Land cover classification by using multi-temporal COSMO-SkyMed data

This paper uses a time-series of COSMO-SkyMed SAR images for land cover classification and soil moisture retrieval over an agricultural area located in Southern Australia. The SAR products analyzed are 11 StripMap Ping Pong images, at HH and HV polarizations, acquired at 21° incidence angle and with a revisiting time of either 8 or 16 days. The classification accuracy has been assessed as a function of the polarization and the number of images analyzed. Results confirm that the temporal information is crucial to improve the classification results. An overall accuracy of approximately 82% was achieved for 10 classes. Moreover, soil moisture (mv) maps over bare or sparsely vegetated areas have been retrieved by means of the SMOSAR-X ('Soil MOisture retrieval from multi-temporal SAR data') algorithm, developed in view of the forthcoming Sentinel-1 data and then adapted to X-band SAR data. The SMOSAR-X algorithm is shown to produce mv maps with an rmse of 6.6% v/v. © 2012 IEEE.