Geoscience Reference
In-Depth Information
evidently has a sense of the spectra density in this class. Hence, the spectrum
providing the maximal spectra density in the class has been chosen as starting.
The formal computer classification has to be followed by a stage of hand
analysis. At this stage, the information concerning measuring factors is ana-
lyzed for every class. Some classes are united after this analysis. The important
result of the cluster analysis is the automatic revealing of the erroneous spectra
thatcanfindawideapplicationintheoperativeprocessingoftheatmospheric
and surface radiative characteristics. Actually the spectrum could be assumed
an erroneous one if there is only one spectrum in the class.
After the classification procedure the mean value and standard deviation is
calculated for every class over all contained spectra. The standard deviation is
mentioned to be sometimes less than the initial random standard deviation s i .
It is suggested that there are mainly two reasons for this: the spectra statistical
averaging, and the yield to the standard deviation of the uncertainty linked
with the surface heterogeneity in the real observational conditions, which
could be less than the average estimation in Table 3.1.
Fig. 3.17. Spectral brightness coefficients (SBC) of the typical natural surfaces. Average
values of the SBC of the correspondent classes and the one standard deviation interval:
1 - pure lake water with low chlorophyll and mineral matter content; 2 - lake water with
high chlorophyll and mineral matter content; 3 - snow; 4 - sand; 5 - black soil; 6 - green
vegetation (grass); 7 - yellow vegetation (ripe grain crop)
 
Search WWH ::




Custom Search