Urban Water Modeling and
Impervious Surface Analysis with GIS

Introduction

This bibliography [for GEO 565, Geographic Information Systems and Science, Winter 2007] provides background information for using GIS to explore the relationships between urbanization and hydrology. Impervious surfaces of urban areas influence hydrology by preventing infiltration, interception and evapotranspiration. Therefore, quantifying impervious surface coverage is crucial to managing stormwater in urban areas, and is one of the major areas where GIS can improve our resource management.

My name is Yarrow Murphy, and I am a graduate student at Oregon State University, studying Water Resources Engineering. My thesis research will examine whether decreasing connectivity of impervious surfaces affect hydrologic response as measured in reduced peak flow volume and extended storm flow duration. Answering this question involves analysis of impervious surfaces, their connectivity and their influence on hydrology. All aspects of this research will incorporate GIS, and draw upon many of the topics discussed in this bibliography. If you would like to contact me, scroll to the bottom of this page.

Annotated Bibliography

Dougherty, M., R. L. Dymond, T. J. Grizzard, A. N. Godrej, C. E. Zipper, and J. Randolph (2007). Quantifying long-term hydrologic response in an urbanizing basin. Journal of Hydrologic Engineering. 12(1), 33-41.

The Occoquan River Watershed is located west of Washington, D. C. This research investigates the relationship between long term changes in land use and changes in hydrology in four study basins. One of these study basins, the Cub Run, has urbanized significantly since the 1970s.

GIS and Landsat images provided 30 m resolution land use data from 1979 to 2000. The percent impervious surface was determined using a classification method in GIS. Land cover classes are based on field samples of reference surfaces, which the model then uses to sort each grid cell by cover class.

Rainfall and basin discharge were analyzed for the same period. Results showed that in years after 1983,when impervious surface area reached 9%, storm runoff per basin surface area in the urbanized basin was greater than in adjacent non-urbanized basins. Those non-urbanized basins had a higher degree of interception, evapotranspiration and infiltration than urbanized areas. Because deciduous species comprised most of the forest cover in the less disturbed basins, interception decreased during winter months, and the proportion of runoff was closer to that in the Cub Run. Evidence supports conclusions of previous work showing that above 10% imperviousness significantly increases storm runoff volumes.

Fankhauser used color and infrared aerial photos to classify impervious surfaces at 0.25 and 0.75 m resolution based on classification algorithms. The results obtained using infrared images were enhanced with color photographs, which improved accuracy slightly. The small improvement in accuracy did not justify the labor investment required for the second set of photographs. Estimated impervious surface based on color photo analysis was within 10% of hand field measurements of impervious surfaces. Estimations based on infrared achieved similar accuracy to those obtained with color photographs. With the development of this automatic estimation of impervious surfaces, hydrologic modelers can now save much time previously taken to perform the same estimations using older methods (statistical sampling or digitizing from maps or aerial photos).

Problems encountered in this classification process were similar to those mentioned in other research on this same topic. Similarity of certain surfaces results in misclassification. Occasional, single or small clusters of pixels are misclassified. Shadow areas are impossible to interpret and must be interpolated. This study encountered misclassification of bare soil into the impervious class which increased error rates. A later study by Tuesink in arid Arizona, USA, also summarized in this bibliography, encountered this problem as well and developed a model to distinguish these land classes.

Research outlined in this article and the work presented by Tuesink present different, but similar approaches to the same problem. They both point out the benefits of more accurate impervious surface estimations compared to the cost of improvements. Even as our computer technology becomes more sophisticated, it continues to be important to consider the costs of advanced sophistication and balance those costs with the level of accuracy and detail required to solve today's spatial problems.

Comparing hydrologic response in two catchments with contrasting degrees of impervious surface connectivity will require accurate measurement of impervious surfaces. As for any research, it is important to consider the accuracy required to obtain meaningful statistical results. This study analyzed impervious surface coverage in 8, 11, 125 hectares, whereas my research proposes to analyze one to two ha study areas. Differences in scale must be considered when determining the accuracy requirements and how errors will influence results.

