Inverse method for development of an advanced vegetation index · Inverse method for development of...

1Envisat Summer School – NG 2

Inverse method for development of an advanced vegetation index

Nadine Gobron

With collaborations of:Bernard Pinty, Michel Verstraete & Jean-Luc Widlowski

What is the main scientific issue in remote sensing?

• Remote sensing instruments on space platforms are but sophisticated detectors recording the occurrence of elementary events (the absorption of photons or electromagnetic waves) as variations in electrical currents or voltages.

• Users, on the other hand, require information on events and processes occurring in the geophysical environment (e.g., atmospheric pollution, oceanic currents, terrestrial net productivity, etc.)

• Extracting useful environmental information from data acquired in space is the main challenge facing remote sensing scientists.

How is remote sensing exploited?

• The signals detected by the sensors are immediately converted into digital numbers and transmitted to dedicated receiving stations, where these data are heavily processed.

• The effective exploitation of remote sensing data to reliably generate useful, pertinent information hinges on the availability and performance of specific tools and techniques of data analysis and interpretation.

• A variety of mathematical models can be used for this purpose; they are implemented as computer codes that read the data or intermediary products and ultimately lead to the generation of products and services usable in specific applications.

Why is remote sensing called an inverse problem?

• Since space borne instruments can only measure the properties of electromagnetic waves emitted or scattered by the Earth, scientists need first to understand where these waves originate from, how they interact with the environment, and how they propagate towards the sensor.

• To this effect, they develop models of radiation transfer, assuming that everything is known about the sources of radiation and the environment, and calculate the properties of the radiation field as the sensor should measure them. This is the so-called direct problem.

Why is remote sensing called an inverse problem?

• In practice, one does acquire the measurements from the satellite, and would like to know what is going on in the environment:

à This is the inverse problem, which is much more complicated: Knowing the value of the electromagnetic measurements gathered in space, how can we derive the properties of the environment that were responsible for the radiation to reach the sensor?

Measurements of interpretation (1)

• A model representing a measurement expresses the dependency of this measurement with respect to the relevant variables and processes:

where sm are the state variables of the system

• Direct problem: if values sm are know, such a model can accurately simulate the observation

• Inverse problem: Quantity is measured, and one seeks information on the states variables sm

),...,( 21 msssfz =

Measurements of interpretation (2)

• If a single state variable accounts for the physics of the measurement, the problem can be solved analytically or numerically, usually with great accuracy:

• In general, more than one state variable is required to describethe physics of the problem. The solution to the inverse problem then requires multiple equations and therefore multiple measurements.

)()( 11 zfssfz )) −=⇒=

Role of independent variables (1)

• To acquire more information on an invariant system, it is necessary to gather different measurements by changing a variable other than the state variables.

where the independent variable x describes the changing conditions of observations.

• If the system of interest can be observed in more than one way, multiple independent variables may be defined.

• In case of RS from space, the useful independent variables are space, time, wavelength, illumination and observation geometry.

),...,;( 21 msssxfz =

Role of independent variables (2)

• Multiple measurements may thus be aquired under different conditions of observation:

sssxxxfz

),...,;,...,,(

Association of physical measurements, models & models

representing the biosphere

The process of fitting data, as seen by Subramanian Raman in Science with a Smile.

Model Inversion

• In general, the system of equations cannot be solved:

o If K < M, the system is still underterminedo If K=M, measurements errors may prevent the

identification of the exact solution.o If K>M, the system is overdetermined.

K depends on space observation strategy

M depend on the RT model

Space Observation Strategies

• Mono-directional multi-spectral scanning sensor on a Sun-synchronous orbit

single view of the same target during a given day (given Sun zenith angle)

Ref: http://rst.gsfc.nasa.gov

Source: http://envisat.esa.int/

• Spinning sensor on a geostationary orbit

multiple views of the same target during a given day (various Sun zenith angles)

• Multi-directional multi-spectral sensor on a Sun-synchronous orbit

Overview of MISR

• 9 cameras at ±70.5, ±60, ±45.6, ±26.1, 0°

• Each camera at 446, 558, 672, and 866 nm

• Spatial resolution: 275 m (250 m nadir)

• Global mode: Full res. nadir and red, 1.1 km otherwise

• Local mode: Full resolution all cameras and all bands

• Swath: 360 km• Coverage: global (9 days)

