Upload
makenna-cott
View
221
Download
0
Tags:
Embed Size (px)
Citation preview
1st year meeting --- Progress Report
wp-2: Satellite observations of vegetation cover, of surface albedo and temperature over the Plateau
Guangjian Yan
State Key Laboratory of Remote Sensing Science,School of Geography, Beijing Normal University
2009-6-29
Objectives
Develop algorithms to retrieve vegetation cover, LAI, surface albedo, emissivity and land surface temperatures from measurements by polar and/or geostationary satellites;
Produce a consistent data set for the land surface variables over the region of Tibetan Plateau.
Partners LSIIT - Université Louis Pasteur (ULP) The International Institute for Geo-information Scien
ce and Earth Observation (ITC) Alterra Green World Research - Wageningen Univer
sity and Research Centre (Alterra) University of Valencia(UVEG) Beijing Normal University(BNU) Institute of Geographic Sciences and Natural Resou
rces Research (IGSNRR) Institute of Remote Sensing Applications(IRSA)
Tasks Task 2.1. A generic algorithm for the retrieval of important surface pr
operties from a multitude satellite sensors (ULP, ITC, IRSA)
Task 2.2 Developing an algorithm for simultaneous retrievals of atmospheric variables, land surface variables using AATSR bi-angular/multi-spectral radiometric observations (Alterra, IGSNRR).
Task 2.3. Improving MODIS LAI and albedo products by combining remote sensing model and dynamic process model (BNU, IRSA)
Task 2.4. Developing an algorithm to produce a consistent LST from polar satellites (LSIIT, UVEG, BNU, IGSNRR, IRSA).
Task 2.5. Estimation of LSE, LST and albedo from Geostationary Satellite (GS) data (LSIIT, Alterra, IGSNRR)
Task 2.1. A generic algorithm for the retrieval of important surface properties from a multitude satellite sensors
Model Algorithm data parameter
Optical/Laser
Albedo
Baysian inversion
Elevation
Synergy of active and passive optical data
Optical
Coupled RT modeling
LAI
Albedo modling for terrain
Albedo retrieval algorithm
★ ★★ Partly finished
★
Task 2.2. Developing an algorithm for simultaneous retrievals of atmospheric variables, land surface variables using AATSR data
Model Algorithm data parameter
LSCT FVC
Algorithm developing for AATSR
Albedo
AATSR
LAI Water vapour AOD
Task 2.3. Improving MODIS LAI and albedo products by combining remote sensing model and dynamic process model
Model Algorithm data parameter
MISR TM
Albedo fCover
Angular effect correction
Assimilation method
LUT based RT model inversion
HJ-1
LAI trend
LAI
MODIS
ASTER
A priori knowledge
based inversion
★ ★ ★★ Partly finished
★ ★
Task 2.4. Developing an algorithm to produce a consistent LST from polar satellites
Model Algorithm data parameter
TDC modeling
Temporal normalization
of LST
AMSR-EHJ-1
LST
Algorithm developing
for HJ-1
RT modeling for canopy and terrain
MODIS
LST retrieval algorithms for all
weather conditions
★ ★★ Partly finished
★
Task 2.5. Estimation of LSE, LST and albedo from Geostationary Satellite (GS) data
LSTAlbedo
Split-window algorithm adaption
Atmospheric correction scheme using TDC model and ECMWF data
ECMWF standard output
Geostationary satellite
Albedo retrieval algorithm
Model Algorithm data parameter
Main goals: measuring ice sheet mass b
alance, cloud and aerosol heights, land topography and vegetat
ion characteristics.
Organization: NASAAltitude: 600kmLaunch: 2003Life time: 5 yearsPayload: GLAS instrument
Elevation retrieval - ICESat Facts
Greenland 5 km DEM
Processing steps
Gaussian FittingPre-processingWaveform Dataset
Normalization
Conversion• Bin-ASCII• Counts-Voltage
Smoothing
GLA01
GLA14
Extraction of Elevation
Points
Initial estimation of
Gaussians
Gaussian Fitting
Parameters
Applications
Parameters
Results of Gaussian decomposition
Red: raw waveformGreen: Gaussian components
First mode: left most Gaussian component Last mode: right most
0 200 400 600-0.01
0
0.01
0.02
0.03
0.04Myfit-date:27-02-2003-49338080:19
Vo
lts
(V)
Relative Time (ns)0 200 400 600
-0.01
0
0.01
0.02
0.03
0.04Myfit-date:30-09-2003-235714742:1
Vo
lts
(V)
Relative Time (ns)
Angular&spectral kernel model to describe land surface BRDF
Albedo retrieval using HJ-1 data
•Here,C0CgCv,are kernel coefficients independent from wavelength and stand for the weights of different scattering parts. •They are all related to the structure of canopy or mixed pixel. •Although all the kernel coefficients have clear physical meaning, they can also be treated as empirical parameters.
