Spectral Measurements of Solar Irradiance and Radiance in Clear and Cloudy Atmospheres - Short-Wave Solar Radiation in the Earth's Atmosphere - page 109

in the spectral range 350-850 nm with a step of 10nm accounting for the
smoothing halfwidth and narrowing the spectral interval of the smoothed
data (Vasilyev A et al. 1997a, 1997b, 1997c).
If the SBC is considered as a reflective characteristic of the surface only (and
not of the system “atmosphere plus surface”), the influence of the atmospheric
layer between the aircraft and surface on the SBC is tobemaximallydiminished.
For this purpose, the observations seem to be conducted at minimal altitudes.
However,thebumpsbecomestrongerwiththealtitudedecreasingandtheyield
of the random item to the total uncertainty caused by the bumps increases.
Thus, it is necessary to choose the optimal altitudes for the observations
so that the influence of the atmospheric layer below the aircraft would be
insignificant and the bumps would not be the main factor determining the
randomuncertainty. The experience of the flights on board of the IL-14 aircraft
has shown that the optimal altitudes above the water surface are 200-300m
and above the ground are 300-500m. However, sometimes the observations
had an occasion to be accomplished at non-optimal altitudes. The approach
excluding the influence of the atmospheric layer below the aircraft on the SBC
has been proposed in Kondratyev and Markaryants (1976), and Vasilyev O
(1986), with carrying out the measurement of the vertical profiles of the SBC.
Such observations were also conducted but their amount in the total quantity
of the spectra was not large thus, the authors have confined themselves to the
analysis of the SBC, measured at the altitudes from 500 m and lower (Vasilyev
A et al. 1997a, 1997b, 1997c). It is necessary to point out that the atmospheric
influence on the SBC is impossible to exclude as a whole even while observing
at low altitudes. It could be displayed as an overstating of the SBC values in the
UV spectral range caused by strong Rayleigh scattering and as an understating
of the SBC values within the oxygen and water vapor absorption bands in the
NIR spectral region.
Studying the SBC spectra dependence upon the surface type, viewing direc-
tion, solar zenith angle etc. is of greatest interest while analyzing the obtained
values SBC. The elucidating of the mentioned dependence is possible only
after statistical processing of the SBC data array with taking into account the
significance of the random item of the observational uncertainty. As there are
numerous and strongly varying factors (for example solar zenith angle varies
constantly) influencing the result, the application of the usual and easiest sta-
tistical methodology is ineffective in that case because the equal observational
conditions are impossible to attribute to large groups of spectra.
To overcome the difficulties mentioned above, cluster analysis (the method
oftheformalclassificationoftheSBCspectra)wasappliedduringthedata
analysis. Its essence is to divide the whole totality of the SBC spectra into classes
(or groups, clusters; Duran and Odell 1974; Gorelik and Skripkin 1989), each
of them forming without any a priori information but only using the principle
of the spectra “closeness”. Thus, the cluster analysis is reducing the problem
to a description of only a few numbers of classes. The metric function, i. e. the
distance between two classified objects is used as a numerical characteristic of
the spectra closeness (Duran and Odell 1974; Kolmogorov and Fomin 1989).
Thealgorithmoftheclassificationhasbeenconstructedrecurrently:letcertain