THE IMPACTS OF URBANIZATION ON THE SURFACE ALBEDO IN THE YANGTZE RIVER DELTA IN CHINA 08/24/2011 Mélanie Bourré
Motivation Since the 20th century, rapid urbanization of the world population. United Nation prediction (2006) : 60% of the world population will live in cities by 2030. In these newly urbanized areas, local climate change : «Heat Island Effect». Although it affects many people, the relationship between urbanization and local climate change is not well understood. However, the role of surface properties on climate has been recognized by many recent studies. Urbanization Surface albedo change Local climate change
Methodology Remote sensing data. Processed and analyzed by ENVY software. Selection of two Landsat TM images One from the late 80s, one from nowadays Urbanization study - Classification of the two images - Post classification comparison Albedo calculation - Atmospheric correction - Conversion of DN to TOA reflectance - Albedo calculation Liang s formula Analysis Albedo change with urbanization
Study area Mouth of the Yangtze River Delta in China Coordinates : Long : 120 39 E Lat : 32 40 N Path : 118 Row : 38 Area : ~25,000 km² www.landsat.org Acquisition date : 08/11/1989 Yangtze River Delta : - One of the most industrialized and urbanized region of China. - Highest population density of China.
Environment Human activities Geology Alluvial plain Elevation : 4m Hydrology Delta Numerous rivers and lakes Maze of intersecting canals Climate Humid subtropical climate Vegetation Subtropical broad-leaf evergreen Primary sector Agriculture Aquaculture Secondary sector Traditional center of textile industry Industrial base advancing new technology Import/export Tertiary sector Commerce and finance Transportation
Satellite Images Source : http://glcf.umiacs.umd.edu/data/ Source : www.landsat.gsfc.nasa.gov/images/media.html Author NASA Landsat Program NASA Landsat program Publication Date 21/10/2008 22/02/2008 Collection Name Landsat 5 TM scene Landsat 7 ETM+ scene SLC-on Image Name ID 201-985 ID 210-284 Processing Level Ortho, GLS 1990 Ortho, GLS2000 Publisher USGS USGS Publisher Location Sioux Falls Sioux Falls Product Coverage Date 08/11/1989 07/03/2001
Image preprocessing Pictures cut To get pictures of same dimension Geometric and radiometric correction Orthorectification and radiometric correction by USGS Atmospheric correction No visible haze or clouds Not necessary for the classification (Song et al. 2001) Image enhancement Automatic linear contrast stretching of 2% by ENVY Software
Classification Land cover mapping - Unsupervised classification - Selection of classes Urbanization study -Post classification comparison - Supervised classification - Accuracy assesment
Unsupervised classification Aim : Evaluate the separability between classes and so guide the supervised classification Principle: - Software groups together pixels with similar spectral pattern ISODATA algorithm - Analyst identifies the classes Chosen parameters : Number of classes : Between 5 and 10 Number of iterations : 15 Change Threshold : 5% Mask applied on water to study only the spectral classes of the land
Results Spectral classes identified : - Water -Agricultural Land - Forest Land - Urban Land : Bright and darker pixels - Intertidal far from the coast Remark Spectral classes identified : most of the pixels having this spectral signature belong to this land cover type but some don t Land cover type with similar spectral pattern - Range Land with Agricultural Land - Intertidal close to the coast/barren land with Urban Land
Class selection USGS «LU/ LC Classification System for Use with Remotely Sensed Data» Level I of Classification Applied to the image : Urban Land, Agricultural Land, Range Land,Forest Land, Water and Barren Land A classification system must be exhaustive -> Add Intertidal Land From the unsupervised classification : 2 classes for urban: Urban Land LA and Urban Land HA From the spectral patterns : No class for Barren Land Spectral separability between Barren Land and Highways Jeffries-Matusita : 1.07993221<1.9 Transformed Divergence : 1.35410740<1.9
Seven classes Class name Urban Land HA Urban Land LA Agricultural Land Range Land Forest Land Water Intertidal Definition Blue and white roofs; Bright quarries Red, brown and grey roofs; Parkings; Highways; Railways; Dark quarries Dry farmland; Paddy fields; Pastures Grass Trees Ocean; Rivers; Lakes; Reservoirs; Aquaculture pounds In between the land and the sea Caution Intertidal and Range Land spectrally similar to other classes
Supervised classification Principle : - Analyst defines the regions of interests (ROI) for each class - Software labels each pixel with the ROI it is closest to Maximum Likelihood Classifier Hypothesis : the distribution of points forming one class is normal - Class combination to form the classes Parameter : Threshold : 5%
Post classification Preliminary accuracy assesment Misclassification between - Urban LA and Urban HA - Range Land and Agricultural Land Combination of the classes in - Urban Land - Agricultural & Range Land Fuzzy picture due to mixed pixels Smoothing Majority minority analysis - Kernel size : 5x5 - Central pixel weight : 1
08/11/1989 Supervised classified image Legend Urban Agricultural & Range Land Forest Land Water Intertidal
07/03/2001 Supervised classified image Legend Urban Agricultural & Range Land Forest Land Water Intertidal
Accuracy assesment For the 2001 image Ground truth data : Google Earth Location : Subset of Shanghai Dates : 11/21/2000 and 07/23/2002 Hypothesis : If a pixel has the same land cover in 2000 and 2002 Same land cover in 2001 Source : Google Earth 11/21/2000 At least 50 GCPs per class (Congalton 1991)
Results Overall accuracy = Number of correctly classified pixels / total number of reference pixels Overall accuracy = 94.35% Kappa coefficient Represents the probable better accuracy of the employed maximum likelihood classification than if the classification resulted from a random assignment instead. Kappa coefficient = 0.87
Class Urban Agri & Range Forest Water Intertidal Producer Acc. 99.48 67.53 80 97.92 82.50 User Acc. 96.62 83.87 82.05 100 100 User accuracy =Nb of correctly classified pixel in the category / total nb of pixels classified in that category Indicates the probability that a pixel classified into a given category actually represents that category on the ground. High for the urban class. Only 80% for the vegetation classes.
