65 
>.065 * T b 5-17.255 
0.094 * T b 5 -25.11 
■ees kelvin) 
I on a linear 
I channel 2 albedos. 
ition 
nvolves assigning a 
'Oint in the 
vigation results, 
ite must be known 
then giving the 
pixel (in 
lishes every other 
ational NOAA- 
e used together 
nline to determine 
lite-point in this 
s of the pixels on 
gational grid can 
can be resampled 
E TEMPERATURES 
three (or two) 
are being used to 
atures. The 
and land surface 
D 
asurements of the 
pace. Nevertheless 
aan be eliminated, 
a emissivity of sea 
stant, radiometric 
an account of the 
temperatures near 
Ltu measurements is 
by far the largest 
.ns in the 
'rection. The 
infra-red channels 
lat atmospheric 
lannel. The 
"window" is 
rapour 
■tical 
near combination 
the number of 
mined, empirically 
imum performance 
ditions believed 
rticular area, or 
he correct 
is critical for 
from satellites, 
e North East 
determined by a 
Laboratory in the 
coefficents are 
ime the "split"- 
ue must be used, 
nergy in channel 3 
triple" window 
ults provided 
Figure 2: Location of the in situ 
stations used in verifi 
cation study. 
department of KNMI has performed some in-situ 
measurements on the North Sea. "Bucket"- 
temperatures are menasured at ± 30 stations (see 
fig. 2). These measurements are compared with AVHRR 
derived SST's. The date of the NOAA-9 image was 
October 25, 3-00 GMT. 
The in-situ measurements started on October 28, 
15.00 GMT and ended on the 30 th , 5.00 GMT. 
The SST's were also calculated using coefficients 
determined by McClain (1981). Figure 2 shows the 
in-situ measurements along with the two different 
satellite SST's retrievals. This graph shows that 
the in-situ spatial structure is well represented 
by the satellite derived SST's and also that the 
Rutherford coefficients perform slightly better 
than the McClain coefficients. The difference 
between in-situ measurements and satellite derived 
SST is never larger then 0.55 for the Rutherford 
coefficients: the mean bias being -0.064 and the 
RMS 0.244. For the Mcclain coefficients the 
statistics are: max. difference is 0.73; mean bias 
is 0.022 and RMS 0.292. 
These statistics are in good agreement with 
published results (Minnett et al, 1984). 
4.2. Land surface temperature (LST) 
4.2.1. Techniques 
For the development of a method to derive LST's out 
of AVHRR-thermal IR-data two more problems, except 
for the atmospheric correction arise: 
1. the variable emissivity of land surface. 
2. Lack of suitable groundtruth data. 
Price (1984) has demonstrated that the use of split 
window techniques originally developed for SST's, 
can also be used for the calculation of true LST's. 
He estimated a 2-3 K error in the estimates of 
LST's, using a SST- split window technique mainly 
caused by uncertainties ih the emissivity small 
compared to satellite-observed temperature 
variations in a certain region. 
t s (c) 
Î 
: in situ "bucket" temperatures 
» « * split window SST (Rutherford coefficients) 
• « : Ü „ „ (McClain „ ) 
Figure 3: Results of verification study for 
split window SST's. 
Oxford, U.K. we investigated the relation between 
split-window LST's and "screen"-temperatures, 
measured at meteorological stations. A data-set 
(day and night NOAA-9 imagery and meteorological 
data of the same times) for the period 14-20 April 
1985 was used. The stations involved are nearly all 
located in the U.K.. LST was calculated, using the 
Rutherford coefficients. It must be added that in 
general, differences between screen- and land 
surface temperatures will depend heavily on the 
actual weather conditions, the nature of the 
underlying surface and the type of vegetation 
cover. This makes the relation rather complicated 
and variable in space and time. The situation is 
still further complicated by the fact that the 
screen temperature measurements are point 
measurements whereas the LST's are representive for 
an area of approx. 1x1 km . this all renders the 
correlation between the two temperatures rather 
poor (regression coefficient 038). However 
byaveraging the surface- and air temperatures, of 
all cloudfree stations at a certain date, it can be 
shown that the average difference between surface- 
and screentemperature is fairly steady from day to 
day (see fig. 4). 
5. CONCLUSIONS 
- Digital AVHRR-imagery can suitable be processed 
in an automatic way to provide Earth-surface 
parameters. 
- The development of the "channel 4- channel 5"- 
technique turned out to be a valuable addition 
to available cloud clearing algorithms 
identifying thin cirrus clouds litherto going 
undetected. 
- Limited research at KNMI has demonstrated that 
the AVHRR can deliver relatively accurate 
surface temperatures (especially for sea 
surface), suitable for use in various 
applications. 
- KNMI now has as the only institute in the 
Netherlands the software available to perform 
all the necessary steps in the automatic 
processing of AVHRR-imagery. 
REFERENCES 
Research 
4.2.2. Validation 
In co-operation with the Met. Office Unit in 
Harries, J.E.,Llewellyn-Jones, D.T., Minnett, P.J., 
Saunders, R.W. and Zavody, A.M., 1983. 
Observations of sea-surface temperature for