VEGETATION COVERAGE MONITORING USING SYNTHETIC APERTURE RADAR IMAGERY

V. Poenaru1,2, A. Badea1, S. Cimpeanu2 , D. Burghila2

1Romanian Space Agency, 2University of Agronomic Science and Veterinary Medicine

Acknowledgements:Radarsat-2 imagery was acquired in the joint ESA-CSA

SOAR Europe-16605 scientific proposal framework. The paper was published under theframe of European Social Fund, Human Resources Development Operational Programme2007 -2013, project no POSDRU/159/1.5/S/132765. This research work was carried outwith the support of Romanian Space Agency and also was financed from Project PN IIPartnership No 171/2013.

IntroductionToday, synthetic aperture radar (SAR) imagery is successfullyused to assess the state of forests and monitoring theirdynamics. In this study, the potential of C-band RADARSAT 2for monitoring vegetation cover is investigated. The testareas are chosen so that we analyze two scenarios underdifferent types of natural surfaces. Ocnele Mari salt miningarea (Fig.1), located in a hilly region, is affected by landdeformations (landslide and subsidence) and vegetationdegradation. The second is Braila agricultural area (Fig.2)covered with crops and forest having a flat surface. PolSARand PolInSAR were applied for a quantitative estimation ofvegetation cover combined with coherent scatteringmodeling.

ConclusionThe PolSAR, PolInSAR and multi-temporal SAR analysis were performed for vegetationcover monitoring. The classification accuracy is improved significantly if the Wishartclassifier is applied on the multi-temporal dual polarized dataset. Since SVM classifieraccuracy depends on training data, a Gaussian kernel function and a training parameterssetted on 0.5 had been used in the soil moisture classification. It observed that theclassification maps does not follow a specific pattern.In the PolInSAR analysis , the best results were obtained at the beginning (April 2015-Braila case) and at the middle of the vegetation season (July 2014- Ocnele Mari case).

ResultsThe classification results are depicted in Fig. 3 and Fig. 4. It is observed that the SAR geometry effectsand topography induce a relative mixture in the scattering mechanism (Fig. 3) whiles higher values of soilmoisture content are obtained in Danube meadow by applying the SVM classifier. PolInSAR techniquewas applied to the interferometric pairs with short temporal baselines to extract vegetation height(Fig.5). The interferometric coherence is optimized by using polarimetric correlations betweenpolarizations so that the polarization which could yield highest coherence is obtained. Thus, vegetationheight estimation is improved, but remains overestimated due to topography (Ocnele Mari case) andhigher moisture content (Braila case).

ResearchPolarimetric SAR allows as to extract information for identification andclassification of different natural features since each polarization is sensitiveto different surface characteristics (shape and orientation) and properties(soil moisture, surface roughness and vegetation cover). In the polarimetricSAR interferometry that combines interferometric and polarimetric SARimaging techniques, information of different polarization channels are usedto investigate the object structure and perpendicular layers of the scatterer.Polarimetric coherence optimization that implies maximization of thesignal-to-noise ratio, usefully in more accurate topography estimation, isapplied to vertically structured media for obtaining dominant scatteringcenters ( Claude et al., 1998; Neuman et al., 2007). A set of Fine Quad anddual polarized RADARSAT-2 data, acquired in ascending and descendingmode, in the joint ESA-CSA SOAR Europe-16605 scientific proposalframework covering July 2015 to May 2015 period, have been used in thestudy. Table 1 lists the image pairs with their respective temporal andperpendicular baselines used in this study.

Interferometric pairs Perpendicular baseline [m]

Height ambiguity [m]

Incidence angle

[0]

Temporal baseline [days]

Polarizatio

n

Mode

02.07.2014-26.07.2014 -2.96 5457.2 36 24 VV+VH Descending

19.082014-12.09.2014 -90.62 178.06 36 24 VV+VH Descending

23.11.2014-17.12.2014 -17.88 902.26 36 24 HH+HV Ascending

31.03.2015-24.04.2015 -79.168 165.46 30 24 Quad Descending

24.042015 – 18.05.2015 76.005 172.356 30 24 Quad Descending

Fig. 1. Ocnele Mari salt mining area

Fig. 2. Braila agricultural area

Table 1. Interferometric pairs from RADARSAT 2 data

Fig.3. Wishard supervised classification. (RADARSAT-2 Data andProducts© MacDonald, Dettwiler and Associates LTD (2014)- All rightreserved. RADARSAT is an official trademark of the Canadian Space

Agency).

Fig.4. Support Vector Machine classification. (RADARSAT-2 Data and Products©MacDonald, Dettwiler and Associates LTD (2014)- All right reserved. RADARSAT isan official trademark of the Canadian Space Agency).

Fig.5. Estimation of vegetation heights: (RADARSAT-2 Data and Products©MacDonald, Dettwiler and Associates LTD (2014)- All right reserved. RADARSAT isan official trademark of the Canadian Space Agency).