Detecting Spatioemporal Dynamics in Satellite Remote Sensing Time Series: Methodological Approach...

DETECTING SPATIOTEMPORAL DYNAMICS IN SATELLITE REMOTE SENSING TIME SERIES: METHODOLOGICAL APPROACH COMBINING OBIA AND DATA MINING TECHNIQUES

Dino Ienco

dino.ienco@irstea.fr

Researcher @Irstea (UMR-TETIS)

Fabio N. Güttler

Jordi Nin,

Pascal Poncelet,

Maguelonne Teisseire

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

I. Climate Change & Remote Sensing

II. Introduction

III. Study Area

IV. Methods

V. Results

VI. Conclusions / Perspectives

OUTLINE

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

I. CLIMATE CHANGES & REMOTE SENSING

Climate changes :

• Important Issue at global/local scale

• Impacts on various sectors: agriculture, fishery,

forestry, water management, health, coastal

environment, natural habitats, etc…

Study what happens in the past to

understand how to act in the future

Remote Sensing analysis constitutes an important tool to

study what happens:

- Capture area characterisitcs (spectral/radiometric)

- Huge quantity of images available soon

- Acquisition of time series over the same area

I. CLIMATE CHANGES & REMOTE SENSING

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

Series of satellite images can help to understand long term behaviors

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

I. CLIMATE CHANGES & REMOTE SENSING

Mining series of satellite images can help to:

Automatically analyse huge data

Study long term behaviour

Extract evolutions of natural/artificial habitats

Methodology to Mine Evolutions/Changes on

Remote Sensing time series data

II. INTRODUCTION

• Spectral response of natural vegetation

• Vegetation types

• Textural response

• Vegetation structure (mapping physiognomic

classes, e.g. shrubland and grassland)

• Temporal response

• Vegetation phenology

and dynamics

MOTIVATIONS – REMOTE SENSING POSSIBILITIES

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

II. INTRODUCTION

• Detect spatiotemporal evolutions on satellite time series automatically

• Provided useful information for natural habitats monitoring and mapping

• Propose a new and adapted method for future high repetitivity RS time series

(i.e. Sentinel-2) that can be employed to understand Climate Changes

OBJECTIVES

Methods describing multi-temporal behaviour are among open

challenges in OBIA (Blaschke et al. 2014 ; Chen et al. 2012)

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

[Blaschke et al 2014] Blaschke et al. Geographic object-based image analysis towards a new paradigm. ISPRS Journal of

Photogrammetry and Remote Sensing 87 (0), 180–191.

[Chen et al. 2012] Chen, G., Hay, G. J., Carvalho, L. M. T., Wulder, M. A., 2012. Object- based change detection.

International Journal of Remote Sensing 33 (14), 4434–4457.

III. STUDY AREAS AND TIME SERIES

Low Aude Valley

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

Area: 4,842 ha

Dominated by

natural habitat

types of

Community

interest (19 in

total)

III. STUDY AREAS AND TIME SERIES

Low Aude Valley

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

IV. PROPOSED METHODOLOGY

MOTIVATIONS

- Approach coupling OBIA (Object-based Image

Analysis) and Data Mining

- Method ables to track object evolution along the time

- Main assumption - Maximal Spatial Extent: For each complex habitat there will be a

timestamp, in the time series, in which we can

observe its maximum extent

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

IV. PROPOSED METHODOLOGY

GENERAL SCHEMA

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

MAIN STEPS

Step 1 •Preprocessing

Step 2

Step 3

Step 4

Step 5

Step 6

- data standardization

- fine geometrical registration

- spatial subset

- spectral indices

- …

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

IV. PROPOSED METHODOLOGY

Step 1 •Preprocessing

Step 2 •Segmentation

Step 3

Step 4

Step 5

Step 6

T0

T1

Tn

…

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

IV. PROPOSED METHODOLOGY

MAIN STEPS

Step 1 •Preprocessing

Step 2 •Segmentation

Step 3 •BB selection

Step 4

Step 5

Step 6

- Retrieve the Maximal Spatial Extents - On the whole study area

- Through the 6 timestamp

- The output is a list of objects

candidateBB

T0

T1

Tn

…

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

IV. PROPOSED METHODOLOGY

MAIN STEPS

Step 1 •Preprocessing

Step 2 •Segmentation

Step 3 •BB selection

Step 4

Step 5

Step 6

- Filtering of candidateBB - Obtain a global cover of the

study area

- Decrease the overlapping

among the BB (redundancy)

- Retrieve the Maximal Spatial Extents - On the whole study area

- Through the 6 timestamp

- The output is a list of objects

candidateBB

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

IV. PROPOSED METHODOLOGY

MAIN STEPS

Step 1 •Preprocessing

Step 2 •Segmentation

Step 3 •BB selection

Step 4 •Graph construction

Step 5

Step 6

- Each BB is exploited to build an

evolution graph

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

IV. PROPOSED METHODOLOGY

MAIN STEPS

Step 1 •Preprocessing

Step 2 •Segmentation

Step 3 •BB selection

Step 4 •Graph construction

Step 5

Step 6

T0

T1

T2

T3

T4

object ID

timestamp

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

IV. PROPOSED METHODOLOGY

MAIN STEPS

Step 1 •Preprocessing

Step 2 •Segmentation

Step 3 •BB selection

Step 4 •Graph construction

Step 5

Step 6

T0

T1

T2

T3

T4

object ID

timestamp

overlap between objects

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

IV. PROPOSED METHODOLOGY

MAIN STEPS

Step 1 •Preprocessing

Step 2 •Segmentation

Step 3 •BB selection

Step 4 •Graph construction

Step 5

•Measuring evolutions and similarity

- Evaluate the temporal evolution of a

natural habitat

- Evaluate the global intensity of

evolution of a graph

- graph ranking for each set of

chosen attributes

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

IV. PROPOSED METHODOLOGY

MAIN STEPS

V. RESULTS

- Total Number of Objects)

3 373 (~ 562 per timestamp)

- Total Number of generated graphs

BPA = 340

Landsat 2009

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

- Each node contains a set of

attributes

- Band1

- Band2

- …

- NDVI

- NDWI

- VSDI

- …

- The graphs are also described

by some features

- N. of nodes

- N. of edges

- N. of sequences

- … Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

V. RESULTS

EVOLUTION GRAPHS

EVOLUTION GRAPHS

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

V. RESULTS

INTENSITY OF EVOLUTION

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

V. RESULTS

VI. CONCLUSIONS

• The proposed methodology :

• Connects objects along the time series monitoring the same phenomenon over time

• Combines OBIA & Data Mining to extract spatiotemporal evolutions

• Can be easily applied to longer time series, no assumption on the number of images is made

• Can be used to generate a ranking of the habitats considering their changes during the time (and the

ranking can be exploited by expert to focus their attention on particular portion of the study area)

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

Tri-National Scientific Workshop, 29 October 2015, Bogor, Indonesia

Thank you for your attention

ANY QUESTION

NDVI and NDWI evolutions

22-Feb

NDVI and NDWI evolutions

27-Feb

NDVI and NDWI evolutions

03-Apr

NDVI and NDWI evolutions

13-Apr

NDVI and NDWI evolutions

18-Apr

NDVI and NDWI evolutions

23-Apr

NDVI and NDWI evolutions

13-May

NDVI and NDWI evolutions

02-Jun