Lee, J. G., and J. P. Heaney (2003). Estimation of urban imperviousness and its impacts on storm water systems. Journal of Water Resources Planning and Management-Asce. 129(5), 419-426.

This paper covered three components of research on imperviousness and its impact on storm water systems; Literature review, long-term hydrologic analysis of an urban watershed, and comparison of imperviousness measurement methods. Impervious measurement methods focused on quantifying directly connected impervious surfaces (DCIA). Impervious surface connectivity remains an area of difficulty in urban hydrological analysis, even with the highest resolution remote sensing methods. Often connectivity is inferred based on road density or land use. Lee and Heaney used five levels of measuring and GIS based analysis to measure DCIA. The GIS only method overestimated DCIA by 1.5 times the value obtained through field surveys, while field methods required more labor resources.

Acceptable error should be considered when deciding which level to use in analyzing impervious surfaces. Street boundary condition (curbed or not) data was lacking from the GIS databases, and this turned out to be a critical source of error in the GIS only analysis. These results underline the need, and resources necessary for field surveys to evaluate connectivity for my thesis research.

Lu, D. S., and Q. H. Weng (2006). Use of impervious surface in urban land-use classification. Remote Sensing of Environment. 102(1-2), 146-160.

Because of the role that impervious surfaces play on urban hydrology and stormwater management, it is important to develop more efficient methods of determining impervious coverage. Lu and Weng present an approach to surface and land use classifications and apply it to Indianapolis, Indiana. The impervious surface classification method incorporates surface temperature and albedo, derived from Landsat 7 ETM+ satellite images. 30 m resolution thermal infrared data was extracted from the ETM+ reflective band. Surface temperatures were derived from the infrared data. The thermal data was combined with albedo to classify surfaces. Once classified, comparison to orthophotos provided validation of the method. Overall, the thermal and albedo method results in 83.8% accuracy, with greater accuracy in areas with more than 30% impervious surfaces. Land use classification combines the impervious surface density obtained through the thermal albedo method and population density, obtained from census data.

Land use type turns out to be more difficult because the classes are less clearly defined and the surface cover and population layers are completely different data structures. The complexity of the urban landscape leads to mixed pixels, which complicate these classification processes. Like other methods for surface classification, surfaces with similar characteristics are misclassified. This confusion often occurs between dry soils and impervious surfaces, tree canopies obscuring the surface, surfaces in shadows or dark surfaces.

This classification is comparable to other methods which use orthophotos or multispectral images. However, some of the other approaches outlined in this bibliography field sampled to validate their classification models, which were based on orthophotos. These tests illustrated that orthophoto interpretation is only about 90% accurate. It would be interesting to see this method field sample rather than using orthophotos for validation.

Qihao, W. (2001). Modeling Urban Growth Effects on Surface Runoff with the Integration of Remote Sensing and GIS. Environ. Manage. V28(6), 737-748.

Integrated remote sensing and GIS provide a set of tools to analyze the progression of urbanization over time. This study examined changing land use and its impact on China's Zhujiang River Delta. Changes the runoff, and the rainfall-runoff relationship were analyzed using GIS with the NRCS SCS curve number method of determining runoff. Remote sensing was used to determine annual changes in urbanization. Surface runoff was modeled in the GIS for each year, and used to find the change in surface runoff from year to year. Much of the spatial information, such as soils and rain gages, had to be digitized from paper maps. Precipitation data was obtained from government operated gages, mostly located in urban areas. The precipitation was interpolated using the Theissen Polygon method in ArcGIS. Combining the land cover image and soil layers generated the curve number layer for the SCS analysis. Ultimately, runoff layers for 1989 and 1997, and a layer of difference in 1989 and 1997 runoff were created. The runoff changes were compared to changes in land use by overlaying the difference layers for urbanization and runoff.

The final remarks in the article point out the need to validate this GIS based model. The one criticism that I have of this paper, is that both the runoff model and the urbanization analysis are based on the same source spatial data. The correlation between the two results then could be explained in the data source. Although the SCS methos is commonly used and well understood, the data used for the analysis could be generating error. Should the results of the runoff model and land use analysis be validated, then the correlation between land use changes and changes in hydrology would be more strongly shown.

