Estimation of total impervious area (TIA) is a pre-requisite for ecohydrological research to allow for a direct prediction on stream ecosystem health within catchments. This paper presents an alternative to using multi-spectral imagery for estimating TIA at a catchment scale, by using high-resolution colour aerial photography. The method was applied to a number of catchments in South East Queensland, Australia, some of which were gauged and some of which were part of an Ecosystem Health Monitoring Program (EHMP). The results from this method were compared to TIA estimates, for some of the same catchments, that were derived through three other techniques, i.e. manual digitization of geo-referenced aerial photos, Brisbane City Council data derived from image analysis using Landsat TM imagery and rainfall runoff depth relationship. The high-resolution colour aerial photography method compared favourably to the other techniques with standard deviations of TIA (%) ranging between 0.8% and 8%. The major constraints were shading effects, particularly on roads and grassed areas, and from the similarity in colours between some surface types, some of which can be reduced by appropriate selection of signature colours and multiple iterations of a supervised classification. It was concluded that while infra-red spectral wave bands could help considerably, the high-resolution colour photography could be applied with confidence to derive catchment-scale TIA estimates.

INTRODUCTION

In hydrological science, a key surrogate measure of urbanisation is the fraction of impervious surfaces, which includes pavement, roof, paved parking, driveways and other sealed surfaces. Stormwater drainage practitioners rely heavily on estimates of impervious area in a watershed (Janke et al. 2011). Impervious areas bear significant environmental concern because they degrade urban waters through generation of urban runoff, primarily from roads, rooftops, and parking lots. The increase in catchment imperviousness increases surface runoff, sediment concentration, associated pollutant discharge and reduces evapo-transpiration and aquifer recharge potential (Han & Burian 2009). Catchment imperviousness changes the duration, frequency, magnitude and quality of urban runoffs (Paul & Meyer 2001; Shuster et al. 2005). An increased frequency, magnitude and peak flows are generally expected from an impervious catchment. Conversely the low flows, which are important in sustaining the aquatic ecology, are also affected by increased catchment imperviousness due to reduced groundwater recharge and the channeling of most rainfall laterally as runoff (Bunn & Arthington 2002; Delleur 2003). Excess nutrients, pollutants and sediments deposited in roads, rooftops and parking lots are rapidly washed into receiving waters by storm drainage systems designed to prevent urban flooding (Janke et al. 2011). Numerous previous studies have delineated catchment imperviousness as a predictor of stream ecosystem health (Beach 2003; Walsh et al. 2005). Impervious surfaces are either hydraulically connected to stormwater pipe inlets or separated by pervious surfaces. Hydraulic connection indicates that runoff follows an entirely sealed pathway prior to entry to stormwater pipes or drains. The hydraulically connected portion of total impervious area (TIA) is known as directly connected impervious area (DCIA) or effective impervious area. Some recent studies have identified a more defined relationship between DCIA and ecosystem health indicators (Walsh et al. 2005). Janke et al. (2011) reported that the DCIA is the most important parameter in determining actual urban runoff. Therefore, one of the primary goals of stormwater management is to reduce DCIA or to disconnect impervious areas from drainage system through applications of Water Sensitive Urban Design (WSUD) or sustainable urban drainage system techniques. However, the connectedness of imperviousness in a catchment is not well understood and there is a lack of standard methods to estimate DCIA.

Several recent studies have attempted to estimate DCIA using a high-resolution digital elevation model, multi-spectral satellite image and digital stormwater drainage pipe network database (Han & Burian 2009; Kunapo et al. 2009). Boyd et al. (1993) and Lee & Heaney (2003) estimated the DCIA by analysing rainfall and runoff data. Han & Burian (2009) and Alley & Veenhuis (1983) estimated the DCIA by using satellite-derived spatial data such as land cover and elevation. Janke et al. (2011) developed a GIS-based tool to estimate DCIA through automated analysis of high-resolution land cover and elevation GIS data derived from satellite and air-based imagery. Their technique was based on the method described in Han & Burian (2009). All of these methods are based on image analysis techniques. However, due to the lack of availability of required data and estimation difficulties, TIA is widely accepted as an indicator of urbanisation for hydrological modelling and ecohydrological studies.

