Commonality of rainfall variables influencing suspended solids concentrations in storm runoff from three different urban impervious surfaces

https://doi.org/10.1016/j.jhydrol.2010.04.008Get rights and content

Abstract

Finding a common set of rainfall variables to explain the concentration of suspended solids in runoff from typical urban impervious surfaces has many applications in stormwater planning. This paper describes a statistical process to identify key explanatory variables to non-coarse particle (suspended solids < 500 μm size) event mean concentrations measured from road, carpark and roof surfaces located in Toowoomba, Australia. The dominant variables for all surfaces were rainfall depth and peak 6-min rainfall intensity. Storm duration, defined as the time period when rainfall intensity exceeds 0.25 mm/h and antecedent storm rainfall were also important predictors, but was less dominant. The regression model fitted to non-coarse particle concentration across all surfaces was proportional to rainfall depth raised to a negative power and peak 6-min rainfall intensity raised to a positive power; the proportionality constant varies by surface type. The form of this common model has a physical basis and is analogous to the modified universal soil loss equation widely used for soil loss estimation for non-urban areas.

Introduction

Unless properly managed, suspended solids in stormwater are a major cause of adverse effects within aquatic ecosystems downstream of urban areas (Borchardt and Sperling, 1997). Low-impact development (LID) has emerged as a management response to address these and other stormwater impacts (Dietz, 2007). Alternative terms for this approach include water sensitive urban design and sustainable urban drainage systems.

LID is transforming the management focus from the reliance on a few large-scale management devices to many controls strategically distributed throughout urban areas. This paradigm shift in stormwater management towards at-source control places greater emphasis on small-scale hydrological processes and, subsequently, more knowledge is needed about stormwater generated from specific types of urban surfaces. This includes techniques to predict suspended solids loads in runoff from impervious urban surfaces such as roads, roofs and carparks.

The estimation of suspended solids in stormwater for the purpose of LID assessment tend to be based on simple methodologies, including buildup–washoff relationships, characteristic concentrations (sometimes with a stochastic component), empirical power rating curves for concentration as a function of discharge, and unit area loadings (Elliott and Trowsdale, 2007).

Consistent with the methodologies used in LID modelling, this paper will focus on the use of simple regression-based relationships to estimate the event mean concentration (EMC) of suspended solids in runoff from impervious urban surfaces. Stormwater data collected by the monitoring of three surfaces (road, roof and carpark) located at Toowoomba, Australia (Brodie, 2007) was used in the EMC analysis.

A prerequisite for any regression analysis is the identification of key explanatory variables. An overview by Vaze and Chiew (2003b) found that a wide diversity of variables have been used to estimate the washoff of particles from urban areas including rainfall intensity, rainfall volume, runoff rate, runoff volume, raindrop impact energy and overland flow shear stress. Most of these variables are to some extent correlated with each other, and thus it is difficult to ascertain the dominant drivers of suspended solids washoff. Key variables are discussed in more detail in Section 2 of this paper, based on a review of the literature.

The main objective of this paper is to establish whether a common set of hydrological variables (such as storm rainfall depth, intensity and duration) apply across the three different surfaces under analysis. In keeping with the ‘dominant processes concept’ used elsewhere in hydrology (Sivakumar, 2004), we aimed to isolate the few dominant variables that capture the essential suspended particle response to storm events. Identification of such dominant predictors avoids overparameterization and leads to more parsimonious regression models. Two different statistical approaches (conventional regression and Bayesian model averaging) were applied to identify dominant predictors and to ascertain whether these predictors are common to all three surface types.

Section snippets

Variables influencing suspended solids washoff from urban surfaces

Regression relationships to estimate total suspended solids (TSS) exports from urban catchments as a whole are available (e.g. Brezonik and Stadelmann, 2002, Driver and Tasker, 1990, LeBouthillier et al., 2000, McLeod et al., 2006, Vaze and Chiew, 2003a). Pollutant generation from urban surfaces in these studies were aggregated as a single land use category (such as ‘residential’ or ‘commercial’), so do not provide information on the individual contributions made by specific types of surfaces.

