Methods InSAR 1

L-band synthetic aperture radar interferograms have the potential to measure geophysical processes such as earthquakes, volcanoes, landslides, and glacier and tectonics movements. The coherence of L-band interferograms is usually higher than that of C- or X-band interferograms thanks to the higher penetration rate of L-band signals, improving the measurements. On the other hand however, lower frequencies are also more sensible to the ionospheric influence. The ionosphere is a layer of the atmosphere where the electron density of ionized gases is high enough to affect the propagation of radiowaves. The effects on L-band signals are a phase advance and a time delay which, converted to meters, can consist of even several meters. Variations of the differential ionospheric total electron content (TEC) between the acquisitions are measured from the interferogram; they are superimposed to topography and ground deformation signals, hindering the measurement of geophysical processes. The ground movements, between acquisitions, in the along-track direction can normally be measured by cross-correlating the SAR images. However, a second ionospheric effect to SAR signals, in the form of azimuth shifts, is generated by the variations of the TEC along the azimuth direction. For this reasons, even if L-band data are more suitable to measure ground movements with respect to C- or X-band data, in particular in difficult conditions like vegetated areas, they are more susceptible to errors due to ionospheric variations. These ionosphere-induced errors have therefore to be removed to enable thorough modelling of geophysical processes.

 In this work we propose a method to estimate and compensate ionospheric effects to interferograms. The method is based on the combination of two ionosphere estimation techniques, with the aim to take advantage of the strengths of each technique, and to obtain a precise and robust result. The two techniques are the split-spectrum method and the azimuth-shift method. The split-spectrum method is based on the dispersive nature of the ionosphere and separates the ionospheric component of the interferometric phase from the nondispersive component related to topography, ground motion, and tropospheric path delay. The split-spectrum method ensures accurate estimation over long wavelengths and can recover range variations, but cannot recover rapid changes of the ionosphere. It is therefore not precise enough to estimate the small-scale ionospheric variations and consequently the ionosphere-induced azimuth shifts. The azimuth-shift method exploits the proportional relation between differential azimuth shift and the azimuth derivative of the differential ionosphere. Being sensitive to local azimuth variations of the ionosphere, the azimuth shifts can be used to estimate the high-frequency components of the ionosphere spectrum but are prone to an increasing error in the long distance and are insensitive to range variations. The proposed method uses three azimuth sub-bands to estimate the azimuth shifts and to separate ground motion from the ionosphere-induced shifts. A Bayesian inverse problem is used to combine the two techniques.

 The results of applying the proposed method to ALOS PALSAR images will be presented in this work. Earthquakes often show distinctive ground motion signals, L-band interferograms of earthquakes which also show ionospheric disturbances are used to demonstrate the potentials of the combined method with respect to the single split-spectrum technique.

In this study, we present a 3D-Variational data assimilation (3DVAR) experiment with precipitable water vapor (PWV) data estimated from high spatial resolution interferometric maps. Improving predictions of rainfall is currently one of the main challenges for numerical weather prediction (NWP) models. Such phenomena often occur in localized areas and have a high temporal variability, associated with numerical and computational limitations; it is difficult to obtain a rainfall forecast with the desired accuracy. Recent assimilation experiments suggest that the PWV values estimated from GPS and interferometric data can play an important role in determining more accurate forecasts of rainfall. In the last decade, the repeat-pass space-borne interferometric synthetic aperture radar (InSAR) technique has been widely used to estimate the variability of the PWV with high spatial resolution and high precision. We estimated interferometric maps using SAR images acquired from the ENVISAT-ASAR mission over the Lisbon region, Portugal. The interferometric SAR images have been used to estimate a map of the temporal changes of PWV, with a spatial resolution of about 160 m and a precision less than 1 mm. We also present a methodology to derive the absolute PWV maps based on the ECMWF/ERA-Interim (with spatial resolution of 14-km), GPS, and meteorological stations data. We use the Weather Research and Forecast Data Assimilation (WRFDA) model, version 3.7, at micro-scale resolutions (1 km) with and without data assimilation, to study the effect of assimilating PWV maps obtained by SAR interferometry on predicting light rain (< 5 mm). It is noteworthy that, in the case studied in this work, the model without assimilation data fails to adequately reproduce the rain quantity and pattern observed in the meteorological radar images. We assessed both forecasts (with and without data assimilation) with PWV values estimated from 12 GPS stations as an independent validation. The results of this experiment show that after the assimilation of InSAR-PWV data, the model improved the predictions of water vapor density, hydrometeors circulation and rain during a minimum of 6-h windows. The results were compared with in-situ meteorological data and with a meteorological radar. The availability of interferometric PWV maps on a routine basis can help to capture the high variability of the water vapor distribution at micro-scales. In this study, we show that the knowledge of the PWV with high spatial resolution, can serve as an additional constraint in variational data assimilation models (e.g. those based on the 3DVAR and 4DVAR analysis), to improve the NWP accuracy.

