Delineating Climate Relevant Structures for the Beijing Metropolitan Area 1,2 3,4 1 Austrian Academy of Sciences GIScience, Salzburg 2 Insitute for Geography, University of Bamberg, Germany 3 ispace, Austrian Research Center, Salzburg, Austria 4 Centre for Geoinformatics,, University of Salzburg, Austria Content Background: 100 cities project Urban Environmental Monitoring (UEM) Area of interest: Beijing, Capitol of China Remote Sensing Image Data Image Analysis: 1) for Land Use Land Cover (LULC) 2) for the differentiation of settlement structure matthias.moeller@uni-bamberg.de thomas.blaschke@sbg.ac.at Research Embedded in 100 Cities Project NASA financed long term research project Development of strategies for comparison of metropolitan areas world wide based [mainly] on ASTER image data Multi-spectral & DEM (bands 3N, 3B) Coordination at Arizona State University Cooperating partners world-wide wide (in 100 cities) Berlin, Germany New Delhi, India Lima, Peru Phoenix, USA Canberra, Australia Chiang Mai, Thailand Unique Classification Schema Reliable Accuracies (1) Overall Classification Accuracy (2) Overall Kappa Statistics Class Hierarchy for images without clouds Class Hierarchy for images with clouds Manila, Philippines Mexico City, Mexico Berlin, Germany 88% 0.8500 Canberra, Australia 83% 0.8000 Lima, Peru 80% 0.7500 Chiang Mai, Thailand 86% 0.8400 New Delhi, India 74% 0.6750 Manila, Philippines 83% 0.8000 Phoenix, USA 88% 0.8500 Mexico City, Mexico 87% 0.8400 1
Statistical Comparison Area of Interest, Beijing China Beijing metropolitan area City of Beijing >15 million people Chinas capital, fast growing, rapid changes Transition process from traditional agricultural society to industrial society Climate/Weather for Beijing Climate/weather: Humid, hot summers and cold, dry winters Precipitation 619 mm/year Sand storms from Mongolian Desert & Increasing traffic, energy consumption leads to air pollution, SMOG, e.g. inversion weather conditions http://static.panoramio.com/photos/original/33219.jpg Beijing Reveals Serious Exhaust Regulations Air pollution and hot and humid August weather conditions in Beijing are the main concerns of the IOC and Chinas national sport association. The IOC is thinking about to change dates for sport events in case the air pollution become worse 2
Special Motivation Area of Interest - Location Beijing hosts the Olympic Games in 2008 An urban climate model was established for: Simulation of air temperature distribution & Ventilation under certain wind conditions Input values: : LULC & settlement pattern ca. 4000 km x 4000 km ca. 250 km x 250 km I Land Use Land Cover Analysis ASTER image recorded 16.03.2006 40 km x 46 km band combination: green-red red-infrared:: blue-green green-redred ATCOR correction geometric correction absolute ref. to UTM image to image referencing to SPOT OBIA LULC mapping Object Based Image Analysis (OBIA) including: Spectral features, Texture parameters, Individual shape parameters, Context information: spatial context,, neighborhood. Shape, Context, Neighborhood, Knowledge Shape, Context, Neighborhood, Knowledge This figure shows 50% of a photo with 50 % transparency! This figure shows 50% of a photo with 50 % transparency! 3
Shape, Context, Neighborhood, Knowledge Context, Neighborhood Example: I arr at Cal int air a few day ago. I arrived at Calgary international airport a few days ago. This figure shows 50% of a photo with 0 % transparency! Three single letters of a word is enough to recognize the word and to understand the meaning of the entire sentences. Neighborhood helps to find the correct meaning. Our Knowledge helps to understand; we use these findings in the OBIA approach. Object Based Image Analysis Process Segmentation on several levels. Image segments, horizontally & vertically connected. Definition of classes. Rule set ontologies: spectral, shape, texture, context. Cognition network of nested rule sets. Classification Evaluation of the results. New definition/refining of rule sets. Final classification. Segmentation & Class Hierarchy Level 4 3 2 1 Scale Parameter 12 50 25 25 Color - Shape 0.9 0.1 0.9 0.1 0.9 0.1 0.5 0.5 Compactness - Smoothness 0.8 0.2 0.8 0.2 0.8 0.2 0.5 0.5 Level 3 was not used for the later classification process. Cognition Network & Classification Schema level 1 developed_sub level 2 level 4 transportation_sub developed not developed transportation hills high density low density vegetation_dev water_dev barren rock vegetation_not_dev water_not_dev 4
Accuracy Assessment Beijing resident (and research assoc.) performed an accuracy assessment based on 50 cm aerials Randomly distributed points verification Additional verification with Google Earth Overall average of the classification accuracy for the Beijing ASTER LULC scene 83% II Texture Analysis for Settlement Areas Objective: Extract building heights and urban canyons for the climate model one LC class & three different settlement categories [1] open space, e.g. water [2] small buildings, traditional, one story [3] medium size, 10-20 years old, 15 20 stories [4] tall buildings, recently build, > 20 stories Some impressions from Beijing Open space, water, class 1 Traditional buildings, class 2 Traditional buildings, class 2 5
Traditional buildings,, ASTER, class 2, 1250 m x 800 m Medium size buildings, class 3 Medium size buildings, class 3 Medium size buildings,, ASTER, class 3, 1000 m x 500 m Tall buildings, class 4 Tall buildings, class 4 6
SPOT pan From the OASIS program SPOT pan 5 m spatial resolution OBIA, segments with a regular grid pattern 50 m cell size (10 x the original pixel extend) GLCM (Grey Level Co-occurrence occurrence Matrix) for each cell & Classification after GLCM values: [1] GLCM > 1 and 3 low texture, open space. [2] GLCM > 3 and 12 = small buildings. [3] GLCM > 12 and 26 = medium texture, medium size buildings. [4] GLCM > 26 = high texture, tall, big buildings. Tall buildings,, ASTER, class 4, 500 m x 500 m SPOT Texture Analysis Legend: open space, water small, traditional buildings medium size buildings tall buildings (missing here) Conclusions ASTER provides image data suitable for urban object mapping (medium scale) OBIA is essential for the classification Future research: Improve SPOT texture analysis Adopt OBIA to more natural environments on earth (frequent updates of urban areas) Compare and rank those urban areas MANY THANKS FOR YOUR ATTENTION matthias.moeller@uni-bamberg.de thomas.blaschke@sbg.ac.at 7