Compute the pixel-average slope and aspect angle for each 500m grid
Compute the subpixel correction factor T for each 5km grid
Compute the pixel-average slope and aspect angle for each 5km grid
90m resolution D
EM
over the Plateau
Store the results in database
Flow chart for setting up a database for topography effect correction.
500m/5km resolution direction reflectance from MODIS or HJ-1A/1B
Correct the pixel level topography effect,using slope and aspect angle.
Inversion of ASK BRDF model
Derive spectral albedo and broadband albedo
500m/5km resolution topography corrected albedo
Database of topographic param
eters
Flow chart for topography effect correction for 500m albedo products.
LAI retrieval -- a priori knowledge based inversion
0
1
2
3
4
5
6
0. 2 0. 4 0. 6 0. 8
NDVI
LAI
LAI = a*e0. 046x
R2 = 0. 9751
0
1
2
3
4
5
6
0 20 40 60
Ral ati ve growth date
LAI
parameter spaceparameter space time seriestime series
0
0. 5
1
1. 5
2
2. 5
11 13 15 17 19Date (Apri l )
ME o
f in
vert
ed L
AINo a pri oriknowl edge
A pri oriknowl edgei ncl ude growthrel ati onshi pA pri oriknowl edgei ncl ude VIrel ati onshi pAl l of the apri oriknowl edge
NW5
NW4
NW3
NW2NW1
Shunyi county, Beijing, 2001.
NW5
NW4
NW3
NW2NW1
(using only red band ) (using only NIR band)
LAI maps without VI-based a priori knowledge
NW5
NW4
NW3
NW2NW1
(using 2 bands with VI-based a priori knowledge)
(using both red and NIR but without VI-based a priori knowledge)
Time series LAI retrieval by coupling crop growth model
Relative leaf area index (RLAI: LAI/LAImax) from 865 ground measure
ments in Shunyi and Changping of Beijing were fitted with relative a
ccumulation temperature (DVS).
0. 0
0. 2
0. 4
0. 6
0. 8
1. 0
1. 2
0 0. 5 1 1. 5 2
DVS
RLAI
2max
2.5
I 1 (7.76 12.75* 5.63* )i
i i
LAI
LA EXP DVS DVS
Empirical LOGISTIC model for crop growth
Time series LAI retrieval -- flowchart
LAImax,A0(A2)
Weather dataC,A0(A2),A1
LOGISTIC
LAImax
A0(A2)
The prior knowledge of LAImax and A0(A2)
Sun Zenith Angle,Sensor Zenith Angle,Sun Azimuth Angle,Sensor
Azimuth Angle,LAD,hotspot factor,the ratio of skylight
SAILH
Leaf ReflectanceLeaf TransmitanceSoil Reflectance
The prior knowledge of Leaf ReflectanceLeaf Transmitance
and Soil Reflectance
Cost Function
Optimized by SCE-UA
Optimal parameter sets:
Leaf ReflectanceLeaf TransmitanceSoil Reflectance
Time Series Reflectance
Measured Times Series Reflectance
Time Series LAI
The initial value and the bound value of the
retrieval parameters
RLAI between 0 and 1
LOGISTIC
Time Series LAI(the first periods)
Obtain the Time Series LAI of the
second ,the third and fourth periods using
the same method
Obtainthe
averagevalue of thefour
periods
Result
Validation using ground measured LAI values in Shunyi, Beijing, 2001
LAI retrieval -- Data assimilation Developing a priori LAI trend from several years’
MODIS LAI product the adaptive Savitzky-Golay filtering to eliminate the contaminat
ed pixels.
the SARIMA method is used to construct the dynamic model.
The Ensemble Kalman Filter technique is discussed to est
imate real-time LAI from time series MODIS reflectance
data.
