172
Picture 7: the same area in two different spatial
resolutions images (30 m and 500 m)
Soft classification allows evaluating the classification
accuracy by a modified confusion matrix that compare,
class by class, not the number of pixels but the
membership degrees. In particularly snow class accuracy
classification is valued as the RMS of differences
between well-known snow degrees and classified ones
for a testing set. Some tests done give snow membership
degree accuracy about 10%; this value is suitable for
small scale snowpack monitoring.
Snow covered area con be evaluated as sum of the
contributions of all the pixels in the image. Each
contribution is in proportion to the pixel snowy condition
in the following way:
Nimage
Snow area = pixel size • £ snowdegree j
i=1
where:
Snow area is the snow-covered area estimated.
Pixel size is the terrain area corresponding to the
sensor IFOV.
Nimage is the number of pixels of image
Snowdegreei is the membership degree to snow
class assigned to pixel i; this value represents the
snow presence percentage into the pixel.
Snow covered area estimation by soft classification is not
underestimated because all the pixels contributions are
considered by the snow membership degrees to snow
class. 6
6. FINAL REMARKS
Snow covered area estimation via remote sensing is not
a new object. At the beginning it was done by
photointerpretation but this is a quite long via and
accuracy can not be evaluated. Estimation by automatic
hard classification (i.e. Maximum Likelihood algorithm)
requires less time but the accuracy is not suitable
because the system is unable to menage mixture and
snowpack borderlines are recorded as mixed pixels. Soft
classification by Fuzzy-statistic algorithm gives good
results (as accuracy of estimation of snowpack
extension) but requires training set about mixture too and
this information can be difficult to find and to valued.
Next step will be use Fuzzy neural network system that
allow supervised soft classification but does not require
training set about mixture.
REFERENCES
1. Brivio, P.A.; Lechi, G.M.; Zilioli, E. (1992) - ||
Telerilevamento da aereo e da satellite, Carlo Delfino
Editore
2. Campbell, J. B.; (1996) - Introduction to remote
sensing - Tylor & Francis, London
3. Colombo, A.: ''Telerilevamento per il monitoraggio del
manto nevoso”. Politecnico di Milano, PhD Thesis
1998
4. Dozier, J.; Stanley, R.; Mc Ginnis, D.F.; (1981)
Effects of grain size and snowpack water equivalence
on visibile and near infrared satellite observations of
snow, Water Resources Research, 17(4), pp 1213-
1221
5. Gartska, W.U; (1964) - Snow and snow survey - In.
Handbook of applied hydrology, McGraw-Hill, New
York, section 10, pp. 10-12
6. Hall, D; Martinec J. (1985) - Remote sensing of ice
and snow - Chapman and Hall LtD -USA.
7. Lucas, R.M.; Harrison, A.R.; Barrett, E.C.; (1989),
Snow monitoring in the United Kingdom using NOAA
AVHRR satellite imagery, Proceeding
WMO/IAHS/ETH, Workshop on precipitation
measurements
8. Martinec, J.; (1973) - Evaluation of air photos for
snow-melt runoff forecast, Unesco-WMO-IAHS
Symposium on the role of snow and ice in hydrology,
Banff 1972, IAHS Publication No 107, Voi. 2, pp. 915-
26
9. Miller, D. M.; Kamkisky, E. D.; Rana, S. (1995) -
Neural network classification of remote sensing data.
Computers and Geosciences, Voi 21, pp 377-386
10. Norwegian Hydropower Companies (1998): Satellite
based snow monitoring, Proceeding II Alpine
Conference Atelier neige et télédétection - SHF,
Grenoble
11. Potts H.L.; (1944) - A photographic snow-survey
method of forecasting runoff modelling Trans. Am.
Geophys. Union, Prt 1, pp. 149-53
12. Ranzi R., B. Bacchi; (1996); Multitemporal monitoring
of the snowpack in large alpine watersheds using
NOAA-AVHRR radiometry and ground
measurements, Technical Report, No 9, 1996,
Dipartrimento di Ingegneria Civile dell'Università agli
Studi di Brescia
13. Wang, Fangju, (1990) - Fuzzy supervised
classification of remote sensing images. IEEE
Transaction on Geoscxience and remote sensing, Voi
28, pp 194-201
