Relating Classification Accuracy Variations with Incidence Angle Changes at C-band
TU Bergakademie Freiberg, GERMANY
Fine quad-polarization experimental mode of RADARSAT-2 [RST, also quad1 (beam mode FQ13; incidence angle of 33.2 degrees) and quad2 (beam mode FQ6; incidence angle of 25.5 degrees)] acquired on September 19 and September 26, 2009 were evaluated for the classification of successional forest stages and land use classes. The Tapajós National Forest (TNF; Eastern Amazon, Brazil) was selected as a study area. It contains shifting cultivation practices and as a result a complex mosaic of successional forest stages. The SAR data were used independently and combined in a temporal approach and together with a clear-sky ALI/EO-1 image acquired July 11, 2009. It was taken as a reference and additional information for the SAR dataset classification scheme. No precipitation events were registered prior data acquisition. Intensity backscattering, polarimetric features, and texture parameters were evaluated for the classification purposes using random forest (RF) classification. Additionally, field measurements were conducted and biophysical parameters such as above ground biomass (AGB) and plant area index (PAI) determined. Results showed that the forest classes were characterized by low temporal backscattering intensity variability and high entropy leading to misclassification between secondary forest classes. Classification accuracy was also affected by the incidence angle changes. Changes varying from 5 to 10% were observed in the individual land use classes. Misclassifications were reduced by integrating optical data. The joint analysis of SAR and their derived parameters with optical data performs even better and thus encourage the further development of joint techniques under the recent political support from the United Nations (UN) endorsing the "Reducing Emissions from Deforestation and Degradation" (REDD+) protocols.