Ref: http://www-misr.jpl.nasa.gov/mission/minst.html

28Envisat Summer School – NG 2 Credit: NASA/GSFC/LaRC/JPL MISR Team

Multiangle animation

Emigrant Gap Fire,California

13 August 2001

9/19/2004 29Envisat Summer School – NG 2

Design of the MERIS Global Vegetation Index

decontaminated from atmospheric and geometrical effects

MERIS/ENVISAT

• Passive optical instrument of Earth Observation• Primary mission: Ocean productivity• Secondary missions: Atmosphere and

land surface characterization• Ground segment support (up to L2)

• Global coverage: = 3 days (depends on latitude)• Swath: 1150 km• Spatial resolution: ± 300 m (FR) & ± 1200 m (RR)

• Spectral band positions, widths and gains are programmable• Radiometric and spectral calibration on-board mechanisms

(white & pink Spectralon, Fraunhofer lines)

Source: http://envisat.esa.int/instruments/meris/

Nominal MERIS spectral bands

Water vapour, land1090015

Atmospheric corrections1088514

Vegetation, water vapour reference2086513

Atmospheric corrections15778.7512

Oxygen absorption R-branch3.75760.62511

Vegetation, clouds7.5737.7510

Fluorescence, atmospheric corrections10708.759

Chlorophyll fluorescence peak7.5681.258

Chlorophyll absorption and fluorescence106657

Suspended sediment106206

Chlorophyll absorption minimum105605

Suspended sediment, red tides105104

Chlorophyll and other pigments104903

Chlorophyll absorption maximum10442.52

Yellow substance and detrital pigments10412.51

ApplicationsWidth [nm]Location [nm]Band

Ref: http://envisat.esa.int/instruments/meris/descr/concept.html

Objective•To provide useful, quantitative, reliable, accurate and low cost information on the state of terrestrial vegetation using remote sensing data

•Development of Optimized Vegetation Indices using a physically-based approach

Approach

Scientific Constraints

• Maximal sensitivity to the Fraction of Absorbed Photosynthetically Active Radiation (FAPAR)

• Minimal sensitivity to atmospheric perturbations

• Minimal sensitivity to soil colour and brightness changes

• Minimal sensitivity to angular effects

SeaWiFS

MODISMERIS

Spectral Sensor Properties

Spectral Leaves Profile

Prospect Model

SeaWiFS

MODISMERIS

Semi-discret Model

SeaWiFS

MODISMERIS

Bidirectional Reflectance Factor at Top Of Canopy : Nadir View

Bidirectional Reflectance Factor at Top Of Canopy

Semi-discret Model

lai =8.lai=5.

lai=4.lai=3.

lai=2.lai=1.5

lai=1.lai=0.5

SeaWiFS

MODISMERIS

Visibility : 151.58 kmVisibility : 38.33 kmVisibility : 20.22 kmVisibility : 7.74 kmVisibility : 5.11 km

Bidirectional Reflectance Factor Top Of Atmosphere: Aerosols Optical Depth

6S Model

SeaWiFS

MODISMERIS

Operational Constraints

• Exploit only the target’s spectral variations as measured by the satellite instruments

• Take the actual spectral response of the sensor into account

• Minimise the computational load• Require data from a single orbit only

Scheme for the algorithm optimization (1)

Radiation Transfer ModelsRadiation Transfer Models

BRF TOABRF TOA BRF TOCBRF TOC FAPARFAPAR

Mathematical Implementation (1)

Medium Variables Range of values Atmosphere τs 0.05, 0.3 and 0.8 Vegetation LAI

0, 1, 2, 3, and 5 0.5 and 2 m 0.01 and 0.05 m Erectophile, Planophile

Soil rs 5 Soil spectra (Price, 1995)

Geophysical properties

Illumination and observation geometries

Angle Values Solar zenith (θ0) Satellite zenith (θv) Sun-Satellite relative azimuth (∆φ)

20 and 50 degrees 0, 25 and 40 degrees 0, 90 and 180 degrees

Models: • 6S atmospheric radiation transfer model (Vermote et al., 1997)• Semi-discrete vegetation radiation transfer model (Gobron et al., 1997)

RT model driven improvement for FAPAR

Rectified RedRectified Red

Angular/Atmospheric effectsAngular/Atmospheric effects

Rectified NIRRectified NIR

),,,,(),,(

icHGiiv

i kF λλλ ρλρ

λρΩΩΩΩΩ

?0 - controls amplitude levelk - controls bowl/bell shapeT - controls forward/backward scattering?C - controls hot spot peak

BRF(z,O0 O) = ?0 MI(k) FHG(T ) H (?c)

