VEGETATION - 2000
Lake
Maggiore - Italy, 3-6 April 2000
SENSITIVITY
ANALYSIS OF COMPOSITING STRATEGIES:
MODELLING AND EXPERIMENTAL INVESTIGATIONS
de
Wasseige Carlos*, Lissens Gil**, Vancutsem Christelle*, Veroustraete Frank**
and Defourny Pierre*
* Department of Environmental Sciences - Université catholique
de Louvain
Croix du Sud, 2 bte 16 B-1348 Louvain-la-Neuve
** Vito - Flemish Institute for Technological Research
Boeretang, 200 B-2400 Mol
Paper
(pdf file, 65 k)
High
temporal resolution satellites, such as VEGETATION provide multiple images
of the same site over short periods of time. Time series constituted of
these individual images are characterised by a lack of signal consistency
since measured radiances generally result from various cloudiness, atmospheric
and geometric conditions. To reduce the related noise, various compositing
techniques are available.
The
pre-launch phase went into a systematic investigation of the main issues
related to the temporal synthesis production using a one-year time series
simulated for the VEGETATION sensor spectral and geometric configuration.
The aim of that investigation was to test globally the sensitivity of
the compositing process to different factors that perturb the signal,
i.e. the sun-target-sensor geometry, the atmospheric conditions and the
surface anisotropy. Perturbing factors have been ranked according to their
impact on the sensor signal. This sensitivity analysis highlighted the
large effect of the viewing angle as opposed to atmosphere variability
with regard to day-to-day variations. However, the perturbing factors
were always manifested as a coupled effect on the sensor signal. The analysis
of the one-year simulated time series showed three nested scales of variation.
A five-day cycle related to the viewing angle and due to the wide swath
of the sensor. A 26-day cycle corresponding to the satellite orbit revisit
time, and the sun annual cycle changing according to latitude. The conclusions
drawn from the pre-launch phase of the VEGETATION programme have resulted
in a proposal for two new image compositing strategies.
The
approach pursued in the pre-launch phase was repeated using actual VEGETATION
data in the post-launch phase. Three decades of global daily VGT-P were
used. Decade 1 from 11/06/98 to 20/06/98, decade 2 from 21/07/98 to 31/07/98
and decade 3 from 11/10/98 to 20/10/98 were selected. A sampling approach
based on the global dataset of VGT-P segments was designed with 50 x 50km
chips to asses the performances of the existing compositing strategies
for the various sun-target-sensor geometries and the different surfaces
of the main terrestrial biomes. The spatial and temporal variability of
the signal was first analysed for the various chips with regard to the
simulation results.
The
current compositing technique for VEGETATION data (VGT-S10 product) shows
radiometric artefacts in the reflective bands that may cause a significant
noise for subsequent retrievals of surface parameters. The performances
of various compositing strategies are assessed as well for the reflective
bands as for the NDVI composites. Dedicated indicators and statistical
analysis are computed to provide quantitative results by zone and by band.
An
innovative strategy such as the Median Composite of FUzzy Multispectral
Estimate (MC-FUME) has been developed as well, to produce composites with
reflectance values independent of the observation/illumination geometry
at the time of measurement. Potential improvements were first tested based
on selected NOAA AVHRR multitemporal time series. The results obtained
using actual VEGETATION data are compared to the current MVC-NDVI approach
and to other documented alternatives. A discussion of the results will
provide suggestions for possible improvements in the VEGETATION processing
chain compositing algorithms. |
