Assessing the Response of Alpine NATURA2000 Habitat Changes by using Dual Polarized (VV/VH) COSMO-SkyMed Imagery
Spindler, Nadine1; Polychronaki, Anastasia2; Notarnicola, Claudia2
1Eurac, ITALY; 2Eurac/ Insitute for Applied Remote Sensing, ITALY
Natural and semi-natural habitats in Alpine regions are experiencing an increasing loss of biodiversity due to human activities, such as tourism, intensification of agriculture and infrastructure development. To protect such ecologically sensitive areas, the NATURA 2000 network was established under the Habitat Directive (92/43/EEC). NATURA 2000 obligates every EU member state to report on the quality status of habitats every six years. For that up-to date maps providing information about the distribution of natural and semi-natural habitats are an important prerequisite.
As yet, the potential of remote sensing has not been fully investigated for mapping habitats in Alpine regions. Moreover, unstable weather conditions of alpine regions cause tremendous data gaps when relying only on optical data. As a result, the aim of this study is to investigate Cosmo - SkyMed PingPong X-band to 1) test the ability to discriminate different Natura 2000 habitats with particular focus on the different grassland types and 2) to derive a correlation with biomass ground measurements to exploit the synergy effect of using optical as well as radar data. The study area of the Nature Park Rieserferner Ahrn is located in the Northeast of South Tyrol, Italy. The park is part of the Natura 2000 framework covering an area of 31.3 ha with elevation ranging from 890 to 3.480 meters above mean sea level. Dominating Natura 2000 types are represented by rocky habitats as well as spruce and coniferous habitats. One fifth of the area is represented by natural and semi natural grasslands whereas less than five percent are covered by shrubs. Cosmo-SkyMed Data was available in Stripmap PongPong Mode with Dual Polarization VV and VH for eight different acquisition dates (03.08.2011, 11.08.2011, 20.08.2011, 27.08.2011, 04.09.2011, 12.09.2011, 20.09.2011, 06.10.2011).
A very detailed reference map indicating the distribution of Natura 2000 habitats was provided on the basis of high resolution optical Rapideye data using an SVM classification approach.
Furthermore, ground measurements on biomass for different alpine grassland types as well as shrubs were available.
The Cosmo-SkyMed L1A Data were pre-processed using the commercial Software SARScape. Initially, the data were multilooked using one look in range and two in azimuth were used for VV/VH polarization. Subsequently, all images were co-registered.
Speckle noise was reduced using the multitemporal De Grandi filter. Following, the data were radiometrically calibrated as well as geocoded using a 2.5m LIDAR DEM. Areas of layover and shadow were masked out.
The Radar information was associated with the both reference data 1) the result of the Natura 2000 habitat classification based on optical data (RapidEye) as well as with 2) the ground measurements providing information of biomass. A mean backscattering value for the two polarizations VV and VH was calculated for each reference polygon.
Associating the backscatter values with the Natura 2000 habitats indicate the ability to separate natural grasslands (6150) from mountain hay meadows (6520) and species-rich Nardus grasslands (6230) as well as the alpine and boreal heaths (4060) in VH polarization. High backscatter values are found for natural grasslands (6150) due to the heterogeneous composition of this type which is mainly influenced by rocky patches. In contrast to this, habitat 6520, 6230 and 4060 are composed of very dense vegetation which lead to a signal decrease causing a difference of 2dB in VH. Significant differences occur occur in the absolute values between wet and dry events whereas the difference between natural grasslands (6150) and the other habitats 6520, 6230 and 4060 remains the same. In contrast, VV polarization highest values are found for mountain hay meadows (6520) whereas the backscatter is less influenced by the other classes 6150, 6230, 4060. As a result of cutting events habitat 6520 shows a clear response to soil whereas the other three classes indicate a low response due dense vegetation in 6230 and 4060 and the high contribution of vegetation and rocks in 6150. Relating the backscatter values to the ground measurements of biomass results verify the above mentioned conclusion. VH polarization indicates a positive correlation between the backscatter values and the biomass values whereas VV polarization reveals a negative correlation.
Future work will focus on the correlation between Cosmo-SkyMed backscatter and NDVI on the one hand. On the other hand the synergy effect of radar data together with optical data in habitat classification processes will be exploited.