Surface Water Quality and · 2019-01-15 · 1. Develop improved and consistent impervious surface estimates using ERDAS Imagine’s Subpixel Classifier® for four separate dates of

PROJECT COMPLETION REPORT

Surface Water Quality and Impervious Surface Quantity: a Preliminary Study

NOAA Grant NA16OC2673

March 2004

James D. Hurd, Research Associate Daniel L. Civco, Principal Investigator

Center for Land use Education And Research

Department of Natural Resources Management & Engineering College of Agriculture and Natural Resources

The University of Connecticut Storrs, CT 06269-4087

Table of Contents

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Goals and Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Impervious Surface Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

Sub-pixel Classifier Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Initial Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

First Iteration of Sub-pixel Classification . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Signature Refinement and Second Iteration of Sub-pixel Classification . . 4 Identification of Impervious Surfaces Through Supervised Classification 6 Application of a 3x3 Majority Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Inclusion of Developed Land Cover Information . . . . . . . . . . . . . . . . . . . . 7 Sub-pixel Classification of 10 Percent Impervious Pixels . . . . . . . . . . . . . 7

Sub-pixel Classification of Other Dates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Creation of Final IS Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Final IS Estimate Adjustment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Water Quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 Need for Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 Appendix A: Examples of 2002 Impervious Surface Estimates

and Land Cover by Watershed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

Appendix B: Per Pixel Comparison Planimetric and Estimated Percent Impervious Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

31

Appendix C: Linear Regression of Water Quality Variables and Impervious Surface . . . 44

i

List of Figures Figure 1. Hydrologic impact of urbanization flowchart . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Figure 2. Overview of procedures flowchart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Figure 3. Example of 1995 initial sub-pixel classification results . . . . . . . . . . . . . . . . . . . 6 Figure 4. Example of 1995 supervised classification results . . . . . . . . . . . . . . . . . . . . . . . 7 Figure 5. Example of inclusion of 1995 land cover information and 1995 final result . . . 8 Figure 6. Examples of classification adjustment and final results . . . . . . . . . . . . . . . . . . . 10 Figure 7. Planimetric validation data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Figure 8. Location and areal extent of study watersheds . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Figure A-1. 2002 impervious surface estimate and land cover for the Farmington

regional watershed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

Figure A-2. 2002 impervious surface estimate and land cover for the Housatonic regional watershed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

25

Figure A-3. 2002 impervious surface estimate and land cover for the Naugatuck regional watershed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

26

Figure A-4. 2002 impervious surface estimate and land cover for the Quinebaug regional watershed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

27

Figure A-5. 2002 impervious surface estimate and land cover for the Quinnipiac regional watershed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

28

Figure A-6. 2002 impervious surface estimate and land cover for the Salmon regional watershed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

29

Figure A-7. 2002 impervious surface estimate and land cover for the Saugatuck regional watershed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

30

Figure B-1. 1990 West Hartford validation data (area 1) and difference graph . . . . . . . . . 32 Figure B-2. 1990 West Hartford validation data (area 2) and difference graph . . . . . . . . . 33 Figure B-3. 1995 Marlborough validation data and difference graph . . . . . . . . . . . . . . . . 34 Figure B-4. 1995 Waterford validation data (area 1) and difference graph . . . . . . . . . . . . 35 Figure B-5. 1995 Waterford validation data (area 2) and difference graph . . . . . . . . . . . . 36 Figure B-6. 2002 Woodbridge validation data and difference graph . . . . . . . . . . . . . . . . . 37 Figure B-7. 2002 Milford validation data and difference graph . . . . . . . . . . . . . . . . . . . . . 38 Figure B-8. 2002 Suffield validation data and difference graph . . . . . . . . . . . . . . . . . . . . . 39 Figure B-9. 2002 Groton validation data and difference graph . . . . . . . . . . . . . . . . . . . . . 40 Figure C-1. Linear regression plots of instantaneous discharge and impervious surfaces . 45 Figure C-2. Linear regression plots of dissolved oxygen and impervious surfaces . . . . . . 46 Figure C-3. Linear regression plots of field pH and impervious surfaces . . . . . . . . . . . . . 47 Figure C-4. Linear regression plots of fecal coliform and impervious surfaces . . . . . . . . . 48 Figure C-5. Linear regression plots of chloride and impervious surfaces . . . . . . . . . . . . . 49 Figure C-6. Linear regression plots of total residue and impervious surfaces . . . . . . . . . . 50 Figure C-7. Linear regression plots of phosphorus and impervious surfaces . . . . . . . . . . . 51 Figure C-8. Linear regression plots of turbidity and impervious surfaces . . . . . . . . . . . . . 52 Figure C-9. Linear regression plots of nitrogen and impervious surfaces . . . . . . . . . . . . . 53 Figure C-10. Linear regression plots of heavy metals and impervious surfaces . . . . . . . . 54 Figure C-11. Linear regression plots of water quality measures and impervious surfaces

for all four dates combined . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

ii

List of Tables

Table 1. Comparison of truth and estimated impervious surfaces for validation areas . . . 13 Table 2. Twelve month average water quality measurements . . . . . . . . . . . . . . . . . . . . . . 14 Table 3. Significance of linear regressions for individual water quality measures by

individual date . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

Table 4. Significance of linear regression for individual water quality measures by combined dates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

17

Table B-1. 1990 impervious surfaces error matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 Table B-2. 1995 impervious surface error matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 Table B-3. 2002 impervious surface error matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

iii

Introduction Nonpoint source pollution (NPS) has been cited as one of the top contributors to water quality problems in the United States (U. S. EPA, 1994). Nitrogen and phosphorus have been identified as the primary nutrients responsible for algal blooms caused by eutrophication which results in fish die-off, endangers human health, and impacts the economic and recreational use of riverine, palustrine, and estuarine waters (U.S. EPA, 1996). Urban runoff has been found to contribute a significant amount of these and other nonpoint source pollutants to our water resources (Beach, 2002; Boyer et al., 2002; U.S. EPA, 2002). It has been well-documented that urbanization increases the volume, duration, and intensity of stormwater runoff (Booth and Reinfelt, 1993). Imperviousness influences hydrology (e.g., an increase in imperviousness is directly related to increase in the volume and velocity of runoff), stream habitat (e.g., the hydrological impacts of increased imperviousness lead to increased stream bank erosion, loss of riparian habitat, and degradation of in-stream habitat), chemical water quality (e.g., increases in imperviousness and runoff directly effect the transport of non-point source pollutants including pathogens, nutrients, toxic contaminants, and sediment), and biological water-quality (e.g., all the above changes have an adverse impact on the diversity of in-stream fauna) (Schueler, 1994; Arnold and Gibbons, 1996). These hydrological impacts of urbanization, particularly impervious surfaces are depicted in Figure 1. Additional research has also suggested that the amount of urban runoff and its impacts on stream conditions and water quality are strongly correlated to the percent area of impervious surfaces within a watershed (Schueler, 1994; Arnold and Gibbons, 1996; Clausen et al., 2003). This strong relationship implies impervious surfaces can serve as an important indicator of water quality, not only because imperviousness has been consistently shown to affect stream hydrology and water quality, but because it can also be readily measured at a variety of scales (i.e., from the parcel level to the watershed and regional levels) (Schueler, 1994). Understanding the degree and location of impervious surfaces and limiting the amount of impervious surface in a watershed is an important component of overall watershed management. Coastal resource and land use managers need to be able to determine the existing percent imperviousness for an area in order to develop appropriate watershed management and/or NPS mitigation plans and to understand the link between water quality and impervious surfaces. Much research has focused on determining the relationship between watershed impervious surface coverage and water resources impacts. Many of these have identified a positive correlation among percentage of urban land or imperviousness and select water quality parameters (Herlihy et al., 1998; Brabec et al., 2002; Boyer et al., 2002; Clausen et al., 2003; Roy et al., 2003). However, work has only recently been undertaken to develop methods to measure impervious surfaces at the watershed scale or larger (Ji and Jensen, 1999; Bird et al., 2000; Wang et al., 2000; Ward et al., 2000; Bird et al., 2002; Justice and Rubin, 2003; Yang et al., 2003). Past efforts to determine watershed imperviousness have been hampered by inconsistent methods and outdated or unavailable data. There is a need for a consistent and replicable technique to calculate easily and quickly watershed imperviousness from readily available and cost effective remote sensing information that achieves an acceptable level of accuracy. This research strives to attain that level of consistency and accuracy in regards to a temporal set of impervious surface estimations and to confirm the link between measured water quality parameters and upstream impervious surfaces.

1

Population density increases

Building density increases

Waterborne waste increases

Water demand rises

Impervious areaincreases

Drainage system modified

Water resource problems

Urban climate changes

Stormwater quality deteriorates

Groundwater recharge reduces

Runoff volume increases

Flow velocity increases

Receiving water quality deteriorates

Base flow reduces

Peak runoff rate increases

Lag time and timebase reduced

Pollution control problems

Flood control problems

Urbanization

Figure 1. Hydrologic impact of urbanization. Gray boxes identify impacts directly related to impervious surfaces (adapted from Hall, 1984). Goals and Objectives The goals and objectives of this research were to:

1. Develop improved and consistent impervious surface estimates using ERDAS Imagine’s Subpixel Classifier® for four separate dates of Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper (ETM) imagery spanning a 17 year period (April 26, 1985, August 30, 1990, August 28, 1995, and September 8, 2002).

2. Relate historical water quality information for select watersheds to the four date watershed-based impervious surface estimates to determine, preliminarily, if a relationship between increased impervious surfaces and decreased water quality can be identified over time.

3. Use the results to substantiate further the often-assumed adverse influence of impervious surfaces on water quality and also to recommend future research directions.

