Conference Agenda

The goal of this work was to examine the potential of the General Empirical Orthogonal Function (EOF), also known as Principal Component Analysis (PCA), in detecting anomalous earthquake precursory variations in Earth’s ionosphere - litosphere geomagnetic system. For this purpose, two major earthquake episodes in China have been selected. M6.0 quake, with epicentre at 39.835°N, 77.108°E, 86 km ENE of Arzak, which occurred on 19.01.2020, and the M6.3 quake, with epicentre at 33.144°N, 86.864°E in western Xizang, on 22.07.2020. The spatial-temporal variability patterns in an ~800 km radius of earthquake epicentres were calculated using geomagnetic data, collected at 1 Hz by Vector Field Magnetometer onboard SWARM A, B and C, ESA’s low earth orbiting satellites. The total magnetic field was calculated, transformed into the NEC frame, and corrected for the impact of Earth’s core magnetism via CHAOS model. Analysis was able to identify eight significant EOF and PC components on a 3 - month and 1- year time scale. The results were observed in earthquake’s seismo-tectonic context, and with planetary geomagnetic storm activity indices K and A, to reduce the possible influence of other anomaly sources. Anomalous patterns in EOF and PC components were revealed along and on borders of the local tectonic fault lines and around earthquake epicentres. At the time when anomalous patterns occurred, examined geomagnetic A and K storm indices were indicative of quiet periods, strongly suggesting the pre-cursory connection between detected geomagnetic anomalies and subsequent earthquake events.

The detection of seismic activity precursor signs as part of an alarm system will provide opportunities for minimization of the social and economic impact seismic events such as earthquakes cause. It has long been theorized that the Earth's electromagnetic field could contain precursor signs before a seismic event with a growing body of empirical evidence. The ability to measure and monitor electromagnetic field activity has increased with each passing year as more sensors and methodologies emerge; missions such as Swarm have enabled researchers to access near-continuous observations of electromagnetic activity at second intervals allowing for more detailed and exciting studies.

This poster presents an approach designed to detect precursor anomalies in electromagnetic field data from Swarm satellites and initial analysis results. This works towards developing a continuous and effective monitoring system of seismic activities based on the already deployed tools of Swarm available. We develop an enhanced form linear probabilistic model based on the martingale probability theories that test the null hypothesis to indicate abnormal changes in electromagnetic field activity. We test this enhanced approach in two experiments; firstly, we perform a quantitative comparison on well-understood and popular datasets alongside the classic approach. We find that the enhanced version does not negatively affect the performance of the underlying theory and instead produces more accurate anomaly detections overall. Secondly, we use three case studies of seismic activity (namely earthquakes in Mexico, Greece, and Croatia) to assess our approach. For each case study, we use two grids (500x500km and 1000x1000km, respectively) centered on the epicenters of the earthquakes and found that our method could detect anomalous phenomena in the electromagnetic data months ahead of the seismic activity when focused on these specific regions, thereby leading up to the study's seismic events.

Sea Crossing Bridge can cross bays, straits, and deep marine water, which are critical nodes to ensure the smooth flow of traffic arteries on the mainland. Even subtle displacements may affect the performance of bridges and cause high maintenance and repair costs. In order to timely and accurately inspect their structural safety status, the Synthetic Aperture Radar Interferometry (InSAR) technique which can quickly extract surface deformation over a large area at fine resolution, with low labor and material costs, and no effects on bridge normal operation, has been established as an effective method for bridge structural health monitoring.

Considering the complex structure and environmental effects of the Sea Crossing Bridge, there are still some issues to be solved to achieve the InSAR based structural health monitoring. First of all, the density and accuracy of point targets (PTs) are low on structures with significant de-coherence effects, leaving how to extract dense and accurate PTs upon partially coherent Sea Crossing Bridge still a problem. Moreover, previous thermal dilation monitoring methods relied on the linear deformation assumption of PTs without considering the actual structural mechanics of bridges, which are not appropriate for long-span Sea Crossing Bridge with long thermal dilation propagation distance. Finally, SAR side-looking imaging geometric distortion and simple analysis of bridge linear deformation maps are difficult for non-expert users to understand, making the InSAR results interpretation difficult.

Aiming at the above issues, the goal of this study is to develop a measurement and analysis method to detect the global deformation of the Sea Crossing Bridge quickly and accurately by introducing the structural principle into the conventional time-series InSAR analysis. As for the PTs selection, the structural coherent and non-coherent information of the bridge are considered to improve the density and accuracy of detectable PTs. In terms of thermal dilation modelling, a structural-driven regression analysis method is applied to establish a quantitative relationship between the temperature and the displacement. In order to provide user friendly results, the 3D visualization of deformation results is realized based on the bridge structural information and local observation geometric parameters, and the 3D visualized analysis results and structural theory are integrated for bridge structural health monitoring.

Taking the East Sea Bridge as an example, the detail deformation along the bridge is estimated by processing Sentinel-1 images from 2015 to 2017 based on the above method. After modelling and separating the thermal dilation along the bridge, the structural risk level is evaluated based on the comprehensive analysis of structural risk indicators and structural deformation. Risk sections along the bridge are identified, which would be helpful in guiding the management and maintenance, as well as disaster prevention of the Donghai Bridge.

The first electromagnetic observation network, consisting of 30 stations and covering the Beijing Capital Area(BCA) and the regions of North South Seismic Belt(NSSB) in China, has been built recently under the support of the National Development and Reform Commission of China. The network is designed to record the natural electromagnetic signal from 1000Hz to 1000s and powered transmitted electromagnetic signal from 300Hz to 0.1Hz, including five components Ex (NS electric field), Ey (EW electric field), Hx (NS magnetic field), Hy (EW magnetic field) , and Hz( vertical magnetic field) at each station of the network. The electromagnetic time series, spectrum, apparent resistivity and impedance phase etc. are the parameters used to study possible anomalies before earthquakes.

In this report, we present the case studies, discovering the precursor of apparent resistivity prior to Yangbi EQ (M=5.1) which occurred on March 27, 2017 near Dali city in Yunnan province. The epicenter depth is about 12km, which was located in the west of Nanjianweishan fault which is part of big Honghe fault belt. The earthquake mechanism solution is dextral strike-slip. Dali station (YX) is the nearest station to the epicenter with 32km distance and locates to the east of the Nanjianweishan fault. The study found that both polarizations of apparent resistivity ρ_xy and ρ_yx for natural signal started to gradually increase in pulse type from January 16, 2017 (about 10 weeks before EQ). They reached the maximum on March 15, 2017 (12 days before the EQ) and then rapidly decreased to the background reference value on March 23-24, 2017 (3-4 days before EQ). At the same time the variation amplitude of anomalous ρ_xy is bigger than that of ρ_yx. For the ρ_xy the values of single pulses of anomalous resistivity are generally 30-80% bigger than the background value. The maximum value is about 130% larger than the background value. Correspondingly, both impedance phase φ_xy and φ_yx firstly gradually decreased and recovered to the background value again before the earthquake. The variation amplitude of anomalous impedance phase φ_xy is larger than that for φ_yx.

Another kind of apparent resistivity anomaly has been observed before the Jinggu earthquake (Ms=5.9) which occurred in Jinggu county in Yunnan province on December 6, 2014. The apparent resistivity ρxy decreased gradually from 80Ω·m before the earthquake. The resistivity suddenly decreased to 60 Ω·m on the day when the earthquake happened. After the earthquake the resistivity remains active for five months.

In addition, spectrum variations of electro-magnetic field with one year period have been observed for almost all stations using the continuous data recorded for several years, which is different from traditional feature of the spectrum variation. According to the powered transmitted data the anomalous increase of electromagnetic field are also found before some earthquake events.

The research has focused on defining new processing methods for high-resolution 3D imaging of the interior of forests and glaciers and the characterization of their temporal behavior, in view of future application in the context of future low frequency Missions such as BIOMASS and Rose-L. The following questions were addressed: i) What will be the impact of changing weather conditions on AGB retrieval based on BIOMASS Tomographic data and Interferometric data? What is the most accurate signal processing approach for forest height retrieval based on tomographic data? What are the driving requirements for very high-resolution tomographic imaging of natural scenarios?

As spaceborne Tomographic data are not available as of today, the research made use of data acquired in the context of dedicated ground-based and airborne campaigns by ESA, including the P-Band airborne campaigns BioSAR 1, AfriSAR, the L-Band airborne campaign AlpTomoSAR, and the P-Band ground-based campaign TropiScat.

Temporal effects were studied on an experimental basis using data from the ground-based campaign TropiSCAT and from the two airborne campaigns BioSAR 1 and AfriSAR. TropiSCAT data were used to directly mimic BIOMASS acquisitions by analyzing tomograms formed by taking the signals acquired by different antennas on different days. Since BIOMASS tomography will be implemented by taking acquisitions each separated by three days, we considered tomograms obtained by taking two TropiS-CAT antennas every 3 days (and three antennas on the last day), which means that any single tomogram can be obtained by mixing seven different days. We refer to these tomograms as “7-day tomograms.” Each 7-day tomogram is associated with an “instantaneous tomogram” obtained by taking all 15 antennas at 6:00 a.m. on the seventh day, which is used as reference. Tomographic processing is then carried out by focusing the acquisitions from seven days in ground range-height coordinates by time-domain back projection. As a result, it was possible to form 66 7-day tomograms, of which 61 included acquisitions with both sunny and rainy days, so as to provide a clear experimental assessment of the radiometric error due to changing weather conditions. The error was then translated to an AGB retrieval error based on the observed sensitivity of tomographic data to forest biomass at the TropiSCAT site, which was known from air-borne measurements.

The evaluation of the impact of temporal decorrelation on AGB retrieval based on in-terferometric data was carried out with reference to the ground-cancellation technique. The technique consists in a coherent combination of two or three SAR images that allows to cancel out the ground echo and emphasize canopy backscattering, and it is currently assumed as the baseline retrieval approach for BIOMASS. Our analysis consisted in implementing the ground-cancellation technique using airborne campaign data acquired on the same day and on different days (time lags of 4, 5, and 9 days). In so-doing, we were able to provide a direct measurement of the radiometric error due to temporal decorrelation.

