Current and Future Approaches to Mapping Canopy Foliage Distribution over National Extents at NRCan, CCRS

1
Current and Future Approaches to Mapping Canopy
Foliage Distribution over National Extents at
NRCan, CCRS

• Richard Fernandes, Sylvain Leblanc, Peter White,
  Goran Pavlic,Josef Cihlar

2
Current Method

• Based on nadir vegetation indices.
• Simple ratio and reduced simple ratio.
• Applied to
• 9 TM/ETM scenes (40 planned)
• 10-day composited AVHRR from 1993-1999, SPOT-VGT
  from 1998-1999, 16 day MODIS composites for 2000
  (in testing)
• 3 Methods Tested Empirical, Physical Model
  Inversion via LUT, Semi-Empirical

3
Empirical Method

• (LAI,Vegetation Index) measured for 3x3 TM pixel
  patches.
• Measurement error modelled.
• Type II regressions applied (errors cannot be
  ignored in both LAI and VI) after tranformation
  of axes for homeoscedacity.
• p(LAIVI) reported but m.l.e. recorded.

4
(No Transcript)
5
(No Transcript)
6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
(No Transcript)
10
(No Transcript)
11
(No Transcript)
12
Physical Models

• Use of validated reflectance models to define
  p(LAISR, parameters x) - for nadir viewing only
  currently.
• Broadleaf - Sellers 1985 Conifer - 5 Scale.
• Simulate over p(x) uniform distn of valid range
  for Canada.
• Populate p(LAISR) using each model run with a
  distribution of LUT entries reflecting model
  errors.
• Reporting m.l.e. of p(LAISR) can be ill-posed.

13
Semi-Empirical Model

• p(x) is a joint p.d.f. of model parameters.
• Previous studies did not use equal steps
  biased p(x).
• Physical model use of p(x)u(x) is wrong.
• Calibrate p(x) by weighting model trajectories
  based on the best match to empirical (LAISR)
  data.
• Assumes we have adequately sampled marginal
  distribution of p(x) in our empirical data.
• calibrates out errors in physics - model is
  physically valid in terms of LAI vs SR.
• Comparison of p.d.f. of empirical and
  semi-empirical model allows check of
  representativeness of data.

14
Conditional Needleaf LAI Distributions Given SR
LAI
SR
15
Conditional Broadleaf LAI Distributions Given SR
LAI
SR
16
Conditional Needleaf LAI Distributions Given SR
Physical Model
Semi-Empirical Model
Empirical Model
SR 5 for solid lines. SR 10 for dashed lines.
17
Conditional Broadleaf LAI Distributions Given SR
Physical Model
Semi-Empirical Model
Empirical Model
SR 5 for solid lines. SR 10 for dashed lines.
LAI
18
Scaling To AVHRR/VGT

• Approach I
• Make TM LAI maps and relate to AVHRR/VGT VI.
• p(LAISRcoarse)p(SRcoarseLAI)P(LAI)/p(SRcoarse)
  
• prior P(LAI) is spatially variable gt bias can be
  serious.
• Approach II
• Relate TM VI to TM SR and apply TM based
• p(SRTMSRcoarse)p(SRcoarseSRTM)P(SRTM)/p(SRcoars
  e)
• p(LAISrcoarse) p(LAISRTM)p(SRTMSRcoarse)
• P(SRTM) is known.
• Larger uncertainty in p(LAISRcoarse) due to
  sensor cross-calibration errors and due to
  scaling errors but hopefully less bias

19
Spectral response function effect in NDVI
observed from AVHRRNOAA-14 and NOAA-15. July 15,
2000. Northern Ontario. Canada.
Green vegetation
model
Channel 2
Channel 1
observations
20
Leaf Area Index - Scaling
LAI AVHRR July 20-30, 1994
21
(No Transcript)
22
1km Water Fraction Mask of Canada from 150000
Topographic Maps
23
Flowchart for LAI from NADIR VI
Subjective
Measurement
(LAI, SRTM)
Objective
Surface LAI TM Nadir VI
Type II Model
Outputs
More data
Is Type II within range of Physical
Stratify by ecozone and lifeform
Constrain Physical Model
p(LAI SR)
Validate model
Model choice
Reflectance Model
Unconstr. LUT
p(LAI SR,u(x))
p(x)u(x)
Scaling relationship between TM and Coarse
ecozone/lifeform.
Apply empirical m.l.e. if sufficent data else
model.
m.l.e. LAISR
Survey parameter ranges
24
Phenology - LAI(t) from VGT
25
Foliage Clumping - Why Monitor It?
26
BRDF Clumping Index
27
Physically Based Estimation of Clumping
NDHD (rHotSpot- rDarkSpot)/(rHotSpot
rDarkSpot)
28
(No Transcript)
29
Clumping Index
30
Future Work

• Simultaneous estimation of sub-pixel land cover
  and LAI.
• Produce LAI from multi-angle data (waiting for
  MISR pre-processing!).
• Compare error budget of multi-angle LAI vs LAI
  from nadir VI and anisotropy index.
• Evaluate sensitivity of LAI estimates to
  variations in leaf biochemistry and leaf/site
  moisture status (Using 5Scale)
• Compare 3D RT with 5-scale in forward and inverse
  mode.