I n s t r u m e n t s

a n d

M e t h o d s

according to TM.These results are consistent with what
the differenceimage analysis has shown.

CONCLUSION

We have examined the spatial,qualitative, andquantita-
tive properties of the accuracy of aSIR-C snow map.The
image is reasonable accurate forthe pure snow case and
thepure snow-freecase, buthasdifficulties withmixed
pixels.There remains much work to be done in this area.
In particular,we plan on investigating the linear overesti-
mation features andperforming some analytical statisti-
cal test on the data, so that misclassification likelihood can
be bettercharacterized. Additionally,the algorithm must
be testedon a differentgeographicareato test its portability.
We arecurrentlypursuingthisgoalforasiteinthe
Himalaya and a site in the Bolivian Andes.

REFERENCES

Congalton, R. G., 1991, A Review of assessing the accuracy of
classifications of remotely sensed data,Remote Sensing of En-
vironment, 37:35-46.

Dozier, J., 1989, Spectral signature of alpine snow cover from
theLandsatThematicMapper,RemoteSensingofEnviron-
ment , 28:9-22.

Rango, A and Itten, K. I., 1976, Satellite potentials in snowcover
monitoring and runoff prediction,Nordic Hydrology , 10:209-
230.

Rosenthal, C. W. and Dozier, J., 1996, Automated mapping of
montane snow cover at subpixel resolution from the Landsat
Thematic Mapper,WaterResources Research ,32:1:115-130.

Rott,H. and Davis,R. E., 1993, Multifrequencyandpolariza-
tion SAR observations on alpine glaciers,Annals of Glaciology,
17, 1993.

Shi J.andDozier,J., 1996, Mapping seasonal snow withSIR-
C/X-SARin mountainous areas,RemoteSensingof Environ-
ment ,in press.

Though limited,there are some areas where SIR-C pre-
dictsmoresnowthanTM.Mostnotable,are thelinear
overestimationfeatures nearthecenter oftheimagebe-
tweenthe shortvegetation /baresurfaceand the forest.
These featuresmay be linked to the transitional vegetation
in these areas. This requires further investigation.
Turning our attention to the confusion matrix analysis,
the overall accuracyfor the binary image is 74%. The pro-
ducer's accuracyis 57%, implying that for this scene, SIR-
C often fails to detect snow where TM finds that it exists.
User's accuracy forthis scene is 71%.Thissuggests that
whatSIR-Cclassifiesassnowisalsolikelytobesnow

Fig.6

132