2

Impervious Surface Estimation Researchers at The University of Connecticut’s Center for Land use Education And Research (CLEAR) have long been involved in investigating ways to measure impervious surfaces (IS), and to use these data in educational programs such as the Nonpoint Education for Municipal Officials (NEMO) Project (Arnold et al., 1993; Stocker et al., 1999). One direction of research has continued to improve upon traditional techniques of assigning percent IS coefficients as a function of land cover type (Prisloe et al., 2000; Sleavin et al., 2000, Prisloe et al., 2001). Another direction of research has been aimed at sub-pixel percent IS modeling directly from Landsat TM and ETM data (Civco and Hurd, 1997; Flanagan and Civco, 2001a, Civco et al., 2002). The research presented here focuses on this latter method of impervious surface estimation and its application to a temporal dataset of Landsat imagery. Sub-pixel Classifier Overview One of the more promising methods for sub-pixel estimation of percent IS has been the application of Leica Geosystems’ ERDAS Imagine Sub-Pixel ClassifierTM (SPC) to Landsat TM and ETM image data. The SPC, engineered by Applied Analysis Inc., is a supervised classifier that enables the detection of materials of interest (MOIs) as a whole or fractional component of an image pixel, with a minimum detectable threshold of 20 percent and in increments of 10 percent (i.e., 20-30%, 30-40%, … 90-100%) (Flanagan and Civco, 2001a; Flanagan and Civco, 2001b). The process consists of identifying and removing the unwanted spectral contribution of materials that make up the background of the pixels and comparing the remaining spectrum to the signature of the material of interest. (Huguenin et al., 2004). The pixel is considered to contain some portion of the MOI if the remaining spectrum matches the signature spectrum. As a brief description of the process, the SPC module is comprised of four required steps. These are image preprocessing, image environmental correction, signature derivation, and MOI classification The first two steps are autonomous – preprocessing resulting in a hidden, companion file to the original image being classified, and environmental correction resulting in a CORENV companion file that contains information pertaining to atmospheric and environmental correction factors. Signature derivation is conducted manually using the Imagine Area Of Interest (AOI) tool to identify pixels with a minimum of 90 percent imperviousness. Because of the diverse reflectance characteristics of impervious surfaces, signatures were individually created for bright, medium, dark, and very dark sub-classes of impervious surfaces. These sub-classes were grouped into a single signature file called a ‘family’ using the optional Signature Combiner function in the SPC. The final step is MOI Classification. This step utilizes the initial preprocessed image, corresponding environmental correction file, and the ‘family’ signature file. Additional parameters that can be set by the user are the classification tolerance and number of output classes. Signature tolerance is a parameter that can be used to adjust the number of MOI detections, and its value can be increased to include more pixels in the classification result, or decreased to reduce false detections. The number of output classes can vary from two, four, or eight. In this project, because of the combination of signatures representing four degrees of impervious surface brightness, the output image contains five thematic layers, one for each of the sub-class signatures and one showing detections for the combined contribution of the four sub-classes. It is this composite layer that is used as the impervious surface estimation.

3

The following sections describe the iterative procedure used to derive impervious surface estimates for the years 1985, 1990, 1995, and 2002. This process is graphically represented in Figure 2. Initial Classification Initial sub-pixel classification was conducted on the August 30, 1995 Landsat TM image. This image was selected based of the existence of planimetric data that corresponded closely to the collection date of the TM imagery. The planimetric data were used to assess the validity of the IS estimate and is discussed further in this report. The remaining three dates (1985, 1990, and 2002) of impervious surfaces were derived based on the results of this 1995 classification. To begin, a principal components image of the entire scene was generated from the 1995 TM image. The first principal component channel (PC1) is known to represent an overall brightness image of the Landsat TM scene and was generated to serve solely as a visual guide for the selection of training pixels representing the four brightness conditions of impervious surfaces. The PC1 image was divided into four levels to identify bright, medium, dark, and very dark reflective pixels within the scene. The four brightness levels represent various spectral characteristics of concrete and asphalt from newly paved conditions to older, more aged conditions, and various reflective characteristics of different roofing and construction materials. It should be noted that these distinctions among impervious surface classes are spectrally-based and do not imply the function of the impervious surface (i.e. roof, road, parking lot, etc…) (Flanagan and Civco, 2001a). From this point, the process of deriving an IS estimate from the 1995 Landsat image was an iterative one, with each step serving to improve the overall result. First Iteration of Sub-pixel Classification Image preprocessing and generation of an environmental correction file were performed as required by the SPC. Training pixels were selected that represent at minimum 90 percent imperviousness. This was done separately for each of the four brightness sub-classes with approximately 50 pixels selected for each sub-class. For a successful sub-pixel classification, the quantity of training pixels is less important then the quality of the pixels (i.e., pixels that represent nearly 100 percent imperviousness), but it is important to try to select pixels that represent the full spectral range of impervious surfaces. The signatures for each sub-class were combined using the SPC Signature Combiner then MOI Classification was performed. Based on tests conducted on a subset of Landsat TM data, a classification tolerance of 1.4 was used. The number of output classes was eight (i.e., class 1 = 20-30%, class 2 = 30-40%, … class 8 = 90-100%). The results were visually examined and it was readily apparent that the classification result needed improvement, due primarily to a number of highly impervious pixels not being detected by the SPC. An example of the results is shown in Figure 3b. Signature Refinement and Second Iteration of Sub-pixel Classification The SPC has a component that allows for signature evaluation and refinement. In this process additional pixels were selected for each of the four sub-classes that represented correctly classified pixels, falsely classified pixels, and missed pixels. These files were created using the AOI tool and applied to the existing signature files to produce a modified signature file for each brightness sub-class. These files were again combined and a new MOI Classification performed using a classification tolerance of 1.4 and eight class output. Visually the results showed a further decline in the number of pixels detected, but, the quality of those pixels detected

4

1995 Landsat TM 1995 Landsat TM D

1995 Landsat TMSupervised Classification

(parallelpiped/100 % pixels)

1995 Landsat TMSub-pixel Classification

(8 impervious levels)

1995 Landsat TMSub-pixel/Supervised

Combination

1995 Landsat TM3x3 Majority Filter

1995 Landsat TMDeveloped Pixels Only

1995 Land Cover

1995 Landsat TMSub-pixel Classification

Class 1 pixels only

Principal Component 1(showing 4 brightness levels)

Sub-pixelClassification

1985

1990

2002

eveloped Land Cover M

ask

Combine Classificationsfor Consistency

AdjustClassifications

FINAL IMPERVIOUSSURFACE ESTIMATIONS

20021995

19901985

Figure 2. Flow-chart overview of procedures followed to produce the four date impervious surface estimates.

5

appeared to improve slightly. However, with the lack of detected pixels experienced, additional steps were needed. Results of this step can be viewed in Figure 3c.

(a) August 28, 1995 Landsat TM image displaying bands 4, 5, 3.

(b) Sub-pixel classification following first iteration.

(c) Sub-pixel classification following signature refinement.

20% – 29 % 30% – 39% 40% – 49%

50% - 59% 60% – 69% 70% – 79%

80% – 89% 90% - 100%

Figure 3. Results of initial Sub-pixel Classification and signature refinement for a portion of Waterford, CT. Identification of Impervious Surfaces Through Supervised Classification To increase the detection of 100 percent impervious pixels that were missing from the sub-pixel classification, a supervised classification approach was applied. A signature file was created that contained signatures of various conditions of 100 percent impervious pixels. Supervised classification was performed using a parallelpiped decision rule. In using a parallelpiped decision rule, any pixel falling within the upper and lower limits of the class signature would be identified as belonging to that particular signature class otherwise the pixel would remain unclassified; a pixel not containing 100 percent imperviousness. If class boundaries overlapped, a maximum likelihood decision rule was followed. The supervised result was combined with the result of the Sub-pixel Classifier to produce a visually improved impervious surface. An example of the results can be viewed in Figure 4a. Application of a 3x3 Majority Filter Because there still existed undetected pixels that should contain high levels of imperviousness, and many of these pixels were surrounded by pixels that were detected as containing high levels of imperviousness through either the sub-pixel process or supervised classification process, a 3 x 3 majority filter was employed to fill in many of these missing pixels. For the majority filtering process, only impervious class values of 0, 5, 6, 7, 8 were used for computation and changes were only applied to pixels with a value of zero. The rationale behind this was that if a pixel was not detected as containing a high level of imperviousness, but was surrounded by pixels with high levels of imperviousness, then the undetected pixel would be given a value similar to its eight neighbors, but not influenced by pixels detected as having lower levels of imperviousness. This allows for a slight gradation of imperviousness if the majority of pixels within the window are not 100 percent. An example of the results of this process is provided in Figure 4b.

6

(a) Classification following addition

of supervised results.

(b) Classification following 3x3 majority filter.

20% – 29 % 30% – 39% 40% – 49% 50% – 59%

60% – 69% 70% – 79% 80% – 89% 90% - 100%

Figure 4. Results of the application of supervised derived 100 percent pixels and subsequent 3x3 majority filter function.

