River Ice Mapping from TERRASAR-X Images
Stéphane Mermoz(1), Sophie Allain(1), Monique Bernier(2) and Eric Pottier(1)
(1) IETR, 262 avenue Général Leclerc, 35042 Rennes, France
(2) INRS ETE, 490 rue de la Couronne, G1K9A9 Quebec (QC), Canada
Rivers certainly play an important role in the water resource distribution, particularly in winter when the land surface runoff is minimal. The development of ice covers affect streamflow, modify the ecosystem, cause flooding, restrict navigation, influence hydropower generation and even impact on the climate. In addition, hydroelectric companies and government services require spatially distributed information about the types and characteristics of river ice. Methodologies have been developed to discriminate ice types at C-band using single-polarized - and fully-polarized SAR data . In order to improve these classifications, the use of polarimetric SAR data at X-band is promising . In fact, at this frequency, electromagnetic waves are more sensitive to volume contribution and should permit to improve the discrimination of ice types.
This paper evaluates the robustness of different classification algorithms to discriminate river ice types using dual-polarized X-band SAR data acquired by the TerraSAR sensor.
The study area is the Saint-François River near the city of Drummondville (P.Q. Canada). Two TerraSAR datasets (HH-VV and HH-HV) were acquired at X-band on March 10 and 13, 2008. Field data were acquired on March 10-14, 2008. Digital airborne photos were taken and 25 ice cores were extracted from 5 sites. For each ice core, the thickness, the porosity, the size and distribution of air bubbles within each ice layer were measured.
Four cover types could be identified using these SAR images. These cover types are: open water, ice without frazil, ice with frazil and consolidated ice (a rougher ice cover easily identifiable). For each class, some ice core localisations are selected as training sites and others as validation sites for classifications.
The two datasets were geocoded and hence jointly used in order to develop more robust algorithms. Different classification algorithms like a supervised Wishart approach have been computed using the two dual-polarized TerraSAR datasets. The results between each data set and each algorithm were compared using confusion matrix.
1. Gauthier, Y., Weber, F., Savary, S., Jasek, M., Paquet, L.M. and Bernier, M. A combined classification scheme to characterize river ice from SAR data. In EARSeL, eProceedings, vol. 5, no. 1, pp. 77-88, 2006
2. Puestow, T.M., Randell, C.J., Rollings, K.W., Khan, A.A. and Picco, R. Near real-time monitoring of river ice in support of flood forecasting in eastern Canada: towards the integration of Earth observation technology in flood hazard mitigation. IEEE Trans. Geosci. Remote Sensing, vol. 4, pp. 2268-2271, 2004.
3. Mermoz, S., Allain, S., Bernier, M. and Pottier, E. River ice mapping from polsar images. Proceedings of IGARSS’2008, Boston, USA, 2008.