inglada-AFrameworkfortheSimulationofHighTemporalResolutionImageSeries.pdf

A Framework for the Simulation of HighTemporal Resolution Image Series

J. Inglada, O. Hagolle, G. Dedieu

25/07/2011

J. Inglada, O. Hagolle, G. Dedieu IGARSS’11, Vancouver 25/07/2011 1 / 33

Outline

1 Introduction

2 Models

3 Example of application

4 Conclusions and future work


Introduction

Outline

1 Introduction

2 Models




Introduction

New sensors

Ï VenusÏ Sentinel (1,2,3)Ï LDCMÏ New applications . . .

... which require to closely monitor the temporal trajectory of thecharacteristics of land surfaces.

Ï real time classificationÏ evolving nomenclatures


Introduction

New sensors

Ï VenusÏ Sentinel (1,2,3)Ï LDCMÏ New applications . . .

... which require to closely monitor the temporal trajectory of thecharacteristics of land surfaces.

Ï real time classificationÏ evolving nomenclatures


Introduction

The VENµS mission

Ï France/Israel cooperationÏ 11 spectral bands (VIS, NIR)Ï 10 m. resolutionÏ 2 day revisit cycle (limited number of sites)


Introduction

Sentinel-2

Ï Optical HR component of ESA’s Sentinel ProgrammeÏ 13 spectral bands (VIS, NIR, SWIR)Ï 10/20/60. m resolutionÏ Earth coverage every 5 days (with 2 sats)


Introduction

Challenges

Ï From the annual classification . . .

Ï ... to the dynamic classification

Inter-crop Stubble disking Deep ploughing

Harrowing Sowing Emergence


Introduction

Challenges

Ï From the annual classification . . .

Ï ... to the dynamic classification

Inter-crop Stubble disking Deep ploughing

Harrowing Sowing Emergence


Introduction

Focus

Applications

Ï Global coverage every few daysÏ Expectations for land cover change monitoringÏ Real-time: update the land-cover maps for every new acquisition

MethodsÏ Describe temporal evolutionsÏ Choose and combine different data sourcesÏ Integration of prior knowledge


Introduction

Objectives

Ï Develop algorithms for high temporal and high spatial resolution imagetime series

Ï Evaluate and compare:Ï algorithmsÏ the sensors

Ï Need for realistic data which are representative of sensors which donot exist

Ï Use physical models as simulation tools


Models

Outline

1 Introduction

2 Models




Models

Essential Climate Variables

Ï For climate change assessment, mitigation and adaptation:Ï River discharge,Ï Water use,Ï Groundwater,Ï Lakes,Ï Snow cover,Ï Glaciers and ice caps,Ï Permafrost,Ï Albedo,Ï Land cover (including vegetation type),Ï Fraction of absorbed photosynthetically active radiation (FAPAR),Ï Leaf area index (LAI),Ï Above-ground biomass,Ï Fire disturbance


Models

Models - Scope

Ï They describe the physical realityÏ Their assumptions/simplifications are clearÏ Naturally use/need ancillary data (meteo, ground measures)

Ï They can be multi-sensor or better . . .. . . Sensor Agnostic

Ï benefit from the synergy between sensorsÏ increase temporal sampling!


Models

Models - Scope

Ï They describe the physical realityÏ Their assumptions/simplifications are clearÏ Naturally use/need ancillary data (meteo, ground measures)

Ï They can be multi-sensor or better . . .. . . Sensor Agnostic

Ï benefit from the synergy between sensorsÏ increase temporal sampling!


Models

Models - Challenges

Ï Areas of interest:Ï hydrology, agriculture, forestry,

Ï Media:Ï Aerial, terrestrial, aquatic, mixed

Ï How to find the good balanceÏ complexity,Ï number of input parameters and variables,Ï computational cost


Models

Open source models - someexamples

Ï Prospect: optical model for estimating leaf-level reflectance andtransmittance

Ï Sail: canopy reflectance modelÏ Daisy: mechanistic simulation model of the physical and biologicalprocesses in an agricultural field

Ï 6s: a basic RT code used for calculation of look-up tables in theMODIS atmospheric correction algorithm

Ï Arts: radiative transfer model for the millimeter and sub-millimeterspectral range.