Inclusion of Developed Land Cover Information Given the similarities in spectral reflectance among many impervious features and barren land (or bare soil), an urban-related land cover mask was employed as an additional processing step to eliminate those pixels that were falsely detected as being impervious to some degree. A 1995 statewide land cover thematic layer that was developed as part of CLEAR’s on-going mission of creating landscape characterization information and derived from the same August 28, 1995 Landsat image, was used to mask out pixels detected as containing imperviousness but not identified as belonging to the developed category. Additionally, pixels that were classified as developed in the 1995 land cover but not detected as containing imperviousness were assigned to a new 10 percent impervious class (class 1). Because the Sub-pixel Classifier only detects the occurrence of component MOI’s to a minimum threshold of 20 percent, it was believed that by including this new, lower level impervious class, that the impervious surface estimation would more closely resemble the true extent of imperviousness within an area. Figure 5a provides an example of this process. Sub-pixel Classification of 10 Percent Impervious Pixels Comparing Figure 4b to Figure 5a, it is apparent that the SPC failed to detect a substantial number of pixels that obviously contain some degree of imperviousness but are now assigned a value of 10 percent impervious following the inclusion of the developed land cover category. The result is an excessive number of pixels on the 10% imperviousness class, which essentially serves to underestimate the total amount of impervious surfaces for a given area. In order to correct for this issue, only pixels identified as 10 percent imperviousness (class 1) were extracted from the Landsat TM image. Image preprocessing was performed, but the original environmental correction file was used since the same image was used. New signatures using pixels that represent only bright and dark occurrences of impervious surfaces were selected using the AOI

7

tool. These signatures were combined into a single signature file and MOI classification performed. The result was an increase in the detection of impervious pixels over the 10 percent value, with the number of pixels in the 20 to 60 percent impervious (classes 2 through 5) ranging showing a significant increase. The result is a more realistic estimate of impervious surfaces for a given area. An example can be viewed in Figure 5b where the increase in impervious levels can be seen, particularly along transportation routes.

(a) Inclusion of developed land cover pixels as 10 percent

impervious class

(b) Sub-pixel analysis of 10 percent impervious class Pixels

(c) Final classification result following classification of other

dates.

10% – 19 % 20% – 29 % 30% – 39%

40% – 49% 50% – 59% 60% – 69%

70% – 79% 80% – 89% 90% - 100%

Figure 5. Results if the inclusion of developed land cover pixels and final sub-pixel analysis. Sub-pixel Classification of Other Dates The remaining three dates of Landsat imagery (April 26, 1985, August 30, 1990, and September 8, 2002) were classified independently following the same procedures used under the first iteration of the 1995 Landsat image. For each date, preprocessing and environmental correction were applied and a principal component image was derived. Signatures were selected for bright, medium, dark, and very dark impervious pixels. These were combined into a single signature file and MOI classification performed. Since these classifications were built upon each other (i.e. the 1990 IS estimate was derived from the 1995 IS estimate, the 1985 IS estimate was derived from the final result of the 1990 IS estimate, and the 2002 IS estimate was derived from the 1995 IS estimate), the classification threshold was set to 1.5 to increase slightly the number of pixels detected as containing impervious surfaces. The resulting IS classification was combined with the corresponding land cover image, classified from the same Landsat image, to mask out erroneously detected impervious pixels and to generate the 10 percent impervious class. Creation of Final IS Estimates A fundamental principle in developing IS estimates for multiple dates was to maintain consistency in IS estimates among the four dates. Models were developed that would combine each adjacent pair of IS estimates to maintain this consistency. The primary concern was to eliminate the chance of a given pixel to fluctuate between dates, such as losing imperviousness between 1985 and 1990, then gaining imperviousness between 1990 and 1995, then losing

8

imperviousness again between 1995 and 2002, or to show unlikely increases or decreases in imperviousness over time. The first step was to derive a difference image between the completed IS estimate (i.e., 1995) and the date of IS to be modified (i.e., 1990). Since there are nine classes for each IS estimate, a constant of 10 was added to the difference to derive positive values ranging from one to 19. If a pixel contained a value of one, then that pixel had a class value of 0 (no imperviousness) in 1995 and a class value of nine (high imperviousness) in 1990, essentially a condition that should not occur and is a case of misclassification in one or both IS estimates. Using the results of the difference image, all pixels in the 1990 classification that deviated one class value below that identified in the 1995 IS estimate (i.e., a class value of 5 in 1995 and a class value of 4 in 1990) were identified. The spatial model then assigned the 1990 IS estimate value to these pixels and the 1995 IS estimate value to the remaining pixels. If a pixel value of zero occurred in the 1990 date, the pixel was assigned a zero value. The result is a final 1990 IS estimate that maintains the 1995 IS estimate levels unless the 1990 pixel was one class value below the 1995 level, based on the original 1990 IS sub-pixel classification and subsequent inclusion of the developed land cover information, or the pixel in 1990 was zero. The reasoning behind this procedure was to maintain a consistent set of impervious surface estimates over time based on the base 1995 IS estimate and eliminate large fluctuations between dates and also eliminate the possibility of a pixel decreasing in imperviousness, although possible, not very likely, at least at large scales. The same procedure was followed for the 1985 date using the 1990 IS estimate as the primary IS source. For the 2002 date, the 1995 IS estimate again served as the primary source, but instead of using a one class value below the 1995 IS estimate, one class value above the 1995 IS estimate was utilized. Also if the 1995 class value was zero, and the 2002 class value was greater then zero, the 2002 value was used. Final IS Estimate Adjustment Visual examination of the resulting IS estimates revealed there still appeared to be an underestimation of overall IS. To improve the overall IS estimate, another set of spatial models was produced that would increase the overall IS by using the results of the 1985 sub-pixel classification and building upon each subsequent date. Essentially what this model did was combine the 1985 final IS estimate derived in the previous step with the original 1985 sub-pixel classification. If a pixel showed a higher value in the original 1985 sub-pixel classification than for the 1985 final IS estimate, then it was assigned the higher value, otherwise it remained as specified in the 1985 final IS estimate. This adjusted 1985 final IS estimate was then applied to the 1990 final IS estimate in a similar manner. This continued for the 1995 and 2002 IS estimates. The result was an increase and more realistic estimate of overall IS for all four dates. Examples are provided in Figure 6 for each year and Appendix A contains the results of the 2002 IS estimate for each of the seven study watersheds in addition to the 2002 land cover and tabular information of general land cover and percent impervious surfaces for each date. Validation Data used to validate the percent impervious estimates derived from the previous steps consisted of high-accuracy planimetric data for several Connecticut towns (West Hartford for circa 1990, Marlborough, Waterford, and Woodbridge for circa 1995, and Groton, Milford, and Suffield for circa 2002). These data were derived from aerial photographs collected within a couple years of

9

Sub-pixel Classification with Land Cover Mask and

Inclusion of Developed Land Cover

Sub-pixel Classification with Inclusion of Developed Land Cover

Adjusted for Consistency

Final Impervious Surface Estimate

1985 1985 1985

1990 1990 1990

1995 1995 1995

2002 2002 2002

10% – 19% 20% – 29% 30% – 39%

40% – 49% 50% – 59% 60% – 69%

70% – 79% 80% – 89% 90% - 100%

Figure 6. Examples of sub-pixel classification and pixel adjustments for each of the four dates of the Manchester, Connecticut area.

10

the Landsat image data used to derive the percent impervious surface estimates. Planimetric data were not available for the 1985 time frame. The planimetric data consist of anthropogenic impervious surfaces such as buildings, roads, sidewalks, driveways, and parking lots. The towns with available planimetric data represent a range of different development densities from rural for Marlborough and Suffield to suburban for Waterford and Woodbridge, to urbanized for Groton, Milford, and West Hartford. The planimetric data were co-registered to the corresponding Landsat TM image. A grid corresponding to the pixel space of the Landsat TM image was then generated for a rectangular area within each town. The grid was then unioned to the planimetric data for each town. From these unioned images, database tables were generated from which the percent of impervious surfaces (from 0 – 100% impervious) was calculated for each grid cell. These tables were joined with the grid file to generate images that represent percent imperviousness. The images were recoded into the nine classes whose values corresponded to the output from the IS estimation procedure. This final, recoded layer for each of the towns comprised the validation data that were compared to the IS estimate on a per pixel basis. The areas for each town used for the validation are displayed in Figure 7. Visual comparisons between the planimetric derived impervious surface and sub-pixel impervious surface estimate along with graphs showing the difference between each image are provided in Appendix B. From a scientific perspective, how well the sub-pixel classification predicted actual percent imperviousness at the sub-pixel level, compared to the calibration planimetric data, is of principal interest. However, even slight mis-registration between the reference planimetric data and Landsat TM data could result in potentially large differences between actual and predicted values. Also, from a management perspective, the assessment of impervious surfaces is more meaningful when reported on a landscape management unit such as a drainage basin. Therefore, in addition to the per pixel-based impervious surface data summarized in Appendix B, percent impervious surface summaries were derived for each validation area. These results are displayed in Table 1. Water Quality Water quality data were acquired from the Water Resources Division of the United States Geological Survey (USGS) at East Hartford, Connecticut via their website1. It was the availability of historical water quality data dating back to at least 1985 that dictated the selection of watersheds used in this analysis. As a result, only seven regional watersheds were identified (Farmington, Housatonic, Naugatuck, Quinebaug, Quinnipiac, Salmon, and Saugatuck) and their locations are shown in Figure 8. Water quality parameters selected to be examined included those most attributed to urban conditions. These were instantaneous discharge (ft3/sec), turbidity (NTU), pH, dissolved oxygen (mg/L), fecal coliform (No. coliforms/100 ml), nitrogen (mg/L), phosphorus (mg/L), chloride (mg/L), and a cumulative total of select heavy metals (cadmium, chromium, copper, lead, and zinc) (µg/L). To decrease the effects caused by storm events, climatic differences and missing data, and to assess best the impact of impervious surfaces on water quality at a specific point in time, water quality data were averaged over the six months prior to and six months following the date of Landsat image collection and resulting impervious 1 http://ct.water.usgs.gov/

11

Marlborough

Waterford West Hartford

Woodbridge

Milford Groton

Suffield

Figure 7. Planimetric data for the seven towns used in the impervious surface validation procedure and the location of the validation areas within each (scale is not constant among the town maps displayed).

12

Table 1. Comparison of calculated percent impervious surfaces from planimetric data and IS estimate for the seven town, nine validation areas.

