8-Band Challenge Worldview 2 -DIGITAL GLOBE

Transcription

1 -Band Challenge Worldview -DIGITAL GLOBE SPATIAL STATISTIC OF BAND WORLDVIEW - MULTISPECTRAL IMAGERY FOR DETAIL URBAN LAND COVER CLASSIFICATION Abdul Wahid HASYIM, Adipandang YUDONO URegSpaceLAB, Urban and Regional Planning Department, Brawijaya University URegSpaceLAB, Urban and Regional Planning Department, Brawijaya University (INDONESIA) s: awhasyim@yahoo.com; adipandang@ub.ac.id Executive Summary Spatial Information about land cover and land use in urban areas is needed various kinds of planners and research professions, such as land use planner, transport planner, architect, hydrologist also urban climate researcher ( accessed 0). According to Eric F. Lambin, Helmut Geist, (00), Land use is a land management information which is obtained from a very detailed analysis. Currently, information on land cover can be easily obtained by using satellite imagery. Through image processing of satellite recordings with sophisticated sensor technology, variety and characteristics of land cover can be known in detail. WORLDVIEW is the image of the New High Resolution Spectral Bands that have excellent band from DigitalGlobe Inc; Coastal Band (00-0nm), Yellow Band (-nm), Red Edge Band (0-nm), and Near Infrared Band (0-00nm), and other standard band of red, green, blue, and near-infrared. These bands very suitable for reading urban land cover classification. This research examine spatial statistic of band- WORLDVIEW by comparing classification techniques using SAM (Spectral Angle Mapper) method and Maximum Likelihood (ML) method to give the best classification results and which most appropriate bands affects the reflection spectrum of urban objects.. INTRODUCTION Urban land utility is physically tangible in land cover (Qihao Weng, 00). Human activities in land surface are varying widely, so as to provide different urban characteristics of each area, especially in Indonesia, which has many traditions. However, a fundamental thing that occurs in complex urban issues is urban space scarcity to provide land supply and demand to perform activities (CHENG et al., 00, Blackman, D, 00). The uniqueness of each area of urban land cover is expected to provide huge information to the planners so as to help them when making decisions. Thus, multispectral image data is required in order to read the details of urban land cover, in order to produce accurate classification with categorization based on pixel values (Brandt Tso, Paul Mather, 009). The development of satellite images technology in such very rapid. Previously, Remote Sensing researchers got a little bit difficulties to analysis land cover observations, particularly for detail land cover classifications, who utilize LANDSAT multispectral imagery products by the year 00 as a result of malfunctioning of SLC (Scan Line Corrector) or SLC Off ( accessed 0). However, today, remote sensing researchers can be happy because of multispectral imagery with New High Resolution Spectral Bands have been present beyond the ability of spectral resolution and spatial resolution of LANDSAT which has band. WORLDVIEW multispectral satellite image that recently launched by a large company DigitalGlobe has bands. New High Resolution Spectral Bands on WV- is the Coastal Band (00-0nm), Yellow Band (- nm), Red Edge Band (0-nm), and Near Infrared Band (0-00nm), and band standard Another is: red, green, blue, and near-infrared. WV- with better resolution (m) than Landsat (0m) further simplify the process of interpretation, so that an error reading image pixel boundaries or areas that often occurs when making the ROI (Region of Interest) can be minimized. Areas of interest study in this research is Sukun district in Malang city-indonesia with land cover characteristics that distinguish built-up area of industrial activities, new housings, roads and un-built up area of vegetation, rivers and rice fields. The boundaries between activities on land cover have a repetitive pattern and a random distribution, making it difficult to distinguish clearly through the manual method. Classification method used in this paper is supervised Maximum Likelihood (ML) and Spectral Angle Mapper (SAM). In the ML method distributed the pixels are considered normal and have the same value, then grouped into class. Maximum Likelihood formula, written (Brandt Tso, Paul Mather, 009): Spatial Statistic of band WorldView -Multispectral Imagery for Urban Land Cover Classification HASYIM & YUDONO

