A deep learning approach to detecting volcano deformation from satellite imagery using synthetic datasets

The detection of early signs of volcanic unrest preceding an eruption, in the form of ground deformation in Interferometric Synthetic Aperture Radar (InSAR) data is critical for assessing volcanic hazard. In this work we treat this as a binary classification problem of InSAR images, and propose a novel deep learning methodology that exploits a rich source of synthetically generated interferograms to train quality classifiers that perform equally well in real interferograms. The imbalanced nature of the problem, with orders of magnitude fewer positive samples, coupled with the lack of a curated database with labeled InSAR data, sets a challenging task for conventional deep learning architectures. We propose a new framework for domain adaptation, in which we learn class prototypes from synthetic data with vision transformers. We report detection accuracy that amounts to the highest reported accuracy on a large test set for volcanic unrest detection. Moreover, we built upon this knowledge by learning a new, non-linear, projection between the learnt representations and prototype space, using pseudo labels produced by our model from an unlabeled real InSAR dataset. This leads to the new state of the art with 97.1% accuracy on our test set. We demonstrate the robustness of our approach by training a simple ResNet-18 Convolutional Neural Network on the unlabeled real InSAR dataset with pseudo-labels generated from our top transformer-prototype model. Our methodology provides a significant improvement in performance without the need of manually labeling any sample, opening the road for further exploitation of synthetic InSAR data in various remote sensing applications.

Tropospheric phase delays pose a major challenge to InSAR (interferometric synthetic aperture radar)‐based studies of tectonic deformation. One approach to the mitigation of effects from tropospheric noise is the application of elevation‐dependent corrections based on empirical fits between elevation and interferometric phase. We quantify the effects of corrections with a range of complexity on inferred earthquake source parameters using synthetic interferograms with known atmospheric characteristics. We infer statistical properties of the stratified component of the atmosphere using pressure, temperature, and water vapor data from the North America Regional Reanalysis model over our region of interest in the Basin and Range province of the western United States. The statistics of the simulated atmospheric turbulence are estimated from InSAR and Global Positioning System data. We demonstrate potentially significant improvements in the precision of earthquake magnitude, depth, and dip estimates for several synthetic earthquake focal mechanisms following a correction for spatially variable atmospheric characteristics, relative to cases where the correction is based on a uniform delay versus elevation relationship or where no correction is applied. We apply our approach to the 1992 M5.6 Little Skull Mountain, Nevada, earthquake and demonstrate that the earthquake source parameter error bounds decrease in size after applying the atmospheric corrections. Our approach for evaluating the impact of atmospheric noise on inferred fault parameters is easily adaptable to other regions and source mechanisms.

Atmospheric effect represents one of the major error sources for interferometric synthetic aperture radar (InSAR), particularly for the repeat-pass InSAR data. In order to further improve the practicability of InSAR technology, it is essential to study how to estimate and eliminate the undesired impact of atmospheric effects. In this paper, we propose the multi-resolution weighted correlation analysis (MRWCA) method between the dual-polarization InSAR data to estimate and correct atmospheric effects for InSAR topographic mapping. The study is based on the a priori knowledge that atmospheric effects is independent of the polarization. To find the identical atmospheric phase (ATP) signals of interferograms in different polarizations, we need to remove the other same or similar phase components. Using two different topographic data, differential interferometry was firstly performed so that the obtained differential interferograms (D-Infs) have different topographic error phases. A polynomial fitting method is then used to remove the orbit error phases. Thus, the ATP signals are the only identical components in the final obtained D-Infs. By using a forward wavelet transform, we break down the obtained D-Infs into building blocks based on their frequency properties. We then applied weighted correlation analysis to estimate the wavelet coefficients attributed to the atmospheric effects. Thus, the ATP signals can be obtained by the refined wavelet coefficients during inverse wavelet transform (IWT). Lastly, we tested the proposed method by the L-band Advanced Land Observing Satellite (ALOS)-1 PALSAR dual-polarization SAR data pairs covering the San Francisco (USA) and Moron (Mongolia) regions. By using Ice, Cloud, and land Elevation Satellite (ICESat) data as the reference data, we evaluated the vertical accuracy of the InSAR digital elevation models (DEMs) with and without atmospheric effects correction, which shows that, for the San Francisco test site, the corrected interferogram could provide a DEM with a root-mean-square error (RMSE) of 7.79 m, which is an improvement of 40.5% with respect to the DEM without atmospheric effects correction. For the Moron test site, the corrected interferogram could provide a DEM with an RMSE of 10.74 m, which is an improvement of 30.2% with respect to the DEM without atmospheric effects correction.

