Pattern-based assessment of the influence of catchment characteristics on urban stormwater quality

It is critical to understand the effects of catchment characteristics on urban stormwater quality. Studies undertaken by several researchers used catchment characteristics such as fraction imperviousness, catchment area and land use, to explain the variability in stormwater quality (Brabec et al. 2002; Göbel et al. 2007; Elizabeth & Brabec 2009; Arora & Reddy 2013; Yazdi et al. 2021; Simpson et al. 2022; Zhao et al. 2022). However, the use of these catchment characteristics may not adequately explain the fundamental relationships explaining the impacts of catchment characteristics on stormwater quality (Liu et al. 2012; Ebrahimian et al. 2018; Lintern et al. 2018). Additionally, catchment hydrology has direct relationships with wash-off and transport characteristics of pollutants (dos Santos et al. 2017; Song et al. 2019; Serrano-Notivoli et al. 2022). Hence, analysis of catchment hydrology based on catchment characteristics can lead to better characterisation of stormwater quality originating from pervious and impervious surface types.

The hydrologic behaviour of a catchment is complex, due to the involvement of heterogeneous catchment and hydrologic factors. The influence of catchment characteristics on catchment hydrology can be understood by analysing the rainfall-runoff relationships (Tetzlaff et al. 2008; Ehret et al. 2014; Farrick & Branfireun 2014; Sivakumar et al. 2015; Merheb et al. 2016; Saft et al. 2016; Gao et al. 2018; Maref et al. 2023; Yilma & Kebede 2023). The conceptual rainfall-runoff relationships for different surface types have been explained by Boyd et al. (1993). As noted by Boyd et al. (1993), there can be three conversion rates of rainfall to runoff representing runoff from directly connected impervious surfaces (segment I), total impervious surfaces (segment II) and a combination of impervious and pervious surfaces, respectively (segment III).

As noted by a range of researchers, the percentages of pervious and impervious areas contributing to runoff can vary with the distance between the drainage system and catchment surface, antecedent dry days (ADD) prior to rainfall events and infiltration capacity of catchment surfaces (Birkinshaw et al. 2021; Yao et al. 2021; Simpson et al. 2022; Zhao et al. 2022). Zhang et al. (2023) argued that the percentage of runoff contributing to catchment areas rises with the increment in rainfall intensity and rainfall depth. This phenomenon is more prominent for larger catchments compared with smaller catchments, due to the availability of different surface types and their wider distribution (Barron et al. 2011; Guan et al. 2016; Yan et al. 2018). More importantly, the percentages of pervious and impervious surfaces contributing to runoff can vary for three different rainfall types, as defined in Chowdhury et al. (2023). The rainfall events that have relatively high intensity in the latter part and comparatively long duration were found to produce a higher volume of runoff from pervious and impervious surfaces with relatively uniform and low event mean concentrations (EMCs) of TSS due to the dilution effect. Those events with relatively high intensity in the initial part and comparatively short duration would produce small runoff volume from impervious surfaces and relatively high EMCs of TSS due to the effects of first-flush. Further, rainfall events with relatively high intensity placed in the mid portion of rainfall events and comparatively short duration could produce variable runoff volume from both pervious and impervious surfaces and EMCs of TSS with a first-flush effect (Chowdhury et al. 2023).

Most previous research studies that focused on evaluating the catchment water quality responses failed to split the effect of catchment characteristics from rainfall characteristics (Kayhanian et al. 2007; Kim et al. 2007; Nicótina et al. 2008; Merz & Blöschl 2009; Liu et al. 2013; Ochoa-Rodriguez et al. 2015; Liu et al. 2016; Gorgoglione et al. 2019; Jaffrés et al. 2021; Müller et al. 2021; Zhao et al. 2022). For example, the attempt made by Liu et al.(2013) to separate the influence of catchment characteristics from rainfall characteristics primarily relied on excluding rainfall variables (intensity, duration and ADD) from the analysis. However, it can be argued that the influence of the rainfall characteristics is still embedded into the stormwater quality variables considered after the removal of rainfall variables. This limits the effective separation of catchment characteristics from rainfall characteristics. In this context, an appropriate approach is necessary for accounting for the pattern-based separation of catchment characteristics from rainfall characteristics, to explore the in-depth understanding of the actual impact of catchment characteristics on urban stormwater quality.