2 -Band Challenge Worldview -DIGITAL GLOBE Where: C j = covariance matrix of class w j with dimension ρ, µ j, is the mean vector of class w j, and determinant. = the probability of coexistence (or intersection) of events x and w = vector transpose denotes the Meanwhile, SAM classification formulated by Kruse (99), a method of classification by comparing the direct interaction of wave spectra image of the unknown (i) in band b of the wave spectra of reference (r) in band b,written as the following equation: Where: n = Number of bands involved Conducted testing of the bands that are owned by the multispectral imagery, namely the New High Resolution Spectral Bands and a few other bands on the WV- courtesy of a big company DigitalGlobe by providing the image we needed for free by joining -band challenge competition. So that, this research will make the reference, especially for researchers in urban (planner) for beyond urban and regional analysis which is not fixated on the standard bands like the previous generation of satellite imagery. Then prepared the following research questions, how much influence each band of spectral reflectance in urban areas?. METHODOLOGY AND DATASET Observations made in the study area of Sukun District in Malang City, with the recording date and Satellite Image number-mds_rc 0JAN00 image-00900_0_p00. Extensive observations of size.0x.99 pixels or,000, pixels. Next, a subset of urban objects that are considered interesting to observe. Subset I (see Figure ) measuring.x pixels or,00, pixels, or km based on spatial resolution WV- (m), subset II (see Figure ) measuring.x0 pixels or 9,90 pixels or.9 km. The procedural steps for land cover classification and the influence of each band of urban objects in this research can be seen in Figure. Figure. Subset, covering an area of km, with a combination of natural bands,, Figure. Subset, covering an area of.9 km, with a combination of natural bands,, Spatial Statistic of band WorldView -Multispectral Imagery for Urban Land Cover Classification HASYIM & YUDONO

3 -Band Challenge Worldview -DIGITAL GLOBE Satellite Imagery WORLDVIEW as new products of DigitalGlobe launched in October, 009. The file contains the image of WV- in a geometric condition has been corrected, so that more and shorten the time to process it immediately. There are types of images in it, namely, Multispectral band and Panchromatic. band multispectral image used for test each band and land cover classification, while the panchromatic imagery with better resolution 0. m be used to assist the process of interpretation (Stuart Murchison, ~ smurchison /, accessed 0). In order to further computational processing performed separation of noise in image data of each band WV- (Boardman and Kruse, 99). This research using MNF (Minimum Noise Fraction) transform to remove the noise data and determine which ones contain coherent band images (by examining the images and eigenvalues), and check the bands that have a value of eigenvalues greater than or contain data. Meanwhile, the band that has a value of eigenvalues narrow or containing noise is not used in this analysis. Through the MNF transform is known, that the eigenvalues all bands WV- is greater than, meaning that all the bands WV- in good quality and it is recommended for beyond process analysis (see Figure ). Figure. Step achieve ML and SAM supervised classification, and testing of each band WV- Spatial Statistic of band WorldView -Multispectral Imagery for Urban Land Cover Classification HASYIM & YUDONO

4 -Band Challenge Worldview -DIGITAL GLOBE, Figure. Each band in WV- stated there is no noise The next stage is band compositing, enhancement and, if necessary panchromatic imagery can be used to facilitate the introduction of land cover object when its interpretation. There were ROI (Region of Interest) are made in the image based on pixel similarity, among others: housing (housing_ and housing_), roads, industry, trees (trees_ and trees_), river, rice fields, wet land (wet and wet land_ land_), grass, and cloud. Testing the adequacy of ROI data for each band WV-, Based on the distribution of pixel data obtained from the standard deviation value. The smaller or lower standard deviation value, getting closer to Mean, means the quality of data is become even better. Mean Equation () and Standard Deviation () is the root of the variance () is written as follows (Steel et. Al, 99), () () () Before doing Urban land cover classification, the entire pixel observations re-tested for each band WV- based on the spatial statistic using index Moran's I to know the relationship neighbourhood spatial (Spatial Autocorrelation) a local homogeneity between pixel neighbors and the mean (Cliff, AD and Ord, JK, 9). Moran's I index values range from + and -, where + = has a strong spatial Autocorrelation positive, 0 = no Autocorrelation, and -= have a strong spatial Autocorrelation negative. Moran's I index equation is written as follows: where, x j is the value of a pixel for band j, x i is the value of a pixel for band i and N = number of bands.. RESULTS AND DISCUSSIONS () Making ROI on the image with the location Sukun Distrit with geographical Location,.0,9,-E up to,9.-s, 9,,9.-E UTM (WGS-) requires precision mainly deal with land cover similar material but no similar activities. Thus, it is not uncommon to distinguish through the coding of the same name but different types, such as housing_ and housing_, trees_ and trees_, wet and wet land_ land_. In housing_ and housing_ using clay base material for the old houses and using ceramics on new houses. Trees_ and trees_ differentiated based on the growing such as near wet areas like surround river areas or dry areas like urban areas. Physically, trees in wet areas or surround river areas have a lot of leaves, compacts, branches with many, strong, and high limbs. Conversely, urban trees are ornamental plants that have leaves no compacts, low position, and the plants branches is less strong. For wet land_ and wet land land_ differentiated by the continuous wet and not continuous wet, for a continuous wet, water generally does not immediately infiltrate into the soil or settle on the ground surface. Malang city specifically Sukun district are still in Spatial Statistic of band WorldView -Multispectral Imagery for Urban Land Cover Classification HASYIM & YUDONO

