Discussion and Conclusions

Abstract

Area 3 was used for investigation of the classification results. The results are shown in Table 2. The table shows that clustering of VV data gave the best results with an error rate of 7.2%. The least good results were obtained for crosspolarization with an error rate of 12.2%. ERS DATAERS-1 SAR PRI datasets from 29 March, 6 June, 11 July and 12 July have been calibrated and processed into terrain corrected images in Universal Transverse Mercator (UTM) map projection by applying high resolution (5m x5m) DEM data and geocoding software [4]. The DEM is derived from airphoto. A 3x3 Lee filter was applied to the data before conversion to dB. In Fig. 3 the mean ERS-1 SAR backscattering coefficient for from, 29 March, 6 June, 11 July and 12 July, respectively, are shown for two areas close to the previously defined areas are used. A decrease of 4 dB in backscattering coefficient is observed for the high mountainous area between 29 March to 6 July. This change is related to the change in snow properties. In 29 March the area is covered with dry snow while in June the area is covered with wet snow. We clearly observe a change between the ascending 11 July and descending 12 July ERS pass. This is caused by the difference in viewing geometry giving rise to different local incidence angle. DISCUSSION AND CONCLUSIONSThe backscatter statistics of two areas with an elevation difference of about 450 m was studied. The difference of the mean of the class snow between the two areas were largest for co-polarization with about 4.4 dB. Correspondingly, it was 1.8 dB for cross-polarization. For bare ground, the corresponding numbers were less than 1.0 and 1.3 dB. Since the ground conditions for snow were very similar in the two areas, the main reason for the change of the backscatter level is probably the incidence angle. For the purpose of classification, a preliminary conclusion is that local classstatistics must be applied. If the reason for variation is mainly due to the incidence angle parametrized class models may be designed. A classification test using K-means clustering showed best results with an error rate of 7.2% for VV polarization. All error rates were between 7.2 and 12.2. An investigation of a larger area is necessary in order to draw more clear conclusions.EMISAR C-band polarization responses from wet snow correspond to theoretical responses from smooth surfaces. The polarization response at L-band show a higher degree of diffuse scattering than C-band. The ground measurements include measurements of snow density, snow grain size, snow liquid water content and surface roughness. Air-and snow temperature data are also available. Several trihedral corner reflectors were deployed within the field for calibration and georeferencing purposes. The field measurements are georeferenced using GPS. One aerial photo of the test site was taken July 14. The field was completely covered with snow on March 22 and on May 1-3, while in July the field was nearly snowfree. REFERENCESThe calibration of the EMISAR data have been verif...