Analysis of High Resolution Multi-frequency, Multipolarimetric and Interferometric Airborne SAR Data for Hydrologic Model Parameterization Martin Herold 1, Volker Hochschild 2 1 Remote Sensing Research Unit, University of Santa Barbara, USA, e-mail: martin@geog.ucsb.edu 2 Department of Geography, University of Jena, Germany, e-mail: c5voho@geogr.uni-jena.de Abstract In this study data of multifrequency, polarimetric and interferometric airborne Experimental Synthetic Aperture Radar (E-SAR) were evaluated and analysed to derive land use information, surface soil moisture, vegetation parameters and topographic information after a standardized processing scheme. The experiments were integrated in two projects of hydrologic and solute transport modelling and simulation in two test areas in Germany. The results show a good representation of the investigated land surface features in the remote sensing data. Under consideration of forthcoming spaceborne multiparameter SAR-systems these techniques can provide miscellaneous spatio-temporal information about the earth surface as required for hydrological applications, i.e. physically-based, distributed hydrological modelling. 1 Introduction The knowledge of the three-dimensional heterogeneity of topographical, pedological, vegetation and land use features of a river catchment is an important requirement of distributed, physically-based representation and modelling of the hydrological system respectively runoff (FLÜGEL 1996). During the last years remote sensing methods in general and radar techniques in particular have shown their potential to derive spatial information for parameterization, calibration and validation of hydrological models (MAUSER et al. 1997, HOCHSCHILD 1999). Multifrequency, polarimetric and interferometric microwave remote sensing data are sensitive for three-dimensional dielectric and structural features of the earth surface allowing the derivation of soil moisture, vegetation parameters, land use and topographic informations as investigated in this studies. In the coming years, due to the launch of spaceborne systems like ENVISAT, TerraSAR or ALOS (ULABY 1998), these techniques will play an important role in estimating areal land surface parameters. Accordingly, this study aims to apply and to evaluate their potential for model parameterization in MMS/PRMS or WASMOD.
In this research project airborne SAR data of two flight campaigns using the Experimental Synthetic Aperture Radar (E-SAR) of the German Space Agency (spatial resolution 1 m) have been evaluated in investigations of a detailed study on hillslope hydrology in the Broel catchment (Rheinisches Schiefergebirge, Germany, FLÜGEL & SMITH 1999) and microscale solute transport simulation for the Thuringian Drinking Water Reservoir Administration (TTV) in Zeulenroda (Thuringia, Germany). The data acquisition took place on the 26 th of June 1996 in the basin of the river Bröl in the Rheinish Slate Mountains and on the 30 th of April 1999 in the Zeulenroda basin. Simultaneously, reference land use mapping and field measurements, including vegetation parameters (biomass, height, plant water content) and soil moisture (tensiometer, TDR and gravimetric sampling, 0-15 cm) were conducted. 2 Image Processing The digital image processing was carried out according to the Level Approach of ULABY et al. (1996). This method involves a hierarchical classification procedure developed particularly for multifrequency and polarimetric data. Figure 1 shows the Level Approach and the image analysis methods applied in this study. All preprocessed channels were used for an unsupervised Level I classification. A separation into the classes of farm- and grassland, forest/settlement and shadow/water was possible by analysing the backscatter characteristics. In Level II the forest and settlement areas were divided due to the spectral and textural characteristics of the L-bands and in a following step the forest class into deciduous and coniferous forests. The quantitative derivation of parameters in Level III relates to the surface soil moisture and the vegetation parameters. Therefore the Principal Components Analysis (PCA) proved to be a suitable method for SAR data analysis, as mentioned previously in VERHOEST et al. (1998). The Principal Components were calculated from the multipolarimetric L-bands in the Level II segmented image areas. From correlation with the field measurements of the surface soil moisture and vegetation parameters like the plant water content, vegetation height, and drymass a quantitative parameterization was performed (Fig. 2). 3 Results of Hydrological Parameter Derivation The hydrological relevance of land use, topography, soil moisture and vegetation parameters are well known. Land use contains spatial and temporal information about vegetation types, degree of imperviousness and solute transports. The soil moisture is a crucial parameter for the energy and water balance between earth surface and the atmosphere as well as for runoff generation. The vegetation parameters yield information about the physiological condition of the plants and system losses due to mowing or harvesting of agricultural crops.