5 -Band Challenge Worldview -DIGITAL GLOBE a state of the rainy season occurs times a day rain, day and night. Thus, not infrequently lands without pavement are always in wet conditions (wet land_). Through a statistical test against ROI based on spectral reflectance and wave length bands of influence, in a sequence based on the standard deviation value arrange as in table. Table. Band which influenced ROI Band Number Region Of Interest (ROI) nm (coastal) rice field, wet land, trees, asphalt, grass, housing_, industrial, river, trees_, housing_, wet land, cloud. 0-0 nm (blue) rice field, wet land, trees, asphalt, grass, housing_, industrial, river, trees_, housing_, wet land, cloud. 0-0 nm (green) housing_, wet land, trees, grass, housing_, river, trees_, wet land, cloud. - nm (yellow) rice field, wet land, trees, grass, river, trees_, wet land, housing_, housing_ nm (red) rice field, wet land, trees, asphalt, grass, housing_, industrial, river, trees_, wet land, cloud, housing_. 0 (red edge) wet land, industrial, river, wet land_. 0 9 (near IR-) wet land, asphalt, housing_, cloud, river, wet land, trees_, housing_ nm (near IR-) cloud, wet land, industrial, wet land_ Results in Table, the graphic can be seen as in table below, Table. The value of standard deviation and the bands at wv- Value and Band Number graphic Ba nd Standard Deviation Value and Band Number graphic Ba nd Standard Deviation Spatial Statistic of band WorldView -Multispectral Imagery for Urban Land Cover Classification HASYIM & YUDONO

6 -Band Challenge Worldview -DIGITAL GLOBE Furthermore, to find out neighborhood spatial (Spatial Autocorrelation) based on similarity of pixels values in an overall observation of,000, pixels, shown by the Autocorrelation Index value in each band (Table ).The table shows that each band has a very strong positive effect (close to +) of spectral reflectance values of each pixel, and neighborhood spatial against the average value (mean), so the size of local homogeneity can be known. The stronger the Autocorrelation Index, provide a level of classification accuracy based on the band and the homogeneity of pixel values become even better. Table. Autocorrelation Index on each band Band Number Autocorrelation Index (Moran s I) Band Band 0.99 Band Band Band 0.99 Band 0.9 Band 0.99 Band Spatial Statistic of band WorldView -Multispectral Imagery for Urban Land Cover Classification HASYIM & YUDONO

7 -Band Challenge Worldview -DIGITAL GLOBE After doing statistical tests on the ROI and the spatial statistics at each pixel, the result of classification based on the ML and SAM, shown in figure,, and. Spatial Statistic of band WorldView -Multispectral Imagery for Urban Land Cover Classification HASYIM & YUDONO

8 -Band Challenge Worldview -DIGITAL GLOBE Figures and then and arrange systematically in stage to make it easier comparing the results of classification through supervised ML method and SAM. It can be seen in the area with black circles, the results of classification on Subset- (ML) in figure in circle is dominated by housing_ while at the same place Subset- (SAM) is dominated housing_ (Figure circle ). In figure circle, the pixels on ML classification trees_ legible and should still have trees_ and rice fields as in Subset- (SAM) infigure -circle. Some comparisons ML and SAM classification results can be seen in figure - circle with -circle, figure circle 0 with figure -circle, figure - circle with figure - circle shows the results of the ML classification less sensitive to differences in common pixel value or range of similarity value of pixels is huge, so it is considered a class. Conversely, in figure -circle and figure -circle shows the results Spatial Statistic of band WorldView -Multispectral Imagery for Urban Land Cover Classification HASYIM & YUDONO