Despite over two decades of work, atmospheric contamination due to spatio-temporal changes in tropospheric temperature, pressure and humidity remains the largest source of error for the measurement of ground motions with InSAR. This is a particular problem when attempting to extract very small deformation signals from InSAR datasets.

Correction methods for tropospheric delay using weather model outputs, GNSS and/or spectrometer data have been applied in the past, but are often limited by the spatial and temporal resolution of the auxiliary data. Alternatively an empirical correction can be estimated from interferometric phase by assuming a linear or a power-law relationship between the phase and topography, but this approach can often erroneously remove ground deformation signals as well.

Here we show the results of a statistical comparison of state-of-the-art tropospheric corrections estimated using our Toolbox for Reducing Atmospheric InSAR Noise (TRAIN) – (Bekaert et al., 2015). This includes correction estimated from the MERIS and MODIS spectrometers, a low and high spatial-resolution weather model (ERA-I and WRF), and conventional linear and new power-law empirical methods. Whilst spectrometers give the largest reduction in tropospheric signal over our three test regions, they are severely limited to cloud-free and daylight acquisitions and none of the other tropospheric correction methods consistently perform best.

However, numerical weather models present an opportunity to correct InSAR data systematically using independent information and a standardized approach. One of the most widely used numerical weather model datasets is ERA-Interim (ERA-I), produced by the European Centre for Medium-range Weather Forecasts, and corrections based on these data have seen rapid uptake by the InSAR community. However attempts to validate the correction method on small test sites show conflicting results; that it works well in some regions but not in others.

We show results of a global validation of wet delays derived from ERA-I against MERIS spectrometer measurements of water vapour acquired during 2003-2010. Comparing the two datasets in 10×10 degree regions, we find significant geographical variation in the quality of ERA-I predicted wet delays with ERA-I retrieving wet delays better at mid-high latitudes than at low latitudes.

There are strong global correlations between the quality of the ERA-I wet delays and both temperature and humidity in the tropospheric boundary layer, enabling the prediction of the efficacy of this method for any given SAR acquisition. We suggest that for automated atmospheric correction of SAR data for future missions, ERA-I is currently the most viable option, but uncertainties should also be estimated for these corrections on the basis of temperature and humidity. We propose a new method to estimate and incorporate these uncertainties in time-series analysis of deformation.

References:

Bekaert, D.P.S., R.J. Walters, T.J. Wright, A.J. Hooper, and D.J. Parker(2015), Statistical comparison of tropospheric correction techniques for InSAR, Remote Sensing of Environment, doi:10.1016/j.rse.2015.08.035.

Modelling the atmospheric path delay (APD) on a per-product basis has been shown to reduce SAR product geolocation error by several metres (typically ~5 – 7 m two-way delay for C-band). The APD magnitude is influenced by meteorological conditions and the local incident angle (i.e. path length). The relevance of APD for geolocation accuracy also depends on the product resolution and coverage, and therefore the acquisition mode and product type.

