Evaluation of an OBIA approach to extract basic urban LC

information from high-resolution optical and LiDAR data

AK Fernerkundung ∙ RU Bochum ∙ October 05, 2012 Christian Berger

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 2 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 2

Outline

1 Introduction

2 Materials & Methods

3 Results & Discussion

4 Summary

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 3 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 3

Introduction

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 4 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 4

Background & Study Objectives

ENVILAND 2/urban WP

develop a method to extract

urban LC from high-res data

six basic target classes

data fusion to exploit multi-

sensoral EO datasets

state-of-the-art algorithms

for practical problems

demonstration of spatio-

temporal transferability

Introduction 1

Urban Density based on LC & height infos.

ENVILAND 2 www.enviland.de

E2 project concept & project partners.

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 5 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 5

Materials & Methods

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 6 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 6

Study Areas & Data Basis

urban LC maps were produced for three (entire) cities

different sets of MS imagery and LiDAR data were used

Materials & Methods 2

Rostock, MV, GER. San Francisco, CA, USA. Erfurt, TH, GER.

Data basis of this study.

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 7 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 7

Overall Workflow

data preprocessing

ATCOR, panfusion, coregistration

derivation of DEMs/DSMs/nDSMs

generation of additional information layers

image brightness, NDVI, nDSM slope

Materials & Methods 2

Workflow of this study.

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 8 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 8

Overall Workflow

extraction of urban LC information

using OBIA and feature fusion

compilation of simple class descriptions

Materials & Methods 2

Workflow of this study. The urban LC classification scheme proposed in this study.

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 9 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 9

Overall Workflow

accuracy assessment

50 random points per class

standard accuracy measures

reference: DOPs & GIS data

transferability assessment

discussion on strenghts and

limitations of the approach

Materials & Methods 2

Workflow of this study.

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 10 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 10

Results & Discussion

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 11 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 11

Urban LC Maps

an overall area of almost 700 km² was classified

Results & Discussion 3

The urban LC maps of

Rostock: QuickBird (left)

and RapidEye (right).

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 12 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 12

Urban LC Maps

an overall area of almost 700 km² was classified

Results & Discussion 3

The urban LC

maps of Erfurt

and San Francisco.

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 13 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 13

Validation Results

The classification accuracies obtained for the urban LC maps.

high overall degree of accuracy across all maps

method performs better for Rostock & San Francisco

UA: 72–100%; PA: 68–100%; OA: 83–91%; Kappa: 0.80–0.89

Results & Discussion 3

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 14 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 14

Major Sources of Errors

ruleset simplicity

extensive shadowing from buildings

bare soil areas vs impervious surfaces

Results & Discussion 3

The urban LC classification

scheme proposed in this study.

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 15 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 15

Major Sources of Errors

nDSM properties (Erfurt)

forests and forested areas

allotment garden cottages

Results & Discussion 3

The urban LC classification

scheme proposed in this study.

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 16 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 16

Methodical Transferability

strengths

simple class descriptions; (re)use of basic features

utilization of the four most common spectral bands

modular ruleset structure; no training data required

Results & Discussion 3

Excerpts from the ruleset.

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 17 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 17

Methodical Transferability

limitations

some ruleset adaptations are necessary

initial setup of variables (MMU, thresholds)

compensating for data properties (nDSM)

Results & Discussion 3

Excerpts from the ruleset.

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 18 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 18

Methodical Transferability

The percentage differences in accuracy (“reference map“: Rostock (QuickBird)).

the presented approach seems promising for…

capturing urban areas with different physical structures

integrating various sets of optical & LiDAR inputs

dealing with changes in illumination and phenology

Results & Discussion 3

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 19 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 19

Summary

Berger et al. | Mapping of Urban Structure Types | 3rd RSS ∙ Wageningen, NL ∙ September 8, 2011 20 Christian Berger | FSU Jena | Evaluation of an OBIA approach for basic urban LC mapping 20

Wrap-Up

a simple OBIA approach for

basic urban LC mapping

application to HR data

acquired over 3 urban areas

user’s/producer’s accuracies

are mainly above 80/90 %

approach can be used as

template or starting point

Geoportal

sf.maps.essi-blog.org

Summary 4

The urban LC mapping result for SF online (J. Eberle).

Contact

Christian Berger

Department for Earth Observation

Institute of Geography

Friedrich-Schiller-University Jena

Löbdergraben 32

D–07743 Jena

Fon: +49(0)3641 948974

Fax: +49(0)3641 948882

E-mail: christian.berger@uni-jena.de

Thank you!