9 -Band Challenge Worldview -DIGITAL GLOBE of SAM classification has a limited range of similarity value of pixels is too narrow so that there are areas do not have class (unclassified). Based on the above case, we can say the results of SAM classification method is still better than the results of classification using the ML method. It still can be shown with standard deviation value of each classification of the ROI in the following table. Table. Results of classification and standard deviation Supervised Classification Standard Deviation ML.99 SAM.. CONCLUSION WV-image- with New High Resolution Spectral Bands have a spatial resolution (m) and is supported panchromatic band with a spatial resolution of 0.m is very suitable for urban planning purposes because it provide information on land cover better than any other images. Multispectral bands in WV- proved to be separate with appropriate spectral reflectance values in accordance with wavelength in each band,so it will produce more accurate classification. Conversely, experiences and meticulous of the researchers when build up or the determination of ROI based on the pixel similarity is to determine quality of final classification of an image. ACKNOWLEDGEMENT This research regarded -Band Challenge Competition to review WORLDVIEW Satellite Imagery that supported by DIGITALGLOBE inc. Special thanks to mr. Ian Gilbert who intensively contact to authors about WORLDVIEW satellite image data availability. Hopefully, this research giving contribution for DigitalGlobe development products. REFERENCES [] Boardman, J. W., and Kruse, F. A., (99), Automated spectral analysis: a geological example using AVIRIS data, north Grapevine Mountains, Nevada: in Proceedings, ERIM Tenth Thematic Conference on Geologic Remote Sensing, Environmental Research Institute of Michigan, Ann Arbor, MI [] Blackman, David, 00, Australia Strives To Ease Growth Pains, Planning, RMIT University. [] Brandt Tso, Paul Mather, 009, Classification Methods For Remotely Sensed Data, Taylor & Francis Group, LLC [] Cliff, A.D. and Ord, J.K. 9. Spatial autocorrelation. Pion, London [] Eric F. Lambin, Helmut Geist, (00), Land-Use and Land-Cover Change: Local Processes and Global Impacts, Springer-Verlag Berlin Heidelberg [] Kruse, F.A., J.W. Boardman, A.B. Lefkoff, K.B. Heidebrecht, A.T. Shapiro, P.J. Barloon, and A.F.H. oetz, 99. The Spectral Image Processing System (SIPS): Interactive visualization and analysis of imaging spectrometer data. Remote Sensing of Environment [] Lillesand, Kiefer, Chipman, 00, Remote sensing and Image Interpretation, Fifth Edition, John Wiley & Sons, Inc, USA. [] Qihao Weng, 00, Remote Sensing of Impervious Surfaces, CRC Press, Taylor & Francis Group 000 Broken Sound Parkway NW, Suite 00 [9] Steel, Robert G.D, James H. Torrie, (99), Principles and Procedures of Statistics, McGraw-Hill Inc. [0] [] [] Spatial Statistic of band WorldView -Multispectral Imagery for Urban Land Cover Classification HASYIM & YUDONO 9

10 Authors Biography: st Author Profile: -Band Challenge Worldview -DIGITAL GLOBE Name: Abdul Wahid HASYIM Author (s) Affiliation:Department of Urban and Regional Planning, Faculty of Engineering, Brawijaya University, INDONESIA Mailing Address: Dept. of Urban and Regional Planning Building, Faculty of Engineering, Brawijaya University, Jl. MT.Haryono no. Malang - East Java, INDONESIA Address: awhasyim@yahoo.com or awhasyim@gmail.com Website: Telephone number (s):+-0 or +- Fax number (s):+-0 Brief Biography: I graduated from Department of Architecture, Faculty of Engineering, Brawijaya University, Malang, Indonesia, as Bachelor of Engineering in 99 then continued Program of Urban and Regional Planning from the Bandung Institute of Technology, Bandung, Indonesia as master degree of development studies in 99. Currently, I working in the Department of Urban and Regional Planning, Faculty of Engineering, Brawijaya University, Malang, INDONESIA as Lecturer, reseacher, and senior planner since 99. In 009, I ve got an award from the President of the Republic of Indonesia as my loyality as lecturer and researcher in Indonesia called Satyalancana Satya. Currently, I pursuing my doctor program in the field of Remote Sensing in Sepuluh November Technology Institute,Surabaya-Indonesia. In the year 00, I have attended short courses with theme 'Urban Land Management and GIS' at KTH (Kungliga Tekniska Högskolan) Sweden. My Research interests are urban land management, urban land use planning, urban growth modeling. nd Author Profile: Name: Adipandang YUDONO Author (s) Affiliation:Department of Urban and Regional Planning, Faculty of Engineering, Brawijaya University, INDONESIA Mailing Address: Dept. of Urban and Regional Planning Building, Faculty of Engineering, Brawijaya University, Jl. MT.Haryono no. Malang - East Java, INDONESIA Address: adipandang@ub.ac.id or adipandang@yahoo.com Telephone number (s):+-0 or +- Fax number (s):+-0 Brief Biography: I graduated from Department of Geography-University of Indonesia in 00 as Bachelor Degree of Science (Geography) and Program of Urban and Regional Planning, School of Built Environment-University of South Australia in 00 as Master degree of Urban and Regional Planning. Currently I work in Department of Urban and Regional Planning, Faculty of Engineering, Brawijaya University, Malang-INDONESIA as Lecturer. Since 009 I ve got promotion as Urban & Regional Space Lab (URegSpaceLAB). I teach Land Use Planning Management, Watershed Management, Coastal Management, Transport Planning and Planning Information System. My Research interests are Disaster Management, Urban Growth Modeling and Public Transport System Modeling. Spatial Statistic of band WorldView -Multispectral Imagery for Urban Land Cover Classification HASYIM & YUDONO 0