The investigation discussed in this paper focused on the pattern-based separation of catchment characteristics from rainfall characteristics and identifying their influences on urban stormwater quality. This was achieved by investigating the effect of catchment characteristics on behaviour of catchment hydrology using rainfall-runoff relationship; the pattern-based separation of catchment characteristics from rainfall characteristics; and their impacts on stormwater quality responses using univariate and multivariate data analysis techniques. The new insights into how stormwater quality varies depending on the changes in catchment characteristics will contribute to generating reliable information about catchment stormwater quality and accurately formulating stormwater quality modelling tools, thereby contributing to designing effective stormwater pollution mitigation measures.

In order to comprehend the effect of catchment characteristics on behaviour of catchment hydrology, the rainfall-runoff relationship graphs were plotted using the events clustered into three, as identified by Chowdhury et al. (2023) for three study catchments. As investigated by Chowdhury et al. (2023), rainfall events can be separated into three clusters and can demonstrate significant differences in terms of rainfall-runoff processes. For example, 139 events were clustered as 35, 41 and 63 into clusters 1, 2 and 3, respectively, for the Alextown catchment. Accordingly, 128 and 133 events were clustered as 18, 44, 66, 37, 17 and 79 into clusters 1, 2 and 3, respectively, for Gumbeel and Birdlife Park catchments. The summary of the analysis undertaken using rainfall-runoff depths for three clustered events of study catchments is shown in Supplementary Table S1. The same data was used to develop rainfall-runoff plots and is presented in Section 3.1. In this regard, a simplified approach was used with line segment I, representing runoff depth from total impervious surfaces, and line segment II, representing the total runoff from a combination of impervious and pervious surfaces.

For the analysis undertaken in this section, the catchment characteristics were detached from rainfall characteristics. This was done by selecting rainfall events common to all three catchments for the analysis. For example, events common to all three catchments for cluster 1 were identified as E2, E4, E13, E18, E83, E105, E122 and E182. Accordingly, 9 and 36 events were found common to all three catchments, representing clusters 2 and 3, respectively, are mentioned as E52, E92, E108, E116, E141, E142, E157, E166, E247 and E8, E17, E19, E30, E36, E41, E45, E50, E55, E58, E62, E64, E65, E73, E75, E77, E86, E89, E91, E100, E102, E111, E135, E137, E145, E147, E163, E169, E188, E191, E198, E201, E220, E225, E231, E241. In addition, the identified 8, 9 and 36 common events representing clusters 1, 2 and 3 were also presented graphically and marked by red, blue and green colours, respectively, in Supplementary Figures S1–S3. The rainfall characteristics for identified common events are similar across Alextown, Gumbeel and Birdlife Park catchments. On this basis, the variation in stormwater quality characteristics between catchments can be directly linked to the variation in catchment characteristics. A similar separation attempt was undertaken by Liu et al.(2013) where only water quality variables (total nitrogen (TN), total phosphorus (TP), total organic carbon (TOC) and total suspended solids (TSS) EMC) were removed from total datasets but the same rainfall events were used in the analysis. The removal of water quality variables does not necessarily separate the catchment characteristics from rainfall characteristics. In addition, EMCs of TN, TP, TOC and TSS were used as lumped stormwater quality characteristics. This study extends the previous knowledge by the pattern-based separation of catchment characteristics from rainfall characteristics based on common events identified from three clusters for the three catchments. In addition, the identified common events were assessed using univariate statistical tools and multivariate principal component analysis (PCA) to explore the impact of catchment characteristics on urban stormwater quality.