Optimal processing for forest height retrieval was addressed by implementing and comparing two different approaches: 1) parametric height estimation by minimizing the least-square problem between random volume over ground (RVoG) model predictions and multibaseline SAR data and 2) thresholding the vertical backscattering obtained by tomographic focusing. Both approaches were applied to the case of P-Band data collected during the AfriSAR campaign, and validated against Lidar measurements.

Optimal processing of tomographic data was addressed by studying imaging arte-facts that arise when conventional one-dimensional (1D) focusing methods, such as beamforming or Capon filtering, are applied to process data collected using large-ly-spaced and/or largely-irregular trajectories. A new processing method was proposed to produce high-quality imaging while largely reducing the computational burden, and without having to process the original raw data. The analysis was validated by results from numerical simulations as well as from real airborne data from the ESA campaign AlpTomoSAR.

Results from TropiSCAT clearly indicate that canopy scattering is more stable than ground scattering level, consistently with the physical explanation that ground scattering is more affected by changing moisture conditions. Based on previous studies at the same test-site, the observed variation of 1–1.5 dB in the 7-day tomograms would entail a bio-mass retrieval error of around 50–80 t/ha, which is on the order about 20% or better. The analysis of ground-cancelled images was particularly relevant as directly linked to AGB retrieval algorithms to be implemented for the BIOMASS Mission. In accordance with theoretical models, experiments indicate that temporal decorrelation mostly affects scarcely vegetated areas, whereas in high biomass regions levels the radiometric error is limited to within 1dB. Overall, these results support the feasibility of extracting accurate biophysical information from repeat-pass BIOMASS data, concerning both the tomographic and the interferometric phase of the Mission.

The research on temporal effects was not limited to a quantification of the resulting errors, but also on the development of signal processing methods to compensate for such effects.

As for forest height retrieval, results indicate that the best approach could result from a joint use of non-parametric and parametric methods, to merge the capability of mod-el-based retrieval to work without requiring parameter tuning and the computational efficiency and accuracy of SAR tomography.

Finally, the work on AlpTomoSAR data fully demonstrates that the quality of a tomographic airborne surveys is not limited in any sense by platform stability during the acquisition, under the condition that tomographic processing is correctly implemented, therefore supporting the possibility of tomographic imaging using small satellite formation.

On 8th August 2017, a magnitude Mw 6.5 earthquake occurred in the County of Jiuzhaigou in Sichuan Province, China (USGS, https://earthquake.usgs.gov/). Because of its high magnitude and shallow epicenter, the earthquake caused grave casualties and property losses. Furthermore, the earthquake triggered numerous secondary mountain disasters such as rockfall and landslide. Shortly after the Jiuzhaigou earthquake, some studies had assessed the number, distribution and other characteristics of the coseismic landslides based on pre- and post-event remote sensing data. However, slow-moving post-earthquake landslides don’t show visible change at short notice and are often neglected by the common change detection method. These slow-moving landslides will pose a long-term potential threat to people’s life and property. Therefore, a detailed monitoring of the slow-moving landslides is crucial to post-earthquake recovery and reconstruction.

Synthetic aperture radar interferometry (InSAR) technology, which has large ground coverage and high spatial resolution, has great advantages for geological disaster observation under all weather conditions. It has been utilized in ground surface deformation measuring of landslide for disaster detection and assessment. In this research, we adopt an optimized strategy that combines D-InSAR technique with SBAS InSAR to accurately detect and monitor the slow-moving post-earthquake landslides in Jiuzhaigou area. Firstly, we carry out a quick detection across wide area using differential InSAR (DInSAR) technique with 6 ALOS‐2 PALSAR‐2 ascending images. To retrieve the temporal evolution of these landslides, a detailed monitoring of specific landslides is carried out using the short baseline subset InSAR (SBAS-InSAR) during the period from 2007 to 2019. Combining with other multi-source data (including field investigation, LiDAR and optical image), we perform an in-depth analysis of the impact of the 2017 Jiuzhaigou earthquake on slow-moving post-earthquake landslides.

The results show that there are 16 slow-moving post-earthquake landslides which can be detected by InSAR analyses in Jiuzhaigou area, including 8 landslides close to residential areas. These landslides are mainly distributed in the NE plate (active plate), and the slide directions mainly are east (including 9 landslides) and southeast (include 5 landslides). Secondly, we found that the earthquake’s impact on the slow-moving landslides can mainly be classified into three categories: (a) accelerating the active historical landslides that were already sliding before the earthquake; (b) reactivating the stable historical landslides which were undisturbed before the earthquake; (c) directly triggering the landslides that had stable geological environment before the earthquake. For each case, we demonstrate some case studies of landslide affected by the earthquake, and perform a detailed analysis combining with multi-source data.

Yongding River Basin Ecological Repair is an important issue. the coordinated development of beijing-tianjin-hebei, South-to-North Water Diversion, and local comprehensive background such as groundwater recharge, along the river yongding river ecological hydrating uncertainty of the problems of groundwater system, ground stability, and urban security risk is an important scientific problems. This topic is comprehensive satellite remote sensing data, traditional hydrogeological formation and groundwater measured data, integrated GIS spatiotemporal analysis, time series decomposition, wavelet analysis and neural network model, etc., analyzing Yongding River (Beijing Section) affects groundwater flow field evolution and surface The deformation response process, and the effect of Ecological compensation on urban safety in the Ecological compensation of the Yongding River is evaluated as an example of building stability, underground rail traffic safety and Beijing Daxing International Airport. The research results of the topic can provide technical support for evaluating the effects of Yongding River ecological water compensation on the stability of coastal formation, providing a scientific basis for urban security risk assessments, providing scientific reference for avoiding the negative impact of water compensation of Yongding River.

Today, more than half of the world population is living in cities. This number is growing rapidly and by the middle of the 21st century, two thirds of the global population is projected to reside in urban areas. Earth Observation (EO) has become a valuable tool to provide up-to-date and reliable maps of urban areas needed to support sustainable city planning. Past studies demonstrated that Synthetic Aperture Radar (SAR) is effective in detecting built-up areas. Recently, deep learning algorithms, specifically Convolutional Neural Networks (CNNs), have gained popularity for built-up area mapping from Sentinel-2 MultiSpectral Instrument (MSI) data. The lack of cloud-free data may, however, hamper the updatability of optical-based built-up area mapping. In contrast to Sentinel-2, the Sentinel-1 SAR is able to acquire data in cloudy conditions as well as during day and night.

In this research, we explored the potential of Sentinel-1 SAR data for built-up area mapping at large-scale using deep learning. Specifically, a light-weight U-Net model was trained in the United States by leveraging open data (i.e., Microsoft building footprints) as labels. We assessed model performance in Beijing, China using 10,000 validation points and compared our results to a state-of-the-art product generated using Sentinel-2 MSI data and deep learning . Our results demonstrate that built-up areas can be accurately mapped by SAR and deep learning (F1 score 0.863). Moreover, the proposed approach using transfer learning produced comparable results in Beijing to the state-of-the-art product using local training. However, further research investigating the model transferability to a diverse set of cities is required.

In the framework of developments of 3D forest [1-3] SAR Tomography (Tomo-SAR), the issue of extensive and detailed characterizations of temporal decorrelation phenomena has emerged, especially for the future spaceborne operations. However, investigation has not been covered by ESA campaigns [4] of height-varying characteristics of decorrelation processes acting at the short-term (fractions of second) scale, for the common case of volumetric scatterers complexly moving in windblown forests.

In this work, a methodology is reported for new analyses of 4D Differential Tomography [5] type of forest temporal decorrelation phenomena, exploiting a ground-based miniradar array with very quick acquisition capabilities. Moreover, corresponding recent real data results are presented, developing characterizations of both height- and time-varying behaviours of the short-term decorrelation in a representative experiment for a stand of trees in Italy (PisaScat experiment), in two different wind conditions and seasons.

Consequently, indications are obtained that short-term decorrelation processes can affect also advanced spaceborne Tomo-SAR systems based on tandem (i.e. formation-flying) configurations [6], depending on the satellite time-lag, and their related products.

This unprecedented characterization concept and the new findings can be useful for current Tomo-SAR and also Pol-InSAR analyses, and most importantly, for development of the advanced mission concepts and programs of SAOCOM-CS-like kind [6], including LuTan-1 [7].

[1] M. Pardini, K. Papathanassiou, “On the Estimation of Ground and Volume Polarimetric Covariances in Forest Scenarios with SAR Tomography,” IEEE GRSL, 14 (10), pp.1860-1864, 2017.

[2] Y. Huang, L. Ferro-Famil, A. Reigber, “Under-foliage Object Imaging using SAR Tomography and Polarimetric Spectral Estimators,” IEEE TGARS, 50 (6), pp.2213-2225, 2012.

[3] X. Yang, M. Liao, L. Zhang, S. Tebaldini, “Forest Height Retrieval in Tropical Areas using P-band Multibaseline SAR Data,” Proc. 6th Asia-Pacific Conference on Synthetic Aperture Radar (APSAR), 2019.

[4] T. Dinh Ho Minh, et al., “Vertical Structure of P-Band Temporal Decorrelation at the Paracou Forest: Results from TropiScat,” IEEE GRSL, 11 (8), pp.1438-1442, 2014.

[5] F. Lombardini, “Differential Tomography: a New Framework for SAR Interferometry,” IEEE TGARS, 43 (1), pp.37-44, 2005.

[6] K. Scipal, M. Davidson, “The SAOCOM-CS Mission: First Bistatic and Tomographic L-band Mission,” Proc. IEEE IGARSS, 2017.

[7] D. Liang, et al., “A High-accuracy Synchronization Phase-compensation Method based on Kalman Filter for Bistatic Synthetic Aperture Radar,” IEEE GRSL, Early Access, 2019.

Landslides are a major cause of damages or losses of property and life around the world. The detection and mapping of active slopes is a critical task for landslide disaster prevention and early warnings development. SAR Interferometry (InSAR) has been proven to be effective for landslides deformation measurement. However there are blind areas, at where the landslides cannot be detected by InSAR. These areas are generally caused by geometric distortion and vegetation decorrelation. In engineering applications, identification of blind areas can not only provide a reference for the selection of SAR data, but also provide guidance for other technical tools such as field geological surveys to improve the identification of potential landslides.

