UNDERSTANDING URBAN ENVIRONMENTS FROM SATELLITE IMAGERY: APPROACHES, DATA EXTRACTION METHODS, AND APPLICATIONS Projo Danoedoro Centre for Remote Sensing and GIS (PUSPICS), Faculty of Geography, Gadjah Mada University Yogyakarta, Indonesia March 2011
Remotely Sensed Data: various platforms, various spectral regions, various sensors, and various scales
Landsat ETM+ Satellite image
Aerial photograph, panchromatic B/W
Quickbird satellite image
Quickbird Satellite image of ricefield area
Quickbird Satellite image of ricefield area
• Different data specifications require different processing methods • General approach to image interpretation: visual (including on-screen), digital – spectral based, digital object-oriented • Urban environments analyses and mapping can make use of various approaches/methods, depending on the types of information to be extracted (even though many practitioners say that only visual interpretation can generate more detailed and accurate information)
Water? Sea? Littoral aquatic environment? Water? Fishponds? Metal surface? Buildings? Industrial area? Herbs and soil? Interleaved planting? Dryland cultivation?
Fiber-cement surface? Buildings? Commercial area? Asphalt surface? Urban infrastructure? Road network?
What are the proper names for those objects? Are those names appropriate for planning purposes?
Land-use as a multidimensional concept (a remote sensing perspective) Cover type
HIG H
Spatial structure, site, shape
SPECTRA L
Development stage, rotation, length of inundation
Ecosystem unit
Land use: function
Land-use Attributes
Legal status
SPATIA L Possibility to extract information from remotely sensed data
Parameters to measure?
TEMPOR AL ECOLOGICA L SOCIOECONOMIC FUNCTION
LEGAL LO W
Complexity and difficulty in information extraction, involving non-remotely sensed data
HIG H
Analysis Methods ?
• Land-use: key information required in physical planning processes • Land-use (function) ≠ land-cover (physical appearance) • Digital remote sensing: rapid method, generates land-cover, but not land-use. Occasionally, it can generate mixed information of land-cover and land-use • Land-use information could be generated by integrating remote sensing and GIS, but normally at limited number of classes
• High spatial-resolution imagery: normally used as a basis for visual interpretation • Multi-spectral classification of high-res imagery creates a lot of misclassified pixels; generates less accurate landcover information • For optimum use of the available multi-spectral bands development of method for extracting land-use information related to socio-economic function is required • Categorization of socio-economic function developed in the Versatile Land-use Classification Scheme could be used to test the method
Original Image
Multispectral classification result
Visual interpretation result
Combination of image processing and GIS
Visual interpretation: spatial aspects included are shape, pattern (including regularity), size, and density Source: Quickbird imagery (0.6 m spatial resolution Method: on-screen digitisation Result: map in vector data model, which can be rasterised Output scale: ~ 1:6,000
Digital Classification: spectral aspect of the land-cover, differentiated based on their reflectance Source: Quickbird imagery (0.6 m spatial resolution Method: Maximum likelihood multispectral classification Result: map in raster dta model Output scale: ~ 1:3,000
Combination of visual and digital analyses temporal aspects included are length o inundation, growth stage, maturity, periodical/seasonal change, old and new, etc. Source: Quickbird imagery (0.6 m spatial resolution Method: RS-GIS integration Result: raster map Output scale: ~ 1:3,000
Combination of GIS and Image Processing Ecological aspect presented is ecosystem unit, both man-made and (semi)natural ecosystems Source: Quickbird imagery (0.6 m spatial resolution Method: RS-GIS integration using knowledge-based techniques Result: raster map Output scale: ~ 1:3,000
Combination of Image Processing and GIS socio-economic function (uses) Source: Quickbird imagery (0.6 m spatial resolution Method: RS-GIS integration using knowledge-based techniques Result: raster map Output scale: ~ 1:4,000