The RPV parametric model

Ref: Rahman et al. (1993) JGR

sI kM −

−−

)cos(coscoscos

);,(θθ

θθθθ

),,;,(~),,,;,( 0000 kzkz cscs ΘΩ→Ω=ΘΩ→Ω ρρρρρρ

);(),();,(),,;,(~0 GHgFkMkz cHGsIcs ρθθρρ Θ=ΘΩ→Ω

]cos21[1

),(Θ+Θ+

Θ−=Θ

c +−

1);(ρ

2/122 ]costantan2tan[tan

cossinsincoscoscos

φθθθθ

• Direct Mode

• Inverse Method when mono-angle data (MERIS)

Ø k, Θ, ρc parameters are optimized with simulated-data for each wavelenght at TOA & TOC: representative values of anisotropic atthe global scale.

sI kM −

−−

)cos(coscoscos

);,(θθ

θθθθ

),,;,(~),,,;,( 0000 kzkz cscs ΘΩ→Ω=ΘΩ→Ω ρρρρρρ

);(),();,(),,;,(~0 GHgFkMkz cHGsIcs ρθθρρ Θ=ΘΩ→Ω

]cos21[1

),(Θ+Θ+

Θ−=Θ

c +−

1);(ρ

cossinsincoscoscos

φθθθθ

• Direct Mode

• Inverse Method when multi-angles data (MISR, SPECTRA)

Ø Resolution of second order equations (Gobron and Lajas, 2002)Ø Users’ controlled accuracy of the fit (model/data)Ø Retrieval of a set of acceptable solutionsØ Selection on the most representative solutions

Ø Resolution of a linear system (Engelsen et al, 1996)Ø No predefined accuracyØ Only one solutionØ Very fast method

Parametric Model: MRPV

• Direct Mode

• Inverse Method

sI kM −

−−

)cos(coscoscos

);,(θθ

θθθθ

),;,(~),,;,( 0000 kbzkbz msms Ω→Ω=Ω→Ω ρρρρ

);();();,(),,;,(~0 GHbgFkMkbz msIms ρθθρρ =Ω→Ω

)cosexp();( gbbgF mm −=

1);(ρ

cossinsincoscoscos

φθθθθ

Rectified NIRRectified NIR)](),([

1 redbluRred g λρλρρ = )](),([~~

2 nirbluRnir g λρλρρ =

),,,,(),,(

icHGiiv

i kF λλλ ρλρ

λρΩΩΩΩΩ

= TOCiblui i

g λρλρλρ ~)](),([~~

• Ratios of polynomials are assumed to be appropriate to generate the “rectified channels” :

)ly,x,(f)yx,(g mn,n =

Where ln,m are the m coefficients of the polynomials, and x and y are either the measured or rectified spectral reflectances,depending on the step

Cost Functions (1)

• The optimisation procedure is applied to the RED and NIR bands to find the coefficients of g1 and g2:

0]~))(~),(~([2 →−ΩΩ∑= TOCjjbluig ???g

xyl)ly(l)lx(lxyl)l(yl)lx(l

)y,x(g10,2

29,28,2

27,26,2

4,23,22

2,21,22

++++++++

xyl)ly(l)lx(l)y,x(g 5,12

4,13,12

2,11,11 ++++=

Rectification Results in the Red Band

Before After

Rectification Results in the NIR Band

Before After

Scheme for the algorithm optimization

),,,,(),,(

icHGiiv

i kF λλλ ρλρ

λρΩΩΩΩΩ

= TOCiblui i

g λρλρλρ ~)](),([~~

Optimization Procedure

FAPAR),(0 →RnirRredg ρρ

• The index formula is generated on the basis of the rectified channel values, ρRred and ρRnir,estimated as follow:

)~,~(g 1 redbluRred ??? = )~,~(g 2 nirbluRnir ??? =

),(0 RnirRredgMGVI ρρ=

Cost Functions (2)

• The coefficients of g0 are optimised so that:

0]FAPAR)([ ,02

→−∑= RnirRredg

3,02,01,0

l)yl()xl(lxlyl

)y,x(go

+−+−−−

MGVI Optimization

Before After

RT model driven improvement for FAPAR

Scheme for the algorithm optimization

030201

),(FAPAR

))(())((),(

),(/),()](),([

llllll

RnirRred

RredRnir

RnirRred

njnninji

jijijin

+−+−−−

ρρρρ

λρλρ

λρλρλλ

λρλρ

λρλρλλ

λλλλλρλρ

),,,,(),,(

icHGiiv

i kF λλλ ρλρ

λρΩΩΩΩΩ

= TOCiblui i

g λρλρλρ ~)](),([~~

Optimization Procedure

FAPAR),(0 →RnirRredg ρρ

Case a Case b

Case c Case d

Sensitivity to scenes heterogeneity

MERIS Global Vegetation Index

Rectified Red Rectified NIR

Angular/Atmospheric `Rectification'

)](),([~~

),,,,(),,(

i kF λλλ ρλρ

λρΩΩΩΩΩ

Fapar),(0 RnirRredg ρρ=MGVI

BRF TOA View angles Sun angles

Satellite Data

L2 products

Conclusions (1)

• Retrieval geophysical parameters is a coupled atmosphere-vegetation problem.