This study divides the wide-area InSAR landslide detection into two parts: Prior identification of undetectable areas and early identification of landslide. First，we use DEM data, landcover map, and SAR acquisition parameters (including satellite heading angle and radar incidence angle) to conduct visibility estimation, sensitivity estimation, and InSAR applicability estimation, then forming a comprehensive evaluation index. The visibility estimation is used to identify layover and shadow areas caused by SAR observation geometric distortion. The sensitivity estimation is used to identify areas that are not sensitive to landslide deformation detection. The applicability estimation is used to identify low-coherence areas due to vegetation coverage. Their comprehensive evaluation indicators are provided to provide the final applicability map of landslide detection. Then, time series InSAR is used to detect the wide-area landslides using the StaMPS/SBAS tool.

A huge landslide occurred on 23^rd July, 2019 in Shuicheng County, Guizhou Province, China, causing huge damage. Taking Shuicheng County as an example (Figure 1), this study focuses on blind areas of landslide detection and early identification of landslides. This place is in the area of high mountains and canyons, with elevation ranging from 700 to 3000 meters above sea level. The landform landscape is dominated by mountains and hills, and there are landforms such as mountain plains, plateaus, and terraces. And the central Shuicheng County is violently eroded by the severe cutting and erosion of the Beipan River, hence the geometric distortion is very serious. The L-band ALOS-2 PALSAR-2 and C-band Sentinel-1 datasets were used. Among them, the ALOS-2 PALSAR-2 data set is in FBD mode, with a total of 16 scenes; the Sentinel-1 data set is in IW wide mode, with 67 ascending scenes and 69 descending scenes, respectively.

The result blind area detection is shown Figure 2. From the visibility of landslide detection (Figure 2(a)), the shadow area is distributed on both banks of the river valley and the edge area of the elevated terrain. Most of the shadow areas of the ascending orbit overlap with the layover areas of the descending orbit. Therefore, some areas are still undetectable even combining both ascending and descending orbits, which needs additional technical means for supplementary investigation. From the sensitivity of landslide detection (Figure 2(b)), the sensitivity to the north-south slope is the lowest, and the east-west slope is highly sensitive. Therefore, special attention should be paid to landslides with low sensitivity. Judging from the applicability of InSAR (Figure 2(c)), , the applicability is poor in most areas due to dense forest cover, which needs long-wavelength SAR data to obtain high coherence.

We detected six active landslides from the deformation rate maps (Figure 3), covering more than 500 square kilometers, which were interactively verified by the three SAR datasets. Most of the landslides are distributed on the edge of the high-level terrain, presenting huge threat to the local residents at the valleys. There are two landslide areas caused by mining deformation (Figure 4). Continuous monitoring of the safety of buildings around the mine is required, although the danger is relatively small.

It should be pointed out that the Jichangzhen Landslide was not identified from the wide-area detection result. The Jichangzhen landslide is located four kilometers west of the Beipanjiang River, on the southern slope of Jichang Town. The landslide descended from south to north, destroying the buildings that on the slope and foothills, causing huge casualties. From the time series deformation of point P3 (Figure 5), no obvious deformation were found. However, there are landslides called Fanazu and Xujiaying landslides that have been detected on the opposite mountain, whose deformation degree exceeds 60mm/y. The time series deformation of points P1 and P2 shows seasonal pattern. The deformation in winter is small, and the deformation in summer is large. Considering that the geological conditions here as same as those of the two nearby landslides, we speculate that the landslide may be a sudden landslide caused by rainfall. Therefore, the rainfall threshold method should be adopted in the region to carry out meteorological and geological disaster monitoring and early warning.

Landslides are a major threat in all mountainous regions throughout the world, causing huge losses of life and property every year. The measurement of landslide surface deformation can facilitate prevention and mitigation of landslide disasters. The potential landslides can be detected through capturing the pre-failure deformation. The short-term emergency monitoring can be carried out to ensure the rescue security after the landslide collapse using ground based InSAR. The satellite InSAR is more suitable for long-term post-failure landslide risk assessment.

This study takes the Xinmo landslide as an example to demonstrate the pre- and post- failure landslide analysis using both satellite and ground based InSAR. The Xinmo landslide happened in the early morning of 24 June 2017 at about 5:38 am local time, and this catastrophic event caused enormous casualties in Xinmo Village, Mao County, Sichuan Province in China.

We use the Sentinel-1 data to capture the pre-failure movement. Both the ascending and descending Sentinel-1 were processed using StaMPS/SBAS. The sliding source area can be identified from their deformation rate map. The accelerated movement of the source area just before the collapse was measured by both the ascending and descending Sentinel-1 data.

The IBIS FL system was used to assess the post-failure stability of the Xinmo landslide, starting at 20:45:41 in 29 June and ending at 08:52:10 in 05 July. We obtained the averaged linear deformation rate from the 927 scenes for the Xinmo landslide. The averaged linear deformation rate map presents two unstable areas moving toward the GBSAR instrument. One is the debris mass on the west flank of the landslide. The whole east side moved with maximum linear deformation rate larger than 40 mm/d. Another is located at the deposit area. The spatial pattern is slender along the elevation direction, indicating the broken rocks slipping along the slope.

The post-failure ascending and descending Sentinel-1 datasets acquired from June 2017 to December 2019 were processed using StaMPS/SBAS method. The ascending and descending orbits present similar deformation distribution. The north, south, and east sides of the debris mass on the west flank are unstable after the landsldie failure as indicated by the Sentinel-1 results. The maximum LOS deformation rate reached -80 mm/yr. In addition, a large number of rocks and gravels loosely stacked at the bottom of the landslide, i.e. the deposit area. These loose deposits are compacted under the force of gravety. The compaction process behaved as subsidence, which can be captured by the Sentinel-1 satellites. The Sentinel-1 results reveal that the subsidence mainly happenedd in the east side of the deposit area.

Large-scale man-made linear hydraulic engineering is a powerful way to improve the ability of water resources allocation and water supply guarantee. It has the features of long line, wide areas, complex engineering layout and many types of functional buildings. The South-to-North Water Diversion (SNWD) Project is a great infrastructure of China to alleviate the serious water shortage in North China and optimize the water supply pattern. It consists of three parts of channels: the Eastern Route Project, the Middle Route Project and the Western Route Project. The total length is 3797 km, involving 15 provinces and 500 million residents. To guarantee the secure operation, it is essential to monitor the geo-stability of such large-scale linear hydraulic engineering. However, traditional ground-based techniques are usually limited by the distribution density of sensors and relocation difficulty of monitoring instruments. Large-scale monitoring of channels and buildings also requires huge monitoring funds.

The Synthetic Aperture Radar Interferometry (InSAR) has been considered as an effective tool for detecting long-term deformation and identifying possible safety problems in civil engineering. The satellite InSAR has the advantages of large-scale coverage, high-precision deformation measurement with low cost. High-frequency coverage of wide SAR images makes this InSAR more suitable for the deformation monitoring of the large-scale infrastructures, such as the SNWD Project.

In this study, a long-term time-series Sentinel-1 IW dataset, including total five frames of Path 11, Path 113 and Path 40, was processed by Persistent Scatters InSAR (PS-InSAR) to monitor the dike of the Middle Route of SNWD Project, as shown in Figure 1. The unstable channel sections were detected and some key sections were analyzed in details.

We obtained the distribution map of deformation rate of the channel in the Henan Province with a length of 730 km, as shown in Figure 2. This result indicates that the dike is generally stable, except for partial canal sections. Some deformation of channel sections is caused by the surrounding ground subsidence, while others are deformation of the channel slope itself.

The Yuzhou-Changge channel section, according to the deformation rate distribution derived by PS-InSAR, crosses through a subsidence region, and the settlement canal is approximately 2.5 km long, as shown in Figure 3. Its maximum deformation velocity is near -20 mm/yr and the profile of the deformation rate along the channel line presents a very uneven deformation of the channel slope. The result is verified by in-situ leveling measurements and they agree well with each other.

The section of the channel in Zhengzhou area also present complex deformation pattern. This channel section passes through deformation areas with uplift and subsidence appearing alternatively. The InSAR deformation time series is consistent with the leveling measurements. The relationship between the special deformation characteristics and the complex distribution of geological faults was discussed.

We carried out an analysis of deformation distribution characteristics of Shahe Aqueduct, one of the most technically complex control engineering in the SNWD Project. The InSAR result reveals that this engineering is relatively stable, except for two design sections. For one of them, the deformation rate of its main body is smaller than that at its entrance. This is related to the engineering features of its lower supporting structure and geological conditions. The other one, the left side of which is excavated from the slope while the right side of which is high fill, is undergoing obvious deformation.

The time-series InSAR technique with high efficiency, high precision makes it a powerful tool for long-term safety monitoring of the SNWD project. It can greatly reduce the monitoring cost for the SNWD Project. It can not only measure the deformation of the channel slope itself, but also detect the stability of the surrounding area of the channel slope. This is more conductive to identify the root causes of deformation and provide guidance for subsequent channel safety maintenance. The InSAR technique can be extended to other routes of the SNWD Project or other water conservancy projects.

On June 17, 2019, an Ms6.0 earthquake occurred in Changning, Sichuan, China (Changning event), which is the largest earthquake within 50 km of the area since records. It attracts great attention as the area has the largest shale gas production and large mineral salt production in China. Based on Interferometric Synthetic Aperture Radar (InSAR), we measure the coseismic deformation and build the fault models of the Changning event and two earlier Ms>5.0 earthquakes (P1:2018/12/16 Ms5.7 and P2:2019/1/3 Ms5.3) using Sentinel-1 and ALOS2 satellite data. The surface deformation caused by the Changning event is mainly uplift with a maximum of 17.2 cm (towards the satellite). Because of the significant non-Double-Couple character of the earthquake, we obtain a double-fault model (FMB) for the event. The final model shows that the Changning event was caused by a small fault (FMB2) and a big fault (FMB1) with a left-lateral strike and thrust slip. The strike of the main fault is 128° with a dip angle of 46°. The total seismic moment obtained by inversion is 6.68×10¹⁷ Nm, corresponding to Mw 5.85. The model is roughly consistent with the double slip model provided by seismology. This provides the geodetic evidence for the double slip of the Changning event. Based on the fault model, we analyze the cause of the Changning earthquake from the local tectonic setting, Coulomb stress change, mining, and fluid injection. The results show that the Changning earthquake is likely to be affected by external stimulation. However, the Changning earthquake does not show the characteristics induced by hydraulic fracturing. The stress change on the main fault after the two earlier Ms>5.0 earthquakes is main positive, with a maximum of 0.09 MPa and the water-flooding position of the salt mine is highly coincident with the main slip area of the fault. Therefore, there is no direct evidence showing that the Changning earthquake was related to hydraulic fracturing. We consider that the event may be related to salt mining. P1 and P2 may also play an important role in advancing the Changning earthquake. The characteristics of aftershock distribution indicate that seismic activities may be controlled by crust heterogeneity and structural complexity.