Adaptive Savitzky-Golay (SG)
0 23 46 69 92 115 138
T ime
0.0
1.0
2.0
3.0
4.0
5.0
6.0
LA
I (m
2 /m
2 )
M O D IS LA I
S G filte ring LA I
0.0
0.2
0.4
0.6
0.8
1.0N
DV
IN D VI
N D VI Enve lope
C ontam inated obs
1 25 49 73 97 121 145 169 193 217 241 265 289 313 337 361
D A Y/2001
NDVI temporal profiles with circles to mark the contaminated data ( ) _ ( ) _ ( )NDVI i NDVI Env i NDVI Env i
Dynamic model the following dynamic model is constructed based on the climatology
to evolve LAI in time and used to provide the short-range forecast of LAI
1LAI LAIt t tF clim
clim
LAI11
LAIt
t
t
dF
dt
Bondville
0.0
1.0
2.0
3.0
4.0
5.0
LA
I (m
2 /m
2 )
1 25 49 73 97 121 145 169 193 217 241 265 289 313 337 361
D A Y/2008
R etrieved LA I
M O D IS LA I
F ie ld LA I
0.0
1.0
2.0
3.0
4.0
5.0L
AI(
m2 /
m2 )
1 25 49 73 97 121 145 169 193 217 241 265 289 313 337 361
D A Y/2001
R etrieved LA I
M O D IS LA I
F ie ld LA ITonzi Ranch
Linze Yingke
(MODIS+MISR)
0.0
1.0
2.0
3.0
4.0
5.0
LA
I(m
2 /m
2 )
1 25 49 73 97 121 145 169 193 217 241 265 289 313 337 361
D A Y/2001
R etrieved LA I-M O DIS+M ISR
R etrieved LA I-M O DIS
M O D IS LA I
F ie ld LA I
M O D IS observations
M ISR observations
Fractional vegetation cover retrieval – angular correction A general formula to calculate fCover
A simple model for mixing pixles:
min
max min
( ) ( )( )
( ) ( )
NDVI NDVIf
NDVI NDVI
( ) ( )f a NDVI b
-LAI G( ) /cos( ) (1 )f S e
eLAI =0.5 LAI 1-S S
θ
Model inversion using multitude remote sensing data with two spatial resolutions MODIS observations are used for their angular
information
Higher resolution images (HJ-1, TM) are used to get S
solve equations for a, b, LAIe, then calculate fcover
e 1-LAI /cos( )1( ) (1 )a NDVI b S e
e 2-LAI /cos( )2( ) (1 )a NDVI b S e
e-LAI /cos( )( ) (1 )m
ma NDVI b S e
Vegetation detection
HJ reflectancesred&NIR
MODIS reflectances red&NIR
fcover retrieval
LAIe
estimate S factor
Geolocation match
fcover
MODIS geometry
Cloud detection
Study area (Heihe basin)
Gebi desert
Zhangye City
Angular effects corrected fCover (Mar. 07 2009)
LAI and fCover generation using high resolution images
To evaluate the uncertainty in the prediction of the biophysical variables (LAI and fCover) due to the coarse spatial resolution of the MODIS sensor by using high resolution optical data
Particular relevance will be given to the temporal profile of LAI and fCover along the different seasons and according to the main and back-up MODIS algorithms.
LAI and fCover estimation procedure
High resolution LAI and fCover ma
psLUT based RTM inversion
(PROSPECT+SAILH)
Sensor Date ASTER 24 July 2001 29 November 2001 12 March 2002 Landsat TM 9 June 2002 28 August 2002 2 December 2002 24 March 2003
High resolution TOC reflectance
Validation procedure
Field measurements
?
High resolution LAI and fCover
maps
Aggregated LAI and fCover maps
Uncertaintiesassessment
Comparison with MODIS LAI and fCover prod
ucts
Sensor Date ASTER 24 July 2001 29 November 2001 12 March 2002 Landsat TM 9 June 2002 28 August 2002 2 December 2002 24 March 2003
More data available ?
Comparison:MODIS – high resolution data
Probability density functions (pdf) of LAI and fCover estimated from high resolution data are determined for a sufficient number of MODIS pixels of different land use.
Based on these analyses, uncertainty assessment of parameter retrievals is carried out separately for the four major native pasturelands, identified from the southeast to northwest: bush-meadow, alpine meadow, alpine grassland and desert grassland.
Physics-based Radiation Transfer Canopy Model for All Growth Stages
LST retrieval -- Modeling
R o w dire c t io n
Inte ge ral r ange
r 0
R o w c e ntral l ine
P
d p
Top view of the wheat canopy
a periodic function to discuss the mutual overlaps between the neighboring rows.
simplify the row planted wheat canopy as foliage gather whose density is unchanged along row direction and in the vertical
direction, and is changed gradually in the crossing-row direction.
Brightness temperature directional distribution simulated by the new model.