Town Planimetric %

Impervious Surface

Estimated % Impervious

Surface Difference

Sample Area Size

(acres)

Groton 21.68 23.65 (year 2002) -1.97 1,412

Marlborough 3.85 2.95 (year 1995) 0.90 6,667

Milford 24.60 27.39 (year 2002) -2.79 3,529

Suffield 7.46 5.94 (year 2002) 1.52 1,985

Waterford (area 1) 4.07 3.91 (year 1995) 0.16 6,697

Waterford (area 2) 8.61 9.31 (year 1995) -0.70 7,052

Woodbridge 6.74 4.27 (year 1995) 2.47 7,394

West Hartford (area 1) 34.06 37.75 (year 1990) -3.69 4,122

West Hartford (area 2) 16.51 14.99 (year 1990) 1.52 4,467

surface estimate. The number of samples averaged for each of these 12 month periods ranged from two samples to 21 samples depending on the year and gauging station. Results for the averaged water quality measurements for each watershed per year are provided in Table 2. The estimated impervious surface data were extracted for the portion of each watershed upstream from the USGS gauging station (Figure 8) for each date and summarized to determine the total amount of impervious surface within each watershed. Linear regression lines of averaged water quality data were then plotted against the estimated impervious surface percentage for each date to determine if a pattern existed between the water quality measure and amount of impervious surface in the watershed. These plots are presented in Appendix C. Also, individual plots were derived for each water quality measure versus all dates of impervious surfaces. These are available in Appendix C. The regressions were tested for significance using the analysis of regression. The results are presented in Tables 3 and 4. It should be noted that comparisons of antecdent and ambient water quality with percent impervious surface area produces only descriptive models and not predictive, causative models (i.e., relationships discovered do not explain how, or even if, impervious surfaces adversely affect water quality, but rather that there is possibly some quantifiable correspondence between the two variables).

13

Table 2. Twelve month averaged water quality measurements per watershed per year.

ed n

L) H

m / de

)

tal uL)

u t)

us

)

en )

vy ls

L)

DisslovOxyge(mg/

p Fecal Colifor(No. coliform

ml) 100

Chlori(mg/L

ToResid(mg/

e Phosphor(mg/L)

s Turbidi(NTU

y InstantaneoDischarge

(ft3/sec

Nitrog(mg/L

HeaMeta(µg/

de 0 0 16 40 00 65 76 61 25

25, 0,

90 USGS co 3 0 40 31,6 9 70,3 6 6

Sum of 101030,1041049, 10

Farmington .1 15 82 .55 55 27 14 64 75 3 10 1 7. 879. 14 84. 0. 1. 428. 0. 023.Housatonic .7 86 00 .91 10 04 28 18 73 .5 10 8 7. 200. 15 137. 0. 1. 2700. 0. 15Naugatuck .0 26 45 .73 18 59 50 45 94 05 11 5 7. 9395. 48 190. 0. 2. 318. 2. 84.Quinebaug .5 97 00 .64 64 13 80 73 81 00 10 7 6. 3070. 12 67. 0. 1. 822. 0. 16.Quinnipiac .7 62 82 .45 55 55 07 40 83 64 10 5 7. 5431. 29 193. 0. 3. 139. 0. 30.Salmon 10.8 04 0 .10 64 02 90 91 56 36 9 7. 068. 14 61. 0. 0. 118. 0. 16.

1985

Saugatuck .14 71 67 .73 27 03 11 6 75 00 11 7. 210. 15 114. 0. 1. 18.0 0. 23.Farmington .4 26 00 .59 00 09 06 91 49 23 10 9 7. 4134. 14 171. 0. 3. 1348. 0. 13.Housatonic 94 71 .00 33 03 20 10 42 64 9.94 7. 131. 14 67. 0. 2. 4402. 0. 14.Naugatuck .7 45 82 .73 82 48 30 57 53 81 10 3 7. 7465. 48 162. 0. 3. 510. 1. 54.Quinebaug .8 60 0 .46 73 06 22 10 65 95 10 9 7. 077. 14 62. 0. 3. 1270. 0. 9.Quinnipiac .6 78 73 .18 82 30 72 38 75 38 10 7 7. 682. 30 166. 0. 3. 197. 0. 19.Salmon 11.1 29 00 .60 36 01 46 09 37 09 3 7. 1924. 14 61. 0. 1. 183. 0. 9.

1990

Saugatuck .15 80 40 .43 91 02 84 9 37 48 11 7. 812. 18 107. 0. 1. 62.8 0. 17.Farmington .1 33 00 .11 6 10 26 67 48 06 10 6 7. 156. 16 79.5 0. 1. 726. 0. 9.Housatonic 78 3 .67 67 04 52 83 33 44 9.70 7. 350. 15 123. 0. 2. 4487. 0. 119.Naugatuck .4 21 75 .38 13 65 94 5 46 1 10 5 7. 915. 58 194. 0. 2. 466.2 2. 334.Quinebaug 33 88 .75 0 05 70 13 51 25 9.83 7. 300. 17 81.0 0. 1. 1089. 0. 10.Quinnipiac 61 38 .43 29 52 93 5 78 69 9.85 7. 7949. 41 205. 0. 2. 134.7 0. 30.Salmon 10.4 33 0 .22 0 01 76 11 13 33 8 7. 134.0 15 59.4 0. 0. 110. 0. 4.

1995

Saugatuck .60 84 29 .00 75 01 26 9 20 63 10 7. 154. 23 130. 0. 1. 24.4 0. 3.Farmington 16 80 .56 80 21 08 20 43 14 8.34 7. 310. 22 100. 0. 2. 462. 0. 7.Housatonic 80 00 .45 00 02 50 50 34 24 9.65 7. 20. 17 128. 0. 1. 3061. 0. 2.Naugatuck .18 57 67 .07 33 59 5 17 50 30 10 7. 823. 50 179. 0. 2. 294. 0. 32.Quinebaug 05 33 .62 33 08 13 50 55 78 8.55 7. 340. 26 95. 0. 4. 1191. 0. 4.Quinnipiac 38 38 20 .72 60 26 98 0 81 16 8. 8. 1081. 45 219. 0. 3. 86.4 0. 11.Salmon 9.70 25 00 .60 50 02 52 .50 25 83 7. 7. 16 69. 0. 2. 90 0. 1.

2002

Saugatuck 95 70 00 .65 00 02 60 .70 33 39 9. 7. 118. 21 134. 0. 1. 33 0. 4.

14

Figure 8. Location and areal extent of study watersheds in Connecticut including the location of USGS gauging stations (yellow points).

15

Table 3. Significance of linear regressions between water quality parameters and percent impervious surface by year (a * denotes significance at α = .05 and ** at α = .01).

% Impervious Surfaces

1985 1990 1995 2002

Water

Quality Measure F-ratio p > F R2 F-

ratio p > F R2 F-ratio p > F R2 F-ratio p > F R2

Chloride 3.8847 0.1058 0.434 3.9784 0.1026 0.443 6.1298 0.0561 0.551 13.0922 0.0152* 0.724

Dissolved Oxygen 0.0279 0.8739 0.006 0.0055 0.9436 0.001 0.1660 0.7005 0.032 0.5039 0.5095 0.092

Fecal Coliform 4.9072 0.0776 0.495 0.2425 0.6433 0.046 29.6409 0.0028** 0.856 39.6324 0.0015** 0.888

Nitrogen 0.6562 0.4547 0.116 1.9793 0.2185 0.284 1.5866 0.2634 0.241 14.2692 0.0129* 0.741

pH 0.1349 0.7284 0.026 0.0535 0.8262 0.011 0.0511 0.8301 0.010 5.1666 0.0722 0.508

Phosphorus 16.1370 0.0101** 0.763 7.1375 0.0443* 0.588 13.7502 0.0139* 0.733 3.0040 0.1436 0.375

Total Residue 7.8069 0.0383* 0.610 4.4395 0.0890 0.470 9.2208 0.0289* 0.648 15.1767 0.0115* 0.752

Turbidity 24.7332 0.0042** 0.832 4.8968 0.0778 0.495 4.5172 0.0869 0.475 2.0886 0.2080 0.295

Heavy Metals 1.3915 0.2912 0.218 1.1038 0.3415 0.181 0.0054 0.9444 0.001 1.8875 0.2279 0.274

Instantaneous Discharge 0.6918 0.4434 0.122 0.8237 0.4057 0.141 0.6869 0.4450 0.121 0.8491 0.3991 0.145

16

Table 4. Significance of linear regressions between water quality measure and percent impervious surface with all dates combined (green cells are significantly correlated, yellow cells are moderately correlated, and red cells are not correlated).