In past studies, we have derived APD estimates from an atmospheric model that has the form of an altitude dependent polynomial requiring three meteorological measurements as inputs: the local temperature, pressure, and relative humidity. Using this model in conjunction with accurate local meteorological measurements, the path delay can be estimated to better than a few cm. However, the model may be fed with purely empirical or nominal values and still provide better ranging accuracy than can be obtained without any APD modelling at all. This could be an advantage in situations where real-time data processing is of great importance. An APD model fed by a combination of nominal/best-guess values and meteorological measurements (such as those that are most readily available on a global scale) may be considered the most useful in the context of near-real-time applications.

Using Sentinel-1A (S-1A) products from Stripmap (SM) and Terrain Observation with Progressive Scans (TOPS) modes as a testbed, detailed analyses were performed on the effect of APD correction on products acquired from different modes. An overview of the current S-1A geometric performance that can be expected with and without APD modelling is provided for different acquisition modes, with offsets expressed in metres and samples. The general applicability of the APD analyses to other spaceborne SAR sensors is also discussed. The expected improvements for different applications, including geocoding, interferometric SAR, change detection, and image fusion are also discussed. This framework is expected to provide users of SAR data with a means to assess the impact of APD correction (or lack thereof) according to acquisition mode, product type and application.

With a view towards potential improvements to SAR data processing, we propose a variety of scenarios whereby the APD could be estimated and incorporated into SAR data products. The auxiliary information needed for global APD modelling is described, and candidates for sources of the meteorological data (pressure, temperature, humidity) and ionospheric TEC content are suggested. Data availability (spatial and temporal) is considered, as well as the requirements for real-time and near-real-time (3h or 24h) processing.

Recommendations are made for the usage of the global models, comparing their quality and latency to results from recent studies over Swiss test sites based on S-1A data takes, whereby meteorological measurements were used in conjunction with the altitude dependent APD model.

Finally, we propose a set of standard APD parameterisations that could be added in a standardised way to SAR product annotations, and describe how they could be used to generate products with inherently more accurate range geolocation, either during post-processing or even during product generation. In the latter case, the annotations would describe the parameters that were used to compensate APD during e.g. geolocation stages.

Over the last decades, several case studies confirmed that it is possible to isolate the atmospheric signal delay from SAR imagery based on radar interferometric analysis. These atmospheric phase screens (APS) can be converted to maps describing the distribution of water vapor in the atmosphere. Although the case studies showed very promising results, the temporal or spatial coverage of the available satellite missions was low. Hence, application was restricted to case studies only. With the launch of the Sentinel-1 satellite a new SAR dataset became available, with unprecedented temporal and spatial coverage. . This new dataset offers a unique opportunity to operationalize the use of SAR based water vapor maps in weather prediction models. In this study we present our approach to provide Sentinel-1 based water vapor maps on an operational basis.

To derive the APS estimates a stack of Sentinel-1 images is processed and continuously updated. The general repeat time of the Sentinel-1a is 12 days at the equator, but by using a combination of overlapping descending and ascending orbits a repeat time of 3 to 4 can be achieved above, for instance, the Netherlands. This repeat time can be reduced to 1 to 2 days once the data from the Sentinel-1b becomes available. To find the individual APS for every acquisition we use different master-slave configurations to split the different contributions of the two acquisitions in every interferogram. Additionally, we model and remove other phase contributors based on a-priori available data. These APS estimates are then converted to water vapor maps based on temperature and pressure measurements in the Netherlands. Although these measurements have a lower resolution than the calculated APS, they can still be used because variations in temperature and pressure fields are generally smaller and more gradual than water vapor estimates.

To illustrate the abilities of this method, a time series of Sentinel-1 based water vapor estimates above the Netherlands will be shown. These water vapor maps give an estimate of the total column water vapor with a very high resolution, in contrast with currently used measurement techniques, which are either based on point measurements or have a low spatial resolution. Therefore, this data source gives us the ability to model and monitor weather processes in a much higher detail than was possible before. The development of these time series is a first step to the generation of an operational water vapor product and the implementation of this product into a weather model. Especially this last step will be done in close cooperation with the Dutch meteorological service, the KNMI.