1 Urban Growth Simulation A Case Study of Indianapolis Sharaf Alkheder & Jie Shan School of Civil Engineering Purdue University March 10, 2005.

Slides:



Advertisements
Similar presentations
Lab for Remote Sensing Hydrology and Spatial Modeling Dept. of Bioenvironmental Systems Engineering, NTU Satellite Remote Sensing for Land-Use/Land-Cover.
Advertisements

Applications of one-class classification
Area and perimeter calculation using super resolution algorithms M. P. Cipolletti – C. A. Delrieux – M. C. Piccolo – G. M. E. Perillo IADO – UNS – CONICET.
Page 1 of 50 Optimization of Artificial Neural Networks in Remote Sensing Data Analysis Tiegeng Ren Dept. of Natural Resource Science in URI (401)
Major Operations of Digital Image Processing (DIP) Image Quality Assessment Radiometric Correction Geometric Correction Image Classification Introduction.
Leila Talebi, Anika Kuczynski, Andrew Graettinger, and Robert Pitt
Multiple Criteria for Evaluating Land Cover Classification Algorithms Summary of a paper by R.S. DeFries and Jonathan Cheung-Wai Chan April, 2000 Remote.
RBF Neural Networks x x1 Examples inside circles 1 and 2 are of class +, examples outside both circles are of class – What NN does.
VALIDATION OF REMOTE SENSING CLASSIFICATIONS: a case of Balans classification Markus Törmä.
Geographic Information Systems
Introduction. Evaluation of ARTMAP classifiers - Dependence on cluster representation Large number of classification algorithms available without detailed.
Lecture 14: Classification Thursday 18 February 2010 Reading: Ch – 7.19 Last lecture: Spectral Mixture Analysis.
Luci2 Urban Simulation Model John R. Ottensmann Center for Urban Policy and the Environment Indiana University-Purdue University Indianapolis.
Spatio-temporal differences in model outputs and parameter space as determined by calibration extent 7 th International Conference on Geocomputation Charles.
Classification of Remotely Sensed Data General Classification Concepts Unsupervised Classifications.
Jacinto C. Nascimento, Member, IEEE, and Jorge S. Marques
Global Land Cover: Approaches to Validation Alan Strahler GLC2000 Meeting JRC Ispra 3/02.
Lecture 23: Brief Introduction to Data quality By Austin Troy Using GIS-- Introduction to GIS.
Radial-Basis Function Networks
Land Use/Land Cover Assessment of Dane County, Wisconsin: Contemporary Trend and Future Projections By Eric Fabian.
Face Recognition Using Neural Networks Presented By: Hadis Mohseni Leila Taghavi Atefeh Mirsafian.
Urban growth simulation using V-BUDEM 1 School of Urban Planning and Design, Peking University 2 Nijmegen School of Management, Radboud University Nijmegen.
Co-authors: Maryam Altaf & Intikhab Ulfat
Geometric Correction It is vital for many applications using remotely sensed images to know the ground locations for points in the image. There are two.
U.S. Department of the Interior U.S. Geological Survey Assessment of Conifer Health in Grand County, Colorado using Remotely Sensed Imagery Chris Cole.
Uneven Intraurban Growth in Chinese Cities: A Study of Nanjing Yehua Dennis Wei Department of Geography and Institute of Public and International Affairs.
Indiana GIS Conference, March 7-8, URBAN GROWTH MODELING USING MULTI-TEMPORAL IMAGES AND CELLULAR AUTOMATA – A CASE STUDY OF INDIANAPOLIS SHARAF.
Seto, K.C., Woodcock, C.E., Song, C. Huang, X., Lu, J. and Kaufmann, R.K. (2002). Monitoring Land-Use change in the Pearl River Delta using Landsat TM.
GIS Data Structure: an Introduction
ASPRS Annual Conference 2005, Baltimore, March Utilizing Multi-Resolution Image data vs. Pansharpened Image data for Change Detection V. Vijayaraj,
Image Classification Digital Image Processing Techniques Image Restoration Image Enhancement Image Classification Image Classification.
Image Preprocessing: Geometric Correction Image Preprocessing: Geometric Correction Jensen, 2003 John R. Jensen Department of Geography University of South.
BUILDING EXTRACTION AND POPULATION MAPPING USING HIGH RESOLUTION IMAGES Serkan Ural, Ejaz Hussain, Jie Shan, Associate Professor Presented at the Indiana.
Fuzzy Entropy based feature selection for classification of hyperspectral data Mahesh Pal Department of Civil Engineering National Institute of Technology.
NEURAL NETWORKS FOR DATA MINING
Image Classification 영상분류
INTERNATIONAL INSTITUTE FOR GEO-INFORMATION SCIENCE AND EARTH OBSERVATION Transition Rule Elicitation Methods for Urban Cellular Automata Models Junfeng.
The ASA-CSSA-SSSA International Annual Meetings (November 12-16, 2006) Urban Growth Modeling Artificial Intelligence Techniques Urban Growth Modeling with.
Change Detection Techniques D. LU, P. MAUSEL, E. BRONDIZIO and E. MORAN Presented by Dahl Winters Geog 577, March 6, 2007.
North American Croplands Teki Sankey and Richard Massey Northern Arizona University Flagstaff, AZ.
Chapter 8 Remote Sensing & GIS Integration. Basics EM spectrum: fig p. 268 reflected emitted detection film sensor atmospheric attenuation.
Digital Image Processing Definition: Computer-based manipulation and interpretation of digital images.
Application of spatial autocorrelation analysis in determining optimal classification method and detecting land cover change from remotely sensed data.
Generation of a Digital Elevation Model using high resolution satellite images By Mr. Yottanut Paluang FoS: RS&GIS.
Feature-based (object-based) Verification Nathan M. Hitchens National Severe Storms Laboratory.
Iterative similarity based adaptation technique for Cross Domain text classification Under: Prof. Amitabha Mukherjee By: Narendra Roy Roll no: Group:
U.S. Department of the Interior U.S. Geological Survey Automatic Generation of Parameter Inputs and Visualization of Model Outputs for AGNPS using GIS.
Downscaling Global Climate Model Forecasts by Using Neural Networks Mark Bailey, Becca Latto, Dr. Nabin Malakar, Dr. Barry Gross, Pedro Placido The City.
Unsupervised Classification
Integrating LiDAR Intensity and Elevation Data for Terrain Characterization in a Forested Area Cheng Wang and Nancy F. Glenn IEEE GEOSCIENCE AND REMOTE.
By: Reid Swanson Sam Soper. Goal: To describe land cover/use changes that have occurred in the Twin Cities Metro-Area from the 1991 to 2005 Quantifying.
Cells - Absolute Values In this instance, the value of the cell represents the value of the phenomenon of interest, e.g. the elevation at that pixel.
26. Classification Accuracy Assessment
Machine Learning 12. Local Models.
Temporal Classification and Change Detection
Pathik Thakkar University of Texas at Dallas
Cluster Analysis II 10/03/2012.
Summary of “Efficient Deep Learning for Stereo Matching”
COMP61011 : Machine Learning Ensemble Models
Evaluating Land-Use Classification Methodology Using Landsat Imagery
LINEAR AND NON-LINEAR CLASSIFICATION USING SVM and KERNELS
Paulo Gonçalves1 Hugo Carrão2 André Pinheiro2 Mário Caetano2
REMOTE SENSING Multispectral Image Classification
Supervised Classification
network of simple neuron-like computing elements
Department of Electrical Engineering
ALI assignment – see amended instructions
Introduction to Radial Basis Function Networks
COSC 4368 Machine Learning Organization
VERY DEEP CONVOLUTIONAL NETWORKS FOR LARGE-SCALE IMAGE RECOGNITION
Presentation transcript:

1 Urban Growth Simulation A Case Study of Indianapolis Sharaf Alkheder & Jie Shan School of Civil Engineering Purdue University March 10, 2005

2 OUTLINE Introduction Introduction Data and preprocessing Data and preprocessing NN approach and implementation NN approach and implementation Results and evaluation Results and evaluation Concluding remarks Concluding remarks

3 INTRODUCTION Urban growth is a complex process Urban growth is a complex process Growth parameters include land use suitability, city development level, economical phase, etc... Growth parameters include land use suitability, city development level, economical phase, etc... A functional model to describe such a process is impossible A functional model to describe such a process is impossible Neural Network (NN) is gaining popularity Neural Network (NN) is gaining popularity

4 INTRODUCTION (Cont’d) Li and Yeh (2002) integrate NN, GIS and CA to simulate different development patterns Li and Yeh (2002) integrate NN, GIS and CA to simulate different development patterns Pijanowskia et al. (2002) integrate Artificial NN and GIS to forecast the change in land use Pijanowskia et al. (2002) integrate Artificial NN and GIS to forecast the change in land use Existing studies Existing studies  Atlanta growth simulation by Yang and LO (2003)  Urban growth prediction for San Francisco and Washington by Clarke and Gydos (1998)

5 STATEMENT OF THE PROBLEM Indianapolis exhibited accelerated urban growth over the last three decades Indianapolis exhibited accelerated urban growth over the last three decades Such a growth makes a small part of Marion County in seventies to cover whole Marion County and parts of neighbor counties in 2003 Such a growth makes a small part of Marion County in seventies to cover whole Marion County and parts of neighbor counties in 2003 The objective of this work is to utilize NN algorithms to simulate urban growth boundaries The objective of this work is to utilize NN algorithms to simulate urban growth boundaries