% Impervious Surfaces

(all dates combined)

Water Quality Measure F-ratio p > F R2

Chloride 28.4447 <0.0001 0.522

Dissolved Oxygen 0.4850 0.4923 0.018

Fecal Coliform 9.3628 0.0051 0.265

Nitrogen 4.3299 0.0474 0.143

pH 2.4297 0.1311 0.085

Phosphorus 35.3110 <0.0001 0.576

Total Residue 41.2260 <0.0001 0.613

Turbidity 17.1158 0.0003 0.397

Heavy Metals 0.9290 0.3440 0.034

Instantaneous Discharge 3.6388 0.0676 0.123

Discussion While the final results provide a consistent set of impervious surface estimates for each watershed for each of the four years, there is noticeable room for improvement, particularly in terms of per-pixel accuracy. As reported in Appendix B, the overall accuracy results varied for each year. For the years 1990, 1995, and 2002, the overall accuracy was 31.02 percent, 76.32 percent and 36.43 percent respectively. These percent accuracies increased substantially when the number of pixels in adjacent percent impervious surface coverage classes were included in the assessment (light green cells in Tables B1 – B3). The overall accuracy in this case increased to 51.70 percent, 88.29 percent, and 57.39 percent for the years 1990, 1995, and 2002 indicating that a significant number of estimated pixels are approximating the truth data, but there is still a need for improvement. The higher accuracy for the 1995 IS estimate was due to the existence of a large number of zero impervious pixels. Out of 121,140 pixels used in the assessment, 88,316 were correctly classified as containing no impervious surfaces (72.90 percent of the total). While this could be considered a minor success by not erroneously identifying impervious pixels where no imperviousness existed, the accuracy of pixels actually detecting the nine classes of imperviousness remained quite low (see user’s and producer’s accuracy for each impervious

17

class in Tables B1 – B3). The 1990 assessment used a total of 37,416 pixels with 7, 994 being correctly classified as containing no impervious surfaces (21.37 percent), and the 1995 assessment used 39,334 pixels with 10,334 being correctly classified as having no impervious surfaces (26.27 percent). Again, it is difficult to compare these results on a per pixel basis due to possible misalignment issues between the planimetric and estimate impervious results. For estimation of impervious surfaces summarized over a given area, the results closely relate to the impervious surface values derived from the planimetric truth data for the same area (Table 1). There does, however, appear to be a noticeable trend in terms of areas that contain lower levels of imperviousness versus those containing higher levels of imperviousness. For those areas having overall lower levels of impervious surfaces (i.e., more rural development densities), the IS estimate tends to under-estimate the actual impervious level with the exception of the Waterford area 2. The reverse is found for the higher impervious areas (i.e., more of an urban development density) with the IS estimate over-estimating the amount of impervious surfaces present with the exception of the West Hartford area 2. This suggests a need to revise further the Sub-pixel Classifier to try to increase the detection and amount of impervious MOIs in lower level impervious pixels and decrease the detection and amount of impervious MOIs in the higher level impervious pixels. This is most likely a function of both the Sub-pixel Classifier not being able to detect levels of MOIs below a 20 percent threshold and therefore relying heavily on the land cover to derive the 10 percent impervious class (class 1), and the overall procedures used to derive the final IS estimates, such as the use of the supervised parallelpiped classification to increase the number of 100 percent impervious pixels. In terms of correlating the level of imperviousness for each watershed and corresponding water quality parameters, having only a sample size of seven watersheds did not allow for a thorough analysis. This is compounded when five of the seven watersheds contain relatively low levels of imperviousness (ranging from 1.5 to 3.5 percent), one watershed had a moderate amount of imperviousness (approximately 7 percent), and one watershed had a higher level of imperviousness (approximately 13 percent). However, as a preliminary assessment using the available data of water quality versus impervious surfaces, some patterns do appear to exist: • The percentage of impervious surfaces does not appear to have an impact on the levels of

dissolved oxygen and pH. This is apparent in all four dates (Tables 3 and 4; Figures C-2 and C-3).

• There appears to be little pattern between heavy metal concentrations and impervious surfaces (Tables 3 and 4; Figure C-10). However, these results may be skewed due to the presence of outliers and/or the selection of heavy metals used in the computation of the water quality measure.

• Water quality parameters that appear to be impacted by the level of impervious surfaces include total residue, turbidity, chloride, phosphorus, and to some extent nitrogen and fecal coliform (Tables 3 and 4; Figures C4-C9). These findings agree with the results identified by Clausen et al. (2003) for 15 watersheds in Connecticut using 1995 IS estimates where total phosphorus, total residue, fecal coliform, dissolved chloride, and total nitrogen all exhibited moderate correlation with impervious surfaces.

• In this research, instantaneous discharge of water shows a negative correlation (Figure C-1). This is possibly attributed more to stream size than it is to impervious surfaces.

18

Conclusions For estimation of impervious surfaces summarized over a given area, the Sub-pixel Classifier is capable of producing adequate results when compared to actual impervious surfaces derived from highly accurate planimetric data. However, a limitation with the Sub-pixel Classifier is that it does not detect MOIs below a 20 percent threshold. Inclusion of land cover information for developed pixels to derive a 10 percent impervious class and also mask erroneously detected impervious surfaces (i.e., bare fields and barren land) was therefore a necessary step to generate a more complete estimate of impervious surfaces. Further work is needed in terms of improving the per-pixel accuracy of the IS estimate. Given the limited number of watersheds included in this study, some trends were identified, if only preliminarily, regarding the concentration and level of potential pollutants and impervious surfaces. To gain a better understanding of these relationships, more watersheds covering a larger range of impervious surface levels are needed. Need for Future Research While the results presented here provide a consistent set of impervious surface estimates for four dates of time over a 17-year period, there is noticeable room for improvement, particularly in terms of per-pixel accuracy. Based on this research, two areas needing improvement are identified. First, as mentioned previously, pixels of lower levels of imperviousness are generally under-estimated while pixels of higher levels of imperviousness are generally over-estimated. Continued refinement of signatures might serve to reduce this error. Additionally, focusing on the use of land cover to extract un-detected impervious pixels then performing sub-pixel analysis on these pixels should also serve to increase the level of impervious estimation of these pixels. The second area in need of improvement is to smooth the overall classification. Referring to Figures 5 and 6, it is apparent that, especially along transportation routes, there is significant variability in the level of imperviousness among neighboring pixels. One would expect a near uniform level of imperviousness along the transportation route with the level dropping off laterally. Options on how to obtain this level of consistency require further examination. Another issue is with the comparison of water quality measures with impervious surfaces. While the preliminary results indicate a pattern does exist between the level of imperviousness with certain water quality measures, not enough sample sites are represented to provide a large, statistically robust analysis. One possible scenario to improve the sample size would be to monitor water quality on a significant number of first order streams with watersheds of small area (approximately 500 acres in area) and containing a range of impervious surfaces. This would allow for a better assessment of the impact of impervious surfaces on water quality over a broader range of impervious surface levels. Additionally, specifically what area used to generate the percent impervious surface variable needs to be examined. For this project, the entire up-stream watershed from the location of the gauging station was included in the calculation of percent impervious surface using the assumption that any water quality pollutant would eventually reach the sampling location, but what is the impact of those areas of imperviousness that are far from the water quality sampling location versus those that are close (i.e., far-range versus near-range effects)? Would a more meaningful percent impervious surface variable come

19

from the examination of impervious surfaces upstream, but within a given distance from the sampling location, or within a given distance from stream channels? These are questions that need further consideration. References Arnold, C.L., H.M. Crawford, C.J. Gibbons, and R.F. Jeffrey, 1993, The Use of Geographic Information System Images as a Tool to Educate Local Officials about the Land Use/Water Quality Connection. Proceedings of Watersheds '93, Alexandria, Virginia, pp. 373-377. Arnold, C.L and C.J. Gibbons. 1996. Impervious surface coverage: the emergence of a key environmental indicator. Journal of the American Planning Association. 62(2), pp. 3243-258. Beach, D. 2002. Coastal Sprawl: The Effects of Urban Design on Aquatic Ecosystems in the United States. Pew Oceans Commission, Arlington, Virginia. Bird, S.L., S.W. Alberty, L.R. Exum. 2000. “Generating High Quality Impervious Cover Data.” Quality Assurance. 8:91-103. Bird, S., J. Harrison, L. Exum, S. Alberty and C. Perkins. 2002. "Screening to Identify and Prevent Urban Storm Water Problems: Estimating Impervious Area Accurately and Inexpensively." In Proceedings for National Water Quality Monitoring Conference, May 20-23, 2002, Madison, WI. Booth, D.B., and L.E. Reinfelt. 1993. Consequences of urbanization on aquatic systems – measured effects, degradation thresholds, and corrective strategies. Proceedings of Watershed 1993: A National Conference on Watershed Management. March 21-24, Alexandria, VA. pp. 545-550. Boyer, E.W., C.L. Goodale, N.A. Jaworssk,and R.W. Howarth. 2002. Anthropogenic nitrogen sources and relationships to riverine nitrogen export in the northeastern USA. Biogeochemistry. 57(1): 137-169. Brabec, E., S. Schulte, and P.L. Richards. 2002. Impervious surfaces and water quality; A review of current literature and its implications for watershed planning. J. of Planning Literature. 16(4): 499-514. Civco, D.L. and J.D. Hurd. 1997. Impervious surface mapping for the state of Connecticut. Proc. 1997 ASPRS/ACSM Annual Convention, Seattle, WA. 3:124-135. Civco, D.L., J.D. Hurd, E.H. Wilson, C.L. Arnold, and S. Prisloe 2002. Quantifying and Describing Urbanizing Landscapes in the Northeast United States. Photogrammetric Engineering and Remote Sensing. 68(10): 1083-1090. Clausen, J.C., G. Warner, D. Civco and M. Hood. 2003. Nonpoint Education for Municipal Officials Impervious Surface Research: Final Report. Connecticut DEP. 18p.