The continuous growing up of Multi-Temporal InSAR (MT-InSAR) methods [1]-[3] have shifted up the reliability and affordability of these techniques. In this context, the new achievable milestones are represented by the challenging tasks of design and tests advanced MT-InSAR for improving the previous techniques, and also for correcting some of the canonical problems that have arisen in the last decades through the well-known interferometric approaches.

The methodology proposed in this abstract is naturally well-posed on the context of the MT-InSAR approaches, and in a specific way, it is designed to correctly work on decorrelated areas, such as the Basilicata region (Italy) and Shanghai bay area (China), which are the two test sites where we have applied the developed framework. Briefly, the test sites are known as problematic and instabilities zones where to perform InSAR analyses, aimed at computing the so-called displacement ground time-series of the investigated targets. The main issues over the aforementioned test sites can be reconducted to the losses of spatial coherence between consecutive SAR acquisitions: These effects are well-known in literature as decorrelation phenomena [4].

The proposed work is based on the use of weighted Least-squares (WLS) approaches for the generation of ground line-of-sight (LOS) displacement time-series through differential interferometric SAR methods. More precisely, in this paper, a WLS method that extends the usability of the Multi-Temporal InSAR (MT-InSAR) Small Baseline Subset (SBAS) algorithm [5] in regions with medium-to-low coherence is presented. The proposed method relies on the adaptive selection and exploitation, pixel-by-pixel, of the medium-to-high coherent interferograms, such as to discard the noisy phase measurements. In such a way, for each pixel, it is possible to obtain several disjoined sets of interferograms, which are then connected by exploiting the weighted singular value decomposition (WSVD) method [6]. As a consequence, the interferogram networks reduction may lead to rejecting some SAR acquisitions from the used dataset, this results in the generation of so-called variable-length displacement time-series. The core of the approach resides in the innovative interferograms inversion step, through the Weighted-SVD approach, which permits us to emphasize the more reliable phase measurements by constraining them with the weights computed as the reciprocal of the phase variance. In a specific way, the strategy adopted for weighting each pixel of every InSAR pairs is straightforward obtained by exploiting the basic principles of the directional statistic [7], in this manner it is possible to compute the variance values, avoiding the computational burden of the standard approach. The solution is now pixel-dependent, because both the selected InSAR data pairs and the associated weights are different from pixel to pixel. In critical regions, it may also happen that the pre-selection of the “good” interferograms, for the given SAR pixel, leads some SAR acquisitions being discarded. In this latter case, the ground displacement time-series can also still calculate but with a reduced length that preserve the reliability of the results, with a significant increase in correctly investigated pixels, at the expense of a less dense temporal sampling.

The experimental results have been performed considering by applying the Weighted Adaptive Variable-lEngth (WAVE) technique [8] to two different SAR datasets collected by the Italian Space Agency Cosmo-SkyMed (CSK) constellation sensors and the Sentinels platform over the Basilicata region (Italy) and the Shanghai bay area (China), respectively. The preliminarily results will be presented and discussed at the next Dragon-IV meeting.

References:

[1] Ferretti, A.; Fumagalli, A.; Novali, F.; Prati, C.; Rocca, V.; Rucci, A. A New Algorithm for Processing Interferometric Data-Stacks: SqueeSAR. IEEE Trans. Geosci. Remote Sens. 2001, 49, 3460–3470.

[2] Fornaro, G.; Verde, S.; Reale, D.; Pauciullo, A. CAESAR: An approach based on covariance matrix decomposition to improve multibaseline multitemporal interferometric SAR processing. IEEE Trans. Geosc. Remote Sens. 2015, 4, 2050–2065.

[3] Pepe, A.; Yang, Y.; Manzo, M.; Lanari, R. Improved EMCF-SBAS Processing Chain Based on Advanced Techniques for the Noise-Filtering and Selection of Small Baseline Multi-look DInSAR Interferograms. IEEE Trans. Geosci. Remote Sens. 2015, 53, 4394–4417.

[4] Zebker, H.A.; Villasenor, J. Decorrelation in interferometric radar echoes. IEEE Trans. Geosci. Remote Sens. 1992, 30, 950–959.

[5] Berardino, P.; Fornaro, G.; Lanari, R.; Sansosti, E. A new algorithm for surface deformation monitoring based on small baseline differential SAR interferograms. IEEE Trans. Geosci. Remote Sens. 2002, 40, 2375–2383.

[6] Van Loan, C. F. Generalizing the singular value decomposition. SIAM J. NUMER. ANAL. 1976, 13, 1.

[7] Mardia, K.V.; Jupp, P.E. Directional Statistics; Wiley: New York, NY, USA, 2000.

[8] Falabella, F.; Serio, C.; Zeni, G.; Pepe, A. On the Use of Weighted Least-Squares Approaches for Differential Interferometric SAR Analyses: The Weighted Adaptive Variable-lEngth (WAVE) Technique. Sensors2020, 20, 1103; doi:10.3390/s20041103

Spaceborne Multi-temporal Synthetic Aperture Radar Interferometry (MT-InSAR) is an advanced technology which is capable of obtaining line of sight (LOS) ground deformation measurements [1, 2]. In order to obtain long-term deformation time-series, data merging methods have been proposed to combine deformation time-series which are independently retrieved by different SAR satellite datasets [3, 4]. Singular value decomposition (SVD) method has been efficiently used for combing two time-overlapped ground deformation time-series [2]. SVD can solve the rank-defect linear solution problem in the combination of two time-overlapped deformation time-series [4]. However, the optimal number of SAR observations during the time-overlapped period for combing two deformation time-series with SVD is still unknown. It is necessary to be investigated.

In this study, two time-overlapped SAR datasets covering Shanghai were used. The first dataset consists of 61 images, which were collected by COSMO-SkyMed (CSK) sensor operating at X-band (Descending, HH polarization) from December 7, 2013 to March 18, 2016. The second dataset consists of 47 images, which were collected by Sentinel-1A(S1A) sensor operating at C-band (Ascending, VV polarization) from February 26, 2015 to January 12, 2019. There are 23 CSK images and 13 S1A images in the time-overlapped period from February 26, 2015 to March 18, 2016. We separately caculated the CSK and S1A LOS deformation time-series through Small BAseline Subset (SBAS) [2]. The CSK and S1A LOS deformation time-series were converted to vertical direction with the satellite incidence angles [5, 6]. Then, we removed the components of the CSK vertical deformation time-series in N1 (1,2…,22) times during the time-overlapped period. Correspondingly, we removed the last N1 components in the N1-th removal operation. Thus, we acquired 22 groups of new CSK deformation time-series. Furthermore, the 22 groups of new CSK time-series were combined with the initial S1A time-series by using SVD method respectively. we obtained 22 groups of CSK+S1A deformation time-series. Similarly, we removed the components of the S1A time-series by N2 (1,2…,12) times, and removed the first N2 components in the N2-th removal operation. We obtained 12 groups of new S1A time-series. After the combination with initial CSK time-series, we obtained another 12 groups of CSK+S1A deformation time-series. Finally, the accuracy of the 34 groups of CSK+S1A time-series were evaluated by using the ground leveling measurement data.

We compared the accuracy of the 34 groups of combined CSK+S1A ground deformation time-series to investigate the optimal number of SAR observations during the time-overlapped period. The results indicate that if the number of overlapped SAR observations is less than 5, the accuracy of combined time-series is dramatically decrease. The accuracy of time-series combined with more than 5 SAR observations in the time-overlapped period is similar with the one obtained with only 5 SAR observations. It suggests that when SVD is used for combing two time-overlapped deformation time-series, the optimal number of SAR observations during the time-overlapped period is approximately 5.

There has been an increase in space-borne sensors for Earth observation in recent years, e.g., the Synthetic Aperture Radar (SAR). These sensors can obtain many images with different resolutions for the entire planet, even for the most remote areas. One of the applications is studying terrain deformations using Interferometric Synthetic Aperture Radar (InSAR) techniques. In 1992, the first demonstration of the InSAR technique, where the ERS-1 satellite generated an image of the deformation caused by the Landers earthquake. Since that time, there have been advances in this technique applied to studying all types of terrestrial deformations, with particular success in studying earthquakes and volcanoes. The success of InSAR has led to the interest of space agencies that have started to create SAR-only missions and make available an exponential increase in the SAR data, which become impossible to analyze manually. Machine Learning (ML) can automatically process large datasets in the most varied areas, including remote sensing data, and it has become an opportunity for earth observation. Recent studies have demonstrated the ML's ability to detect visible fringes in InSAR images caused by deformations such as earthquakes deformations. However, InSAR data is frequently unbalanced - deformations are sparse compared to those that do not have deformation, and it needs special attention for training ML models. There are several techniques to approach data unbalancing, such as synthetic data and focal loss. In this work, we created an earthquake dataset from the LICS database with 3100 images and are exploring unbalanced technics to improve the train for ML models to detect earthquake fringes.