This research illustrates a method to relate urbanization and hydrology at the large scale. Some points from this article to remember: Runoff coefficients increase with size of storm event (flood). Studied the effect of urbanization with runoff coefficient curve patterns (how much runoff coefficient changes with storm size changes). Included diagram illustrating the GIS implementation.

Sample, D. J., J. P. Heaney, L. T. Wright, and R. Koustas (2001). Geographic Information Systems, Decision Support Systems, and Urban Storm-Water Management. Journal of Water Resources Planning and Management. 127(3), 155-161.

Sample et al demonstrate using GIS for urban watershed management. From the authors' perspective, the best use of GIS for water resource applications, considering the time investment required, is when a set of advance tools to support complex decision-making is integrated with the GIS and simulation model. This integrated tool set is referred to as a Decision Support System, or DSS. The research presented works at the neighborhood level in the urban environment. GIS in the urban environment and neighborhood scale differs from the most common use of GIS for watershed resources for hydrology at the basin scale, often in a less disturbed landscape. Urban land uses are complex, which complicates the hydrologic processes associated with them. One interesting point made in this literature review is that vector based data is often more suitable for representing the urban landscape, whereas rasters can represent the more uniform characteristics of natural landscapes.

This work provides an example of a conceptual DSS framework for small scale urban stormwater management. A parcel level GIS was developed for a hypothetical 43 ha study area. Based on the principle that micro-storms make up the majority of urban storm runoff and that it is important for urban stormwater management to control this runoff, because in an undisturbed state, these storms would mostly be abstracted by the soil surface, and would result in minimal runoff. The study considers multiple options for controlling this runoff using a DSS from the perspective of a land developer. The analysis considers economic optimization for various BMP approaches. The resulting method incorporates a database, a mathematical model and an evaluation tool (ie cost optimization), which are the three components of a DSS.

This article relates to my research in a few ways. It is one of the few articles working with urban hydrology on such a small scale. The use of the simple hydrological model provides an example of a hydrological model for urban areas. Lastly, this work illustrates how using optimization with a mathematical model and GIS can help in making complex decisions.

Seth, I., P. Soonthornnonda, and E. R. Christensen (2006). Use of GIS in Urban Storm-Water Modeling. Journal of Environmental Engineering. 132(12), 1550-1552.

This article presents an overview of GIS applications in water resources. The purpose of this article is to introduce these ideas to practicing professionals who may not yet be using these tools and to encourage the entire profession to work to resolve the issues currently limiting GIS use for watershed management.

At this time, one of the major limitations to widespread and efficient use of GIS is the availability of the right kind of data. This brief overview of the uses of GIS in water resources management discusses this and other aspects of integrating GIS into making decisions about water. Once the necessary data is developed, managers can perform such tasks as determining stormwater runoff or evaluating stormwater mitigation. This article recommends specific actions to be taken by water resource professionals, including working closely with GIS specialist to gain the skills needed to use GIS tools. Furthermore, the author strongly recommends developing more robust local and national data repositories. This data must be compatible with updated software, and available to the appropriate users.

It is helpful to think about the larger context of GIS and watershed management integration as well as the technical implementation details. Although broad, this article touches on issues discussed in other, more specific issues.

Tuesink, M., R. Chasan, N. Thomas, C. Lovely, and M. Ledbetter (2001), Beyond TM: Making High Resolution Imagery Work for Urban Applications, paper presented at Twenty-First Annual ESRI User Conference, ESRI.

