Presentation is loading. Please wait.

Presentation is loading. Please wait.

Temporal Characterization of Impervious Surfaces for the State of Connecticut James D. Hurd & Daniel L. Civco Center for Land use Education And Research.

Similar presentations


Presentation on theme: "Temporal Characterization of Impervious Surfaces for the State of Connecticut James D. Hurd & Daniel L. Civco Center for Land use Education And Research."— Presentation transcript:

1 Temporal Characterization of Impervious Surfaces for the State of Connecticut James D. Hurd & Daniel L. Civco Center for Land use Education And Research (CLEAR) Department of Natural Resources Management & Engineering The University of Connecticut U-4087, Room 308, 1376 Storrs Road Storrs, CT 06269-4087 May 26, 2004

2 Center for Land use Education And Research (CLEAR) A collaboration of the Land, Sea and Space Grant programs at the University of Connecticut NEMO LERIS GVI Forestry National NEMO Network GTP To provide information, education and assistance to land use decision makers, in support of balancing growth and natural resource protection. MISSION:

3 CLEAR Publications Tech tools Web Presentations Demonstration Projects

4 CLEAR http://CLEAR.uconn.edu

5 CLEAR Maps and Statistics on: - Land Cover - Land Cover Change - Forest Fragmentation - Urban Growth - Impervious Surfaces

6 Impervious Surfaces Materials like: - cement, - asphalt, - roofs, that prevent percolation of runoff into the ground.

7 Impervious Surfaces Why should we care? Local impacts to water quality

8 Impervious Surfaces Adapted from Schueler, et al., 1994 8070605040302010 0 WATERSHED IMPERVIOUSNESS (%) DEGRADED PROTECTED IMPACTED WATERSHED WATER QUALITY Water Quality Impacts 25

9 Impervious Surfaces Urbanization Adapted from Hall, 1984 Population Density Increases Building Density Increases Urban Climate Changes Waterborne Waste Increases Water Demand Rises Water Resource Problems Base Flow Reduces Groundwater Recharge Reduces Stormwater Quality Deteriorates Receiving Water Quality Deteriorates Pollution Control Problems Peak Runoff Rate Increases Flood Control Problems Runoff Volume Increases Drainage System Modified Lag Time and Time Based Reduced Flow Velocity Increases Impervious Area Increases

10 Quantifying Imperviousness Previous examples of impervious surface estimation in Connecticut: Impervious Coefficients estimates are calculated by multiplying a land use specific percent impervious coefficient by the total area of that land use within an area. Results in an estimate, does not identify specific locations if impervious surfaces. Does not account for small scale variance within a particular land use. Direct Estimation from Remote Sensing Imagery Neural Networks Sub-pixel Classifier

11 Technical Session 97: Wed., May 26 th 4:30 - 5:30 Room: Governor’s Square 14.

12 Quantifying Imperviousness Software: ERDAS IMAGINE’s Subpixel Classifier ® Add-on module developed by Applied Analysis Inc. Able to detect features smaller than the spatial resolution of the sensor. Provides the percentage of the material within each pixel. Output: raster image overlay depicting per pixel IS occurrences greater than 20%.

13 Pixel Reflectance Values 100% Impervious Pixel Value B1 - 141 B2 - 70 B3 - 255 B4 - 155 B5 - 210 B7 - 80 50% Impervious Pixel Value B1 - 102 B2 - 45 B3 - 161 B4 - 169 B5 - 171 B7 - 47 50% Impervious Pixel Value B1 - 113 B2 - 47 B3 - 174 B4 - 138 B5 - 131 B7 - 40 0% Impervious Pixel Value B1 - 98 B2 - 42 B3 - 165 B4 - 148 B5 - 206 B 7- 53 Landsat ETM 4,3,2 ADAR

14 Sub-pixel Classifier Procedures Automated process. - performed prior to signature development and classification. Image Preprocessing Environmental Correction Process that compensates for unwanted spectral variations (i.e. haze and clouds). - In scene - Scene to scene

15 Signature Selection Material of Interest (MOI) is impervious surfaces. Bradley Int’l. Airport Windsor Locks, CT April 26, 1985

16 Impervious Pixel Spectral Variability bright medium dark very dark Principal Components 1

17 Signature Selection Bradley Int’l. Airport Windsor Locks, CT April 26, 1985 Select pixels that Represent 100% IS: -Very dark pixels -Dark pixels -Medium pixels -Bright pixels