In the 20th century, scientists discovered that earthquakes occur due to the release of energy that was accumulated in tectonic faults. These natural phenomena occur with different intensities and scales. They can be felt from a few meters to kilometers from their epicenter depending on their magnitude and depth and cause significant damage and fatalities. Knowledge of the affected area is essential to avoid risks after the event, such as the collapse of buildings in short/medium term after the event. Since 1992, with the development of Interferometric Synthetic Aperture Radar (InSAR) techniques, it is possible to generate images of the deformations and visualize areas on the globe that have been deformed. With this great success, European Space Agency (ESA) developed Sentinel-1 with C-band and provided free access to SAR data bulks. The Centre for Observation and Modelling of Earthquakes, Volcanoes & Tectonics (COMET) make InSAR calculations and make them free available (https://comet.nerc.ac.uk/comet-lics-portal/). As this amount of data has become impossible to process manually and Machine Learning (ML) start to be used to process these data automatically. Several studies have been applying ML's ability to detect InSAR images with the fringes in InSAR images; however, no studies have been found where the area is isolated using segmentation techniques. In this work, we apply ML segmentation models to automatically identify areas affected by earthquakes in interferograms.

Changbaishan Tianchi volcano is located at the border of China and North Korea, which is the most complete Cenozoic polygenetic composite volcano preserved in China, and also the most active volcano with eruption potential in China. In history, the volcano has experienced several major eruptions, amongst which the Millennium eruption about 1000 years ago is one of the largest eruptions in the world during the past 2000 years. During the period from 2002 to 2005, the frequency and magnitude of seismicity in Tianchi volcano area increased significantly, and the gas geochemistry at several hot springs also changed significantly, indicating the volcanic activity entered an active period. After the active period, volcanic activity has returned to previous level. However, a volcanic swarm suddenly appeared on December 22, 2020, with 38 volcanic seismicity events of various types occurred. On March 5, 2021, another earthquake with magnitude of ML3.1 occurred, which is the largest VT (Volcanic-tectonic) type earthquake event after active volcanic disturbance period. These phenomena show that the volcanic seismic activity after December 2020 is beyond the background level, and monitoring of Tianchi volcano should be conducted continuously.
Traditional surface deformation monitoring methods, such as leveling and GPS observation, are limited by shortcomings of sparse space points and low accuracy, which are not able to reveal the spatial distribution of deformation information over a large area. On the other hand, Interferometric Synthetic Aperture Radar (InSAR) technology makes up for the shortcomings. Advanced time-series InSAR methods are able to overcome the influence of orbital errors, atmospheric artifacts, topographic errors and other errors in interferometric phase by analyzing long time series SAR data, and obtain large-scale and high-precision deformation monitoring results. Among the commonly used time-series InSAR methods, the Small Baseline Subset time-series InSAR (SBAS-InSAR) can effectively reduce the influence of temporal and spatial decorrelation due to its high image utilization, high spatial density and short baselines. It is more suitable for deformation monitoring of Tianchi volcano area, which long-time snow coverage and vegetation.
In this study, 19 ALOS-2 images from November 2018 to October 2020 were processed to estimate the surface deformation of Tianchi volcano area. Due to the large topographic inequality in the volcanic area, obvious vertical stratified atmospheric phase and topographic residual are detected and removed from the interferometric phase. The results show that the ground surface of Tianchi volcano is uplifting slightly during this period, and the closer to the crater, the more obvious the uplift. Geophysical modeling based on Mogi point source was also conducted using the SBAS-InSAR results, indicating a point source at the depth of approximately 3km. The modeled depth is in agreement with the depth of seismic events, which is possibly where hydrothermal activity is happening in the shallow magma chamber. Generally, this study provides data support for the monitoring of volcanic activity and future disaster assessment in Changbaishan Tianchi volcano, and verifies the feasibility of time series InSAR and geophysical modeling in applications of active volcanoes.

Surface displacement monitoring of open-pit mines is of vital importance for the safety of field operators and mine production. With the Development of Synthetic Aperture Radar (SAR) remote sensing technology, Time series Interferometric SAR (TS-InSAR) has been widely used in wide-area displacement monitoring in recent decades. However, limited by the side-looking geometry, the estimated displacements in TS-InSAR are only in the satellite’s line-of-sight (LOS) direction, which are completely different from real displacements in areas with large topographic inequality, since displacements may occur in the three dimensional space in such areas. In translational landslide monitoring, an assumption that displacements generally happen parallel to the largest slope gradient is made. In order to resolve slope displacements of landslides, an accurate digital elevation model (DEM) aided slope displacement inversion method is proposed. The accurate DEM is generated by Light Detection and Ranging (LiDAR) scanning, used to minimize the topographic residual TS-InSAR analysis, and then in inversion of slope displacements from those in LOS.

In this paper, three stacks of Setinel-1 images (186 scenes) acquired from both ascending and descending orbits are used for SBAS analysis over Qidashan and Anqian open-pit mines, with temporal coverage from December 2016 to May 2019. In the SBAS processing, LiDAR DEM of the mining pits and TanDEM-X DEM with 3 arc second spatial resolution are used to remove the topographic phase. Accurate slope angles and aspects derived from the LiDAR DEM are used to estimate the slope displacements. In order to ensure high temporal sampling rate and high point density, a multi-master interferometry mode is adopted in the small baseline subsets (SBAS) method, with reduced spatio-temporal decorrelation problems. In SBAS, firstly, all the slave images are registered with the super master image, and small baseline differential interferograms are generated. Then, a minimum cost flow algorithm is used to unwrap the differential interferograms. After that, singular value decomposition method is adopted to combine the unwrapped differential interferograms and generate a time series which includes atmosphere, residual terrain and deformation signal. After that, the residual topography and atmosphere signals are eliminated by a second-order unwrapping and filtering, and finally the deformation rates and the cumulative deformation are retrieved. With the assistance of high-resolution LiDAR DEM, the deformation parameters in LOS are converted to slope direction using the radar incidence angle, heading angle, azimuth angle, slope aspects and angles.

The derived results are highly consistent with on-site GPS measurements and precipitation data. Based on the analysis of local precipitation changes, it is found that there is a correlation between deformation and precipitation. The results show that the stability of open-pit mines is mainly affected by rock mass structure, lithology and precipitation. SBAS method can be used as a routine tool to monitor the stability of open-pit mines, and provide technical support for disaster prevention and safe production.

Landslide inventory mapping is an essential asset for hazard assessment and mitigation. Despite significant progress in the development of automated image processing chains for landslide inventory mapping, tedious manual interpretation of aerial and satellite images remains the de facto standard (European Centre on Geomorphological and Seismological Hazards, 2017). Optical remote sensing is until today not used as an operational tool in contrast to SAR interferometry and LiDAR. The reasons are a lack of adapted image processing techniques that address the specific information needs of landslide investigations while remaining sufficiently generic to a broad range of different landslide types and environmental settings. This study investigates in automated change-detection workflows using annual Sentinel-2 (S2) composites and S2 multi-temporal imagery. Landslide areas in the time period 2015-2019 were analyzed based on already-known landslide location points, downslope-oriented moving windows and supervised classifications of change-vector-intensity and -angle using Receiver Operating Characteristic (ROC) curves. Subsequently, time-series analysis was applied on the resulting change-pixels to derive temporal breakpoints (i.e. the timing of the landslide occurrence). Our findings highlight that out of 65 reported landslide locations in South Tyrol, only 9 (13.8%) are recognizable by means of S2 imagery. Large landslides, however, were detectable both spatially and temporally by means of the multi-temporal change-detection approach. By applying a quantitative accuracy assessment for the independent test site in Longnan, China, our results show that the approach is highly transferable with minimal adjustments and is suitable for efficient spatial-temporal landslide mapping across large areas.

One of the limitations in the characterization of Earth surface deformation lies in the compromise between wide spatial coverage, high spatial resolution, and high temporal resolution. Today, Sentinel-1 constellation overrides this limit by providing global and systematic coverage, opening up new perspectives for monitoring surface deformation.

We focus our study on the oriental part of the Tibetan plateau, where major strike-slip and thrust faults combine to accommodate its overall eastward extrusion.

Here, we present the first FLATSIM automated time series analysis of Sentinel-1 data, acquired over the period 2014-2020 along 1200 km-long tracks on seven ascending and seven descending orbits. FLATSIM (ForM@Ter LArge-scale multi-Temporal Sentinel-1 InterferoMetry) is an automated Sentinel-1 InSAR processing chain based on the NSBAS approach (Doin et al., 2011, Grandin, 2015), implemented at CNES high-performance computer center in Toulouse in the framework of the french Data and Services Center ForM@Ter.

Our study area covers in total 1 700 000 km2, We provide the phase evolution through time for this area as well as average velocity maps with a 160 m spatial resolution.

After time series inversion, we proceed to a source separation to identify seasonal signals and perform a referencing in ITRF2008 of the average velocity maps on all tracks in eastern Tibet with a new GPS-InSAR adjustment method using a minimum number of parameters. These referenced InSAR data are then decomposed into horizontal and vertical motion contributions and inverted with the latest published GPS velocity field using the T-DEFNODE elastic block model. This allows to discuss strain localization and rates at a regional scale in the light of the tectonic and geological evolution of this area.

Urbanization accelerates the deterioration of the urban thermal environment, which is characterized by increased spatial variability. Consequently, sustainable urbanization strategies require air temperature information in high-resolution for the provision of effective solutions and countermeasures to thermal stress. The existing networks of in-situ air temperature observations fail to describe effectively the thermal spatial variability and thus statistical modelling approaches based on Earth Observation (EO) data can contribute towards this end.

The proposed data-driven modelling framework/tool can be used to provide insight on the state of the thermal environment in high-resolution and consists of the following:

1. Downscaling of medium resolution EO data.
2. Development of a regional air temperature database.
3. Artificial Neural Network (ANN) modelling approach.
4. Construction of high-resolution spatio-temporal air temperature fields.

The framework is based on using the high temporal resolution thermal infrared imagery in combination with high spatial resolution data, also in the thermal infrared. The initial step is the downscaling of Land Surface Temperature (LST) to higher spatial resolution via a disaggregation method to obtain sub-daily and high spatial resolution LST information. In addition, the modelling framework supports the use of multi-source meteorological data that includes in-situ observations from weather stations and field experiments/campaigns, along with high-quality reanalysis products. The ERA5-Land dataset (produced by ECMWF as part of the Copernicus Climate Change Service), which provides hourly high-resolution climate information can be used upon validation, as an auxiliary dataset for enhancing the available observational databases with high-quality data. The use of ANNs facilitates the development of a statistical model that incorporates the non-linear interactions that formulate the relationship between EO data with the near-surface air temperature in the urban environment. The complexity of this relationship is highly dependent on the surface energy budget and requires the application of sophisticated data-mining approaches. The use of the trained ANN models results into the construction of long-term spatio-temporal air temperature fields that provide the required level of information regarding the local thermal conditions and thus can be effectively used in urban strategies for enhancing resilience to thermal risks.