The interferometric synthetic aperture radar (InSAR) is a powerful space geodetic tool to observe surface displacement with a wide spatial coverage and high spatial resolution. However, InSAR often suffer from the atmospheric noise due to the microwave propagation delay effect. Because the ionospheric delay effect has a frequency-dispersive nature, the range-split spectrum method that uses two band-splitted SAR data was recently shown as an effective correction method for the ionospheric delay noise. On the other hand, the neutral atmospheric delay has a non-dispersive nature and is mainly caused by the spatial heterogeneity of atmospheric water vapor that shows high variability in space and time, resulting in the difficulty of precise modelling. For more than two decades, although numbers of studies proposed and applied atmosoheric noise corrections, there has been no decisive methods for InSAR atmospheric noise correction until now. Mainstreams of the InSAR neutral atmospheric delay correction are the timeseries approach and the use of external data. I'm now developing a new InSAR delay correction model based on GNSS observation data, whose preliminary result showed a better correction effectiveness than numerical weather model approaches.

Assessment of ERA-Interim and ERA5 reanalysis data on atmospheric corrections for InSAR

Spaceborne Interferometric Synthetic Aperture Radar (InSAR) is a well established technique useful in many land applications, such as tectonic movements, landslide monitoring and digital elevation model extraction. One of its major limitation is the atmospheric effect, and in particular the high water vapour spatial and temporal variability which introduces an unknown delay in the signal propagation. This paper describes the general approach and some results achieved in the framework of an ESA funded project devoted to the mapping of the water vapour with the aim to mitigate its effect in InSAR applications. Ground based (microwave radiometers, radiosoundings, GPS) and spaceborne observations (AMSR-E, MERIS, MODIS) of columnar water vapour were compared with Numerical Weather Prediction model runs in Central Italy during a 15-day experiment. A dense network of GPS receivers was deployed close to Como, in Northern Italy, to complement the operational network in order to derive Zenith Total Delay as well as Slant Delay which can support InSAR processing. A comparison with Atmospheric Phase Screens (APS) derived from a sequence of Envisat multi pass interferometric acquisitions processed using the Permanent Scatters technique on the two test sites has been also performed. The acquired experimental data and their comparison give a valuable idea of what can be done to gather information on water vapour, which, besides InSAR applications, plays a fundamental role in weather prediction and radio propagation studies. The work has been carried out in the framework of an ESA funded project, named “Mitigation of Electromagnetic Transmission errors induced by Atmospheric Water Vapour Effects” (METAWAVE). This paper presents the general approach an the various methodologies exploited in the project, together with the overall intercomparison of the results. In deep details on the comparison with the InSAR APS maps derived by the PS technique, as well as on GPS receiver processing and water vapour tomography are reported in two companion papers.

Large-scale and high-intensity mining underground coal has resulted in serious land subsidence. It has caused a lot of ecological environment problems and has a serious impact on the sustainable development of economy. Land subsidence cannot be accurately monitored by InSAR (interferometric synthetic aperture radar) due to the low coherence in the mining area, excessive deformation gradient, and the atmospheric effect. In order to solve this problem, a novel phase unwrapping method based on U-Net convolutional neural network was constructed. Firstly, the U-Net convolutional neural network is used to extract edge to automatically obtain the boundary information of the interferometric fringes in the region of subsidence basin. Secondly, an edge-linking algorithm is constructed based on edge growth and predictive search. The interrupted interferometric fringes are connected automatically. The whole and continuous edges of interferometric fringes are obtained. Finally, the correct phase unwrapping results are obtained according to the principle of phase unwrapping and the wrap-count (integer jump of 2π) at each pixel by edge detection. The Huaibei Coalfield in China was taken as the study area. The real interferograms from D-InSAR (differential interferometric synthetic aperture radar) processing used Sentinel-1A data which were used to verify the performance of the new method. Subsidence basins with clear interferometric fringes, interrupted interferometric fringes, and confused interferometric fringes are selected for experiments. The results were compared with the other methods, such as MCF (minimum cost flow) method. The tests showed that the new method based on U-Net convolutional neural network can resolve the problem that is difficult to obtain the correct unwrapping phase due to interrupted or partially confused interferometric fringes caused by low coherence or other reasons in the coal mining area. Hence, the new method can help to accurately monitor the subsidence in mining areas under different conditions using InSAR technology.

