International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
3.3 Predictions and Validation
Predictive equations were determined from the relationships
between the spectral indices (NDVI, MSAVI2 and MTVI2)
using ground data collected over peas, beans, and corn
canopies. The overall best fits were given by an exponential fit
with a coefficient of determination (r^) depending on the index
of interest and the crop type. Indeed, for the three indices, we
have obtained values around 0.91, 0.92, and 0.70 for beans,
peas and corn, respectively. Equations obtained were applied to
CASI hyperspectral images to map LAI status over agricultural
fields seeded with corn, wheat, and soybean. Results were
validated using ground truth measurements collected during the
field campaigns of 1999, 2000, and 2001.
In this paper, we present preliminary results based on the
use of predictive equations derived from the measurements over
bean and pea canopies (Figure 3). The aim is to show the
relative dependency of prediction algorithms on crop type.
Work is in progress to evaluate overall results concerning corn
canopies, as well as predictive algorithms determined from data
representing all the three canopies.
6.00 - —À pe tet p a
$9 MTVI2
D MSAVI2
À NDVI
8
3.00
Estimated LAI
1.00
0.00
0.00 1.00 2.00 3.00 4.00 5.00 6.00
a Measured LAI
6.00 r
© MTV | |
5.00 TVI |
C) MSAVI2! |
A NDVI |
[7 B |
4.00 |
2 3.00 - |
2.00 - |
|
|
1.00 |
|
|
0.00 * |
0.00 1.00 2.00 3.00 4.00 5.00 6.00
b Measured LAI
Figure 3. Comparison Measured LAI — Estimated LAI from
CASI images over corn and soybean, using predictive equations
determined from pea (a) and bean (b) data.
112
Figure 3 compares LAI estimations from CASI reflectance data
and LAI measurements in the field and laboratory. In general, it
reveals very good agreement between the predictions and the
ground truth. It shows, however, significant differences in
indices’ performances, and the specific predictive equation by
crop type.
First, the crop type influence on the predictions is well
illustrated by the difference in LAI estimation between the
equations established from pea data (a) and bean data (b). The
latter tends to underestimate canopy LAI over intermediate and
high density canopies (LAI > 3).
Second, these preliminary results lead to the following
remarks:
- NDVI, MSAVI2, and MTVI2 have similar predictive
power for LAI estimation in the case of low to
intermediate canopy densities (LAI < 3);
- NDVI is not adequate for LAI predictions of
intermediate to dense canopies. It exhibits a clear
saturation when LAI exceeds the value of 3;
- MSAVI2 and MTVI2 have similar behaviors,
following the one-to-one slope for the pea-based
equation (Figure3a);
- MSAVI2 seems to have the best overall performance
regarding the underestimation issue, though it has the
highest level of overestimation for intermediate LAI
values.
These results contrast with those based on predictive equations
derived from simulated data using PROSPECT and SAILH
(Habaoudane et al., 2004). Indeed, using same indices (MTVI2,
MSAVI2), we noticed that algorithms based on ground
measurements tend to underestimate LAI, while algorithms
based on simulated data have a tendency to overestimate LAI.
4. CONCLUSIONS
This paper presents the results from a study that focused on
using ground measurements of spectral and biophysical
properties over crop canopies in order to develop predictive
equations for LAI estimation from CASI hyperspectral images.
Based on recommendations from previous studies, three indices
(NDVI, MSAVI2, and MTVI2) were evaluated regarding their
potential for LAI prediction from remotely sensed data.
Comparison between CASI-estimated LAI and ground truth
from different sites, with different crop types (soybean and
corn) has led to the following conclusions:
- Relationships between measured LAI and spectral
indices from measured spectra are crop type-
dependent;
- Use of different predictive equations resulting from
different crop types influences estimations results at
medium to high LAI levels;
= In comparison with NDVI, MSAVI2 and MTVI2
showed no saturation effects even when LAI values
exceed 5.
5. REFERENCES
Baret, F., and Guyot, G. (1991), Potentials and limits of
vegetation indices for LAI and APAR assessment. Remote Sens.
Environ. 35:161-173.
Internatio
Boardm:
mapping
Annual
Publicat
Broge, ?
and stab
for estin
density.
Broge, I
area ind
spectral
Cassel,
Assessin
station |
Agron. J
Chen, J.
boreal cı
Sens. En
Daughtr
and Ku:
radiation
canopies
Daughtn
Colstoun
leaf ch
reflectan
Fassnach
F Vv, à
index of
Mapper.
Gitelson.
Stark, R.
vegetatic
Geospati
Coloradc
Habouda
Tejada, |
green LA
heteroget
Recent /
Spain, or
Habouda
and Stra
novel al;
modeling
Remote S
Hu, B., ^
Retrieval
tower sit:
Environ.
Huete, A
Remote S
