Search

Monitoring volcanic deformations allows us to track dynamic states of a volcano and to know where an eruptions could happen. Spaceborne Synthetic Aperture Radar (SAR) and SAR interferometry (InSAR) techniques created an opportunity to track volcanoes globally, even in inaccessible regions without ground measuring stations.This paper proposes a convolutional neural network (CNN) for detection of volcanic deformations in InSAR velocity maps. We had only a small amount of velocity maps over the region of central South American Andes, therefore the synthetic data are used to train the model from scratch. In the region of interest, the velocity maps contain the patterns of salt lakes and slope induced signal which confuse CNN models trained on synthetic data.In order to bridge the gap between the synthetic and real data, the hybrid synthetic-real data set is used for fine-tuning the model. The hybrid set consists of the real background signal data and synthetic volcanic data. Four fine-tuning sets which were created by different combinations of the original hybrid data, the filtered hybrid data, and simulated data have been used and compared with each other. Besides, we compared four fine-tuning approaches to determine where and how to fine-tune the model. Results show significant improvement in performance by majority of the approaches, and training the last or last two layers have given the best results. In addition, using the FT1 (containing only hybrid set), and FT4 (containing all sets) improved the area under the curve receiver operating characteristic (AUC ROC) from 55% to 86% and 88% respectively.

TecVolSA (Tectonics and Volcanoes in South America) is a project with a goal of developing intelligent Earth Observation (EO) data processing and exploitation for monitoring various geophysical processes in central south American Andes. Large amount of Sentinel-1 data over the period of about 5 years has been processed using mixed Permanent Scatterer and Distributed Scatterer (PS/DS) approaches. The received products are velocity maps with InSAR relative error in the order of 1 mm/yr on a large scale (>100km). The second milestone of the project was automatic extraction of information from the data. In this work, the focus is on detecting volcanic deformations. Since the real data prepared in such manner is limited, to train a deep learning model for detection of volcanic deformations, a synthetic training set is used. Models are trained from scratch and InceptionResNet v2 was selected for further experiments as it was found to give best performance among the tested models. The explainable AI (XAI) techniques were used to understand and analyze the confidence of the model and to understand how to improve it. The models trained on synthetic training set underperformed on real test set. Using GradCAM technique, it was identified that slope induced signal and salt lake deformations were mistakenly identified as volcanic deformations. These patterns are difficult to simulate and were not contained in synthetic training set. Bridging this distribution gap was performed using hybrid synthetic-real fine-tuning set, consisting of the real slope induced signal data and synthetic volcanic data. Additionally, false positive rate of the model is reduced using low-pass spatial filtering of the real test set, and finally by adjustments of the temporal baseline received from a sensitivity analysis. The model successfully detected all 10 deforming volcanoes in the region, ranging from 0.4 - 1.8 cm/yr in deformation.

Interferometric Synthetic Aperture Radar (InSAR) is widely used to measure deformation of the Earth's surface over large areas and long time periods. A common strategy to overcome coherence loss in long-term interferograms is to use multiple multilooked shorter interferograms, which can cover the same time period but maintain coherence. However, it has recently been shown that using this strategy can introduce a bias (also referred to as a “fading signal”) in the interferometric phase. We isolate the signature of the phase bias by constructing “daisy chain” sums of short-term interferograms of different length covering identical 1-year time intervals. This shows that the shorter interferograms are more affected by this phenomenon and the degree of the effect depends on ground cover types; cropland and forested pixels have significantly larger bias than urban pixels and the bias for cropland mimics subsidence throughout the year, whereas forests mimics subsidence in the spring and heave in the autumn. We, propose a method for correcting the phase bias, based on the assumption, borne out by our observations, that the bias in an interferogram is linearly related to the sum of the bias in shorter interferograms spanning the same time. We tested the algorithm over a study area in western Turkey by comparing average velocities against results from a phase linking approach, which estimates the single primary phases from all the interferometric pairs, and has been shown to be almost insensitive to the phase bias. Our corrected velocities agree well with those from a phase linking approach. Our approach can be applied to global compilations of short-term interferograms and provides accurate long-term velocity estimation without a requirement for coherence in long-term interferograms.

We scrutinize the reliability of multilooked interferograms for deformation analysis. Designing a simple approach in the evaluation of the accuracy of the estimated deformation signals, we reveal a prominent bias in the deformation velocity maps. The bias is the result of propagation of small phase error of multilooked interferograms through the time series and can sum up to 6.5 mm/yr in case of using the error prone short temporal baseline interferograms. We further discuss the role of the phase estimation algorithms in reduction of the bias and put recommend a unified intermediate InSAR product for achieving high-precision deformation monitoring.

International audience; A recent GPS study suggests interseismic coupling variations along the Main Himalayan Thrust in Bhutan, with implications on the regional seismic potential (Maréchal et al., 2016). We use the complete Sentinel-1 radar data acquired since 2014 along descending and ascending orbits to provide average velocity maps across the front range in Bhutan, based on radar interferometry (InSAR). We compare results from time series analysis from a Small Baseline Subset approach (NSBAS, Doin et al., 2011, Grandin, 2015) and Distributed Scatterers Interferometry techniques to better assess and mitigate the various sources of noise in our data set (Ansari et al., 2018, De Zan et al., 2015). We correct tropospheric delays using global atmospheric models (ERA5, ECMWF) and ionospheric delays with split-spectrum technique (Gomba et al., 2016), to improve the quality of the deformation signal especially at large scale. We finally confront GPS and InSAR velocity profiles across range through simple 2D elastic modelling to better constrain the displacement rate across the range and discuss the existence of shallow creep in eastern Bhutan, as suggested by GPS data.