• Inverse methods depend on the space remote sensing data types (multi/mono spectral & multi/mono angular).

• Retrieval of geophysical parameters values are always associated to accuracies.

• MGVI (instantaneous) can be read in the LEVEL 2 land product (TOAVI) and represents FAPAR.

Conclusions (2)

• Remote sensing products require a validation exercise through:

• Ground measurements• Inter-comparison with other sensor derived products

• Temporal & spatial scale production also depend on specific applications (global or regional).

Inverse method for development of an advanced vegetation index · Inverse method for development of...

Documents

Section 4-5 Inverse Functions · Inverse Notes •The inverse of a function f is written −1and is read “f inverse” • −1( )is read, “f inverse of x” •If point ( , )is

FUNCTIONS AND INVERSE FUNCTIONSgifs.africa/.../uploads/2020/06/Grade-12-Mathematics-Inverse-Functi… · MATHEMATICS GRADE 12 INVERSE FUNCTIONS 1 FUNCTIONS AND INVERSE FUNCTIONS A

Lecture 6 - Section 7.7 Inverse Trigonometric Functions Section 7… · 2008-02-05 · Inverse Trig Functions Hyperbolic Sine and Cosine Inverse Sine Inverse Tangent Inverse Secant

Inverse Functions – Worksheet Partmrsstillman.com/Mrs._Stillmans_Class/A2H_files/Inverse Functions … · 10. The inverse is a function because f(x) passes the HLT. 11.The inverse

Inverse Problem-- Leaks Carrie Nugent December 6th, 2007 Inverse Theory Carrie Nugent December 6th, 2007 Inverse Theory

Radiation transfer model intercomparison (RAMI) exercise · Radiation transfer model intercomparison (RAMI) exercise Bernard Pinty, 1 Nadine Gobron, 1 Jean-Luc Widlowski, 1 Sigfried

Technical Datasheet - Veracious Inc · Inverse Characteristics Curve [Over Current IDMT]: Very Inverse Long Inverse Standard Inverse Extremely Inverse α C 0.02 1 2 1 0.14 13.5 80

Section 9.1 – Inverse Functions. DOES an inverse function exist? IF YES, you can find the inverse function

GLOBAL LEAF AREA INDEX PRODUCT VALIDATION - … · GLOBAL LEAF AREA INDEX PRODUCT VALIDATION Good Practices ... Sylvain LeBlanc, ... Jean-Luc Widlowski, Miguel Roman,

Inverse and Optimization Problems in Heat Transfer Inverse ... · Inverse and Optimization Problems in Heat Transfer This paper presents basic concepts of inverse and optimization

Inverse Stieltjes-like Functions and Inverse Problems for Systems

4.7 Inverse Trigonometric Functions inverse trig.pdfInverse Trigonometric Functions NASA 4.7 Definition of Inverse Sine Function The inverse sine functionis defined by if and only

4.7 Identity and Inverse Matrices -Identity matrices -Inverse matrix (intro) -An application -Finding inverse matrices (by hand) -Finding inverse matrices

Algorithmic properties of tree-like inverse monoids · Nóra Szakács Algorithmic ropperties of tree-like inverse monoids. Inverse monoids and languages ree-likTe inverse monoids

Math for Game Programmers: Inverse Kinematics - …dtecta.com/files/GDC14_VanDenBergen_Gino_Math_Tut.pdfMath for Game Programmers: Inverse Kinematics ... Inverse Kinematics? Problem

Inverse Variation Investigate inverse variation through real- world problems Learn the basic inverse variation equations Graph inverse variation functions

Inverse Reinforcement Learning - Peoplecbfinn/_files/bootcamp_inverserl.pdf · Apprenticeship Learning via Inverse Reinforcement Learning. Good introduction to inverse reinforcement

Inverse Modelling and Inverse Simulation for System ...theses.gla.ac.uk/2/1/2007luphd.pdf · Inverse Modelling and Inverse Simulation for System Engineering and Control Applications

SECTION 6-D Inverse Trig Derivatives. Inverse Trig functions

Inverse and Optimization Problems in Heat Transfer Inverse and