284
F. Giittsche, F.-S, Olesen
calculations with ENZO-NN are estimated to be of the order of 10 4 times faster than
with MODTRAN-3: this allows atmospheric corrections of long time series of IR
satellite data, e.g. the 25 years of METE OS AT single channel IR data.
2 Thermal Surface Parameters
2.1 The Data Basis
For quantitative exploitation METEOSAT IR data have to be calibrated in
brightness temperatures. Furthermore, for processing algorithms to produce
meaningful results, the data have to be cloud-screened. The calibration information
is taken from the EUMETSAT web-pages (EUMETSAT, 1998). The cloud
screening is based on the information of the IR and VIS channel during daytime and
the IR-channel alone during night time (Schadlich et aI., 200 I). This information
is utilised to deduce monthly temporal and spatial dynamic thresholds. It is assumed
that the monthly minimum VIS/maximum IR pixels of each slot are cloud-free
values. In order to find a correlation between thermal behaviour and land-surface
characteristics, diurnal temperature cycles (DTC' s) are extracted from sequences of
the pre-processed METEOSAT IR measurements. Data gaps due to cloud-screening
and errors are identified and limited to a duration which allows small cloud-fields
to pass through an observed scene, but rejects permanently-clouded areas. This
process may be regarded as a cloud-clearing correction scheme.
2.2 Modelling of Diurnal Temperature Cycles
Ideally, after calibration and cloud masking only cloud free pixels remain for the
determination ofLST. For each pixel location more than 10.000 pixels are processed
for one vegetation period and about half of them remain after the cloud masking. For
such an amount of data tools must be developed to make interpretation feasible. The
first step is to model the diurnal variation ofLST. The model consists of a harmonic
and an exponential term (Fig. 2), describing the effect of the sun and the decrease
of the surface temperature at night, respectively. It is fitted automatically to the
temperature waves by a Levenberg-Marquardt least-squares scheme (Gottsche and
Olesen, 2001). Therefore, model parameters can be determined for data of arbitrary
size. Ideally, the diurnal thermal behaviour of every METEOSAT pixel is
completely characterised by a set of parameters.
For data of high quality the method works sufficiently well and yields useful
parameters. In a case study for the 19.4.96 - a selected day with nearly cloud-free
conditions in northern Italy - it was shown that the model is able to describe thermal
F. Giittsche, F.-S, Olesen
calculations with ENZO-NN are estimated to be of the order of 10 4 times faster than
with MODTRAN-3: this allows atmospheric corrections of long time series of IR
satellite data, e.g. the 25 years of METE OS AT single channel IR data.
2 Thermal Surface Parameters
2.1 The Data Basis
For quantitative exploitation METEOSAT IR data have to be calibrated in
brightness temperatures. Furthermore, for processing algorithms to produce
meaningful results, the data have to be cloud-screened. The calibration information
is taken from the EUMETSAT web-pages (EUMETSAT, 1998). The cloud
screening is based on the information of the IR and VIS channel during daytime and
the IR-channel alone during night time (Schadlich et aI., 200 I). This information
is utilised to deduce monthly temporal and spatial dynamic thresholds. It is assumed
that the monthly minimum VIS/maximum IR pixels of each slot are cloud-free
values. In order to find a correlation between thermal behaviour and land-surface
characteristics, diurnal temperature cycles (DTC' s) are extracted from sequences of
the pre-processed METEOSAT IR measurements. Data gaps due to cloud-screening
and errors are identified and limited to a duration which allows small cloud-fields
to pass through an observed scene, but rejects permanently-clouded areas. This
process may be regarded as a cloud-clearing correction scheme.
2.2 Modelling of Diurnal Temperature Cycles
Ideally, after calibration and cloud masking only cloud free pixels remain for the
determination ofLST. For each pixel location more than 10.000 pixels are processed
for one vegetation period and about half of them remain after the cloud masking. For
such an amount of data tools must be developed to make interpretation feasible. The
first step is to model the diurnal variation ofLST. The model consists of a harmonic
and an exponential term (Fig. 2), describing the effect of the sun and the decrease
of the surface temperature at night, respectively. It is fitted automatically to the
temperature waves by a Levenberg-Marquardt least-squares scheme (Gottsche and
Olesen, 2001). Therefore, model parameters can be determined for data of arbitrary
size. Ideally, the diurnal thermal behaviour of every METEOSAT pixel is
completely characterised by a set of parameters.
For data of high quality the method works sufficiently well and yields useful
parameters. In a case study for the 19.4.96 - a selected day with nearly cloud-free
conditions in northern Italy - it was shown that the model is able to describe thermal