le averaged 
lished from 
)n satellite 
oach. Even 
ation of the 
approached 
' data in the 
i change of 
ows one to 
i from high 
is also been 
ta, and the 
tellite data 
r to make it 
method of 
and Melia, 
y corrected 
he value of 
nt does not 
pixels to be 
iterpolation 
changes in 
he viewing 
responding 
rived from 
lution, e.g., 
on satellite 
irements in 
d dataset is 
: method of 
used to test 
et al„ 1986 
e and from 
re analyzed 
laracteristic 
; a function 
used as an 
usions may 
erences are 
erization of 
rithin-pixel 
ale, for the 
ibtained for 
Barrax area 
lility on the 
d deviation 
1 values of 
lationships 
ogarithmic 
elationship 
D and low 
ionships to 
aeous areas 
n values by 
Fig. 8 . Comparison of the NDVI value derived from ground reflectance measurements with the values 
derived from NOAA AVHRR data (with two different interpolation approaches) and with those obtained by 
simulation from TM data (considering the particular viewing geometry of the AVHRR data), as a function 
of the spatial scale for each data type. 
A similar analysis has been carried out for the temperature field. Particularly interesting in the 
EFEDA case is the up-scaling analysis for thermal infrared observations, due to the importance of 
temperature in agrometeorological models. The avaliabily of thermal data for 25 m (airborne TMS), 120 m 
(LANDSAT TM), 1.1 km (NOAA AVHRR) and 5 km (METEOSAT) allows one to analyse the 
temperature field as the spatial scale changes. In this case, the variability of temperature values derived from 
different spatial scales is directly related to the local variance of the ground temperature, but the coupling of 
emissivity-temperature spatial variations needs to be considered in future research. One important problem 
to be solved previously is the necessity of accounting for the spatial variability of atmospheric water vapour 
and the consequences of this variability on the corrections for deriving ground temperature maps from 
aircraft and satellite data. All the available estimates of water vapour (radiosounding, aircraft LIDAR, 
AVIRIS water vapour channels and split-window channels of AVHRR data) indicate the importance of 
spatial variability. Then, the coupling emissivity/ground temperalure/atmospheric water vapour spatial 
variability needs to be properly considered in modeling approaches. 
The Free University of Berlin carried out spectral reflectance and albedo measurements to validate 
LANDSAT-TM data. Although their aspect angles may be different, the agreement between ground and 
atmospherically corrected satellite data is sticking for different surfaces like bare soil, maize and alfalfa, but 
the scatter may be large. The TM data sets for 12 and 28 June may not be sufficient for a general validation. 
They also compared LANDSAT-TM and NOAA-AVHRR data quantitatively, once both data sets could be 
superimposed. The histogram analysis shows that AVHRR-1 and TM-3 and -4 have much in common for 
the Tomelloso area but the AVHRR results are shifted to higher values. This can be understood if one or the 
other instrumental calibration factor is incorrect. Since the TM data had partially been validated by means 
of ground based measurements, not showing a tendency to provide too low values, the possible error in the 
AVHRR data has to be investigated further. 
Apart from spatial integration, another aspect to be considered in the study of scaling effects is 
spectral integration. Since most satellite sensors use broad-band approaches for measuring with the required 
signal-to-noise level, the derived values really correspond to spectrally integrated values. This integration is 
coupled with spatial integration and therefore, both should be well understood when using low resolution 
data. A problem to this is the different spectral bands actually used for each satellite to measure in each 
spectral region. Although NDVI or temperature values can be derived from different kinds of satellite data, 
these data sets are not directly comparable due to the different spectral bandpass used by each sensor. 
Unfortunately, the technical failure of TIMS made it impossible to analyse these aspects for thermal 
data, but the successful AVIRIS overflight (although with technical problems in the D spectrometer as well) 
permits a good analysis of spectral integration effects in the range 0.4 - 2.5 (im. 
The high radiometric quality of AVIRIS data (10 bits) and its outstanding calibration permits the 
simulation of broad-band measurements and check their representativity. Besides this, the availability of 
simultaneous (identical viewing geometry and atmospheric conditions) data with different spectral 
resolution from AVIRIS and TMS observations, allows also to test the conclusions derived from AVIRIS 
simulations by using real TMS data (Moreno ct al., 1993). An important problem found in this work carried 
out by the University of Valencia was the lack of good calibration for the TMS sensor. Although calibrated 
before the field campaign, the results of the analysis by comparing TMS-AVIRIS data with simultaneous 
327