• SPECTRAL INDEX APPROACH: Combining two or more spectral bands which are sensitive to particular phenomena (vegetation density, biomass content, impervious surface, soil moisture, etc.) in order to generate new maps with more representative pixel values called ‘index’ of phenomena of interest
VEGETATIONINDEX INDEX VEGETATION
Near Infrared-Red/Near Infrared + Red Urban area Semarang 1994 Vegetation Index Semarang 1994
Urban area Semarang 2002
Vegetation Index Semarang 2002
RED Band (ETM Band 3)
Bright, can be either bare soil or cloud
NearInfrared Band (ETM Band 4)
Dark, can be either vegetation or water
Bright, can be either bare soil, vegetation or cloud
Dark: water
Ratio Veg Index: B4/B3
The brighter, the denser the vegetation cover
Black/dark: no vegetation
Normalised Difference Vegetation Index (NDVI)
Field samples of biomass content 4 1
Biomass content map (kg/m2)
12 8
2
16
17
20 5 1 3
9
19
8
15 19
11
14 6
7
13 10
Regression inversion using map calculator Biomass = (NDVI - 0.011)/ 0.040
Y = 0.011 + 0.040X r = 0.951
Can we apply a non-linear regression equation? What’s the consideration behind the
URBAN INDEX
Middle Urban area Semarang 1994
Urban area Semarang 2002
Urban Index Semarang 1994
Infrared-Near Infrared/Middle Infrared + Near Infrared
Urban Index Semarang 2002
• FRAGMENTATION ANALYSIS Neighbourhood analysis for measuring fragmentation level of an ecosystem
THE EFFECT OF SCALES IN THE ANALYSIS OF AGRICULTURAL LANDUSE FRAGMENTATION BASED ON SATELLITE IMAGE OF SEMARANG AREA, INDONESIA Projo Danoedoro
PUSPICS/Department of Geographical Information Science and Regional Development Faculty of Geography, Gadjah Mada University, Indonesia
LAND FRAGMENTATION • Land fragmentation a stage/condition that shows a certain level of heterogeneity of a landscape feature, based on a particular observation scale
Massive/ Compact
Fragmented, two predominant features
Fragmented, three Very fragmented, pre-dominant two predominant features features
Very fragmented, four-five predominat features
Fragmentation as a MAUP Phenomenon Map of Variable A (20m)
Map of Variable B (20m)
Scale Effect: Correlation coefficient: 0.645 Linear equation: B = 3.55 + 0.452*A Map of Variable A (40m) Map of Variable A (40m)
Correlation coefficient: 0.988 Linear equation: B = 0.309
The use of different scales/spatial resolutions causes different conclusions
Fragmentation as a MAUP Phenomenon
Zoning Effect: The use of different spatial zone/ aggregational zone causes different conclusions
Research Methods
Maximum Likelihood, 2002, overall acuracy 94.52% (0,9366)
Problem 1 • Digital image processing of remotely sensed imagery cannot directly generate land-use information. It produces spectral- related land-cover classes • Class merging/regrouping should be carried out to derive more meaningful land-cover classes
Land-cover
Terrain unit map
Land-use
Problem 2 • Land-cover classes are not land-use categories. A terrain unit map containing land characteristics can be integrated in a raster GIS environment for generating landuse classes.
Land-use fragmentation in Semarang area as a result of modelling at various observation window sizes
3x3 pixel analysis unit (0.81 ha)
5x5 pixel analysis unit (2.25 ha) 7x7 pixel analysis unit (4,41 ha)
Problem 3 • Fragmentation index formula adopted by various softwares does not work with a fragmentation phenomea involving limited number of classes • The land-use map should be simplified into a new map containg less number of relevant classes • The fragmentation index map is then superimposed with the new land-use map to produce new fragmentation index occuring in the classes of interest.
A Simplified land-use categorisation used as a basis for fragmentation index computation on agricultural land-use only.
Fragmentation levels of agricultural land-use, processed using 3x3, 5x5, and 7x7 window sizes. See the difference between those images
3 x 3 pixel (0.81 ha)
5 x 5 pixel (2.25 ha)
7 x 7 pixel (4.41 ha)
CONCLUDING REMARKS
Fragmentation is a phenomenon which is sensitive to observation scale match with the Modifiable Areal Unit Problem (MAUP)
In the study area, the MAUP phenomenon represents the scale effect, and the observation at approximately 1 hectare can give more clear fragmentation due to the average land parcel size, which is much smaller than 1 hectare.
The agricultural land-use fragmentation in Semarang mostly took place in the alluvial and coastal alluvial lands due to the expansion of settlement and industrialisation.