In this study, TIA fractions were estimated using image analyses of high-resolution colour orthophotos in the South East Queensland (SEQ), Australia. It was necessary to use the available colour orthophotos, as opposed to an alternative source with multi-spectral resolution, as they have already been utilised for assessing urban expansion for that date and the models requiring TIA estimates were to be run for the same point in time as the imagery. Some manual digitizing estimates of TIA had already been acquired from some of the images and, as more colour orthophotos will be acquired by the Brisbane City Council (BCC) in the future, it would therefore be advantageous to compare the results of this method acquisition with a more automated methodology. This would establish whether such a method could be applied accurately for future TIA estimates. Finally, the available research budget did not provide the resources to acquire additional imagery with multi-spectral capability, which limited the ability to distinguish between pervious green vegetation and impervious green roofs and concrete (e.g. tennis courts).

The results were compared with two other methods, the rainfall–runoff relationship method and manual digitisation of orthophotos method. Results from all of these methods were also compared with existing imperviousness data from the BCC. However, the BCC data cover only eight catchments out of 24 Ecosystem Health Monitoring Program (EHMP) catchments. If the results from this method are favourable, compared with other results, then it will prove to be a viable solution for estimating TIA in any urban environment.

STUDY AREA

The study area is located within 50 km to the south and south-east of Brisbane, Australia and includes eight gauged catchments out of 24 EHMP catchments in the SEQ region (Figure 1). The combined area of all catchments is approximately 570 square kilometres. The SEQ region is sub-tropical with an average annual rainfall and potential evapotranspiration of 1,150 mm and 1,450 mm, respectively (BoM 2012). Rainfall exhibits a seasonal pattern in the region and the highest rainfall occurs during the summer (December to February). These catchments include semi-urban and forested catchments ranging from 1.44 km2 to 27.85 km2 (Table 1) (Aryal et al. 2016). All eight gauged catchments contain ephemeral streams with different degrees of intermittency. The average elevations of the catchments range from 0 m to less than 300 m above the Australian Height Datum. The catchments are characterised into three general types of land use (a) Reference catchment – undisturbed and forested (b) Urban catchment – with traditional urban development, and (c) Mixed catchment – urban development with some water sensitive urban design (WSUD) features.
Table 1

Eight gauged study catchments and their key features (Aryal et al. 2016)

Creek and Catchment NameLocation in SEQArea (km2)Key features
Scrubby Creek Karawatha Forest 1.44 
  • Reference catchment

  • Old metal mine present at upstream but unlikely to overflow

  • DCIA absent

 
Sheepstation Creek Parkinson 1.90 
  • Urban catchment; turbid water flow is observed;

  • Residential land uses upstream;

  • Sedimentation pond upstream

  • Direct connection to stormwater pipes present

  • Constructed pond/wetland system present upstream

 
Upper Yaun Creek Coomera 3.62 
  • Reference catchment

  • DCIA largely absent

 
Pimpama River Kingsholme 4.15 
  • Reference catchment

  • Waterholes (depression) present; creek water not usually turbid

  • DCIAs absent

 
Stable Swamp Creek Sunnybank 4.42 
  • Urban catchment

  • Mixed residential and industrial land uses upstream

  • Directly connected stormwater pipes; Non turbid flow

 
Blunder Creek Durack 5.63 
  • Urban catchment; residential land uses upstream

  • Non-turbid water flow is observed

  • Directly connected stormwater pipes present

  • Open concrete lined channel at gauging site

 
Blunder Creek Carolina Parade 21.76 
  • Mixed catchment

  • WSUD features such as swales, sediment ponds and wetland present

  • Healthy wide riparian zone

  • Residential land uses upstream,

  • Limited DCIA

  • Earthworks started early 2008 at south east of sampling site

  • Located downstream of Forest Lake and large forested catchment area of military reserve

 
Tingalpa Creek Sheldon 27.85 
  • Reference catchment

  • Waterholes (depression) present; creek water not usually turbid

  • Some rural residential properties present at upstream

  • DCIA absent except some road run-off

 
Creek and Catchment NameLocation in SEQArea (km2)Key features
Scrubby Creek Karawatha Forest 1.44 
  • Reference catchment

  • Old metal mine present at upstream but unlikely to overflow

  • DCIA absent

 
Sheepstation Creek Parkinson 1.90 
  • Urban catchment; turbid water flow is observed;

  • Residential land uses upstream;