18 Signature Derivation

19 Signature Combiner

20 MOI Classification

21 Initial Subpixel Classification 20% - 29% 30% - 39% 40% - 49% 50% - 59% 60% - 69% 70% - 79% 80% - 89% 90% - 100% Manchester, CT August 30, 1995

22 Signature Refinement

23 Manchester, CT August 30, 1995 20% - 29% 30% - 39% 40% - 49% 50% - 59% 60% - 69% 70% - 79% 80% - 89% 90% - 100%

24 Supervised Classification Manchester, CT August 30, 1995 20% - 29% 30% - 39% 40% - 49% 50% - 59% 60% - 69% 70% - 79% 80% - 89% 90% - 100% Create signatures that represent 100% IS. Used Parallelpiped decision rule to identify only those pixels within upper and lower limits.

25 3 x 3 Majority Filter 20% - 29% 30% - 39% 40% - 49% 50% - 59% 60% - 69% 70% - 79% 80% - 89% 90% - 100% Manchester, CT August 30, 1995 Used 3x3 majority filter. Applied to class values 0, 5, 6,7, 8 only.

26 Inclusion of Land Cover Manchester, CT August 30, 1995 10% - 19% 20% - 29% 30% - 39% 40% - 49% 50% - 59% 60% - 69% 70% - 79% 80% - 89% 90% - 100% Extract pixels classified as developed from land cover. Mask the sub-pixel classification to identify only pixels from the developed category. Pixels classified as developed but not identified as containing IS make 10% class.

27 Classification of 10% IS Pixels Extract pixels identified as 10% IS. Perform sub-pixel Analysis on extracted pixels. Overlay on previous sub-pixel classification. Manchester, CT August 30, 1995 10% - 19% 20% - 29% 30% - 39% 40% - 49% 50% - 59% 60% - 69% 70% - 79% 80% - 89% 90% - 100%

28 Initial 1995 IS Estimate Manchester, CT August 30, 1995 10% - 19% 20% - 29% 30% - 39% 40% - 49% 50% - 59% 60% - 69% 70% - 79% 80% - 89% 90% - 100%

29 Subsequent Dates (1985, 1990, 2002) Perform same analysis operations on 1985, 1990, and 2002 Landsat images. 19851990 2002 10% - 19% 20% - 29% 30% - 39% 40% - 49% 50% - 59% 60% - 69% 70% - 79% 80% - 89% 90% - 100%

30 Deriving Final IS estimates (1985, 1990, 1995, 2002) A fundamental principal in developing a temporal IS dataset was the need for consistency between dates. Needed to eliminate the possibility of fluctuation in IS% between dates. Needed to eliminate the decrease in IS% over time. Combine IS results from adjacent dates by using the difference between dates.

31 Deriving Final IS estimates (1985, 1990, 1995, 2002) For Example to generate the final 1990 IS estimate: There are nine IS classes 1 = 10% - 19% 2 = 20% - 29% 3 = 30% - 39% 4 = 40% - 49% 5 = 50% - 59% 6 = 60% - 69% 7 = 70% - 79% 8 = 80% - 89% 9 = 90% - 100% Take the difference 1995 IS – 1990 IS (add constant of 10 to maintain positive values) Pixel 2: (5 – 1) + 10 = 14 Pixel 3: (5 – 4) + 10 = 11 If the difference value was 11, then the 1990 IS value was maintained, otherwise the 1995 IS value was used. If the 1990 IS value was 0 then the 0 value was maintained. 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. Same procedure used between 1990 and 1985. Reverse procedure used between 1995 and 2002. Pixel 1: (5 – 8) + 10 = 7 19951990

32 Deriving Final IS estimates (1985, 1990, 1995, 2002) Examination of results with validation data still resulted in under-estimate of IS. Needed to increase the overall IS levels. Beginning with the 1985 initial IS estimate and adjusted 1985 IS estimate, if IS values per pixel were higher in the initial 1985 estimate, then maintained higher values. Applied this 1985 IS estimate to the 1990 adjusted estimate. Continued for 1995 and 2002.

33 Final IS Estimates (1985, 1990, 1995, 2002) 10% - 19% 20% - 29% 30% - 39% 40% - 49% 50% - 59% 60% - 69% 70% - 79% 80% - 89% 90% - 100% Manchester, CT 1985 Manchester, CT 1990 Manchester, CT 1995 Manchester, CT 2002

34 Validation Areas 1:140,000 West HartfordGroton Marlborough

35 Planimetric Data West Hartford, CT Generate 30-meter grid that corresponds to Landsat pixel grid. Calculate actual IS.