Fig. 1 Methods, techniques and accuracy of SAR data processing according to the level approach (mp = multipolarimetric) Fig. 2 Correlation between the Principal Components and in situ field measurements
3.1 Land use The land use map was derived by aggregation of the seven Level II classes. The overall classification accuracy was estimated with 88.3 % using ground truth data. The map represents the model-required hydrological land use categories as well as agricultural pattern (Fig. 3). This information has been used for the derivation of Hydrological Response Units (HRU), areas of homogeneous hydrological dynamics (FLÜGEL 1996). 3.2 Topography Interferometric X-band SAR data were used to calculate a digital surface model with 5 m horizontal and 0.5 m vertical resolution. For correction of the vegetation heights the land use information and radar backscatter information were included in the retrieval of a digital elevation model as required for various hydrological models. 3.3 Soil moisture For determing spatial soil moisture distribution the lower frequencies (e.g. L-Band, 23 cm wavelength), which partly penetrate through vegetation into the soil surface (penetration depth 5-10 cm), are most effective. The soil and vegetation induced backscatter components were separated by the calculation of principal components (PC) from the L- band polarizations. A regression between the tensiometer field measurements and the values of the first PC (R=0,79) were applied to estimate a surface soil moisture map (Fig. 4). The spatial soil moisture distributions were compared with other measurements derived (a) from the DEM (multiple flow topographical index) and (b) with geostatistical methods (kriging) from data of 51 TDR-measurement points (2-3 measurements at each point). It demonstrates the good representation of topography-induced soil moisture variations with wet drainage lines visible in the microwave data, while the slopes are dryer (Fig. 4). At the Kiefer test site, there is a saturated zone in the middle of the slope resulting from a dip in the underlying geology. Interflow from the upper slope is accumulating in this dip and then linearly draining to the valley floor. This saturated area is visible in the radar backscattering (Fig. 4 a) and the TDR soil moisture sampling (Fig. 4 c), but not in the mf-topographical index (Fig. 4 b), because this feature is not represented in surface topography. 3.4 Vegetation parameters The derivation of vegetation parameters of short vegetation was performed by correlation of radar backscatter and field measurements of plant water content (PWC), biomass and vegetation height (correlation with field data: R = 0,95-0,98). Figure 3 shows the spatial distribution of the plant water content. Low values are found for freshly mowed grasslands, higher plant water contents could be found for crops or non-cut meadows.
Fig. 3 Land use map and distribution of plant water content of short vegetation in the Broel-testsite Fig. 4: Surface soil moisture distribution at the test sites Kiefer (top) and Simon (bottom) derived from radar backscatter (a), result of the multiple flow topographical index (b) and interpolated from TDR measurements (c), (Note: The varying values of the soil moisture scale result from different field methods)
4 Conclusion The results of this study show the potential of multi-frequency, multi-polarimetric and interferometric SAR data for derivation of hydrological land surface parameters such as land use, vegetation, soil moisture and topography. The most useful Radar information were found in the L-Band data that provide both vegetation and soil moisture signals with distinctive differences in the different polarizations (see also Herold et al. 2000). Adequate data will be available from future space-borne systems. However, the SAR data analysis will get more complicated if structural land surface features (surface roughness, different plant structures) are more heterogeneous than in this study area. This issue should be addressed and investigated in further research activities. References Flügel, W.A. (1996): Hydrological Response Units (HRU s) as Modelling Entities for Hydrological River Basin Simulation and their Methodological Potential for Modelling Complex Environmental Process Systems Results from the Sieg Catchment. Die Erde, 127, 43-62, Berlin. Flügel, W. A. & Smith, R. E. (1999): Integrated process studies and modelling simulations of hillslope hydrology and interflow dynamics using the HILLS model. In: Environmental Modelling and Software, 14, 153-160. Herold, M., Hochschild, V. & Schmullius. C. (2000): Multifrequente und multipolarimetrsiche Radarfernerkundung hydrologischer Parameter der Landoberfläche, in Photogrammetrie, Fernerkundung, Geoinformation, 355-360. Hochschild, V. (1999): Parameterization of Hydrological Models: The Contribution of Remote Sensing to Water Resources Management. Proceedings of the MODSIM 99, International Congress on Modelling and Simulation, 06.-09.12.99, Hamilton, NZ. Mauser, W., Bach, H., Demicran, A. Eibl, B., Riegler, G. & Schneider, K. (1997): The Contribution of Microwave Data to Distributed Hydrologic Modeling. Proceedings of the Third ERS Symposium, Florence. Ulaby, F. T., Dubois, P. C. & van Zyl, J. (1996): Radar Mapping of Surface Soil Moisture. Journal of Hydrology, 184, 57-84. Ulaby, F. (1998): SAR Biophysical Retrievals: Lessons learned and Challenges to Overcome. Proceedings of the 2. International Workshop on Retrieval of Bio- and Geo-Physical Parameters from SAR Data for Land Applications, ESTEC, Noordwijk. Verhoest, N. E. C., Troch, P. A., Paniconi, C. & De Troch, F. P. (1998): On the retrieval of bio-physical parameters from multi-temporal series of ERS-SAR PRI images. Proceedings of the 2 nd International Workshop on Retrieval of Bio- and Geo-physical Parameters from SAR Data for Land Applications, ESTEC, Noordwijk.