Ï etc.Ï have a look at ecobas.org


Example of application

Outline

1 Introduction

2 Models





Purpose

Which is the best sensor to recognize these:

60

40

20

0

0,4 0,6 0,8 1,0 1,2 1,4 1,6 1,8 2,0 2,2 2,4 2,6

sol nu sec

végétation

eau

visible proche infrarouge moyen infrarouge

Longueur d'onde (µm)

Ré

fle

cta

nce

(%

)



Purpose

Ï Or these



Principle



Results

0

0.2

0.4

0.6

0.8

1

Vegetation

Soils

Man-m

ade

Minerals

Acc

urac

y

Spot 5QuickbirdPleiades

Landsat TMIkonos

FormosatMeris



Results

0

0.2

0.4

0.6

0.8

1

RoadConcretes

Constructions

RoofIgneous

Metam

orphic

Sedimentary

Alfisol

Aridisol

Entisol

Inceptisol

Mollisol

Acc

urac

y

Spot 5QuickbirdPleiades

Landsat TMIkonos

FormosatMeris



But we said HTR . . .

Ï How to simulate a multi-t mission?Ï Venus, Sentinel-2

Ï Realistic temporal evolutionsÏ Use existing image time series

Ï Formosat-2Ï 8 m., 4 bands (B,V,R,NIR), 3 days



Example of series

March 14, 2006



Example of series

July 17, 2006



Example of series

November 2, 2006



Available data

Ï 49 images in 2006Ï Orthorectification OKÏ Radiometric corrections OK

Ï TOC and aerosol correctionsÏ Cloud screening

Ï Land-cover map availableÏ Leaf pigments data base for several vegetation types (LOPEX’93)



Spectral responses

500 1000 1500 2000wavelength

0.0

0.2

0.4

0.6

0.8

1.0Fo

rmos

at-2

Relative Spectral Responses

500 1000 1500 2000wavelength

0.0

0.2

0.4

0.6

0.8

1.0

Venu

s

500 1000 1500 2000wavelength

0.0

0.2

0.4

0.6

0.8

1.0

Sent

inel

-2



Simulator architecture

Formosat-2 Input Series

LAI (t)

Land Cover Map

Cab

Car

N

PROSPECT+SAIL Full Spectra Venµs RSR

Formosat-2 RSR

Sentinel-2 RSR




Mar 2006

Apr 2006

May 2006

Jun 2006Jul 2006

Aug 2006

Sep 2006

Oct 2006

Nov 20060.0

0.2

0.4

0.6

0.8

1.0

Cloud %40 dates30 dates

0 10 20 30 40 50Number of dates

0.40

0.45

0.50

0.55

0.60

0.65

0.70

0.75

0.80

Kapp

a In

dex

FSAT-2VenusSentinel-2


Conclusions and future work

Outline

1 Introduction

2 Models





As a conclusion

Ï New missions in the coming yearsÏ Venus, Sentinel, LDCM

Ï Nowadays: Formosat-2Ï How to prepare the use of future systems

Ï Algorithm design and validationÏ Understanding phenomena

Ï Use of simulationÏ Completely syntheticÏ From very high resolution(s) dataÏ The third way!

Ï Use real time seriesÏ but with lower resolutions

Ï Use physical models



What we’ve got

Ï Source code available for many simulatorsÏ Ongoing work for

Ï Prospect, Sail & Daisy integrationÏ new hyper/multi- spectral/temporal algorithm integration

http://www.orfeo-toolbox.org


http://www.orfeo-toolbox.org


What we need

Engineering - Development

Ï Improve image simulation: MTF, realistic landscapesÏ Hide physical models under common interfaces

ResearchÏ Learn to select the best model set for a given problemÏ Incorporate domain expert knowledge



Creative Commons Attribution-ShareAlike 3.0 Unported License


inglada-AFrameworkfortheSimulationofHighTemporalResolutionImageSeries.pdf

Business

dedieu igarss11

future work

use physical models

models open source models

optical model

landcover maps

physical reality

introduction challenges