  • Sedimentation pond upstream

  • Direct connection to stormwater pipes present

  • Constructed pond/wetland system present upstream

 
Upper Yaun Creek Coomera 3.62 
  • Reference catchment

  • DCIA largely absent

 
Pimpama River Kingsholme 4.15 
  • Reference catchment

  • Waterholes (depression) present; creek water not usually turbid

  • DCIAs absent

 
Stable Swamp Creek Sunnybank 4.42 
  • Urban catchment

  • Mixed residential and industrial land uses upstream

  • Directly connected stormwater pipes; Non turbid flow

 
Blunder Creek Durack 5.63 
  • Urban catchment; residential land uses upstream

  • Non-turbid water flow is observed

  • Directly connected stormwater pipes present

  • Open concrete lined channel at gauging site

 
Blunder Creek Carolina Parade 21.76 
  • Mixed catchment

  • WSUD features such as swales, sediment ponds and wetland present

  • Healthy wide riparian zone

  • Residential land uses upstream,

  • Limited DCIA

  • Earthworks started early 2008 at south east of sampling site

  • Located downstream of Forest Lake and large forested catchment area of military reserve

 
Tingalpa Creek Sheldon 27.85 
  • Reference catchment

  • Waterholes (depression) present; creek water not usually turbid

  • Some rural residential properties present at upstream

  • DCIA absent except some road run-off

 
Figure 1

Study area.

Figure 1

Study area.

The EHMP delivers one of the most comprehensive marine, estuarine and freshwater monitoring programs in Australia (www.ehmp.org). The goal of this program is to investigate whether the health of SEQ's waterways is improving or declining, and also discovers the issues impacting on waterway health. The waterways of the SEQ region provide important ecosystem values, they also play an important role in providing commercial resources such as drinking water supply, fishing, aquaculture, agriculture and industrial use. Local authorities commenced an integrated regional ecosystem health assessment program to protect these important aquatic resources. The selected eight gauged catchments were also used as test catchments for SEQ stormwater harvesting research (Aryal et al. 2016). Estimation of imperviousness of these catchments was considered an important input for ecosystem health and stormwater harvesting research in the region. More details of the EHMP is available at their website (www.ehmp.org).

DATA AND METHODOLOGY

Data preparation

A set of 75 geo-referenced colour aerial orthophotos captured in 2009, at 0.5 m resolution, were supplied along with GIS layers containing the EHMP and gauged catchments polygons. The images were converted from tiff image format to the native format of the image analysis software being utilised, i.e. ERDAS Imagine v9.1, format (IMG). Mosaics were then created for each catchment area.

Cadastral filtering

The Digital Cadastral Database (DCDB) GIS layer was obtained, and was clipped to the extent of each EHMP and gauged catchment using ArcGIS v9.3. The resulting DCDB catchment layers were then used to identify the urban areas within each catchment. The large parcels of non-urban areas were selected and removed, leaving DCDB polygons only covering the urban areas (Figure 2). Each of the image mosaics were then clipped to the extent of the new DCDB catchment layers to create images that only covered the urban areas of each catchment. It is important to note that the EHMP and gauged catchment boundaries were not necessarily hydrologically accurate, i.e. they had originally been defined using cartographic products without accurate drainage boundaries, and there was a mismatch of scales between the boundaries and the high-resolution images in some areas. This was not considered a serious issue given that the TIA estimates were being derived at a whole of catchment scale and the boundary errors mostly fell in areas with no urban cover.
Figure 2

DCDB filtering of non-urban areas.

Figure 2

DCDB filtering of non-urban areas.

Training data

Training data were developed by taking signature samples from each image mosaic. While the spectral resolution of the imagery was restricted to red, green, and blue visible wavebands, with no infra-red wavebands, the 0.5 m spatial resolution allowed for correct visual identification of surface features, without the need for ground truthing. In addition, the urban catchments were within a close distance, enabling easy access with one author of this paper even having his home within one of the catchments which provided an excellent ground truthing opportunity. As infra-red wavebands were not available to differentiate between features such as green roofs and green grass, or pavements and high-albedo compacted soils, it was necessary to obtain samples from a wide range of surface types. These included different coloured rooftops, different shades of bitumen roads, light and dark shades of grassed surfaces and trees and other common surface types such as water in large ponds. The signature sampling was the most critical aspect to obtaining accurate classification results and it was therefore given a significant proportion of the total effort.

Supervised classification

An initial supervised classification, using the parallelepiped non-parametric rule in ERDAS Imagine, was run for each mosaic using the respective signature sets as training data. During the supervised classification process, performance quality was checked against the original mosaics to identify errors, with some additional signature class sampling, where spectral confusion occurred, such as for different coloured road surfaces and where roof colours were confused with non-impervious areas, i.e. a green roof and green grass.