SAR interferometry is increasingly accurate in retrieving the surface deformation history over large areas of the Earth. Advancements have occurred in atmosphere delay compensation, both ionospheric and tropospheric: the remaining contributions of atmospheric propagation to the error budget are a few cm for deformation or a few mm/yr for deformation rates, for a typical Sentinel-1 case. We think that today a crucial aspect that needs consideration is the choice of interferograms to be processed to estimate the deformation. Many groups are limiting the processing to high-coherent pairs with short temporal separation: however, short temporal interferograms tend to be biased. This can be understood observing the presence of closure phases. Examining three subsequent images and building the closure phase, one typically founds inconsistencies in the order of a few degrees: scaling the error to the time span of a year the final deformation rate bias can easily reach several mm/yr. A possible physical interpretation of non-zero closure phase is related to moisture variations in semi-transparent media. The short term interferograms are biased because of the asymmetry of the moisture cycles combined with a non-linearity in the phase variations w.r.t. moisture. In our experiments with Sentinel-1 with a limitation of the temporal baseline at 72 days, we have observed biases of ~7 mm/yr on distributed targets. Here permanent scatterers served as the gold standard since their closure phase is always zero. We consider two main approaches to the closure phase problem in InSAR time series: using full-covariance estimators, based on the temporal covariance matrix, and modelling the underlying moisture process. Our preliminary experiences with full covariance matrices indicate a small bias (< 1 mm/yr) with respect to PS processing. Given the low-resolution of available moisture products, we are working on the inversion of moisture levels directly from closure phases. First L-band results are encouraging, compared to ground truth. If models and inversions are correct, the compensation of the modelled phase should reduce the path-dependency in the temporal integration and consequently the biases.

Wide-swath satellite missions with short revisit times, such as Sentinel-1 and NISAR, provide an unprecedented wealth of interferometric time series and open new opportunities for systematic monitoring of the Earth surface. The process-ing of the emerging Big Data with the state-of-the-art InSAR time series analysis techniques is, however, computa-tionally challenging. A new demand has emerged for the analysis of the fast growing data volumes specifically for systematic near real-time (NRT) monitoring of the Earth surface. We have addressed this demand by the proposal of two efficient alternative estimators for NRT processing of the emerging Big Data in [1, 2]. In this contribution, a hybrid approach based on the proposed estimators is introduced and applied in efficient wide area processing of two-year archive of Sentinel-1 data over eastern part of the Trans-Mexican Volcanic Belt.

Signal decorrelation poses a limitation to multipass SAR interferometry. In pursuit of overcoming this limitation to achieve high-precision deformation estimates, different techniques have been developed; with SBAS, SqueeSAR and CAESAR as the overarching schemes. These different analysis approaches raise the question of their efficiency and limitation in phase and consequently deformation estimation. This contribution firstly addresses this question and secondly proposes a new estimator with improved performance. Called Eigendecomposition based Maximum-likelihood-estimator of Interferometric phase (EMI), the proposed estimator combines the advantages of the state-of-the-art techniques. Identical to CAESAR, EMI is solved using Eigendecomposition; it is therefore computationally efficient and straightforward in implementation. Similar to SqueeSAR, EMI is a maximum-likelihood-estimator; hence it retains estimation efficiency. The computational and estimation efficiency of EMI renders it as an optimum choice for phase estimation. A further marriage of EMI with the proposed Sequential Estimator of [1] provides an efficient processing scheme tailored to the analysis of Big InSAR Data. EMI is formulated and verified in relation to the state-of-the-art approaches via mathematical formulation, simulation analysis and experiments with time series of Sentinel-1 data over the volcanic island of Vulcano, Italy.

Interferometric synthetic aperture radar (InSAR) measurements are merely sensitive to the deformation along the Line of Sight (LOS) direction of the sensor. To improve the geometrical sensitivity and retrieve the three-dimensional deformation, the integration of InSAR from non-coplanar acquisitions as well as fusion with resolution-scale SAR image shift measurements has become a standard approach. Using different statistical measures, we assess and compare the influence of different image acquisition strategies as well as data fusion on the performance of InSAR in 3D deformation retrieval. Integrating nominal InSAR acquisitions, i.e. a set of measurements from ascending and descending tracks acquired from right-looking geometry, a strong correlation between the retrieved 3D parameters in the local vertical-north plane is observable. This correlation is sought to be decreased by non-nominal acquisitions; i.e. left-looking or squinted observations. These acquisition strategies are discussed for consideration in the future L-band mission Tandem-L.

The capacity of SAR interferometry to measure surface deformation with accuracy of 1 mm/year or better are well known. However this is typically limited to relative motion at short distance. Thanks to several advances in SAR sensor quality, data availability, orbit determination, processing, and calibration of atmospheric delays it is now possible to achieve that accuracy even across large distances of hundreds of kilometers. In this paper we revise the main contributions to the large scale error, considering available mitigation techniques. We provide a first validation for a processing based on Sentinel-1 data, by comparing our results with GNSS stations. For future SAR's operating at lower frequencies, it is vital to consider ionospheric corrections and likely also the influence of moisture ariations in natural scatterers. The choice of processing algorithms, though typically not discussed, can also have a significant effect on the quality of the result.