The interferometric synthetic aperture radar (InSAR) technique is widely utilized to measure ground-surface displacement. One of the main limitations of the measurements is the atmospheric phase delay effects. For satellites with shorter wavelengths, the atmospheric delay mainly consists of the tropospheric delay influenced by temperature, pressure, and water vapor. Tropospheric delay can be calculated using numerical weather prediction (NWP) model at the same moment as synthetic aperture radar (SAR) acquisition. Scientific researchers mainly use ensemble forecasting to produce better forecasts and analyze the uncertainties caused by physic parameterizations. In this study, we simulated the relevant meteorological parameters using the ensemble scheme of the stochastic physic perturbation tendency (SPPT) based on the weather research forecasting (WRF) model, which is one of the most broadly used NWP models. We selected an area in Foshan, Guangdong Province, in the southeast of China, and calculated the corresponding atmospheric delay. InSAR images were computed through data from the Sentinel-1A satellite and mitigated by the ensemble mean of the WRF-SPPT results. The WRF-SPPT method improves the mitigating effect more than WRF simulation without ensemble forecasting. The atmospherically corrected InSAR phases were used in the stacking process to estimate the linear deformation rate in the experimental area. The root mean square errors (RMSE) of the deformation rate without correction, with WRF-only correction, and with WRF-SPPT correction were calculated, indicating that ensemble forecasting can significantly reduce the atmospheric delay in stacking. In addition, the ensemble forecasting based on a combination of initial uncertainties and stochastic physic perturbation tendencies showed better correction performance compared with the ensemble forecasting generated by a set of perturbed initial conditions without considering the model’s uncertainties.

Large‐scale ground deformation in Iceland is dominated by extensional plate‐boundary deformation, where the Mid‐Atlantic Ridge crosses the island, and by uplift due to glacial isostatic adjustment (GIA) from thinning and retreat of glaciers. While this deformation is mostly steady over multiple years, it is modulated by smaller‐scale transient deformation associated with, for example, earthquakes, volcanic unrest, and geothermal exploitation. Here, we combine countrywide Sentinel‐1 interferometric synthetic aperture radar (InSAR) data (from six tracks) from 2015 to 2021 with continuous Global Navigation Satellite System observations to produce time series of displacements across Iceland. The InSAR results were improved in a two‐step tropospheric mitigation procedure, using (a) global atmospheric models to reduce long‐wavelength and topography‐correlated tropospheric signals, and (b) modeling of the stochastic properties of the residual troposphere. Our results significantly improve upon earlier countrywide InSAR results, which were based on InSAR stacking, as we use more data, better data weighting, and advanced InSAR corrections to produce time series of ground displacements instead of just velocities. We fuse the three ascending and three descending track results to estimate maps of East and Up velocities, which clearly show the large‐scale extension and GIA deformation. Using revised plate‐spreading and GIA models, based on these new ground velocity maps, we remove the large‐scale and steady deformation from the InSAR time series and analyze the remaining transient deformations. Our results demonstrate the importance of (a) mitigating InSAR tropospheric signals over Iceland and of (b) solving for time series of deformation, not just velocities, as multiple transient deformation processes are present.