20

Flanagan, M. and D.L. Civco. 2001a. Subpixel impervious surface mapping. Proc. 2001 ASPRS Annual Convention, St. Louis, MO. 13 p. Flanagan, M and D.L. Civco. 2001b. Imagine Subpixel Classifier Version 8.4: Software Review. Photogrammetric Engineering and Remote Sensing. 67(1): 23-28. Hall, M.J. 1984. Urban Hydrology, Elsevier Applied Science Publishers, Northern Ireland. 299p. Herlihy, A.T., J.L. Stoddard, and C.B. Johnson. 1998. The relationship between stream chemistry and watershed land cover data in the mid Atlantic region, U.S. Water Air and Soil Pollution. 105 (1-2): 377-386. Huguenin, R.L., M.H. Wang, R. Biehl, S. Stoodley, and J. N. Rogers 2004. Automated subpixel photobathymetry and water quality mapping. Photogrammetric Engineering and Remote Sensing. 70(1): 111-123. Ji, M.H., and Jensen, J.R. 1999. “Effectiveness of Subpixel Analysis in Detecting and Quantifying Urban Imperviousness from Landsat Thematic Mapper Imagery”, Geocarto International, Vol. 14(4), pp. 31-39. Justice,D. and F. Rubin. 2003. Developing impervious surface estimates for coastal New Hampshire. A Final Report to the New Hampshire Estuaries Project. The University of New Hampshire, Durham, NH. 25 p. Prisloe, M. C.L. Arnold, D.L. Civco, and Y. Lei. 2000. A simple GIS-based model to characterize water quality in Connecticut watersheds. Proc. 4th Intl. Conf. On Integrating GIS and Environmental Modeling Problems, Alberta, Banff, Canada. 9 p. Prisloe, S., Y. Lei and J.D. Hurd. 2001.Interactive GIS-based Impervious Surface Model. Proc. 2001 ASPRS Annual Convention, St. Louis, MO. 9 p. Roy, H.A., D.A. Rosemond, J.M. Paul, D.S. Leign, and B.J. Wallace. 2003. Stream macroinvertebrate response to catchment urbanization. Freashwater Biology. 48: 329-346. Schueler, T. R., 1994. The Importance of Imperviousness. Watershed Protection Techniques, vol. 1(3): pp. 100-11. Sleavin, W., S. Prisloe, L. Giannotti, J. Stocker, D.L. Civco. 2000. Measuring Impervious Surfaces for Non-point Source Pollution Modeling. Proc. 2000 ASPRS Annual Convention, Washington, D.C. 11 p. Stocker, J. W., C. Arnold, S. Prisloe and D. Civco. 1999. Putting Geospatial Information into the Hands of the “Real” Natural Resource Managers. Proceedings of the 1999 ASPRS Annual Convention, Portland, Oregon. pp. 1070-1076.

21

U.S. Environmental Protection Agency (EPA). 1994. The Quality of Our Nation’s Water: 1992 Report. EPA-841-S-94-002. Office of Water Quality. U.S. Environmental Protection Agency. Washington D.C. 20460. U.S. Environmental Protection Agency (EPA). 1996. The National Water Quality Inventory Report to Congress, 1996. Office of Water Quality. U.S. Environmental Protection Agency. Washington, D.C., 200 p. U.S. Environmental Protection Agency (EPA). 2002. National Water Quality Inventory. 2000 Report. EPA-841-R-02-001. Office of Water. Washington D.C. 20460. Wang, Y.Q., Zhang, X. and W. Lampa. 2000. “Improvement of Spatial Accuracy in Natural resources Mapping using Multisensor Remote Sensing and Multisource Spatial Data”, in Proceedings, the 4th International Symposium on Spatial Accuracy Assessment in natural Resources and environmental Sciences, Amsterdam, Netherlands, July, 2000. pp. 723-730. Ward, D., Phinn, S.R. and A.T. Murry. 2000. Monitoring growth in rapidly urbanized areas using remotely sensed data”, Professional Geographer, Vol. 52(3), pp. 371-386. Yang, L., G. Xian, J.M. Klaver and B. Deal. 2003. Urban land-cover change detection through sub-pixel imperviousness mapping using remotely sensed data. Photogrammetric Engineering and Remote Sensing 9(9):1003-1010.

22

Appendix A

Examples of 2002 Impervious Surface Estimates and Land Cover by Watershed

23

Farmington Regional Watershed Upstream from USGS Gauging Station

2002 Impervious Surfaces 2002 Land Cover

Stat 89995 s): 286,0

ion No: 011 Area (acre 56 Developed N

ation ationImpervious

Surfaces on-forested

Veget Forested

Veget Acres % Acres % Acres % %

1985 32,167 11.24 30,858 10.79 201,289 70.37 3.11 1990 35,581 12.44 31,361 10.96 197,074 68.89 3.28 1995 36,244 12.67 31,931 11.16 196,013 68.52 3.36 2002 37,808 13.22 31,889 11.15 194,367 67.95 3.59

Figure A-1. Impervious surface and land cover information for the Farmington regional watershed.

24

Housatonic Regional Watershed Upstream from USGS Gauging Station


Station No: 01205500 Area (acres): 208,287

Developed Non-forested Vegetation

Forested Vegetation

Impervious Surfaces

Acres % Acres % Acres % % 1985 16,937 8.13 30,486 14.64 146,999 70.58 1.67 1990 18,257 8.77 30,843 14.81 144,967 69.60 1.76 1995 18,825 9.04 31,316 15.03 143,867 69.07 1.83 2002 19,895 9.55 31,180 14.97 142,889 68.60 2.02

Figure A-2. Impervious surface and land cover information for the Housatonic regional watershed

25

Naugatuck Regional Watershed

Upstream from USGS Gauging Station


Station No: 01208500 Area (acres): 167,180


Forested Vegetation

Impervious Surfaces

Acres % Acres % Acres % % 1985 30,397 18.18 22,632 13.54 104,188 62.32 6.90 1990 33,406 19.98 22,636 13.54 100,904 60.36 7.17 1995 34,148 20.43 23,106 13.82 99,649 59.61 7.37 2002 35,137 21.02 23,068 13.80 98,705 59.04 7.71

Figure A-3. Impervious surface and land cover information for the Naugatuck regional watershed.

26

Quinebaug Regional Watershed Upstream from USGS Gauging Station


Station No: 01127000 Area (acres): 237,602 Developed Non-forested

Vegetation Forested

Vegetation Impervious

Surfaces Acres % Acres % Acres % %

1985 20,898 8.80 33,745 14.20 159,135 66.98 2.11 1990 23,155 9.75 35,811 15.07 154,330 64.95 2.24 1995 23,961 10.08 36,946 15.55 151,712 63.85 2.34 2002 25,226 10.62 38,813 16.34 148,128 62.34 2.50

Figure A-4. Impervious surface and land cover information for the Quinebaug regional watershed.

27

Quinnipiac Regional Watershed Upstream from USGS Gauging Station

2002 Impervious Surfaces 2002 Land Cover Station No: 01196500 Area (acres): 72,852


Forested Vegetation

Impervious Surfaces

Acres % Acres % Acres % % 1985 20,801 28.55 15,681 21.52 31,032 42.60 12.11 1990 23,044 31.63 15,244 20.92 29,252 40.15 12.55 1995 23,517 32.28 15,373 21.10 28,662 39.34 12.91 2002 24,349 33.42 15,164 20.82 27,977 38.40 13.60 Figure A-5. Impervious surface and land cover information for the Quinnipiac regional watershed.

28

Salmon Regional Watershed Upstream from USGS Gauging Station



Vegetation Forested



1985 8,033 11.47 7,736 11.05 48,983 69.95 2.26 1990 8,704 12.43 8,208 11.72 47,882 68.37 2.39 1995 9,029 12.89 8,348 11.92 47,267 67.50 2.47 2002 9,475 13.53 8,888 12.69 46,326 66.15 2.66

Figure A-6. Impervious surface and land cover information for the Salmon regional watershed.

29

Saugatuck Regional Watershed Upstream from USGS Gauging Station



Vegetation Forested



1985 1,284 9.69 1,247 9.41 9,714 73.30 1.80 1990 1,360 10.26 1,255 9.47 9,633 72.69 1.88 1995 1,373 10.36 1,282 9.68 9,605 72.48 1.90 2002 1,413 10.66 1,280 9.66 9,593 72.39 2.03

Figure A-7. Impervious surface and land cover information for the Saugatuck regional watershed.

30

Appendix B

Per Pixel Comparison Planimetric and Estimated Percent Impervious Surfaces

31

West Hartford Validation Area 1

Planimetric Truth Data Impervious Surface Estimate

Reference IS - Estimated IS

62 68 121293

552

887

1179

1680

23442214

1963

1641

1097

583

22286 34 31 27

0

500

1000

1500

2000

2500

-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90

Difference (% impervious)over estimate under estimate

Num

ber o

f Pix

els

Figure B-1. Comparison of planimetric impervious data to the 1990 impervious surface estimate for a portion of West Hartford, Connecticut. The bar graph displays the pixel value difference between both data sets excluding pixels that contained a zero value in both dates.

32

West Hartford Validation Area 2



12 21 28 104287

464634

942

18831680

2163

1642

1083

524

17552 22 21 7

0

500

1000

1500

2000

2500

-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90


Num

ber o

f Pix

els

Figure B-2. Comparison of planimetric impervious data to the 1990 impervious surface estimate for a portion of West Hartford, Connecticut. The bar graph displays the pixel value difference between both data sets excluding pixels that contained a zero value in both dates.

33

Marlborough Validation Area



6 1 6 26 35 50 132264

11551193

1887

849

328

77 16 12 3 2 20

400

800

1200

1600

2000

2400

-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90


Num

ber o

f Pix

els

Figure B-3. Comparison of planimetric impervious data to the 1995 impervious surface estimate for a portion of Marlborough, Connecticut. The bar graph displays the pixel value difference between both data sets excluding pixels that contained a zero value in both dates.

34

Waterford Validation Area 1

Planimetric Truth Data

Impervious Surface Estimate


33 17 26 43 100187

276385

1521

614

1078

794

391

19780 33 8 7 11

0

200

400

600

800

1000

1200

1400

1600

1800

-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90


Num

ber o

f Pix

els

Figure B-4. Comparison of planimetric impervious data to the 1995 impervious surface estimate for a portion of Waterford, Connecticut. The bar graph displays the pixel value difference between both data sets excluding pixels that contained a zero value in both dates.