Measured suspended solids data

Runoff samples were collected from three different urban impervious surfaces located at Toowoomba, Australia (Table 2). Toowoomba is located within a temperate climate region of South East Queensland subject to mild to warm summers and mild, dry winters. A flow splitter device described by Brodie (2005) was used to obtain flow-weighted composite samples in response to 35 storms during the period December 2004 to January 2006. Rainfall was recorded by a 0.25 mm tipping bucket pluviometer

Step 1 – selection of the NCP dependant variable

As is the case for all pollutants, NCP can be expressed as a concentration (EMC, mg/L) or as a mass load (L, kg/storm or mg/m2/storm). In their study of the US National Urban Runoff Program data compiled for total phosphorus, May and Sivakumar (2004) found that regression models using load as the dependent variable had errors 50% higher than the concentration-based models. NCP EMC was selected as the dependent variable in the statistical analysis to limit these potential errors. EMC is based on

A common NCP regression model

Both statistical approaches adopted suggest the common explanatory variables for determining NCP EMC are ln(RD) and ln(PI), with the constant term in the model varying by the surface type. In addition, both approaches suggest ln(SD) has a more complicated relationship with NCP and perhaps is only useful for the roof surface. BMA also suggests that AR has a role to play, albeit a less important role than the other predictors.

The goal for this paper is to identify the dominating hydrological

Conclusions

Two statistical approaches (conventional regression and Bayesian model averaging) were applied to identify key hydrological factors of suspended solids runoff from representative urban impervious surfaces. Non-coarse particle (NCP, <500 μm) event mean concentration (EMC) data in runoff collected from a road, a carpark and a roof located in Toowoomba, Australia were used in the analyses.

Both statistical approaches isolated rainfall depth and peak 6-min rainfall intensity as the dominant

References (56)

  • H. Akaike

    A new look at the statistical model identification

    IEEE Transaction on Automatic Control

    (1974)
  • American Society for Testing and Materials (ASTM), 2002. Standard test method for determining sediment concentration in...
  • R.T. Bannerman et al.

    Sources of pollutants in Wisconsin stormwater

    Water Science and Technology

    (1993)
  • Brodie, I.M., 2005. Stormwater particles and their monitoring using passive devices. In: 10th International Conference...
  • Brodie, I.M., 2007. Investigation of stormwater particles generated from common urban surfaces. PhD thesis, University...
  • I.M. Brodie et al.

    Suspended particle characteristics in storm runoff from urban impervious surfaces in Toowoomba, Australia

    Urban Water

    (2009)
  • I.M. Brodie et al.

    Using soil loss models to estimate suspended solids concentrations in stormwater from pre-urban areas

    Australian Journal of Water Resources

    (2008)
  • T.W. Chui et al.

    Pollutant loading model for highway runoff

    Journal of the Environmental Engineering Division, American Society of Civil Engineers

    (1982)
  • A. Deletic et al.

    Modelling of storm wash-off of suspended solids from impervious surfaces

    Journal of Hydraulic Research

    (1997)
  • M.B. Desta et al.

    Highway runoff quality in Ireland

    Journal of Environmental Monitoring

    (2007)
  • M.E. Dietz

    Low impact development practices: a review of current research and recommendations for future directions

    Water, Air, and Soil pollution

    (2007)
  • Driver, N.E., Tasker, G.D., 1990. Techniques for the estimation of storm-runoff loads, volumes and selected constituent...
  • Duke, L.D., Mbah, A., Yanev, G.P., 2007. Statistical relationships of storm runoff constituents with storm event...
  • Goyen, A.G., O’Loughlin, G.G., 1999. Examining the basic building blocks of urban runoff. In: 8th International...
  • J.A. Hoeting et al.

    Bayesian model averaging: a tutorial

    Statistical Science

    (1999)
  • INSA/SOGREAH, 1999. CANOE User Manual. ALISON, INSA de Lyon, Villeurbanne,...
  • L.B. Irish et al.

    Use of regression models for analyzing highway storm-water loads

    Journal of Environmental Engineering

    (1998)
  • Kerri, K.D., Racin, J.A., Howell, R.B., 1985. Forecasting pollutant loads from highway runoff. Transportation Research...
  • Cited by (0)

    1

    Tel.: +61 7 5456 5085; fax: +61 7 5430 2896.

    View full text