InSAR monitoring of creeping landslides in mountainous regions: A case study in Eldorado National Forest, California

For the past two decades, interferometric synthetic aperture radar (InSAR) has been used to monitor ground deformation with subcentimetric precision from space. But the applicability of this technique is limited in regions with a low density of naturally-occurring phase-coherent radar targets, e.g. vegetated nonurbanized areas. Third-party end-users of InSAR survey results cannot, in a systematic way, determine a priori whether these coherent targets have adequate spatial distribution to estimate the parameters of their interest. Additionally, InSAR deformation estimates are referred to a local datum, meaning that the technique is sensitive only to the relative deformation occurring within the SAR images. This makes it difficult to compare these estimates with those from other techniques, e.g. historical levelling data or changes in the sea level. Here we propose the design of a geodetic network for InSAR, aimed at densifying the naturally-occurring measurement network and converting from a local datum to a global one. A practical solution for improving spatial sampling is to deploy coherent target devices such as corner reflectors or transponders on ground, tailored to the specific monitoring application under consideration. The proposed method (1) provides a generic description of any deformation phenomenon; (2) determines whether the naturally-occurring InSAR measurements are adequate in terms of user-defined criteria; (3) finds the minimum number of additional devices to be deployed (if required); and (4) finds their optimal ground locations. It digests, as inputs, any prior knowledge of the deformation signal, the expected locations and quality of the existing coherent targets, and the quality of the deployed devices. The method is based on comparing different covariance matrices of the final parameters of interest with a criterion matrix (i.e., the ideal desired covariance matrix) using a predefined metric. The resulting measurement network is optimized with respect to precision, reliability and economic criteria; this is demonstrated via synthetic examples and a case of subsidence in the Netherlands. As a basis for the choice and number of deployed devices, we evaluate the measurement precision of compact active transponders and demonstrate their viability as alternatives to passive corner reflectors through three field experiments, using different satellite data and geodetic validation techniques. Transponders are shown to be usable for subcentimetre-precision geodetic applications, while improving upon the drawbacks of corner reflectors in terms of size, shape, weight and conspicuousness. For transforming the spatially-relative InSAR deformation estimates (local datum) to a standard terrestrial reference frame (global datum), we introduce a new concept involving the collocation of transponders with Global Navigation Satellite System (GNSS) measurements. The displacement of such a transponder is consequently determined in the standard reference frame used by GNSS, eliminating the need for any assumptions on reference-point stability in applications where the InSAR deformation estimates are compared with results from other techniques. The considerations, results and practical lessons learnt at several permanent GNSS stations in the Netherlands are described.

Correcting for tropospheric delays is one of the largest challenges facing the interferometric synthetic aperture radar (InSAR) community. Spatial and temporal variations in temperature, pressure, and relative humidity create tropospheric signals in InSAR data, masking smaller surface displacements due to tectonic or volcanic deformation. Correction methods using weather model data, 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 a correction can be estimated from the interferometric phase by assuming a linear or a power-law relationship between the phase and topography. Typically the challenge lies in separating deformation from tropospheric phase signals. In this study we performed a statistical comparison of the state-of-the-art tropospheric corrections estimated from the MERIS and MODIS spectrometers, a low and high spatial-resolution weather model (ERA-I and WRF), and both the conventional linear and new power-law empirical methods. Our test-regions include Southern Mexico, Italy, and El Hierro. We find spectrometers give the largest reduction in tropospheric signal, but are limited to cloud-free and daylight acquisitions. We find a ~ 10–20% RMSE increase with increasing cloud cover consistent across methods. None of the other tropospheric correction methods consistently reduced tropospheric signals over different regions and times. We have released a new software package called TRAIN (Toolbox for Reducing Atmospheric InSAR Noise), which includes all these state-of-the-art correction methods. We recommend future developments should aim towards combining the different correction methods in an optimal manner.