Case-study results and validation, demonstrate the efficiency of the framework for providing accurate estimates that are useful for improving the understating on how LST relates with the near-surface air temperature in urban areas. The EO retrievals of LST could be used to create high-resolution air temperature estimates that are useful in urban climate science and for supporting smart urbanization plans. Furthermore, as a tool towards sustainable urbanism, the framework can be also applied even in areas with sparse weather observation networks and can provide accurate estimates of the local variation of near-surface air temperature for studying critical phenomena such the urban heat island.

Abstract: Heilongjiang Sanjiang Plain has naturally unique advantage in development of model agriculture with abundant land resources, large connected arable fields. In recent years, the emergence of various kinds of high spatial and temporal resolution satellites data has strongly promoted the integration and combination of agricultural remote sensing, global navigation technology and Internet of things technology and then played an important role in support of crop classification, accurate agricultural production management, yield estimation and pest control. In this study, the Chuangye farm in the plain was selected as the study area. The sentinel 2 time series images during the growing season of 2020 were downloaded from the official website https://scihub.copernicus.eu. The samples were collected from Google Earth high-resolution image, query with farmers and field survey. The Random Forest algorithm was applied to implement the classification. Finally, the error matrix revealed that the overall accuracy of rice variety classification in the study reached 70%.

Key words: Rice; Crop classification; Agricultural Remote Sensing; Sentinel Satellite; Dragon Program

The re-analysis data can provide global gridded surface temperature, albedo, atmospheric profile and other parameters in continuous spatial and temporal dimensions. The atmospheric profile information derived from re-analysis data had been used in radiometric calibration and validation in infrared band by some researchers, together with the support of sea surface temperature (SST) measured by buoys. However, the spatial distribution of buoy data is uneven and there are many interference factors in the observation of buoy data. In order to reduce the cost and the complexity of calibration procedure, at the same time to increase the calibration frequency, it is a feasible way to use both the SST and atmospheric profile information extracted from the re-analysis data. However, one of the prerequisite is to check the accuracy of re-analyzed SST and consistency with other observations, and subsequently the reliability of the simulated top-of-atmosphere (TOA) based on the re-analysis data should also be verified. This study takes the ocean as research target to carry out the comparison and verification of SST and TOA brightness temperature from different sources, including the re-analyzed data from the European Center for Medium-Range Weather Forecast (ECMWF) ERA5, the buoys measured data, the satellite observations and related SST product. This study provides support for thermal infrared radiation calibration based on re-analysis data.

First, the comparison of ERA5 SST and ARGO buoy STT was carried out. ARGO buoy data was used in the South China Sea from 2008 to 2018. The results show that the annual maximum RMSE between ERA5 SST and ARGO SST is 0.53K, and the minimum RMSE is 0.42K. Next, comparisons between ERA5 SST and MODIS SST were carried out, with both the daily data covering the whole global area, and the annual data in the ocean area surrounding China. The results show that these data are highly consistent when the range of SST is big enough. The correlation coefficient between the two types of daily data is above 0.99, and the RMSE is 0.4K-0.65K. Finally, ERA5 SST and atmospheric profile data were used to simulate MODIS 31 band TOA brightness temperature, using MODTRAN atmospheric radiometric transfer model, and those simulations were compared with MODIS observations. The results show that the annual RMSE is within 0.7K in the marine areas surrounding China.

Pseudo invariant calibration sites (PICSs) have the advantages of high surface reflectance, low latitude, low aerosol thickness, low cloud cover frequency and so on, so they are very suit for monitoring the stability of sensors. In recent years, the TOA reflectance variation and modeling of PICS have attracted the attention of many researchers and organizations including CEOS WGCV/IVOS. Algeria 3, Algeria 5, Libya 1, Libya 4, Mauritania 1 and Mauritania 2 were identified as pseudo invariant calibration sites during the CEOS IVOS-19 meeting in 2008. However, PICS are currently distributed in the Sahara Desert in North Africa and the Middle East. It is difficult to apply them to the Chinese land satellites, which are mostly imaged in China. Therefore, this study analyzes and evaluates the site characteristics of the high-altitude Golmud Gobi site located around the Qaidam Basin in Qinghai Province, China, and proposes a TOA reflectance modeling method for inland stable targets. The TOA reflectance model of the Golmud Gobi site was established by using the long-time TOA reflectance product of MODIS, and the applicability and error analysis of the model were carried out, including:

Ⅰ)Based on the ECMWF re-analysis database, the changes of atmospheric parameters such as water vapor content, ozone content, and aerosol optical thickness in the Qaidam Basin of Qinghai Province were analyzed. The Golmud Gobi site with the best atmospheric and surface stability conditions was then selected for the measurements of site characteristics, including directional characteristics observation with UAV, spectral observation with field spectrometer SVC, and aerosol observation with CE318, etc. Finally, the surface BRDF data of the site was obtained.

Ⅱ)The Sentinel2A/2B cloudless surface reflectance data from April to September 2020 were analyzed. The surface reflectance data of B2/B3/B4/B8 with 10m resolution in each landscape image was screened with the spatial uniformity <3% as the constraint condition. Finally, a 5km*5km Gobi site with both spatial uniformity and temporal stability better than 3% was obtained.

Ⅲ)The TOA reflectance model of Golmud Gobi land site was finally established by analyzing the regularity between the MODIS TOA reflectance time-series data from 2010 to 2019 and SZA/VZA/RAA/DOY. When VZA is less than 35 degrees and VZA minus SZA is less than ±30 degrees, the mean and standard deviation of the relative error between the modeled TOA reflectance and observed TOA reflectance in each band are better than 0.08% and 2.7% respectively. Meanwhile, compared with the TOA reflectance measured by Sentinel2B/MSI sensor for in B2/B3/B4/B8A/B12 bands, the relative error of the TOA reflectance predicted by the model is 0.01%~4%.

Digital Terrain Model (DTM) is one of the main products of Earth observation from space, and spaceborne SAR interferometry is a unanimously recognized technique for accurately mapping ground topography at a wide scale. Nevertheless, such approches meet difficulties over densely forested areas due to both wave penetration problems and signal decorrelation with time. In this context, the BIOMASS mission, which will operate the first orbital P-band radar in order the map forest biomass at global scale, is also expected to provide a global DTM over forested areas, as its larger wavelength and short revisit time allows to overcome both the aforementioned penetration and decorrelation issues. This poster proposes a technique for retrieving the topography of the ground lying below dense forest covers, such as tropical forests, based on the use of P band Polarimetric Tomographic SAR acquisitions (PolTomoSAR). It is based on the regularization of tomographic estimates performed according to geophysical constraints on the estimated surface. In particular, this techniques aims to prevent discontinuities due sudden and significant variations of scattering mechanisms and to avoid over filtering, generally associated to the estimation of PolTomoSAR covariance matrices. The performance of the proposed technique is assessed over data acquired by the ONERA RAMSES sensor at P band over the test site of Paracou, in French Guiana.

With the rapid development of spaceborne SAR systems, temporal polarimetric SAR (PolSAR) image data has become an important research topic. The temporal data contain helpful information about the scene and are also related to the time evolution of the scene. Analyzing and developing this information in SAR time series is an essential and challenging task. Based on the time series polarization SAR data structure, this paper constructs the time series- polarimetric features. Through the characteristic analysis and dynamic analysis of the covariance matrix, the time-polarization features of the PolSAR image can be obtained. The time-polarimetric feature provides a feature reference basis for the temporal PolSAR classification.

A multi-level model (MLM) for forest height inversion is introduced and investigated using X-band single-pass interferometric synthetic aperture radar (InSAR) data. Compared with the two-level model (TLM), as a generalized model, the MLM is more in line with the characteristics of forest structure and the scattering mechanism for X-band data. Based on the MLM model, three simplified MLM models were derived: TLMm, SINC and MTND. An improved method of calculating coherence considering the assumptions of the MLM model is proposed. Airborne X-band single-pass InSAR data and LiDAR H100 data were used to verify the proposed new approach. The results showed that the MTND model can obtain more reliable and accurate inversion results compared to the SINC model and the TLMm. And the inversion accuracy of forest height can be effectively improved by ensuring that the theoretical models and calculations of coherence obey the same assumptions.

The Xi’an, China has suffered from severe land subsidence and ground fissure hazards since the 1960s. With the progress of economic development and urbanization, groundwater was over-exploited for more than 50 years. Consequently, it had caused the formation of fourteen ground fissures accompanying land subsidence throughout the city. The maximum land subsidence rate reached 300 mm/year in 1996, and the maximum cumulative subsidence reached 3 m approximately in last 60 years, which has threatened and will continue to threaten the safety of urban artificial constructions.

InSAR time series deformation monitoring plays an important role in revealing the historical displacement of the Earth’s surface. In order to study the spatiotemporal characteristics of subsidence and ground fissures, previous studies show that there existed a close spatiotemporal relationship between land subsidence and the formation of earth fissures. The degradation of the aquifer system is the one of the key factors to these typical deformations. In order to alleviate the land subsidence and ground fissures caused by over-extraction of groundwater, the pumping groundwater was restricted in 1996, meanwhile, water was supplied from Heihe River. Moreover, cumulative volume of 1552800 m3 had been recharged in Xi'an from 2009 to 2014.

In order to retrieve the history of land subsidence in Xi’an, we employed the sequential estimation method to update the time series deformation dynamically, which is an efficient InSAR tool to update the surface deformation as quickly as possible, when the SAR image is acquired one by one. In the experimental section,we take 83 Sentinel-1A images acquired from 20 June 2015 to 17 July 2019 to show the performance of the method and to analyze the evolution of land subsidence in Xi'an, China. For the sequential InSAR time series deformation processing, we divide SAR data into two groups. The first group is the archived SAR data for parameters initialization. While the second group is newly received SAR image (i.e., a new SAR acquisition) to update the new deformation parameter.