36 Validation Results West Hartford, CT circa 1990 Planimetric Derived ISLandsat Derived IS Excluding 0 value pixels in both dates.

37 Validation Results West Hartford, CT circa 1990 Planimetric Derived ISLandsat Derived IS Excluding 0 value pixels in both dates.

38 Validation Results Marlborough, CT circa 1995 Planimetric Derived IS Landsat Derived IS Excluding 0 value pixels in both dates.

39 Validation Results Waterford, CT circa 1995 Planimetric Derived IS Landsat Derived IS Excluding 0 value pixels in both dates.

40 Validation Results Waterford, CT circa 1995 Planimetric Derived IS Landsat Derived IS Excluding 0 value pixels in both dates.

41 Validation Results Woodbridge, CT circa 1995 Planimetric Derived ISLandsat Derived IS Excluding 0 value pixels in both dates.

42 Validation Results Milford, CT circa 2002 Planimetric Derived IS Landsat Derived IS Excluding 0 value pixels in both dates.

43 Validation Results Suffield, CT circa 2002 Planimetric Derived IS Landsat Derived IS Excluding 0 value pixels in both dates.

44 Validation Results Groton, CT circa 2002 Planimetric Derived IS Landsat Derived IS Excluding 0 value pixels in both dates.

45 Validation Results Misregistration 100% 0% 50% error 50% error 30 m ½ pixel shift 100% 0% 75% error ½ pixel x ½ pixel shift 100% 0% 100% error 100% error 1 pixel shift can significantly affect classification accuracy even at small scales.

46 Validation Results Town Planimetric % Impervious Surface Estimated % Impervious Surface Difference Sample Area (acres) Groton21.68 23.65 (year 2002) -1.971,412 Marlborough3.85 2.95 (year 1995) 0.906,667 Milford24.60 27.39 (year 2002) -2.793,529 Suffield7.46 5.94 (year 2002) 1.521,985 Waterford (area 1) 4.07 3.91 (year 1995) 0.166,697 Waterford (area 2) 8.61 9.31 (year 1995) -0.707,052 Woodbridge6.74 4.27 (year 1995) 2.477,394 West Hartford (area 1) 34.06 37.75 (year 1990) -3.694,122 West Hartford (area 2) 16.51 14.99 (year 1990) 1.524,467

47 Conclusions The ERDAS Imagine Sub-pixel Classifier was designed to identify sub-pixel contributions of MOIs, therefore is suitable for deriving this type of product. Care needs to be taken in generating appropriate signatures due to the complex spectral variability of impervious surfaces. –Identify signature pixels, classify, refine, classify, refine, etc….

48 Conclusions Areal results of IS estimates compare favorably with actual IS. Per-pixel accuracy is less then desirable with: –1990 overall accuracy 31%, n = 37,416 52% if include +/- 1 class increment –1995 overall accuracy 76%, n = 121,140 88% if include +/- 1 class increment –2002 overall accuracy 36%, n = 39,334 57% if include +/- 1 class increment There is still work to be done…

49 Future Research Want to improve per-pixel accuracy. Want to create a more uniform IS estimate. Westbound I-84 3 lanes Eastbound I-84 3 lanes Access road Rest area Decrease variability along interstate

50 Future Research Focus sub-pixel analysis on groupings of pixels from previous IS estimate. Extract IS class values 7, 8, 9Extract IS class values 4, 5, 6Extract IS class values 1, 2, 3

51 Expanded Area Phase 1, Greater Connecticut Phase 2, Long Island & Westchester County Land Cover Land Cover Change Forest Fragmentation Urban Growth Impervious Surfaces

52 Acknowledgements Funding from the NOAA Coastal Services Center (CSC). Phase 1. Funding from the EPA Office of Long Island Sound Programs. Phase 2.

53 Temporal Characterization of Impervious Surfaces for the State of Connecticut James D. Hurd James.hurd_jr@uconn.edu Center for Land use Education And Research (CLEAR) Department of Natural Resources Management & Engineering The University of Connecticut U-4087, Room 308, 1376 Storrs Road Storrs, CT 06269-4087 May 26, 2004


Download ppt "Temporal Characterization of Impervious Surfaces for the State of Connecticut James D. Hurd & Daniel L. Civco Center for Land use Education And Research."

Similar presentations


Ads by Google