Confusion matrix Class Urban Agri & Range Forest Water Intertidal Unclassified 0 9.09 0 0 0 Urban 99.48 14.29 2.5 2.08 17.50 Agri & Range 0.52 67.53 17.5 0 0 Forest 0 9.09 80 0 0 Water 0 0 0 97.92 0 Intertidal 0 0 0 0 82.50 Total 100 100 100 100 100 Most of the pixels wrongly classified as agricultural are forest, And vice versa.
Class statistics The Water, Intertidal and Unclassified pixels in 1989 or in 2001 are masked. 9000 8500 8000 7500 7000 6500 Surface (km^2) 509 1470 8343 7382 1989 2001 Urban Vegetation The urban ratio change is +0.109.
Albedo calculation Method Conversion of DN to radiance Conversion of radiance to TOA reflectance Atmospheric correction Histogram matching. Geometric regression Surface albedo calculation Liang s formula
Conversion from DN to radiance Where Quantized calibrated pixel value [DN] Spectral radiance at the sensor s aperture [W/ m^2 sr µm] Band-specific rescaling gain factor [ (W/ m^2 sr µm)/dn ] Band-specific rescaling bias factor [W/ m^2 sr µm] Band specific rescaling gain and bias factors from paper Chander 2009
Conversion from radiance to TOA reflectance Planetary TOA reflectance [unitless] Spectral radiance at the sensor s aperture [W/(m^2 sr µm)] Earth-sun distance [ astronomical units] Mean exoatmospheric solar irradiance [ W/(m^2 µm)] Tables from paper Chander 2009 Solar zenith angle [degrees] Cos solar zenith angle= sin solar elevation angle (stored in L1 product header file)
Atmospheric correction Method : Multiple-date Empirical Radiometric Normalization Relative atmospheric correction Reference image : 2001 Principle : Selection of bright and dark pseudo invariant features (pixels of constant reflectance) Assumption : their difference in DN values between 1989 and 2001 is linear for each band Geometric regression to normalize
Albedo calculation For heat surface budget, surface shortwave broadband black sky albedo Black sky albedo : (Schaepman 2006) Ratio of the radiant flux reflected into the view hemisphere by a unit surface area to the incident radiant flux coming from a single direction Shortwave broadband : albedo within the range [Δλ=0.25µm ; 5µm] Source : http://ec.europa.eu/index_en.html Liang s formula (2000) to calculate shortwave broadband albedo from narrowband albedo Where is the spectral albedo in band i If the surface is Lambertian, surface narrowband albedo = surface reflectance =atmospherically corrected TOA reflectance
Results Map of albedo change -0.29 to -0.1-0.1 to -0.05-0.05 to +0.05 0.05 to 0.1 0.1 to 0.4 The Water, Intertidal and Unclassified pixels are masked.
Land cover change Albedo change Std dev All -0.006 0.04 Agri&Range to Agri&Range -0.012 0.03 Urban to Urban +0.017 0.03 Agri&Range to Urban +0.018 0.04 Agri&Range to Agri&Range: Decrease Summer crops growth between July and August Urban to Urban : Increase New materials Agri&Range to Urban : Increase Contradictory with previous studies
Study of Agri&Range to Urban pixels Residential Residential with planted vegetation 2001 image Residential 2001 image Albedo stable or decreased
Non residential zone Shanghai Pudong International Airport 1989 image Industrial zone 2001 image Albedo change 1989 image 2001 image Albedo change High increase in albedo (>0.1) Affect greatly the overall value
Albedo sensitivity Over the whole scene, Albedo sensitivity = -0.054 City Urban ratio change Albedo change Albedo sensitivity Chongming 0.137 0.003 0.025 Haimen 0.118 0.010 0.089 Qidong 0.0676 0.006 0.073 Rudong 0.162-0.012-0.077 Taicang 0.312-0.006-0.021 Tongzhou 0.075-0.029-0.394 Sanchangzhen 0.077 0.001 0.015
Conclusion The urbanization effect on albedo : Previously urbanized area: Increase due to new materials Newly urbanized area : Depends locally on the type of urban feature built - Bright surfaces : High albedo increase - Darker surfaces : Decrease Not possible to make a correlation between urbanization and albedo for now. Need more cities and data from different dates.
Thank you