Waterford Validation Area 2

35



107 52 81 132 227427

630789

2409

1244

1567

1178

701

347136 56 36 24 22

0

400

800

1200

1600

2000

2400

2800

-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90


Num

ber o

f Pix

els

Figure B-5. Comparison of planimetric impervious data to the 1995 impervious surface estimate for a portion of Waterford, Connecticut. The bar graph displays the pixel value difference between both data sets excluding pixels that contained a zero value in both dates.

36

Woodbridge Validation Area



5 5 12 42 77202

343511

1935

1093

3039

2026

981

348122 31 22 11 9

0

500

1000

1500

2000

2500

3000

3500

-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90


Num

ber o

f Pix

els

Figure B-6 Comparison of planimetric impervious data to the 2002 impervious surface estimate for a portion of Woodbridge, Connecticut. The bar graph displays the pixel value difference between both data sets excluding pixels that contained a zero value in both dates.

37

Milford Validation Area



102 77 125207

364

562

760

1010

1577

13141458

1123

694

391

21592 45 37 12

0

400

800

1200

1600

2000

-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90


Num

ber o

f Pix

els

Figure B-7 Comparison of planimetric impervious data to the 2002 impervious surface estimate for a portion of Milford, Connecticut. The bar graph displays the pixel value difference between both data sets excluding pixels that contained a zero value in both dates.

38

Suffield Validation Area



2 5 16 2353

91130

198

366

206

474

350

227

8951

26 20 14 23

0

100

200

300

400

500

600

-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90


Num

ber o

f Pix

els

Figure B-8. Comparison of planimetric impervious data to the 2002 impervious surface estimate for a portion of Suffield, Connecticut. The bar graph displays the pixel value difference between both data sets excluding pixels that contained a zero value in both dates.

39

Groton Validation Area



41 30 40 58102

168230

348

750

467509

407

286

165

8536 19 15 6

0

100

200

300

400

500

600

700

800

-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90


Num

ber o

f Pix

els

Figure B-9. Comparison of planimetric impervious data to the 2002 impervious surface estimate for a portion of Groton, Connecticut. The bar graph displays the pixel value difference between both data sets excluding pixels that contained a zero value in both dates.

40

Table B-1. Error matrix for the 1990 impervious surface estimate based on data from the town of West Hartford TRUTH

CLASS 0 1 2 3 4 5 6 7 8 9 Commission User’s Accuracy

0 7,994 1,499 1,412 1,115 675 290 103 43 35 34 5,206 60.561 862 593 824 839 687 295 71 26 13 17 3,634 14.032 159 146 272 252 271 126 32 3 1 0 990 21.553 172 241 393 516 502 283 89 27 10 8 1,725 23.034 231 287 495 674 702 480 162 43 27 23 2,422 22.475 209 307 481 693 831 568 234 82 36 51 2,924 16.276 92 172 288 393 513 492 223 95 70 84 2,199 9.217 20 52 68 139 143 193 150 86 74 164 1,003 7.908 14 17 22 33 49 57 63 61 59 166 482 10.919 12 42 56 73 113 130 145 191 595 762 43.85

Omission 1,771 2,763 4,039 4,211 3,671 2,329 1,034 525 457 547

EST

IMA

TE

Producer’s Accuracy 81.86 17.67 6.31 10.92 16.05 19.61 17.74 14.08 11.43 52.10 Overall

31.02 n = 37,416 pixels Correctly classified = 11,608 pixels (green cells) Overall accuracy = 31.02% (green cells) Overall accuracy combining {-10%, 0%, and +10} ∆IS classes = 51.70% (green and light green cells)

41

Table B-2. Error matrix for the 1995 impervious surface estimate based on data from the towns of Marlborough, Waterford, and Woodbridge TRUTH


0 88,316 4,506 2,890 1,610 654 224 85 49 23 44 10,085 89.751 5,299 2,430 2,105 1,449 585 217 100 35 16 21 9,827 19.832 567 327 371 296 175 57 20 6 4 4 1,456 20.313 650 484 499 486 315 131 55 25 8 8 2,175 18.264 540 348 425 435 339 165 82 23 21 16 2,055 14.165 248 134 170 239 216 129 68 46 23 32 1,176 9.896 138 54 73 81 93 83 69 45 34 48 649 9.617 56 19 21 37 40 33 45 29 28 40 348 8.338 32 7 10 17 12 23 18 15 23 43 177 11.509 151 43 62 76 99 70 69 90 255 268 915 22.65

Omission 7,681 5,922 6,255 4,240 2,189 1,003 542 334 412 256

EST

IMA

TE



42

Table B-3. Error matrix for the 2002 impervious surface estimate based on data from the towns of Groton, Suffield, and Milford. TRUTH


0 10,334 1,207 1,108 910 546 289 129 56 43 45 4,333 70.461 1,378 504 608 702 570 367 140 52 37 40 3,894 11.462 389 162 248 246 265 173 44 13 8 5 1,305 15.973 460 257 414 444 461 337 116 41 18 24 2,128 17.264 515 340 549 659 708 555 246 65 54 61 3,044 18.875 450 315 439 640 770 627 306 131 78 86 3,215 16.326 266 215 285 394 543 571 323 169 115 163 2,721 10.617 113 90 101 183 192 267 242 162 144 263 1,595 9.228 57 34 40 48 73 80 105 121 133 232 790 14.419 205 118 139 162 204 246 264 288 354 846 1,980 29.94

Omission 3,833 2,738 3,683 3,944 3,624 2,885 1,592 936 851 919

EST

IMA

TE



43

Appendix C

Linear Regression of Water Quality Variables

and Impervious Surface

44

Instantaneous Discharge 1985

y = -84.252x + 1010.1R2 = 0.1215

0.00

1000.00

2000.00

3000.00

4000.00

5000.00

0.00 5.00 10.00 15.00

% Imperviousness

Dis

char

ge (f

t3 /s

ec)


y = -142.95x + 1777.7R2 = 0.1414

0.00

1000.00

2000.00

3000.00

4000.00

5000.00

0.00 5.00 10.00 15.00

% Imperviousness

Dis

char

ge (f

t3/s

ec)


y = -133.07x + 1617.4R2 = 0.1211

0.00

1000.00

2000.00

3000.00

4000.00

5000.00

0.00 5.00 10.00 15.00

% Imperviousness

Dis

char

ge (f

t3/s

ec)


y = -96.52x + 1215.9R2 = 0.1453

0.00

1000.00

2000.00

3000.00

4000.00

5000.00

0.00 5.00 10.00 15.00

% Imperviousness

Dis

char

ge (f

t3/s

ec)

Figure C-1. Instantaneous discharge of water as a function of impervious surfaces over four dates.

45

Dissolved Oxygen 1985

y = 0.0064x + 10.728R2 = 0.0053

8.00

8.75

9.50

10.25

11.00

11.75

0.00 5.00 10.00 15.00

% Imperviousness

Dis

solv

ed O

xyge

n (m

g/L)


y = -0.0032x + 10.727R2 = 0.001

8.00

8.75

9.50

10.25

11.00

11.75

0.00 5.00 10.00 15.00

% Imperviousness

Dis

solv

ed O

xyge

n (m

g/L)


y = -0.0155x + 10.223R2 = 0.031

8.00

8.75

9.50

10.25

11.00

11.75

0.00 5.00 10.00 15.00

% Imperviousness

Dis

solv

ed O

xyge

n (m

g/L)


y = -0.0555x + 9.5208R2 = 0.0912

8.00

8.75

9.50

10.25

11.00

11.75

0.00 5.00 10.00 15.00

% Imperviousness

Dis

solv

ed O

xyge

n (m

g/L)

Figure C-2. Dissolved oxygen concentration as a function of impervious surfaces over four dates.

46

Field pH 1985

y = 0.0146x + 7.3102R2 = 0.026

7.00

7.25

7.50

7.75

8.00

8.25

8.50

0.00 5.00 10.00 15.00

% Imperviousness

Fiel

d pH

Field pH 1990

y = 0.0067x + 7.561R2 = 0.0103

7.00

7.25

7.50

7.75

8.00

8.25

8.50

0.00 5.00 10.00 15.00

% Imperviousness

Fiel

d pH

Field pH 1995

y = -0.0059x + 7.5181R2 = 0.0095

7.00

7.25

7.50

7.75

8.00

8.25

8.50

0.00 5.00 10.00 15.00

% Imperviousness

Fiel

d pH

Field pH 2002

y = 0.0755x + 7.1904R2 = 0.5067

7.00

7.25

7.50

7.75

8.00

8.25

8.50

0.00 5.00 10.00 15.00

% Imperviousness

Fiel

d pH

Figure C-3. Field pH as a function of impervious surfaces over four dates.

47

Fecal Coliform 1985

y = 638.14x + 19.403R2 = 0.4951

0.00

1000.00

2000.00

3000.00

4000.00

5000.00

6000.00

7000.00

8000.00

9000.00

10000.00

0.00 5.00 10.00 15.00

% Imperviousness

Feca

l Col

iform

(No.

/100

ml)

Fecal Coliform 1990

y = 145.82x + 1524.2R2 = 0.0464

0.00

1000.00

2000.00

3000.00

4000.00

5000.00

6000.00

7000.00

8000.00

9000.00

10000.00

0.00 5.00 10.00 15.00

% Imperviousness

Feca

l Col

iform

(No.

/100

ml)

Fecal Coliform 1995

y = 650.5x - 1610.8R2 = 0.8553

0.00

1000.00

2000.00

3000.00

4000.00

5000.00

6000.00

7000.00

8000.00

9000.00

10000.00

0.00 5.00 10.00 15.00

% Imperviousness

Feca

l Col

iform

(No.

/100

ml)

Fecal Coliform 2002

y = 90.23x - 53.698R2 = 0.888

0.00

1000.00

2000.00

3000.00

4000.00

5000.00

6000.00

7000.00

8000.00

9000.00

10000.00

0.00 5.00 10.00 15.00

% Imperviousness

Feca

l Col

iform

(No.