The fragmentation index formula using standard procedure of image processing and GIS do not sufficient to describe fragmentation of particular land-uses/ecosystem units. There is a need to develop method for more representative models based on satellite remeotely-sensed imagery
Tsunami Run-up Risk of Non-settlement Building (contributed by Rudiansyah Putra, MSc Thesis)
Building identification and mapping using High-spatial resolution imagery (Visual interpretation)
Material types
Floor condition
Number of story
Building orientation
Building vulnerability
Land-use
Surface roughness
Elevation
Slope steepness
RUN-UP SIMULATION
NON-SETTLEMENT BUILDING RISK AT VARIOUS RUN-UP HEIGHT
Remote Sensing and GIS for Analysing Tuberculosis Contagion Risk in Urban Area (materials supported by Satya Erlangga ~ MSc Thesis in Remote Sensing and GIS, Gadjah Mada University)
Dibuat Oleh : Satya Erlangga, Tahun 2009 Program Studi Penginderaan Jauh Program Pascasarjana Universitas Gadjah Mada
Sistim Proyeksi UTM Sistim Proyeksi UTMZone Zone49S 49S Sumber : Pengolahan Citra QuickBird th 2003 Survei Lapangan tahun 2008 Peta Administrasi Kota Yogyakarta
Gambar 4.29. Peta Penggunaan Lahan Tingkat II 135
Dibuat Oleh : Satya Erlangga, Tahun 2009 Program Studi Penginderaan Jauh Program Pascasarjana Universitas Gadjah Mada
Sistim Proyeksi UTM Zone 49S Sistim Proyeksi UTM Zone 49S Sumber : Pengolahan Citra QuickBird th 2003 Survei Lapangan tahun 2008 Peta Administrasi Kota Yogyakarta
Gambar 4.30. Penggunaan Lahan Tingkat III 136
Method for identifying building dnsity using visual interpretation and GIS
Legenda
Conversion of ponit-based building identification to gridbased building density
PETA SEBARAN KEPADATAN BANGUNAN KOTA YOGYAKARTA
Dibuat Oleh : Satya Erlangga, Tahun 2009 Program Studi Penginderaan Jauh Program Pascasarjana Universitas Gadjah Mada
Sistim Proyeksi UTM Zone 49S Sumber : Pengolahan Citra QuickBird th 2003 Survei Lapangan Tahun 2008 Peta Administrasi Kota Yogyakarta
Batas Kelurahan
MAPPING OF SPATIAL DISTRIBUTION OF TUBERCOLOSIS INCIDENCE IN 2004 Gambar 4.42 Peta Sebaran Kejadian Penyakit TB Tahun 2004 Dibuat Oleh : Satya Erlangga, Tahun 2009 Program Studi Penginderaan Jauh Program Pascasarjana Universitas Gadjah Mada
SistimProyeksi ProyeksiUTM UTMZone Zone49S 49S Sistim Sumber : Data Register Pasien TB (Dinas Kesehatan Kota Yogyakarta) Survei Lapangan Tahun 2008 Peta Administrasi Kota Yogyakarta
153
Accessibility map (from the road and river network) Dibuat Oleh : Satya Erlangga, Tahun 2009 Program Studi Penginderaan Jauh Program Pascasarjana Universitas Gadjah Mada
Sistim Proyeksi Sistim ProyeksiUTM UTMZone Zone49S 49S Sumber : Pengolahan Citra QuickBird th 2003 Survei Lapangan tahun 2008 Peta Administrasi Kota Yogyakarta Hasil Analisis Spasial
Gambar 4.45 Peta Jarak Terhadap Sungai 167
Dibuat Oleh : Satya Erlangga, Tahun 2009 Program Studi Penginderaan Jauh Program Pascasarjana Universitas Gadjah Mada
Gambar 4.46 Peta Jarak Terhadap Jalan Utama
Sistim Proyeksi UTM Sistim Proyeksi UTMZone Zone49S 49S Sumber : Pengolahan Citra QuickBird th 2003 Survei Lapangan tahun 2008 Peta Administrasi Kota Yogyakarta Hasil Analisis Spasial
Dibuat Oleh : Satya Erlangga, Tahun 2009 Program Studi Penginderaan Jauh Program Pascasarjana Universitas Gadjah Mada
Sistim SistimProyeksi ProyeksiUTM UTMZone Zone49S 49S Sumber : Pengolahan Citra QuickBird th 2003 Survei Lapangan tahun 2008 Peta Administrasi Kota Yogyakarta Hasil Analisis Spasial
Gambar 4.47 Peta Jarak Terhadap Pusat-Pusat Kegiatan 169
Accessibility map (from the centres of activities)
Kelas Kerentanan
Tuberculosis vulnerability zones
Dibuat Oleh : Satya Erlangga, Tahun 2009 Program Studi Penginderaan Jauh Program Pascasarjana Universitas Gadjah Mada
Sistim Proyeksi UTM Sistim Proyeksi UTMZone Zone49S 49S Sumber : Pengolahan Citra QuickBird th 2003 Survei Lapangan tahun 2008 Peta Administrasi Kota Yogyakarta Hasil Pemodelan Spasial
Dibuat Oleh : Satya Erlangga, Tahun 2009 Program Studi Penginderaan Jauh Program Pascasarjana Universitas Gadjah Mada
Gambar 4.70 Peta Kerentanan Terhadap Penyakit TB 196
Sistim SistimProyeksi ProyeksiUTM UTMZone Zone49S 49S Sumber : Pengolahan Citra QuickBird th 2003 Survei Lapangan tahun 2008 Peta Administrasi Kota Yogyakarta Hasil Pemodelan Spasial
Tubercolosis risk map
Gambar 4.71 Peta Resiko Terhadap Penyakit TB
Accuracy assessment: 77.8%
Thank you