PCA is a widely used multivariate data analysis technique to identify the relationship among different variables and similar object cluster groups. It is a data retrenchment process commonly used to decrease a large number of variables to a lesser number of self-reliant new variables known as principal components (PCs). The PCs can be represented as a function of an eigenvector matrix, where rows and columns are formed corresponding to loadings of each PC and initial data variables, respectively. These loading coefficients are applied to convert the source data into PCs (Abdi & Williams 2010). The weights of the eigenvector corresponding to each PC can be plotted to determine the significant number of PCs commonly known as the Scree plot method. The Scree plot can be defined as the graphical expressions of eigenvalues in decreasing trends corresponding to respective PCs. The desired number of PCs for effective analysis is determined by identifying the point where the Scree plot significantly changes the slope (Jackson 2005).

Before performing the PCA analysis, the preparation of a data matrix is necessary. This was done by combining the identified common events together on a cluster basis from three catchments. For example, a total of 24 events can be found by combining 8 common events from cluster 1 of three catchments. Accordingly, a total of 27 and 108 common events were found corresponding to clusters 2 and 3, respectively, from three catchments. A pollutant load vs. runoff volume (MV) curve, for identified common events, was plotted and the selected 10 stormwater quality parameters (P01, P12, P23, P34, P45, P56, P67, P78, P89 and P910) were extracted from each event as presented in Supplementary Figure S4. An MV curve is a graphical illustration where (M) denotes the pollutant load as a percentage with respect to the total and (V) denotes the runoff volume as a percentage with respect to the total (Chowdhury et al. 2023). Accordingly, three matrices of (24 × 10), (27 × 10) and (108 × 10) were formed for clusters 1, 2 and 3, respectively, by combining the dataset of stormwater quality variables (P01, P12, P23, P34, P45, P56, P67, P78, P89 and P910) as shown in Supplementary Tables S2–S4. The data matrix was prepared by defining the columns and rows as variables and data objects, respectively. A standard pre-treatment method is often used to reduce the bias outcomes of PCA since the units and magnitudes of the variables are different. Column standardisation, mean centring and auto-scaling are the commonly used pre-treatment methods before applying PCA to the datasets (Keenan & Kotula 2004; Tyler et al. 2007). Column standardisation is done by taking the ratio of each cell value and the standard deviation of the respective column. This enables the variable to be weighted equally where the standard deviation is obtained as one (Tyler et al. 2007). Mean centring is the method of deducting each variable mean from each cell value in each column. Auto-scaling is the term commonly used to represent the combined pre-treatment of column standardisation and mean centring (Settle et al. 2007).

The output of PCA is the scores of data objects and loadings of variables. The visualisation of the relationships among the variables is represented by the loading plot while the score plot represents the structural similarity among the object groups. The combined graphical representation of loading and scores in a single plot is known as a PCA biplot (Abdel-Fattah et al. 2021; He 2024). PCA biplot is more informative in understanding the correlations between variables and data objects. For example, a narrow angle between the variables represents a strong relationship while a wide angle represents a weak relationship between the variables. Similarly, objects in the same group represent similar structural properties while objects in scattered form represent structural dissimilarity among the objects (Dutta & Das 2019; Cao et al. 2020). The correlation among the variables can be viewed quantitatively by a correlation matrix. A larger correlation value indicates a strong relationship while a smaller or negative value indicates a weak or negative correlation, respectively. In this study, PCA analysis was performed using the StatistiXL Version 2.0 software (StatistiXL 2016).

It can be noted that rainfall characteristics among the group events are similar across the three catchments but catchment and resultant water quality characteristics are different. The outcomes of these analytical analyses are presented in Section 3.2.

As seen in Figures 2–4, the rainfall depth versus runoff depth variations were found to be different for the three catchments. This is particularly due to the dissimilarities in characteristics of study catchments. The impervious surface initial loss for Gumbeel catchments was found to be 2.3 mm, which is slightly lower than Alextown (2.4 mm) and Birdlife Park (2.5 mm) catchments as discussed in Chowdhury & Egodawatta (2022). This suggests the presence of relatively rough and flat impervious surfaces in Alextown and Birdlife Park compared with the Gumbeel catchment.