The result was then checked again followed by further sampling of signatures and a re-run of the classification to correct false negatives/positives if necessary. Once a classification result was deemed final, the image was converted to ArcGIS GRID format. The GRID was then re-classified into impervious and pervious classes within ArcGIS (Figure 3). The TIA was estimated for each catchment by viewing summary statistics to obtain the number of impervious grid cells (each cell = 0.25 m2) for reclassified catchment grid) and subtracting number of false-positive areas from the digital cadastre comparison.
Figure 3

TIA estimation using image analysis of aerial orthophotos. Aerial orthophoto and impervious signature colours are shown in the left-hand side and identified impervious areas are shown in the right hand side (‘1 = red colour’ and ‘2 = blue colour’ are impervious and pervious area, respectively).

Figure 3

TIA estimation using image analysis of aerial orthophotos. Aerial orthophoto and impervious signature colours are shown in the left-hand side and identified impervious areas are shown in the right hand side (‘1 = red colour’ and ‘2 = blue colour’ are impervious and pervious area, respectively).

The TIA estimates from the image analysis of areal orthophotos method were compared to TIA estimates, for some of the same catchments, that were derived through three other techniques, i.e. (a) manual digitization of geo-referenced aerial photos; (b) BCC data derived from image analysis using Landsat TM imagery; and (c) rainfall runoff depth relationship. The manual digitization was conducted for few small sample areas by estimating roof, parking, road and other impervious areas and then averaged over the whole catchment area. The rainfall and runoff data for the study catchments were collected from the SEQ stormwater harvesting project (Aryal et al. 2016) and analysed using the method of Boyd et al. (1993). The method is considered hydrologically accurate, but it is often difficult to get rainfall and runoff data for many catchments (Janke et al. 2011).

Constraints

The Major constraints of the automated image analysis technique were from the effects of shade and different colours of impervious surfaces, which required more signature sampling in order to capture a representative range of most impervious surface types (Figure 4). However, while the shading effects cannot easily be corrected, the effects of coloured surface confusion, e.g. green roof and forest, can be reduced with careful selection of signature colours and several iterations of the supervised classification.
Figure 4

Two major constraints in the image classification process were the shades and multiple roof colours, as shown in the figure.

Figure 4

Two major constraints in the image classification process were the shades and multiple roof colours, as shown in the figure.

RESULTS AND DISCUSSION

The results for the TIA estimates were compared to existing estimates, where they existed, as shown in Figure 5. The high-resolution colour aerial photography method compared favourably to the other techniques with standard deviations of TIA (%) ranging between 0.8% and 8%. In general, the image analysis of areal orthophotos method predicted lower values of impervious area, except for the lower Yaun creek and the Blunder creek catchments. The accuracy of the image analysis of areal orthophoto method is completely dependent on the ability to correctly represent the amount of roof, parking lot, drive way, road and other impervious areas in a catchment.
Figure 5

Estimated TIA for two sets of catchments located in the SEQ. Different methods are expressed as Manual (manual digitization of aerial photos, RR (Rainfall–runoff depth relationship), BCC (Brisbane City Council data), and GIS (image classification and cadastral filtering technique). For Blunder Creek (Durack), Stable Swamp Creek (Rocklea) and Lower Yuan Creek (Coomera), rainfall–runoff linear relationships were absent. The Lower Yuan Creek is located outside of BCC impervious area data. (a) Urban Catchments. (b) Urban Catchments with Water Sensitive Urban Development (WSUD) Features.

Figure 5

Estimated TIA for two sets of catchments located in the SEQ. Different methods are expressed as Manual (manual digitization of aerial photos, RR (Rainfall–runoff depth relationship), BCC (Brisbane City Council data), and GIS (image classification and cadastral filtering technique). For Blunder Creek (Durack), Stable Swamp Creek (Rocklea) and Lower Yuan Creek (Coomera), rainfall–runoff linear relationships were absent. The Lower Yuan Creek is located outside of BCC impervious area data. (a) Urban Catchments. (b) Urban Catchments with Water Sensitive Urban Development (WSUD) Features.

The rainfall–runoff method (Boyd et al. 1993) is relatively simple to implement, as it did not require familiarity with specialized software tools (e.g. ArcGIS) and it is considered hydrologically accurate, however availability of rainfall and runoff data is not guaranteed in most of the cases. The manual digitization technique is time and effort intensive, and practically not feasible. By contrast, the areal orthophoto image analysis using the ArcGIS software has the advantage of being applicable to ungauged catchments. The TIA estimates of this study have already been used in the stormwater harvesting research in SEQ (Aryal et al. 2016). Aryal et al. (2016) showed that as the extent of impervious areas across the catchments increased, the duration of stream flow under high flow conditions increases together with increases in the mean flow and the frequency of runoff events. However, many hydrological responses to increasing imperviousness were found linked to the physical characteristics of catchments, and to the spatial pattern of imperviousness. The TIA information is needed in all hydrological models. The technique developed through this study can efficiently be applied in any hydrological study in any catchment.