Results show three surface deformation phenomena occurred in Xi’an city from 20 June 2015 to 17 July 2019, i.e. continuous land subsidence, slow uplift and rebound after long-term subsidence, which can be explained as the response to the underground water changes in different regions. As for the response tothe artificial water injection, the rebound pattern can be further divided into immediate elastic recovery deformation and time-dependent visco-elastic recovery deformation.

Weihe Basin is located in Shaanxi Province, central China, bordering the south of Ordos block in the Weibei uplift belt, the north of North Qinling orogenic belt, the east of the southwestern margin of the Ordos arc structural belt, and the west of the Shanxi uplift belt, which plays an important role in the well-known Fen-Wei seismic belt, spanning 350km from east to west. Hundreds of active faults were discovered over the basin in history, and most of these deeply buried faults criss-crossed the basin . The main faults that control the sedimentary basin formation include Beishan Piedmont fault, Guanshan fault, Qinling North Piedmont fault, Chang'an-Lintong fault, and Lishan Piedmont fault. Weihe Basin is characterized by intense crustal activity, and more than 25 earthquakes have been recorded with magnitude larger than 5.0 since 1177. One M_S 8.0 earthquake occurred in Huaxian was recorded in 1556, however, there are no documented cases greater than M_S 5.5 since the 20th century.

In order to understand the structure and slip rate of the known fault zones, identify potential active faults as well as their activity intensities, and compensate for the existing discrete located GPS and leveling measurements in Weihe Basin, 92 scenes ascending Sentinel-1 (06/20/2015-05/30/2019) and 6 scenes descending ALOS-2 ScanSAR (04/01/2015-12/19/2018) datasets were utilized to derive LOS deformation velocity respectively. The geoid offset correction between EGM96 datum and WGS84 datum was applied to the 1-arc-second (~30 m) Shuttle Radar Topography Mission (SRTM) DEM, that is used as an external DEM to remove the topographic phase from the interferograms for both the Sentinel-1 and ALOS-2 ScanSAR processing. GACOS atmospheric products are employed to reduce the effects of tropospheric delay. Then the deformation components in the east-west and vertical directions are estimated using the descending and ascending InSAR images together over the entire basin. The standard deviation of vertical deformation between InSAR and leveling measurement is around 2 mm/yr. InSAR measurements allow us to position not only the previously known faults, but also new active fault not previously revealed, which indicates that the complex internal motion within the Weihe Basin could be controlled by multiple faults. Finally, we calculate the slip rates and blocking depths of the Longxian-Mazhao fault, Puyang-Lantian, the southern part of Weihe fault, and the Kouzhen-Guanshan fault, and evaluate their potentials to arise large earthquakes.

At 10:47 on January 19, 2019(Beijing time), the Xuelong ship, which carried out the 35th Antarctic survey mission, sailed in the dense ice area of the Amundsen Sea. Due to the influence of dense fog, Xuelong collided with the iceberg at 69 ° 59.9 ′ south latitude and 94 ° 04.2 ′ west longitude. A SAR image covering the same area was acquired by Sentinel-1 with HH polarization and pixel resolution of 40m at 10:41 on the same day. The research objective is to evaluate the potential of Sentinel-1SAR images to identify icebergs, especially whether the icebergs that hit the Xuelong can be detected to issue early warnings in advance. In addition to the Sentinel-1 data, the research data also includes the location information of the Xuelong provided by the ship's automatic identification system. First the Xuelong ship is detected according to AIS data. Then area of interest around the ship in the image is cropped. CFAR algorithm fit for iceberg detection is select to automatically identify the iceberg. Finally by the visual interpretation, the result is to determine whether the iceberg hitting Xuelong ship can be detected using Sentinel-1 data and the smallest size iceberg.

Nowadays, conventional Delay-Doppler processors (DDP) are continuously operating with data from satellites such as Sentinel-3 and CryoSat-2, providing surface waveforms with along-track resolution around 300m. Such a resolution is the maximum achievable with DDP methods, which process the echoes received within a burst in a coherent way but then combine the echoes from multiple burst incoherently. However, new Fully-Focused SAR (FF-SAR) techniques are expected to improve along-track resolution down to around 1m by coherently combining all the echoes received during the whole observation time. Such a step ahead in resolution opens a new set of possibilities to study the impact of along-track resolution in the evaluation of some sea state parameters.

Indeed, parameters such as the Significant Wave Height (SWH) when retrieved with DDP analytical retrackers often present residual bias that are related to how the illuminated scatters recombine within the 300m DDP footprint into the final waveform. In this sense, usage of FF-SAR processors offers the opportunity to study the SWH dependence on waveforms computed with along-track resolutions much smaller than 300m. Still, current radar altimeters operate at close-burst, which is a non-optimal mode for FF-SAR processing and therefore the presence of along-track replicas affects the measurements.

In this contribution we therefore present a comparison analysis of sea state variables retrieved by evaluating the Fully-Focused SAR processor with different along-track resolutions and multi-looking parameters. We have processed a set of Sentinel-3 passes over the Yellow Sea with a local Fully-Focused SAR processor implemented in time domain. The retracker considered is based on an analytical retracker developed specifically for DDP and adapted to operate with FF-SAR.

Sea ice thickness is one of the important environmental parameters describing the material balance of sea ice and controlling regional heat exchange, which has an extremely important impact on global climate change, resource development and shipping in polar regions. In the past three decades, with the development of sea ice remote sensing technology, satellite altimeters can achieve sea ice thickness observation on a global scale, and a variety of satellite sea ice thickness data products have been publicly released. These products usually provide complete coverage of the polar region for one month, which is limiting for polar sea ice monitoring and sea ice forecasting applications. The recently launched CryoSat-2 and Sentinel-3A, both carrying SAR mode sensors operating in Ku-band, are able to invert sea ice thickness more accurately. Due to the different orbital configurations, the two satellites have a low spatial overlap, making the fusion of sea ice thickness data possible. In this study, we propose to use the spatial complementarity of CryoSat-2 and Sentinel-3A to realize a half-month sea ice thickness fusion product in the Arctic region. The results are also compared and analyzed with Operation IceBridge (OIB) airborne measurement data and Beaufort Gyre Explorer Project (BGEP) elevation sonar actual measurement data to verify the accuracy of this fusion product.

This paper demonstrates the capability and performance of sea surface wind speed retrieval in coastal regions (within 200 km away from the coastline) using spaceborne Global Navigation Satellite System Reflectometry (GNSS-R) data from NASA's Cyclone GNSS (CYGNSS) mission. The wind speed retrieval is based on the Artificial Neural Network (ANN). A feedforward neural network is trained with the collocated CYGNSS Level 1B (version 2.1) observables and the wind speed from European Centre for Medium-range Weather Forecast Reanalysis 5th Generation (ECMWF ERA5) data in coastal regions. An ANN model with five hidden layers and 200 neurons in each layer has been constructed and applied to the validation set for wind speed retrieval. The proposed ANN model achieves good wind speed retrieval performance in coastal regions with a bias of −0.03 m/s and a RMSE of 1.58 m/s, corresponding to an improvement of 24.4% compared to the CYGNSS Level 2 (version 2.1) wind speed product. The ANN based retrievals are also compared to the ground truth measurements from the National Data Buoy Center (NDBC) buoys, which shows a bias of −0.44 m/s and a RMSE of 1.86 m/s. Moreover, the sensitivities of the wind speed retrieval performance to different input parameters have been analyzed. Among others, the geolocation of the specular point and the swell height can provide significant contribution to the wind speed retrieval, which can provide useful reference for more generic GNSS-R wind speed retrieval algorithms in coastal regions.

Data preparation is critical to the development of deep learning methods and can mean the difference between an inaccurate or highly accurate model. In standard deep learning image segmentation applications, data preparation normally takes the form of standardising input image sizes, colour and contrast adjustments, and augmentation such as randomised rotation and image cropping to vary the training process and improve accuracy while also reducing overfitting to training datasets.

Multi- and hyper- spectral satellite earth observation data offer many more channels than the consumer cameras used to train the majority of deep learning approaches. Standard pre-processing techniques, such as atmospheric correction, can be applied to earth observation data but, without consideration to their effects, they may remove or alter information that could prove valuable in a deep learning context. Theoretically, a deep learning neural network should be able to make use of all available waveband channels without alteration to achieve best accuracy.

The process of atmospheric correction uses dedicated wavebands and fitting procedures to separate the contributions of water, atmosphere and adjacent land prior to the interpretation of biogeochemical water quality from satellite imagery. The atmospheric correction combines known physics of the interaction of light with the various media and spectral optimisation approaches to remove up to around 90% of the at-sensor signal. This raises two questions. The first is whether atmospheric correction may remove features that deep learning networks can exploit to distinguish between objects. Reciprocally, it is worth exploring whether atmospheric correction over water could benefit deep learning networks by contributing to the level of harmonization in the input data, as the variable atmospheric influence is (in the ideal situation) removed.

To evaluate the need for atmospheric correction when applying deep learning for segmentation of objects in water, we prepared a training dataset processed to three levels of atmospheric correction. These are: (1) no correction (i.e., top of atmosphere), (2) correction for Rayleigh scattering to produce bottom-of-Rayleigh reflectance and (3) atmospheric correction using the POLYMER atmospheric correction tool. The dataset consists of coastal and inland waterbodies assembled from 150 Sentinel-2 MSI observations divided into 55,000 image tiles of varying sizes covering multiple sites across North America, South Africa and China and extensively labelled for the presence of phytoplankton blooms, land, water and cloud to be compatible with Deep Learning methods. This dataset was used to train a variation of Mask-RCNN to segment algal blooms.

We find that a priori atmospheric correction provides significant accuracy gains. Using all 13 MSI wavebands provides maximum accuracy, while subsets of 5 and 3 wavebands can be used with an accuracy loss of less than 10%, which suggests transferability of the method to sensors with alternative spectral capabilities may be possible.

Tropical cyclone (TC)-induced coastal flooding is a dangerous natural hazard to coastal areas. It is the combined influence of storm surge-caused ocean water inundation and heavy rain-caused river water flooding. TC-induced coastal flooding causes massive loss of life and property in coastal areas, where most of the world's population and mega-cities are densely distributed.