It is also noticeable that Alextown and Birdlife Park show clear pervious surface runoff contribution (see Figures 2 and 4), whereas the Gumbeel catchment does not represent runoff from pervious surfaces (see Figure 3). This is suggested since there is no deviation in the rainfall-runoff relationship to form segment II of the straight line for the Gumbeel catchment. This is because the Gumbeel catchment is situated on a ridge where pervious surfaces (e.g. lawns, gardens) are situated at a lower elevation than the road surface (Supplementary Figure S7). Such geographic feature allows pervious surface runoff to flow in an opposite direction to the drainage path. In the Gumbeel catchment, the main contributor to the runoff is the access road and residential roof runoff. In addition, some grass belts located close to the road kerb may also contribute to the runoff.

As seen in Figures 2–4, events belonging to all three clusters generally produce runoff from impervious surfaces. Only cluster 1 and 3 events are capable of generating runoff from pervious surfaces. Cluster 1 events are long durational and can produce rainfall depths that are sufficient to meet the pervious surface initial losses. Cluster 2 events have not created pervious surface runoff. This is because cluster 2 events are short durational, with high ADD, which does not have the potential to satisfy higher moisture deficits in pervious surfaces. Cluster 3 events are high-intensity events with low ADD, which have mostly created high runoff depths compared with cluster 1 and 2 events.

As seen in Figures 2 and 4, the rainfall depth required to initiate pervious surface runoff is different for Alextown and Birdlife Park catchments. For Alextown, pervious runoff was initiated at around 47 mm rainfall depth, whereas for the Birdlife Park catchment, it was at around 35 mm. This is due to differences in catchment slope, soil type and vegetation in Alextown and Birdlife catchments. The commencement of pervious area runoff was found different from the results reported by Liu et al. (2012). This is due to the influence of the rainfall temporal pattern on pervious runoff processes. The location of pervious surfaces for Alextown and Birdlife Park catchments are shown in Supplementary Figures S5 and S6.

The above discussion suggests that not only an antecedent dry period is responsible for differences in the commencement of pervious area runoff, but also the location of pervious areas has a prominent influence on the hydrologic behaviour of these catchments. Based on the above analysis, it can be concluded that catchment characteristics have a significant influence on the catchment hydrology of the study areas.

Outcomes of the analysis undertaken in Section 3.1 suggest that catchment characteristics significantly influence catchment hydrology. Due to the direct relationships between catchment hydrology, and wash-off and transport characteristics of pollutants, the influence can propagate to stormwater quality. The impacts of catchment characteristics on stormwater quality can be further complicated due to the wash-off of heterogeneous pollutants from pervious areas. This demands a proper understanding of the impacts of catchment characteristics on urban stormwater quality.

As seen in Figure 5(a), Alextown and Birdlife Park catchments produce closely similar EMCs, where EMCs for the Gumbeel catchment were found slightly higher for cluster 1 events. This could be due to the non-availability of pervious surface runoff contribution for the Gumbeel catchment, as shown in Figure 3. In contrast, Alextown and Birdlife Park may generate large runoff volumes due to contributions from both pervious and impervious surfaces, and these create dilution effects when mixed up together. This result reduced the overall TSS EMCs for Alextown and Birdlife Park catchments compared with the Gumbeel catchment.

As seen in Figure 5(b), mean and median TSS EMC was found higher for Gumbeel compared with other catchments for cluster 2 events, while EMC for the Birdlife Park catchment was found slightly higher than the Alextown catchment. This is due to the dissimilarity in catchment surface slope, urban formation and the impervious areas of the three catchments, as presented in Figure 1 (Section 2.1). As discussed in Section 3.1, the Gumbeel catchment is situated on a ridge, hence slope of the catchment is comparatively steep. Additionally, cluster 2 event does not produce pervious runoff. This results in higher TSS EMC for Gumbeel than other catchments. In addition, due to urban form characteristics, the traffic density is comparatively higher in Birdlife Park compared with the Alextown catchment. These resulted in the generation of high pollutant load during large ADD of cluster 2 events and resulted in the higher TSS EMC of Birdlife Park than Alextown catchments.