CONCLUSION

Imperviousness is considered as a surrogate measure of urbanization. The adverse impact of increasing urbanisation on in-stream and downstream ecosystems is a major and ongoing environmental concern. The TIA is an essential input in all ecohydrological modelling studies. However, estimation of TIA of catchments is still a challenge for practitioners. The TIA of a catchment can be estimated from their rainfall versus runoff plot, but rainfall and runoff data is not always available especially for ungauged catchments. Image analysis is an alternate method to accomplish this task. Through this study, a method was developed for image analysis of areal orthophoto using the AcGIS software. The method was applied to eight gauged catchments located in SEQ, Australia. The selected catchments were used as test beds for ecohydrological studies in that region. The image analysis using colour orthophotos was found to estimate the TIA reasonably accurately, a maximum of 8% variation was found when compared with other techniques (rainfall and runoff plot, manual digitization and previously estimated value). Since the technique reasonably estimates impervious area, it was applied in ecohydrological studies in that region (Aryal et al. 2016). The effects of shade and different colours of impervious surfaces (e.g., roofs) were the major constraints identified in the technique. The effects of coloured surface confusion (e.g., green roof and forest) can be minimised with careful selection of signature colours and several iterations of the supervised classification. Further investigation is recommended for minimising shading effects.

ACKNOWLEDGEMENTS

The authors would like to acknowledge the South-East Queensland Urban Water Security Research Alliance (SEQ UWSRA) for funding the work as well as Dr Meng Chong and Dr Peter Hairsine for reviewing the manuscript. Also, thanks to the Mr. Richard Gardiner (Department of Science, Information Technology and Innovation, QLD, Australia) for providing us the images including the Brisbane City Council data.

REFERENCES

Alley
W. M.
Veenhuis
J. E.
1983
Effective impervious area in urban runoff modelling
.
Journal of Hydraulic Engineering
109
(
2
),
313
319
.
Aryal
S. K.
Ashbolt
S.
McIntosh
B. S.
Petrone
K. P.
Maheepala
S.
Chowdhury
R. K.
Gardener
T.
Gardiner
R.
2016
Assessing and mitigating the hydrological impacts of urbanisation in semi-urban catchments using the storm water management model
.
Water Resources Management
DOI 10.1007/s11269-016-1499-z
.
Beach
D.
2003
Coastal Sprawl: The Effect of Urban Design on Aquatic Ecosystems in the United States
.
Pew Oceans Commission
,
Virginia
,
USA
.
Boyd
M. J.
Bufill
M. C.
Knee
R. M.
1993
Pervious and impervious runoff in urban catchments
.
Hydrological Sciences Journal
38
(
6
),
463
478
.
Bureau of Meteorology, BoM
2012
Average Annual and Monthly Evapotranspiration
. .
Delleur
J. W.
2003
The evolution of urban hydrology: past, present, and future
.
Journal of Hydraulic Engineering
129
,
563
573
.
Han
W. S.
Burian
S. J.
2009
Determining effective impervious area for urban hydrologic modelling
.
Journal of Hydrologic Engineering
14
(
2
),
111
120
.
Janke
B.
Gulliver
J. S.
Wilson
B. N.
2011
Development of Techniques to Quantify Effective Impervious Cover
,
Research Report CTS 11-20
.
Center for Transportation Studies, University of Minnesota
,
USA
.
Kunapo
J.
Chandra
S.
Peterson
J.
2009
Drainage network modelling for water sensitive urban design
.
Transactions in GIS
13
(
2
),
167
178
.
Lee
J. G.
Heaney
J. P.
2003
Estimation of urban imperviousness and its impacts on storm water systems
.
Journal of Water Resources Planning and Management
129
(
5
),
419
426
.
Paul
M. J.
Meyer
J. L.
2001
Streams in the urban landscape
.
Annual Review of Ecology, Evolution, and Systematics
32
,
333
365
.
Shuster
W. D.
Bonta
J.
Thurston
H.
Warnemuende
E.
Smith
D. R.
2005
Impacts of impervious surface on watershed hydrology: a review
.
Urban Water Journal
2
,
263
275
.