6 Simple Adaptive Linear NN (SALNN) and Back Propagation (BPNN) algorithms were used Simple Adaptive Linear NN (SALNN) and Back Propagation (BPNN) algorithms were used Different city centers were selected Three different datasets were used fortraining Three different datasets were used for training Short and long-term predictions were made Short and long-term predictions were made Methods

7 STUDY AREA/ INDIANAPOLIS DATA PREPARATION

8 Six historical satellite images for Indianapolis Six historical satellite images for Indianapolis over 30 years were used: - One Landsat MSS 60-meter image (1973) - One Landsat MSS 60-meter image (1973) - Five Landsat TM 30-meters image (1982, - Five Landsat TM 30-meters image (1982, 1987, 1993, 2000 and 2003) 1987, 1993, 2000 and 2003) All images were rectified and registered to Universal Transverse Mercator (UTM) NAD1983 All images were rectified and registered to Universal Transverse Mercator (UTM) NAD1983 DATA

9 Although all images are georeferenced, co- registration is still needed Although all images are georeferenced, co- registration is still needed Second order polynomial transformation function was used Second order polynomial transformation function was used Projected images were resampled to 30 meters Projected images were resampled to 30 meters 12 control points were used per image 12 control points were used per image The Landsat TM 2000 georefrenced image was used as the reference image The Landsat TM 2000 georefrenced image was used as the reference image IMAGE REGISTRATION

10 EXAMPLE OF REGISTERED IMAGES

11 A panchromatic 15-meter resolution image was fused with the 2003 XS low resolution images A panchromatic 15-meter resolution image was fused with the 2003 XS low resolution images Fusion is to produce an image with high both spectral and spatial resolution Fusion is to produce an image with high both spectral and spatial resolution Multiplicative method was used for fusion using all image bands Multiplicative method was used for fusion using all image bands In fused image In fused image spatial resolution is improved spatial resolution is improved spectral resolution may deteriorates in certain areas such as roads and residential areas spectral resolution may deteriorates in certain areas such as roads and residential areas IMAGE FUSION

12 FUSION RESULTS OriginalFused

13 FUSION RESULTS Spatial resolution improvement examples Spatial resolution improvement examples Fused Original Fused Original

14 FUSION RESULTS Spectral resolution deterioration examples Spectral resolution deterioration examples Fused Original Fused Original

15 Fused images and original images are respectively used for classification Fused images and original images are respectively used for classification Same training and testing conditions with 1:4 ratio were implemented for both classifications Same training and testing conditions with 1:4 ratio were implemented for both classifications Classification method: maximum likelihood; supervised Classification method: maximum likelihood; supervised High resolution orthophotographs and USGS land classification maps were used as ground references High resolution orthophotographs and USGS land classification maps were used as ground references Seven classes were specified in the images: Seven classes were specified in the images: - Water - Road - Residential - Commercial - Water - Road - Residential - Commercial - Forest - Pasture/grasses - Row crops - Forest - Pasture/grasses - Row crops IMAGE CLASSIFICATION

16 CLASSIFICATION RESULTS Original Fused

17 Some areas in the fused images were classified better than the original images, e.g. forest class Some areas in the fused images were classified better than the original images, e.g. forest class Other areas were deteriorated e.g. commercials Other areas were deteriorated e.g. commercials Classification accuracy of the original 2003 images was 89.14%, while it was 84.00% for the fused images Classification accuracy of the original 2003 images was 89.14%, while it was 84.00% for the fused images Higher overall classification accuracy is achieved for original image Higher overall classification accuracy is achieved for original image CLASSIFICATION RESULTS (Cont’d)

18 The six years historical urban growth boundaries of Indianapolis area were measured. The six years historical urban growth boundaries of Indianapolis area were measured. DATA FOR NN SIMULATION

19 Two centers were selected Two centers were selected For every configuration, six measurements were recorded at each 3 degrees angle interval For every configuration, six measurements were recorded at each 3 degrees angle interval A matrix of 120 by 6 measurements was prepared A matrix of 120 by 6 measurements was prepared DATA FORNN SIMULATION DATA FOR NN SIMULATION

20 Three datasets were prepared for NN training: Three datasets were prepared for NN training: - Real data set without interpolation - Real data set without interpolation - 5 year interpolated data set - 5 year interpolated data set - 1 year interpolated data set - 1 year interpolated data set RBFN algorithm RBFN algorithm NEURAL NETWORK ALGORITHMS