As seen in Figure 5(c), mean and median TSS EMC were found quite similar in terms of cluster 3 events for Alextown and Birdlife Park catchments where the Gumbeel catchment produces relatively higher EMCs. This is due to the difference in impervious area percentage, road surface characteristics and pervious surface runoff among the three catchments. The impervious area of Gumbeel (40.7%) is comparatively less than Alextown (57.2%) and Birdlife Park (45.8%) catchments (see Section 2.1). The catchment containing smaller impervious areas can generate high TSS EMC due to less runoff volume. As Egodawatta et al. (2007) noted, rainfall should have adequate kinetic energy to remove pollutants from road surfaces. As discussed in Chowdhury et al. (2023), cluster 3 events associated with high intensity can produce more pollutant load from the smooth roads and roof surfaces of Gumbeel, compared with Alextown and Birdlife Park catchments. In addition, cluster 3 events produce previous runoff for Alextown and Birdlife Park, as shown in Figures 2 and 4. This results in the generation of high TSS EMC for Gumbeel, compared with other catchments.

These preliminary findings confirm two main aspects. First, catchment characteristics have a significant influence on urban stormwater quality. Second, stormwater quality characteristics can vary with different catchment characteristics, even though rainfall characteristics are similar for the three study catchments. These findings directed this research to undertake analysis using multivariate analytical tools so that the magnitude of the impact of catchment characteristics on stormwater quality can be understood.

As seen in Figures 6–8, similar events of the three catchments are scattered for three clusters and oriented along different water quality vectors. This suggests that stormwater quality characteristics for the three catchments are different and significantly influenced by the catchment characteristics, although residential land use and rainfall characteristics are the same. The details of the PCA analysis for each cluster are discussed in the following sections.

As seen in Figure 6, Alextown and Gumbeel rainfall events are clustered closely together and projected towards positive and negative PC2 axes while Birdlife Park rainfall events are relatively scattered and projected towards the negative PC2 axis. This means that the generated pollutant load is highly variable for the Birdlife Park catchment at different percentages of runoff volume than for Alextown and Gumbeel catchments. It is also noticeable that some of the events that belonged to Alextown and Gumbeel catchments are positively clustered along pollutant load vectors P01 and P910, while the Birdlife Park events are comparatively negatively scattered and independent of the pollutant load vectors. This suggests that Alextown and Gumbeel catchments produce high pollutant loads corresponding to the first and last 10% of runoff volume, compared with the Birdlife Park catchment. This is due to different catchment areas, impervious percentages and location of pervious areas, which resulted in different wash-off processes for Alextown and Gumbeel as compared with the Birdlife Park catchment.

Alextown (1.79 ha) and Gumbeel (2.106 ha) are smaller catchments compared with the Birdlife Park (8.88 ha) catchment. As discussed in Section 3.1, the Gumbeel catchment contributes runoff only from impervious surfaces. Additionally, the drainage system of Gumbeel is comparatively simpler than other catchments. This results in quick drainage of generated wash-off from impervious surfaces and hence possesses less variability in pollutant concentration. Alextown generates large runoff volumes due to higher impervious surfaces for long durational cluster 1 events. In Alextown, the pervious area is located at far distances from the drainage system (see Supplementary Figure S5), hence pervious area runoff does not greatly influence the pollutant wash-off processes, due to the continuing loss that occurred in travelling a longer distance to reach the drainage system. This results in less variability in pollutant wash-off processes. In contrast, pervious areas located close to the drainage system have comparatively steep slopes for the Birdlife Park catchment (see Supplementary Figure S6). These results for both pervious and impervious surfaces contribute to runoff at low rainfall depth. As noted by Miguntanna et al. (2010), pervious area runoff generates a heterogeneous mixture of pollutants and concentration comparatively lower than the impervious area. In addition, long durational rainfall reduces the pollutant concentration by providing a large runoff volume due to a large catchment area. Due to these factors, dilution effects occur when the runoff travels a longer drainage path. Hence, overall pollutant concentration is comparatively lower and highly variable for the Birdlife Park catchment than for other catchments (Herngren et al. 2010).