Directional thermal radiation from rugged terrain
affect spatial distribution of vegetation
obstruct sky radiance
geometric effect on surface radiation
shadowing effect
increase environmental radiation
temperature difference on shadowed and sunlit surfaces
A parameterization scheme for coarse resolution pixels
1 ) emitted radiation of surface
2 ) environmental and sky radiation
i iR
i
L AL
A
( )
( )
i i i sky i cH
i
t sky c
C L Vd L AL
A
C L VdL
calculate the statistics in a coarse pixel
( )i RL L
introduce an effective view angle
average ratio that obstruction surroundings occupied in view hemisphere of surface
averaged sky factor
account for emitted directional radiation of a coarse pixel
( ) ( )
cos cos( ) (1 )LAI G LAI G
v sL L e L e
2 / 2
0 0
2 / 2
0 0
1/ 22
arccos cos ( )
arccos cos cos sin sin cos( ) ( )
arccos cos 1 tan
a S dSd
S S A a S dSd
S
1. θ
1. S
a simple model for angular anisotropy effect
supposed L(θ) has a linear relationship with cos , with Hapke formula (1993)
parameterization for sky factor vd and terrain configuration factor Ct
slope S, azimuth A is the horizon angle for direction ( Dozier and Frew, 1990 )。 integrating over the coarse pixel
2 / 2 01
2 0 0 / 2( ) ( ) ( )vd vd k H dH P d a S dSdA
H
pdf of elevation
/ 2
0
1/ 22
1 cos( )
2
1 11 tan
2 2
t
SC a S dS vd
S vd
pdf of horizon angle Hφ pdf of slope angle
preliminary validation
error caused by simplized form of surface emitted radiance (K)
error of radiance measured at surface (K)
1 、 Followed by a statistical parameterization scheme, a model was developed for remote sensing retrieval. 2 、 All the simplifications calculate statistics of the topographic effects exerted on radiative transfer.3 、 The parameterized model cost much less time with an acceptable accuracy lost.
HJ-1 150m LST products was proposed by the IRSA/CAS as a daytime land product over the Tibet Plateau.
A view angle dependent single channel LST algorithm has been developed for correcting atmospheric and emissivity effects for all land cover types.
HJ satellite constellation
LST retrieval using HJ-1 data
CCDresolution : 30mbands : 0.43-0.52μm , 0.52-0.60μm , 0.63-0.69μm , 0.76-0.90μmswath : ≥ 700kmIRSresolution : 150m ( NIR MIR ) /300m ( TIR )bands : 0.75-1.10μm , 1.55-1.75μm , 3.50-3.90μm , 10.5-12.5μmswath : 720km
Supposing that the atmosphere is uniform on spherical surface and layered vertically, the atmospheric transmittance can be expressed as (Fan et al, 2007):
where η is the path factor, R is the earth radius, s0 is the atmosphere layer thickness, θ is the view zenith angle.
(1) Obtaining the atmospheric parameters
•simulation results indicate that there are strong linear relationships between different L↑(θ) and L↑(0) (nadir view)
HJ-1 LST
Plots of Atmospheric parameters against the water vapor content
(a) atmospheric transmittance (b) atmospheric upwelling radiance(c) atmospheric downwelling radiance
(2) Obtaining the land surface emissivity For HJ-1 IRS data we will use the NDVI method to estimate land
surface emissivity. (3) Retrieving the land surface temperature
HJ-1 LST
Validation related works
Ground-based LAI measurement methods comparison
LAI-2000
Direct LAI measurement
Indirect LAI measurement
Digital Hemisphreical Photography (DHP)
The comparision of LAI2000 data and FishEye data with moving average fitting. Obviously , the variance of LAI2000 data is littler than that of Hemispherical photography data.
:is the woody-to-total area ratio, it was used to correct the woody components effects in LAI measurements. :the effective LAI :the needle-to-shoot area ratio :is the elements clumping index
(1 ) /e e eL L
eL
e
e
Indirect estimation of forest canopy LAI correct forest LAI by Multispectral Canopy
Imager (MCI)
1st year meeting --- Progress Report
VIS NIR
Classified image
Sky fractionWoody components fractionLeafy fraction
statistical methods for:
:the effective plant area index at zenith angle :the effective woody area index at zenith angle:the proportion of elements within the image:the proportion of woody components within the image
Sky Fraction, Woody components Fraction, Leafy Fraction (VZN 0~90°interval 10°) :
/2
0
( ) 2 ln[ ( )] sin( )e tPAI p d
The effective PAI:
/2
0
( ) 2 ln[ ( )] sin( )e wWAI p d
The effective WAI:
10
1
2e i iP PW
( )ePAI
( )eWAI
( )tp ( )wp
:the effective PAI at zenith angle iP :the weight at zenith angleiW
/WAI PAI
Stand 1: 0.1909Stand 10: 0.1798α : 0.1816
:the woody area index:the plant area index
WAI
PAI
1st year meeting --- Progress Report
Thank you!