THE DEVELOPMENT OF A HIGH PRECISION TROPOSPHERE EFFECT MITIGATION PROCESSOR FOR SAR INTERFEROMETRY Nico Adam German Aerospace Center (DLR), Remote Sensing Technology Institute ABSTRACT Troposphere effect mitigation based on numerical weather prediction (NWP) is an actual research topic in SAR interferometry (InSAR) and especially in persistent scatterer interferometry (PSI). This is the reason, a scientific troposphere effect mitigation processing system has been developed. The objective of this paper is to provide the methodology of four developed algorithms, demonstrate application examples, discuss the methods characteristic and recommend techniques for operational systems. Index Terms—persistent scatterer interferometry (PSI), atmospheric phase screen (APS), numerical weather prediction (NWP), wide area processing (WAP) 1. INTRODUCTION Troposphere artefacts are the most dominant error source in InSAR data. In recent years, their mitigation has attracted strong interest starting with [1], [2]. Subject is to improve the measurement precision and therefore provide operatio- nally troposphere corrections with all SAR scenes as proposed by [3] or provide an independent operational service [4]. However, troposphere effect mitigation based on numerical weather prediction (NWP) is still a research topic. This is the reason, an independent scientific troposphere effect mitigation processor (TEMP) has been developed by the author. Four mitigation techniques are implemented and were demonstrated and validated in projects e.g. Terrafirma and a pilot study [5]. Each of these methods has its own characteristic, advantages and limitation. Subject of this paper is to provide the methodology, demonstrate application examples and discuss the methods characteristic. 2. METHODS The Smith-Weintraub equation forms the basis of the NWP based troposphere effect mitigation [6]. () () () () () () () () () r T r e k r T r e k r T r P k r T r P k r N w w c d          2 3 2 4 1 ⋅ + ⋅ + ⋅ + ⋅ = (1) It models the scaled-up atmospheric refractivity () r N  by the partial pressures of dry air P d and of carbon-dioxide P c , absolute temperature T and the water vapor partial pressure e w . Rüger established the scaling constants k 1 , k 2 , k 3 and k 4 [7]. The range error d NWP can be estimated by the integrated scaled-up refractivity () r N  along the wave propagation path i.e. the line of sight (LOS) () ∫ ⋅ = − sensor scatterer R R NWP r d r N d     ) ( 10 6 (2) Via the radar wavelength λ, the atmospheric phase screen (APS) φ NWP is calculated by the straight forward relation NWP d NWP λ π ϕ 4 = (3) 2.1 Master selection support Using the WRF system [8], the atmosphere state is com- puted at the time of the SAR acquisition with a resolution of 3 km x 3 km. Now, the scaled-up atmospheric refractivity () r N  is computed separately for the dry (terms with P d and P c ) and the wet (terms with e w ) component using Eq. (1). For the characterization of the atmosphere, Eq. (2) needs to be implemented. In practice, the zenith direction integration path (Fig. 3) is convenient. A five-point Newton-Cotes integration formula integrates the tabulated data (i.e. the WRF grid, blue points in Fig. 3) fast and without heavy CPU load. A small value of the wet component in the scenes area indicates a suitable master scene. Alternatively, the wavelet transform provides an estimate of the structure function. 2.2 Vertical stratification mitigation At few scene locations, vertical profiles of the troposphere effect are calculated (Fig.1). For this method, the integration along the LOS is required and Eq. (2) is implemented by the Gauss–Kronrod quadrature formula. In practice, the increa- sed computation time of this quadrature algorithm is uncri- tical because a coarsely sampled grid of vertical profiles (e.g. every 10 km) is adequate. Starting from points along the vertical profiles (e.g. every 50m), the troposphere range error is calculated as a function of altitude for the master and slave atmosphere independently. Of course, the height de- pendent interferometric vertical stratification correction is calculated from the difference of these two values (Fig. 2 left). Typically, a third order polynomial can model this