated to complement the method used in future studies. In this way, uncertainties associated with the statistical method can be reduced, and the results can be made more robust.

For example, a practical application for future studies would be an intelligent system that takes rainwater quality measurements and uses such results to define the volume of the first flush and the number of chemicals (mainly chlorine) to be used for disinfection. This could connect the different points used to manage rainwater quality and ensure continuous monitoring and control. All these discussions involve the study of first flush in the context presented in this paper.

5. Conclusions

The main objective of this study was to characterise the first flush from a fibre–cement roof qualitatively. For this purpose, samples of the first five millimetres of roof runoff were collected, along with environmental data. The sampling was conducted over six months, encompassing eleven rainfall events, of which ten were suitable for first flush analysis (with 48 runoff samples and ten samples of rainfall collected directly from the sky). The water quality analyses revealed that the building’s roof is a significant source of pollution for harvested rainwater. The parameter with the highest percentage increase compared to the average result of samples collected directly from the sky was the concentration of E. coli (41,655.6%), indicating that runoff rainwater tends to be contaminated and may offer health risks to users.

The exponential regression analyses did not show an adequate fit for the models, including the collected first flush and the dependent variables’ pH, total suspended solids, total coliforms, and E. coli. For such parameters, no consistent trend in behaviour was found according to the amount of first flush drained. However, the exponential regression models obtained for turbidity, apparent colour, and conductivity were more explanatory, indicating a decreasing trend in the concentration of such pollutants with the amount of first flush drained. It is estimated that diverting the first millimetre of runoff reduces the turbidity of runoff by 23.7% and conductivity and apparent colour concentration by 20.5%. According to the models, a 50% reduction in turbidity would require diverting 2.5 millimetres of initial runoff. This diversion could reduce conductivity and apparent colour by 43.7%. Therefore, through the results obtained in this study, one may start to consider guidelines for quality improvement, such as the requirement of 2.5 mm of first flush if turbidity and apparent colour are sought to be diminished in the non-potable water.

The t-tests and Mann–Whitney tests showed the presence of a dry weather period effect for almost all water quality parameters analysed, except for conductivity. In general, higher pollutant concentrations were observed during longer dry weather periods. Regarding E. coli, the hypothesis tests showed that the median concentration of such microorganisms was higher during short dry weather periods. However, the data indicated that none of the rainfall events with a short antecedent dry weather period exceeded the limits of the E. coli quantification test (with quantities above 2419.6 MPN/100 mL). However, this did occur for samples collected during an event with a long antecedent dry weather period. Therefore, further investigation is recommended to understand the influence of the antecedent dry weather period on E. coli concentrations in rainwater runoff. Conducting more research will enable the obtaining of more precise answers that can lead to the study of solutions to minimise the concentration of such microorganisms.

The Kruskal–Wallis test showed a significant effect of rainfall intensity on the turbidity and conductivity parameters. The Dunn post hoc test identified statistically significant differences between the median turbidity results for events of strong intensity (higher median turbidity) and moderate intensity (lower median turbidity). The Dunn test also indicated that conductivity tends to be lower in events of strong intensity compared to events of moderate and weak intensity.

The principal component analysis enabled the extraction of three components that collectively explained over 74% of the data variance. The most influential variables for components one, two, and three were, respectively, apparent colour, rainfall height, and initial rainfall intensity. Strong correlations were observed between the antecedent dry weather period and the pH, turbidity, apparent colour, and E. coli parameters; between pH and turbidity, apparent colour, and E. coli parameters; between turbidity and conductivity parameters; E. coli and turbidity parameters; and apparent colour and E. coli parameters. No strong correlations were detected between rainfall intensity and the water quality parameters analysed.

Regression analyses established linear relationships between the turbidity and E. coli, apparent colour and E. coli, and turbidity and apparent colour. Investigating the behaviour of E. coli concentration in relation to the antecedent dry weather period is essential. Based on the data collected, it is not possible to affirm that diverting a specific portion of runoff will guarantee that the rainwater collected will meet the quality recommendations established in NBR 15527 [24], especially for E. coli concentration. However, diverting the initial portion of runoff tends to reduce the contaminant load, particularly those that change the rainwater colour, turbidity, and conductivity.