Dr Andrew P' Rodger

1
Dr Andrew P. Rodger
2
NDVI

• NDVI(?NIR-?R)/(?NIR?R) or NDVI(LNIR-LR)/(LNIRL
  R)
• The Normalized Difference Vegetation Index
  (NDVI), is useful as a tool for estimating net
  primary production over various biomes (Huete
  Liu, 1994), but has a number of deficiencies.
  These include,
• sensitivity to soil variants (particularly dark
  and/or wet) (Karnieli et al, 2001)
• saturation of the index in the case of densely
  vegetated areas (Karnieli et al,2001 Huete et
  al, 2002, Unsalan Boyer, 2004)
• variation due to solar viewing geometry (aka
  BRDF effects). This is more pronounced on sensors
  with large swath widths such as MODIS (Huete et
  al, 1999 Chilar et al, 1994).
• atmospheric contamination (i.e. aerosol
  loadings, path length, cloud cover etc) (Kaufman
  Tanre, 1992 Huete Liu, 1994 Miura et al,
  2001 Karnieli et al, 2001).

3
NDVI
Of the two NDVI forms, reflectance radiance (?
L), numerous studies have shown the reflectance
form compares more favourably with in-situ NDVI
measurements (Kaufman Tanre, 1992 Huete Liu,
1994 Miura et al, 2001 Karnieli et al, 2001
Tachiiri, 2005). Further, soil-adjusted
vegetation indices, such as the SAVI (Huete,
1988 Huete Liu, 1994) have also proven to
increase the reliability of vegetation indices.
This concept has been expanded upon to include
the afore mentioned atmospheric effects (aka
SARVI also called ARVI) (Kaufman and Tanre,
1992). Lastly, the inherent BRDF nature of
vegetation has been addressed by using
semi-empirical BRDF models by various researchers
and includes the MODIS EVI and NDVI products
MOD18 (Huete et al, 1994).
4
NDVI Example
As an example, NDVI images are currently produced
for NOAA satellite as 14 day composites where a
basic atmospheric compensation is performed, no
BRDF adjustment is made and no soil adjustments
are made. Besides the cloud (right hand image),
the two images are over a time frame when
rainfall is minimal and solar zenith angles are
small. Yet significant differences may still be
found in the NDVI. Why?
5
NDVI Crop Yield

• In-Season Prediction of Potential Grain Yield in
  Winter Wheat Using Canopy In-Season Prediction of
  Potential Grain Yield in Winter Wheat Using
  Canopy Reflectance (Agron J. 93131-138)
• W.R. Raun, J.B. Solie, G.V. Johnson, M.L. Stone,
  E.V. Lukina, W.E. Thomason and J.S. Schepers
• Use of Indirect Measures for Grain Yield
  PredictionThe normalized difference vegetation
  index (NDVI), calculated with measurements of
  reflected light from the red and near-infrared
  bands, has long been used as an indirect measure
  of crop yield, including that of wheat (Colwell
  et al., 1977 Tucker et al., 1980 Pinter et al.,
  1981).
• Aase and Siddoway (1981) confirmed the
  relationship of NDVI to wheat grain yield but
  noted that the relationship deteriorated rapidly
  as wheat ripened.
• Soil background, view and solar angles,
  atmospheric conditions, and crop canopy
  architecture are also important factors affecting
  NDVI (Huete, 1987 Jackson and Huete, 1991).
• Pinter et al. (1981) reported that summing NDVI
  values from late-season (Feekes 10.5, flowering
  to grain fill) spectral measurements was useful
  in predicting wheat grain yield.
• Bartholome (1988) reported that accumulated NDVI
  was a more stable predictor of millet and sorghum
  grain yields than a single spectral measurement.
• Rasmussen (1992) calculated a sampling-interval
  weighted average NDVI by integrating
  multi-temporal spectral measurements with time,
  which improved millet grain yield estimates from
  a single spectral measurement.
• Smith et al. (1995) reported that sensing twice
  and combining NDVI using a linear model improved
  correlation with wheat grain yield compared to
  sensing once.
• Rasmussen (1998) failed to improve the
  correlation of NDVI to grain yield by integrating
  the product of multi-temporal NDVI measurements
  and photosynthetically active radiation (PAR).
• Different forms of NDVI will give differing
  results. Therefore, is NDVI the best metric to
  use when considering crop biomass?

6
NDVI Crop Yield

• The NDVI is an indicator of the available biomass
  in a given pixel. However, for LAIgt3 the NDVI
  will tend to exhibit asymptotic behaviour
  (Serrano et al, 1999), and yield information may
  be lost.
• Does greater spectral coverage yield better
  results?
• To derive crop yield generally requires
  seasonally adjusted empirical equations that may
  be combined with crop models and ancillary
  remotely sensed data.
• Maselli et al (1999) suggest that the effect of
  soils should decrease in a linear fashion as the
  vegetation increases over the growing season.

7
Possible Spectral Approach for Coverage
Information

• NDVI can be used to locate important temporal
  events in the crop cycle.
• The minimum NDVI, at t0, over a given year
  should correspond to bare soil, while the maximum
  NDVI, at tT, should correspond to the time of
  maximum crop foliage (this does not account for
  saturation).
• At t0 the surface reflectance, as calculated
  from the atmospherically corrected satellite
  radiance, may be written as ?0,
• At tT the surface reflectance will have changed
  due to the presence of the crop and is written as
  ?T

8
Possible Spectral Approach for Coverage
Information
If a linear mix (single scattering only) of soil
and vegetation is assumed (aka Maselli et
al,1999) then the surface reflectance, as
calculated from the satellite radiance, is the
sum of two surfaces, ?(?,T) Swi(?,T) ?i(?,T)
w1(?,T) ?0(?,T) w2(?,T) ?v(?,T)
(1) Where, w1 and w2 are weights, and where
Swi1 and ?v is the surface reflectance of the
species we are interested in. In the simplest
case, a sample of ?v at time T is measured and
the value of w1 modified until ?2 ? ?m(T) - ?
(T) is minimized, where ?m(T) is the measured
satellite spectral reflectance at time t and the
summation is performed over the available
spectral bands.
9
Possible Spectral Approach for Coverage
Information
At some time in the growing season it may be
reasonable to expect the spectral signature of
the vegetation in question to effectively become
stable, or near constant. Again, if we know, or
may reasonably estimate what the final spectral
signature of ?v is we now have one equation with
one unknown per spectral channel. Serrano et al
(2000) show an example of winter wheat grown at
the experimental station of Mas Badia, Girona,
Spain. The resulting spectra appear to show that
linear mixing is dominating, and that the
spectral signature of the wheat, is essentially
unchanged beyond a certain time.
10
Simple Linear Mixing
Tie Point
The simplistic linear mixing of a grass spectra
and soil spectra (right hand side) shows a
general similarity to the measured spectral
signatures of wheat and soil (left hand side).
11
Simple Linear Mixing

• If a linear mixing model was to used to estimate
  the coverage of a particular crop what inputs do
  we require?
• The baseline surface spectra before planting
  commences.
• An estimate of when the vegetation spectra will
• Get the farmers to send in samples for spectral
  analysis.
• Assume a generic spectral model
• Use a crop model (will probably still need grower
  input)
• Use retrospective data in combination with
  observed biomass and/or yield.