As seen in Figure 7, Alextown, Gumbeel and Birdlife Park have different pollutant wash-off characteristics at different percentages of runoff volume. Most of the events belonging to the Alextown catchment are directed towards the positive PC1 axis, whereas events belonging to Gumbeel and Birdlife Park catchments are directed towards the negative PC1 axis. This suggests that Alextown produces a higher pollutant load corresponding to the first 10% of runoff volume, compared with Gumbeel and Birdlife Park catchments for cluster 2 events. Gumbeel and Birdlife Park events are comparatively scattered along the P12–P910 and projected towards positive and negative PC2 axes. This suggests that pollutant load generation and wash-off processes for Gumbeel and Birdlife Park are more variable than for the Alextown catchment. It is also noticeable that cluster 2 rainfall events start with high initial intensity, which produces a highly concentrated pollutant load at the first 10% of runoff volume, as discussed in Chowdhury et al. (2023). This type of rainfall event is more likely to create a first-flush effect. Most of the Alextown events were projected towards the P01 pollutant load vectors, although the same rainfall characteristics were used for Gumbeel and Birdlife Park catchments. This suggests that Alextown rainfall events are more likely to produce a first-flush effect due to the influence of catchment characteristics. These findings can be supported by different impervious fractions, areas and pervious area locations within the three catchments.

The Alextown area (1.79 ha) is comparatively less than Gumbeel (2.106 ha) and Birdlife Park (8.88 ha) catchments. It can be noted that the impervious area fraction for Alextown (57.2%) is comparatively higher than that of Gumbeel (40.7%) and Birdlife Park (45.8%) catchments. This suggests that a smaller catchment containing a large impervious surface fraction can produce more pollutant load than a large catchment area of less impervious fraction. Additionally, Gumbeel and Birdlife Park produce highly variable pollutant loads, compared with the Alextown catchment. This is because pollutant build-up on a smaller percentage of impervious surfaces is limited. Hence, variability in wash-off processes increases with the increased runoff volume. In addition, Birdlife Park has an extended pervious surface due to its large area and is directly connected with a rectangular mesh drainage system (see Supplementary Figure S6). This enables shorter durational rainfall events to contribute to runoff more quickly. This result variability of pollutant concentration increases due to the mixing of highly concentrated impervious surface runoff with comparatively low concentrated pervious surface runoff, and decreases the overall concentration. This suggests that only a percentage of impervious surface is not capable of describing the total features of stormwater quality characteristics. Hence, in catchment modelling, both pervious and impervious areas should be taken into account, to reduce the variability of stormwater quality.

As seen in Figure 8, rainfall events for three catchments are spread at different water quality vectors. This means that pollutant wash-off behaviour is significantly influenced by catchment characteristics, although the same rainfall characteristics were applied to three catchments. Most of the events belonging to the Birdlife Park catchment are directed towards the positive PC2 axis while most of the events belonging to Gumbeel are projected towards the negative PC2 axis. Events belonging to Alextown are comparatively dispersed and spread in both positive and negative PC2 axes. This suggests that cluster 3 rainfall events of Birdlife Park contribute more pollutant load compared with Gumbeel, while the pollutant load is highly variable for Alextown. This suggests that the Alextown catchment pollutants experienced different wash-off processes at different percentages of runoff volume whereas Gumbeel and Birdlife Park catchments experienced comparatively consistent wash-off processes. It is also noted that Birdlife Park events are oriented along pollutant load vectors P01, P12, P23, P34, P45, P56 and P67 hence producing a high fraction wash-off corresponding to 10–70% runoff volume due to high-intensity cluster 3 events. In contrast, the Gumbeel catchment produces a comparatively low fraction of wash-off and is oriented along P78, P89 and P910 pollutant load vectors. This suggests that even though three catchments experienced the same rainfall characteristics, water quality characteristics significantly vary due to the influence of catchment characteristics.

