ground measurements have demostrated that significant errors on the TMS channel calibration exist. A comp 
procedure for re-calibration of TMS has then been developed, using calibrated AVIRIS data as reference. anc * c 
A significant problem encountered is the spatial variability of atmospheric conditions (mainly water 
vapour content) when developing atmospheric correction methods for deriving ground reflectance values 
from aircraft radiance data. Even by using simultaneous radiosounding data, the atmospheric correction 4- FI 
carried out with LOWTRAN 7 (Fortea et al„ 1993a) does not reproduce the ground measurement, being the 
most significant differences located in the water absorption bands at 0.9 and 1.1 pm. This may be partly due Four i 
to the known non-fully-adecuacy of LOWTRAN as a radiative transfer code for atmospheric correction, a reg 
especially for high spectral resolution data. The AVIRIS water vapour absortion bands themselves may be imprc 
used to derive atmospheric water vapour content, and software for this do exist such as the ATREM model propc 
developed at CSES, Colorado, but it does not give enough satisfactory results for accurate modelling either. signif 
Some research is now in progress trying to incorporate MODTRAN as the radiative transfer code for its at rej 
more adequate modelling of the water absorption effects. empii 
The wavelength dependence of spatial integration has also been analysed by using AVIRIS data, but take i 
further work is still required for an adecuate modelling of the coupling between spatial variabilities of the Tome 
ground surface and of the atmosphere, in order to properly model the effects associated to spectral scaling. comb 
A multiresolution database design has been elaborated in order to make it possible to study scaling editec 
effects on different resolution aircraft/satellite data in a four-level approach (Moreno and Melia, 1993c). 
The first level corresponds to ground data, and the other levels correspond successively to aircraft and high- 
/ low-resolution satellite data. Only model-independent corrections are applied to the data, but they are 5 - At 
geometrically integrated and resampled to a common UTM grid to facilitate the comparison with ground 
data. Multiresolution display is possible, allowing for a visual glance of the effects of changing spatial scale. The \ 
Numerical calculations can also be performed because each data is mapped simultaneously in two contri 
consecutive levels of resolution to allow transition from one level to another. 
Power spectrum analysis based on Fourier transform techniques have also been applied to Radii 
multiresolution actual data to test for the quality of geometric integration (Moreno and Melia, 1993b). 1‘ke 11 
Following such an approach, autocorrelation functions for NDVI values derived from LANDSAT TM and Bolle 
NOAA AVHRR data have been calculated for June 12, 1991. This is actually equivalent to the estimation of (Win; 
the information loss when using low resolution (AVHRR) data as compared with the information provided Franc 
by high resolution (LANDSAT) data. This kind of analysis, by using actual data, is indeed required to Thon 
properly understand spatial scaling of basic measurements, and to model the scaling of derived quantities to menti 
be used as inputs in global scale models. Headi 
The Free University of Berlin carried out an intercomparison of AVIRIS and LANDSAT-TM data 
for Holm oak. Although the data were not corrected for atmospheric effects, there is good agreement 
between the results from both sensors. 6 * 
3.2.2. Surface Parameter Retrieval From The Interpretation Of Multispectral Images. The University of 
Bristol obtained Quercus reflectance spectra derived from AVIRIS after application of the ATREM 
algorithm, (ATmosphcric REMoval, CSES (1992)). In order to estimate SLA from AVIRIS data, pixels 
containing pure semi-natural vegetation were extracted from the image. The results from factor analysis 
obtained from field spectroscopy were applied to the AVIRIS reflectance spectra. This enables the 
calibration between the spectroradiometer data and the SLA parameter to be applied to the AVIRIS data. In 
order to verify the results, two different measures of biomass were calculated. The first was an NDVI 
derived from the AVIRIS image, and the second is a measure of projected ground vegetation cover from the 
airborne video imagery. Despite the relatively small sample size, the results appear extremely promising, 
suggesting that it is possible to predict SLA (through the relationship with the second eigenvector) from the 
AVIRIS data. 
3.3 - Rainfall-NDVI Correlation Studies For The Year 1991 In The EFEDA Area 
The University of Valencia carried out a retrospective climatological study of the area of Castilla-La 
Mancha (Lopez-Baeza et al. (1993a)) in order to be able to characterize the last years (1987-1991) from 
precipitation data recorded at 40 thcrmopluviomctric stations that the Spanish National Institute of 
Meteorology has scattered over the whole area. This characterization aimed to determining whether 
different zones within the area were VERY DRY, DRY, NORMAL, HUMID or VERY HUMID.Once this 
character was established for a specific period of time, correlations between precipitation data and NDVI 
values obtained from AVHRR data could be carried out for the last five years. This work was developed in 
close colaboration with the Free University of Berlin. 
For 1991 the correlation was approached in two different but complementary ways: on the one hand, 
point to point correlations wete obtained between each station precipitation value and the averaged value of 
NDVI, as derived from AVHRR HRPT data, over an area of 25 km^ around the station site. On the other 
hand, an image to image correlation was also performed between the NDVI map, also derived from 
AVHRR HRPT data, and a precipitation map generated from the meteorological stations network, now 
328