/100

ml)

Figure C-4. Fecal coliform bacteria concentrations as a function of impervious surfaces over four dates.

48

Chloride 1985

y = 2.2428x + 11.986R2 = 0.4371

0.00

10.00

20.00

30.00

40.00

50.00

60.00

0.00 5.00 10.00 15.00

% Imperviousness

Chl

orid

e (m

g/L)

Chloride 1990

y = 2.1612x + 12.491R2 = 0.4438

0.00

10.00

20.00

30.00

40.00

50.00

60.00

0.00 5.00 10.00 15.00

% Imperviousness

Chl

orid

e (m

g/L)

Chloride 1995

y = 3.0011x + 12.994R2 = 0.5512

0.00

10.00

20.00

30.00

40.00

50.00

60.00

0.00 5.00 10.00 15.00

% Imperviousness

Chl

orid

e (m

g/L)

Chloride 2002

y = 2.6725x + 15.647R2 = 0.7236

0.00

10.00

20.00

30.00

40.00

50.00

60.00

0.00 5.00 10.00 15.00

% Imperviousness

Chl

orid

e (m

g/L)

Figure C-5. Chloride concentrations as a function of impervious surfaces over four dates.

49

Total Residue 1985

y = 10.977x + 74.291R2 = 0.6094

50.00

100.00

150.00

200.00

250.00

0.00 5.00 10.00 15.00

% Imperviousness

Tota

l Res

idue

(mg/

L)

Total Residue 1990

y = 8.8037x + 74.968R2 = 0.4704

50.00

100.00

150.00

200.00

250.00

0.00 5.00 10.00 15.00

% Imperviousness

Tota

l Res

idue

(mg/

L)

Total Residue 1995

y = 11.108x + 73.754R2 = 0.6483

50.00

100.00

150.00

200.00

250.00

0.00 5.00 10.00 15.00

% Imperviousness

Tota

l Res

idue

(mg/

L)

Total Residue 2002

y = 10.388x + 81.761R2 = 0.7518

50.00

100.00

150.00

200.00

250.00

0.00 5.00 10.00 15.00

% Imperviousness

Tota

l Res

idue

(mg/

L)

Figure C-6. Total residue concentrations as a function of impervious surfaces over four dates.

50

Phosphorus 1985

y = 0.0557x - 0.0049R2 = 0.7626

0.00

0.25

0.50

0.75

1.00

0.00 5.00 10.00 15.00

% Imperviousness

Phos

phor

us (m

g/L)

Phosphorus 1990

y = 0.0344x - 0.0115R2 = 0.6005

0.00

0.25

0.50

0.75

1.00

0.00 5.00 10.00 15.00

% Imperviousness

Phos

phor

us (m

g/L)

Phosphorus 1995

y = 0.0561x - 0.0601R2 = 0.7362

0.00

0.25

0.50

0.75

1.00

0.00 5.00 10.00 15.00

% Imperviousness

Phos

phor

us (m

g/L)

Phosphorus 2002

y = 0.0297x + 0.0284R2 = 0.3781

0.00

0.25

0.50

0.75

1.00

0.00 5.00 10.00 15.00

% Imperviousness

Phos

phor

us (m

g/L)

Figure C-7. Phosphorus concentrations as a function of impervious surfaces over four dates. 51

Turbidity 1985

y = 0.1905x + 0.8697R2 = 0.8311

0

2

4

6

8

10

0.00 5.00 10.00 15.00

% Imperviousness

Turb

idity

(NTU

)

Turbidity 1990

y = 0.1488x + 2.023R2 = 0.4952

0.00

2.00

4.00

6.00

8.00

10.00

0.00 5.00 10.00 15.00

% Imperviousness

Turb

idity

(NTU

)

Turbidity 1995

y = 0.147x + 1.234R2 = 0.4742

0.00

2.00

4.00

6.00

8.00

10.00

0.00 5.00 10.00 15.00

% Imperviousness

Turb

idity

(NTU

)

Turbidity 2002

y = 0.1328x + 1.9692R2 = 0.2941

0

2

4

6

8

10

0.00 5.00 10.00 15.00

% Imperviousness

Turb

idity

(NTU

)

Figure C-8. Fecal Turbidity levels as a function of impervious surfaces over four dates. 52

Nitrogen 1985

y = 0.0729x + 0.7388R2 = 0.1156

0.00

0.50

1.00

1.50

2.00

2.50

3.00

0.00 5.00 10.00 15.00

% Imperviousness

Nitr

ogen

(mg/

L)

Nitrogen 1990

y = 0.0544x + 0.4133R2 = 0.2846

0.00

0.50

1.00

1.50

2.00

2.50

3.00

0.00 5.00 10.00 15.00

% Imperviousness

Nitr

ogen

(mg/

L)

Nitrogen 1995

y = 0.0952x + 0.2617R2 = 0.2385

0.00

0.50

1.00

1.50

2.00

2.50

3.00

0.00 5.00 10.00 15.00

% Imperviousness

Nitr

ogen

(mg/

L)

Nitrogen 2002

y = 0.0374x + 0.2753R2 = 0.7515

0.00

0.50

1.00

1.50

2.00

2.50

3.00

0.00 5.00 10.00 15.00

% Imperviousness

Nitr

ogen

(mg/

L)

Figure C-9. Nitrogen concentrations as a function of impervious surfaces over four dates. 53

Heavy Metals 1985

y = 2.9326x + 17.245R2 = 0.2175

0.00

25.00

50.00

75.00

100.00

125.00

0.00 5.00 10.00 15.00

% Imperviousness

Hea

vy M

eals

(um

/L)

Heavy Metals 1990

y = 1.6788x + 12.3R2 = 0.1813

0.00

25.00

50.00

75.00

100.00

125.00

0.00 5.00 10.00 15.00

% Imperviousness

Hea

vy M

eals

(um

/L)

Heavy Metals 1995

y = -0.3317x + 31.769R2 = 0.0011

0.00

25.00

50.00

75.00

100.00

125.00

0.00 5.00 10.00 15.00

% Imperviousness

Hea

vy M

eals

(um

/L)

Heavy Metals 2002

y = 1.2933x + 2.8194R2 = 0.2741

0.00

25.00

50.00

75.00

100.00

125.00

0.00 5.00 10.00 15.00

% Imperviousness

Hea

vy M

eals

(um

/L)

Figure C-10. Cumulative heavy metal (cadmium, chromium, copper, lead, and zinc) concentrations as a function of impervious surfaces over four dates. 54

Instantaneous DischargeAll Dates (1985, 1990, 1995, & 2002)

y = -113.81x + 1403.3R2 = 0.1229

0.00

1000.00

2000.00

3000.00

4000.00

5000.00

0.00 5.00 10.00 15.00

% Imperviousness

Dis

char

ge (f

t3/s

ec)

Dissolved OxygenAll Dates (1985, 1990, 1995, & 2002)

y = -0.0273x + 10.342R2 = 0.0182

8.00

8.75

9.50

10.25

11.00

11.75

0.00 5.00 10.00 15.00

% Imperviousness

Dis

solv

ed O

xyge

n (M

g/L)

Field pHAll Dates (1985, 1990, 1995, & 2002)

y = 0.0251x + 7.3887R2 = 0.0852

7.00

7.25

7.50

7.75

8.00

8.25

8.50

0.00 5.00 10.00 15.00

% Imperviousness

Fiel

d pH

Fecal ColiformAll Dates (1985, 1990, 1995, & 2002)

y = 355.69x + 53.291R2 = 0.2648

0.00

1000.002000.00

3000.004000.00

5000.00

6000.007000.00

8000.009000.00

10000.00

0.00 5.00 10.00 15.00

% Imperviousness

Feca

l Col

iform

(No.

/100

ml)

55

ChlorideAll Dates (1985, 1990, 1995, & 2002)

y = 2.5697x + 13.095R2 = 0.5227

0.00

10.00

20.00

30.00

40.00

50.00

60.00

0.00 5.00 10.00 15.00

% Imperviousness

Chl

orid

e (m

g/L)

Total ResidueAll Dates (1985, 1990, 1995, & 2002)

y = 10.364x + 75.993R2 = 0.6131

50.00

100.00

150.00

200.00

250.00

0.00 5.00 10.00 15.00

% Imperviousness

Tota

l Res

idue

(mg/

L)

Phosphorus

All Dates (1985, 1990, 1995, & 2002)

y = 0.0431x - 0.0095R2 = 0.5798

0.00

0.25

0.50

0.75

1.00

0.00 5.00 10.00 15.00

% Imperviousness

Phos

phor

us (m

g/L)

TurbidityAll Dates (1985, 1990, 1995, & 2002)

y = 0.1566x + 1.5117R2 = 0.3962

0

2

4

6

8

10

0.00 5.00 10.00 15.00

% Imperviousness

Turb

idity

(NTU

)

56

NitrogenAll Dates (1985, 1990, 1995, & 2002)

y = 0.0611x + 0.4378R2 = 0.1427

0.00

0.50

1.00

1.50

2.00

2.50

3.00

0.00 5.00 10.00 15.00

% Imperviousness

Nitr

ogen

(mg/

L)

Heavy MetalsAll Dates (1985, 1990, 1995, & 2002)

y = 1.241x + 16.588R2 = 0.0345

0.00

25.00

50.00

75.00

100.00

125.00

0.00 5.00 10.00 15.00

% Imperviousness

Hea

vy M

eals

(um

/L)

Figure C-11. Linear regression plots for each of the 10 water quality measures as a function of impervious surfaces for all four dates combined.

57

Documents

Surface Water Quality and · 2019-01-15 · 1. Develop improved and consistent impervious surface estimates using ERDAS Imagine’s Subpixel Classifier® for four separate dates of