Aerial multispectral images were taken and used to determine actual impervious area of the city of Scottsdale, AZ. The amounts of impervious surface were then compared to empirical values obtained from look-up tables based on urbanization. At a 7000 feet altitude, an ADAR 5500 multi-spectral digital camera system took images at one to two meter GSD resolution. Each digital frame covered 1 km x 1.5 km, had a 1-2 meter GSD, and contained four individual images representing either reflected blue, green, red, or near-infrared light. The vignetting produced by the camera was removed. The images were geometrically corrected and the bands were co-registered. Classification sorted land surfaces into two impervious (rooftops and pavement) and three pervious (vegetation, bare soil, and water) land cover classes. The initial classification produced 50 classes, which were then identified based on field sites. Other field sites were also later used to validate the classifications. Then, using statistical analysis (minimum distance Euclidian clustering), the incorrectly classified signatures were determined and corrected whenever possible. Models were constructed to determine the remaining confused pixels. For example, a pixel within a buffer around street centerlines could be classified as pavement. Other model bases included land use zoning and proximity to swimming pools. The paper further describes the method used to calculate percent impervious surface using ArcMap and Spatial Analyst.

Ultimately, this impervious surface data was used to re-evaluate stormwater infrastructure, and resulted in reducing the capacity of several pipe construction projects in Scottsdale. The combined savings achieved through downsizing these projects was nearly $100,000. However, this study also found that an equal amount of storm runoff estimations are underestimated, mainly in older residential areas. Overall, 60% of storm drainage facilities were improperly sized, according to the results of this study. In areas with undersized infrastructure, flooding risk increases.

This project provides a good example on which to base impervious surface classifications. Because this project took place in the desert, the landscape classes had more spectral similarities than in a more humid climate such as Portland, Oregon. Bare ground and pavement, for example, could be easily confused in the desert. The classification process could comprise an entire master's thesis, depending on the extent of such a project.

Wolock, D. M., T. C. Winter, and G. McMahon (2004). Delineation and evaluation of hydrologic-landscape regions in the United States using geographic information system tools and multivariate statistical analyses. Environ. Manage. 34, S71-S88.

The USGS has an ongoing project called the National Water Quality Assessment Program. The purpose of this program is to assess the trends and status of ground and surface water quality in the United States. In the face of reduced funding, the number of watersheds that could be continually monitored must be minimized. This paper presents the results of a project to analyze the characteristics of watersheds at the scale of 200 km2 and classify those watersheds based on similar surface form, geologic texture and climate into 20 Hydrologic Landscape Regions (HLR). Each HLR is a set of watersheds with similar characteristics, but not necessarily in the same geographic region. The surface form analysis was done using a USGS digital elevation model with resolution 1 km. This study used GIS extensively for statistical analysis. After the watersheds were statistically sorted using GIS, the results were then analyzed for means, standard deviations and ANOVA. Fish species richness and nitrogen transport efficiency were used as water quality indicators. These water quality parameters were correlated against the regional hydrologic framework to test the correlation power of the HLRs.

For my thesis, I will be sorting micro-scale catchments using landscape characteristics. This study illustrates a method to classify areas of hydrologic similarity using GIS. My project will require excluding areas that do not fit certain criteria (such as above and below a slope range). The remaining areas will be grouped based on impervious surface area, soil and slope.

Xu, Z. X., K. Ito, G. A. Schultz, and J. Y. Li (2001). Integrated Hydrologic Modeling and GIS in Water Resources Management. Journal of Computing in Civil Engineering. 15(3), 217-223.

GIS has increased the accuracy and efficiency of watershed modeling for the purposes of managing water resources. This article reviews previous watershed modeling efforts and then uses GIS to represent watershed complexities and incorporate them into a previously developed watershed model. GIS was used before modeling to calculate watershed parameters for the hydrologic model. GIS was again used at the end of the modeling process to present results. The model used here is a "tank" type model, which means that water in the watershed in stored in various tanks, which represent surface, subsurface, groundwater and river or stream. GIS layers included stream hydrography, elevation, land use classification, hydrologic soil classification, surface geology classification and rain gage points. Movement of water is determined by parameters of the tanks. When using GIS, each grid cell has associated with it a set of tanks. Because multiple parameters can be described for each grid cell, when the model is applied using GIS, the variation within a watershed is more accurately represented. For example, a watershed model incorporating GIS can be used to test the effects of land use changes on runoff processes.

Using GIS for modeling is very powerful because parameters can be described for each grid cell in a layer, and multiple layers can add even more complexity. Models can be spatially varied, which makes their representation of reality more accurate. Using GIS to display results makes it easier to see landscape scale watershed process patterns.