This behaviour can be described by differences in urban form for the three catchments. For example, the Birdlife Park catchment consists of high socio-economic detached occupancy residences where Gumbeel and Alextown are duplex housing and townhouse developments, respectively, as presented in Section 2.1. Due to the geographic location, the road layout, traffic volume and density are relatively higher in Birdlife Park than Alextown and Gumbel catchments. Moreover, water quality characteristics significantly vary with traffic-related activities (Herngren et al. 2010). This causes a high amount of pollutant generation in Birdlife Park compared with Alextown and Gumbeel catchments. In addition, high-intensity rainfall (cluster 3) is capable of producing more wash-off load due to high kinetic energy from impervious surfaces (Egodawatta et al. 2009). This could further lead to the generation of high pollutant load from Birdlife Park compared with Alextown and Gumbeel catchments. This finding suggests that urban form is also an essential catchment characteristic that has a significant influence on urban stormwater quality.

From the above discussion, it is clear that stormwater quality characteristics are significantly different for the three rainfall clusters at the study catchments, even though rainfall characteristics for the three clusters are similar. This indicates that catchment characteristics significantly influence the urban stormwater quality. Based on the analysis results of Section 3.2.1, TSS EMC was found higher for Gumbeel catchments for all three rainfall clusters even though catchment characteristics are different for the three catchments. This confirms that treatment system design depends on the lumped characteristics of pollutant EMC and does not account for the variations of catchment characteristics on stormwater quality accurately. This results in the inefficiency in treatment system design.

The rainfall-runoff relationship for Alextown, Gumbeel and Birdlife Park was investigated based on three rainfall clusters. Additionally, the catchment characteristics were separated from rainfall characteristics based on three clusters for three study catchments. Finally, the impact of catchment characteristics on stormwater quality was assessed, depending on three separate sets of common rainfall events for three clusters. Based on the investigation, the following conclusions were made.

In current practices, evaluation of catchment water quality responses failed due to the use of ineffective approaches for the separation of catchment characteristics from rainfall characteristics. This creates complexity in understanding the actual influences of catchment characteristics on stormwater quality. This study adopted a pattern-based separation approach to identifying common events from three rainfall clusters where rainfall characteristics for identified common events are similar, but catchment characteristics are different. The adopted approach overcomes the separation issues and enables comprehension of the fundamental relationship between catchment and water quality characteristics required for improving the effectiveness of stormwater treatment system design.

Conventional catchment characteristics, including fraction imperviousness and land-use type, are not adequate to explain the behaviours of catchment hydrology and their influence on urban stormwater quality. This study identified that the locations of pervious surface and urban form are also important catchment characteristics that have a significant influence on rainfall-runoff processes and can alter the pollutant transport processes through changing runoff characteristics.

The contribution of pervious area to runoff is significant for clusters 1 and 3 rainfall events associated with low ADD. In contrast, impervious areas only can contribute to runoff for cluster 2 events associated with short-duration rainfall and high ADD. Pervious area situated near the drainage structure contributes to runoff at low rainfall depth and dilutes the stormwater quality, especially for cluster 1 events associated with long duration and low intensity rainfall.

Catchment topography can control the pervious surface runoff. The catchment area situated on a ridge topographic zone does not contribute pervious surface runoff for three rainfall cluster events.

Pollutant wash-off processes at different runoff fractions vary with catchment characteristics, even if rainfall characteristics are similar for different catchments. A catchment containing a large impervious area can contribute more pollutant load than a large catchment area with less impervious fractions. High socio-economic developed urban form can contribute more wash-off load due to traffic-related behaviour than townhouse and duplex housing developments.

The authors would like to acknowledge the contributions from the Bureau of Meteorology (BOM), Australia for providing historical rainfall data for this study.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.