Repeat-pass spaceborne interferometric synthetic aperture radar (InSAR) is commonly used to measure surface deformation; phase delays due to atmospheric water vapour may have significant impact on the accuracy of these measurements. In recent years, there has been a growing interest in using forecasts and analyses from numerical weather prediction (NWP) models – which can provide good estimates of the atmospheric state – to correct for atmospheric phase delays. In this study, three separate estimates of atmospheric water vapour content from NWP output are used in combination with Environmental Satellite (Envisat) Advanced Synthetic Aperture Radar (ASAR) data over the Pearl River Delta region in South China to mitigate atmospheric distortion. The NWP-based estimates are derived from: (1) interpolation of National Centers for Environmental Prediction (NCEP) Final Operational Global Analysis (FNL) data; (2) Weather Research and Forecasting (WRF) model simulations initialized with FNL analysis without additional data assimilation; and (3) WRF simulations initialized with a three-dimensional variational (3DVar) data assimilation system that ingests additional meteorological observations. The accuracy of the atmospheric corrections from these different NWP model outputs is further verified quantitatively with precipitable water vapour (PWV) data from several ground-based global positioning system (GPS) stations in Hong Kong. Inter-comparison shows a good agreement between the PWV derived from the WRF-3DVar simulations and the GPS measurements, suggesting that atmospheric correction by convection-permitting WRF simulations initialized with mesoscale data assimilation may effectively mitigate atmospheric distortion in InSAR measurements, especially for coastal areas.

With the rapid development of interferometric synthetic aperture radar (InSAR) measurement technology, its measurement accuracy requirements are increasing. Atmospheric delay errors must be corrected, especially in the case of crustal deformation monitoring, the 20% variation of tropospheric water vapor among InSAR pairs generally produces range from 10 cm to 14 cm deformation errors. Such errors can be of the same magnitude as the annual changes in crustal deformation, or even greater, masking crustal deformation information and seriously affecting the results of crustal deformation monitoring. Therefore, in order to obtain a more accurate InSAR atmospheric delay correction model, this paper calculated and integrated atmospheric delays that were estimated by different sources, including the 37 pressure levels of the fifth generation of the European Centre for Medium-Range Weather Forecasts (ECMWF)) numerical weather prediction model, ECMWF Reanalysis v5 (ERA5), and Global Navigation Satellite System (GNSS) measurement data from the crustal movement observation network of China, based on the variance component estimation (VCE) weighting method. The results showed that the integrated model, based on the VCE method, is better than the generic atmospheric correction online service (GACOS) model for InSAR measuring of crustal deformation. The precision in monitoring crustal deformations was improved by approximately 5 mm, the correlation coefficient of atmospheric delay errors and crustal deformations improved from 0.287 to 0.347, and accuracy improved by approximately 25%. However, the improvement in accuracy was limited because of system error decoherence that was induced by atmospheric noise caused by abundant vegetation or snow cover. Therefore, in order to achieve more accurate results, we recommend the adoption of the multi-source integrated atmospheric delay correction model, based on the VCE method, for InSAR high-precision measuring of crustal deformation and seismic activities.

Interferometric Synthetic Aperture Radar (InSAR) is a type of active remote sensing whereby a satellite transmits electromagnetic radiation (microwaves) at the ground and measures the differential phase of the reflected signal over multiple images (or multiple antennas on a single satellite). InSAR has the potential to provide centimeter and even millimeter-scale measurements of displacement over time, but is sensitive to vegetation, topography, and atmospheric effects. We consider herein, the application of InSAR at two known landslides on the Enbridge pipeline system, and discuss the strengths, weaknesses, values, and limitations of its application in the Geohazard Management of landslides impacting pipeline ROW’s. We compare information provided at each site by InSAR (both L-band and X-band) to data derived by mapping using Light Detection and Ranging (LiDAR) or air photographs, to differential LiDAR techniques, and to data derived from subsurface measurements (slope inclinometers). In doing so we find that L-Band data can be an effective tool to establish the extent or footprint of movement (or lack of movement) at known landslide locations, extending the interpretive power of a specialist and the understanding of event magnitude, and potentially affecting the mitigation options. Further, L-Band InSAR can be used in a supporting role to pre-screen areas for active landslides along the right of way (ROW), however, data gaps, a lack of explanatory power, and considerable noise in the results mean that a user step that further considers the terrain, other sources of data, and the identified magnitude, is essential. X-Band InSAR appeared impractical for ROW monitoring where vegetation prevented coherence between images, however, X-Band InSAR was able to detect small displacements at above ground infrastructures.