21 Two of the well-known NN algorithms were trained using the three prepared datasets for every center configuration Two of the well-known NN algorithms were trained using the three prepared datasets for every center configuration The adaptive linear NN as well as BP algorithms were used The adaptive linear NN as well as BP algorithms were used Radial growth distance was predicted as a function of angular distribution and years Radial growth distance was predicted as a function of angular distribution and years Short (3 years, for 2003) and long term (7 years, for 2000) predictions Short (3 years, for 2003) and long term (7 years, for 2000) predictions NEURAL NETWORK ALGORITHMS

22 SALNN Structure NEURAL NETWORK ALGORITHMS

23 BPNN Structure NEURAL NETWORK ALGORITHMS

24 For every center configuration, we produced the following outputs: For every center configuration, we produced the following outputs: - SALNN & BPNN long term prediction (2000 based on 1973 to 1993) - SALNN & BPNN long term prediction (2000 based on 1973 to 1993) - SALNN & BPNN short term prediction (2003 based on 1973 to 2000) - SALNN & BPNN short term prediction (2003 based on 1973 to 2000) NEURAL NETWORK GROWTH SIMULATION

25 SALNN vs. BPNN long term prediction (2000)/Center(a) SALNN BPNN SALNN BPNN Better Performance for SALNN for real data only Better Performance for SALNN for real data only Close performance at the third dataset with SALNN being better Close performance at the third dataset with SALNN being better BPNN didn’t perform well at real data BPNN didn’t perform well at real data Noticeable discrepancy between real and long-term predicted boundaries Noticeable discrepancy between real and long-term predicted boundaries RESULTS (1)

26 SALNN vs. BPNN short term prediction (2003)/Center(a) SALNN BPNN SALNN BPNN Better match between predicted and real boundaries Better match between predicted and real boundaries SALNN perform better than BPNN for all of the three data sets SALNN perform better than BPNN for all of the three data sets RESULTS (2)

27 SALNN vs. BPNN long term prediction (2000)/Center(b) SALNN BPNN Some effect of center is clear on the predicted results Some effect of center is clear on the predicted results Third dataset produce the best results Third dataset produce the best results SALNN performs better SALNN performs better RESULTS (3)

28 SALNN vs. BPNN short term prediction (2003)/Center(b) SALNN BPNN Better performance for SALNN Better performance for SALNN Center effect is less than for long term prediction Center effect is less than for long term prediction RESULTS (4)

29 Urban growth rate is faster in certain directions due to driving factors such as development probability Urban growth rate is faster in certain directions due to driving factors such as development probability Weighted radial growth as a function of radial measurement and growth direction was used Weighted radial growth as a function of radial measurement and growth direction was used Threshold should be met to implement the weighted growth modification Threshold should be met to implement the weighted growth modification Better results were obtained were the real boundaries match the predicted ones very closely Better results were obtained were the real boundaries match the predicted ones very closely WEIGHTED NN URBAN GROWTH

30 Weighted SALNN short term prediction (2003) Weighted SALNN short term prediction (2003) Center (a) Center (b) Center (a) Center (b) Very close match between real and predicted boundaries Very close match between real and predicted boundaries The effect of center on prediction results minimized The effect of center on prediction results minimized RESULTS (5)

31 The best growth prediction for the two algorithms and centers achieved using the third dataset The best growth prediction for the two algorithms and centers achieved using the third dataset For both centers, the results showed that SALNN gave better results compared to the BPNN results For both centers, the results showed that SALNN gave better results compared to the BPNN results Under the limitation of the availability of the data the SALNN works better than BPNN. Under the limitation of the availability of the data the SALNN works better than BPNN. Results of predictions is somewhat independent on the centers location especially for the third dataset Results of predictions is somewhat independent on the centers location especially for the third dataset Weighted NN results are the best in term of matching the real and predicted boundaries Weighted NN results are the best in term of matching the real and predicted boundaries RESULTS SUMMARY

32 SALNN algorithm produced better results than BPNN given the limited size of the available data SALNN algorithm produced better results than BPNN given the limited size of the available data Prediction results improved as the interpolation interval between the real data points gets smaller. Prediction results improved as the interpolation interval between the real data points gets smaller. City center location has certain effect on the predicted urban growth pattern City center location has certain effect on the predicted urban growth pattern Weighted NN improved the prediction results and minimized the effect of center location Weighted NN improved the prediction results and minimized the effect of center location CONCLUSIONS

33 CURRENT AND FUTURE WORK Urban growth prediction using (X,Y) coordinates of the boundaries Urban growth prediction using (X,Y) coordinates of the boundaries Weighted NN simulation for the fact that growth is not the same in all directions Weighted NN simulation for the fact that growth is not the same in all directions Urban growth errors NN training Urban growth errors NN training Growth errors statistical modeling as a function of the radial distance, time and angles of growth Growth errors statistical modeling as a function of the radial distance, time and angles of growth Cellular Automata and Fuzzy Logic simulation Cellular Automata and Fuzzy Logic simulation

34 QUESTIONS