Accurate mapping of coastal flooding is important in several aspects: (1) it can assist the management to accurately evaluate the loss and make more suitable disaster relief plans; (2) it can help the scientists to better understand the flooding mechanisms and develop more precise forecasting models; and (3) the flooding extent information can be converted into the flooding depth information.

Synthetic aperture radar (SAR) is a suitable remote sensing means for coastal flooding mapping, since it can provide a day-and-night, all-weather sensing ability and high-resolution images of the flooded scene. In addition, SAR is sensitive to surface flooding, and sometimes it can extract flooding under vegetation. The traditional methods for coastal flooding mapping in SAR images are mainly based on image processing techniques by using backscattering, statistical, and polarimetric information. The traditional methods use pre-defined features and/or rules based on human observation and design. Due to the limitations of human-designed features and rules, it is difficult for them to provide robust performance under various influences, including speckle noise, SAR imaging system parameters, temporal mis-registration, meteorological conditions, and environmental differences.

The rapidly-developed deep learning technology in the big data era provides a promising solution for robust coastal flooding mapping in SAR imagery. In the deep learning paradigm, the deep convolutional neural networks (DCNN) are suitable for image information mining. In the DCNN models, the features for robust image pattern recognition and classification are mined from a large volume of input data, instead of being pre-defined with human's observation of a limited amount of data. These data-driven models with the help of domain knowledge have the advantages of providing robust features for pattern recognition and classification under various influences. In our previous work, we proposed an improved DCNN model which provides robust performance for coastal flooding mapping from bi-temporal and dual-polarization SAR images, and verified this model is useful for geospatial and temporal analysis of TC-induced coastal flooding.

Currently, there exist some DCNN-based models for coastal flooding mapping in SAR images. Most of them are based on supervised DCNN models. The supervised DCNN models need a large volume of ground-truth labels matching the observation data. The ground-truth labels are generated by human experts with the help of multiple data sources and scene information. This process is very time-consuming, costly, and subject to human bias. In addition, the trained supervised model should be fine-tuned for new conditions, if the new conditions are not covered by the data set used in the training. This may affect the generalization abilities of the DCNN model. Therefore, in this study, we propose to explore the DCNN models in another direction, that is, the models without human supervision, for coastal flooding mapping in SAR images.

The proposed design is based on deep convolutional autoencoder (DCAE). The DCAE models can fuse the backscattering, polarimetric, spatial contextual, and temporal information of SAR images together, and mine the most salient features in an unsupervised way from the high-dimensional input image data. In this study, we exploit this ability of the DCAE models, and then perform unsupervised clustering on the DCAE-generated features for coastal flooding mapping. We find the following interesting characteristics of the proposed framework: (1) The framework is robust under local interference, such as speckle noise and temporal mis-registration, since the convolutions in the DCAE can incorporate the spatial contextual information. (2) The framework is robust under meteorological influence, since it deals with the mapping problem case by case. The features for flooding mapping are generated from the data under the same meteorological condition. (3) The framework can not only extract flooded areas, but also classify the scene. Since the flooded areas are linked to the scene, we can get richer information from the flooded areas. For example, the flooded areas can be categorized as flooded areas along water areas, around built-up areas, or over open areas. (4) The framework can not only find surface flooding, but also have the potential to find flooding under vegetation at the same time. Under certain circumstances, the flooding under vegetation can have a distinct pattern in SAR images, since the flooded areas cause the increase of double-bounce scattering. This pattern can be extracted in the proposed framework along with the pattern from surface flooding. Therefore, we find the proposed framework is promising for robust coastal flooding mapping in SAR imagery, and it can get a more comprehensive understanding of the coastal flooding.

We perform some experiments with the Sentinel-1 SAR data during the passages of 2017 Harvey and 2019 Hagibis. The validation data are generated from human labeling with the help of history information and high-resolution optical data, if available. The preliminary results verified the aforementioned characteristics of the proposed framework. The unsupervised framework is suitable for natural disaster impact, such as coastal flooding, information mining.

Ocean monitoring is of paramount importance for maritime surveillance, climate change, environmental purposes and for the blue economy. When considering the microwave range of the electromagnetic spectrum, ocean shows specific scattering characteristics. In this domain, the Synthetic Aperture Radar (SAR) is becoming a well-established tool for maritime applications, including wind speed retrieval, target detection and pollution monitoring.

In this study, the spectral signatures of low-backscattering sea areas in Sentinel-1 SAR imagery, namely the autocorrelation function evaluated along the azimuth direction (AACF), are analyzed. Typically, low-backscattering sea areas are associated to oil spills due to the damping effects they have on the wind-driven short sea waves that result, in the gray-tones intensity SAR images, in patches darker than the surrounding sea. This study aims at analyzing the sensitivity of the AACF over different low-backscattering areas. A dataset composed of six Sentinel-1 SAR scenes, collected in Interferometric Wide Swath VV+VH imaging mode under low-to-moderate wind conditions over low-backscattering sea areas of known origin, is considered.

Experimental results show that the AACF is sensitive to low-backscattering sea areas of different nature, i. e., oil spills, algal blooms, etc. The largest departure, with respect to the AACF evaluated over sea surface, is observed over oil slicks. The outcomes of this sensitivity analysis can support the development of improved sea oil spill detection algorithm.

As a final milestone of the ESA-MOST Dragon-4 project, the papers published on peer-reviewed international journals must be mentioned, as follows:

V. Corcione, F. Nunziata, M. Portabella, G. Grieco, X. Yang, and M. Migliaccio, "SAR Azimuth Cut-off to Estimate Wind Speed under High Wind Regimes," Chinese Journal of Geodesy and Geoinformation Science, vol. 4. no.1, pp. 30-37, 2021.

V. Corcione, G. Grieco, M. Portabella, F. Nunziata and M. Migliaccio, “A novel azimuth cut-off implementation to retrieve sea surface wind speed from SAR imagery,” IEEE Transactions on Geoscience and Remote Sensing, vol. 57, no. 6, pp. 3331-3340, 2018.

V. Corcione, G. Grieco, M. Portabella, F. Nunziata and M. Migliaccio, “A novel azimuth cut-off implementation to retrieve sea surface wind speed from SAR imagery,” IEEE Transactions on Geoscience and Remote Sensing, vol. 57, no. 6, pp. 3331-3340, 2018.

G. Benassai, D. Di Luccio, V. Corcione, F. Nunziata and M. Migliaccio, “Marine Spatial Planning using high resolution SAR measurements,” IEEE Journal of Oceanic Engineering, vol. 43, no. 3, pp. 586-594, 2018.

Abstract: As the nontrivial instrument to probe at high spatial resolution under extreme wind conditions, C-band synthetic aperture radar (SAR) is able to collect high resolution measurements covering with tropical cyclones (TCs) to support the modeling of TC wind speed. The tangential wind profile model is one of the effective and widely used methods to reconstruct the radial wind speed of TCs. In this research, a tangential wind profile model (TWP) in the form of Gaussian function is proposed based on the retrieval wind speed of cross-polarized SAR data. The proposed model is to solve the common defects existing in the present models, such as the unsmooth transition of the wind speed profile in the high wind speed area (near the eyewall) and the poor performance of wind estimation in the TC eye area. In order to respectively estimate the wind speed inside and outside eyewall, the model function is designed as piecewise one with the maximum tangential wind as the threshold. Then, the model parameters can be determined by fitting the model functions to the azimuthal-mean wind speed derived from SAR observations. For model validation, we choose hurricane Arthur (2014) as a study case and compare the TWP model with the widely used single-modified Rankine vortex (SMRV) model, the results show that as the distance from the TC center increases, the tangential wind speed reconstructed by the TWP model changes relatively smoothly in the high wind speed area (near the eyewall). Besides, compared to linear fitting of SMRV in the aera inside eyewall, nonlinear estimating result of TWP performs better. Moreover, both the results of TWP and SMRV model have been validated with collocated Stepped-Frequency Microwave Radiometer (SFMR) measurement. For TWP model, the root means square error (RMSE), bias, and correlation coefficient are respectively 1.01 m/s, 0.58 m/s, and 0.99, while the same statistics are only 2.22 m/s, 1.50 m/s, and 0.96 for the SMRV model, respectively. Finally, the value distribution ranges of parameters are discussed by utilizing flight-level observations collected from Atlantic and eastern Pacific storms in the period 1977–2001, and the results of discussion can be the references for determining value of parameters in model application.

Keywords: C-band synthetic aperture radar (SAR), tropical cyclone (TC), tangential wind speed, parametric model.

Acknowledgments: This work was supported by the ESA-MOST Dragon 4 cooperation project ID 32235 “Microwave satellite measurements for coastal area and extreme weather monitoring”.

Emulsion oil slick is one of the most common types of oil spills in the marine environment. Oil slicks can be observed as “dark” patches on SAR images because: a) oil slicks effectively damp capillary and short gravity sea waves responsible for backscattering energy; b) the relative permittivity of the contaminated area decreases significantly due to the smaller oil permittivity. In this paper, to take both effects into consideration, the backscattering from the oil-covered sea surface is predicted using the Advanced Integral Equation Method (AIEM) model which is augmented with: a) the composite reflection coefficient of the two-layer medium that consists of adding an oil layer on top of the sea surface, where the seawater volume fraction in the oil emulsion modifies the dielectric properties; b) the sea spectrum model combined with the hydrodynamic model of local balance (MLB) which is employed to describe the damping of the small-scale roughness by an oil film. An accurate description of geometrical and dielectric properties for oil-covered sea surface is provided according to a layered medium and used to predict the radar backscattering according to oil’s thickness and oil/water mixing for different scenarios.

According to the model's predictions, the sensitivity of oil-covered sea surface backscattering to oil thickness and water content of emulsion increases as the increasing radar frequency with a reduced L-band sensitivity. The backscattering signals exhibit a nonlinear behavior with respect to oil thickness because oil films affect the backscattering in a twofold way: through the damping of the small-scale roughness as well as the composite reflection coefficient resulting from the three-layer medium. The incidence angle has a relatively minor impact on deviating the contaminated sea's backscattering, and the high wind speed can generally narrow the difference between the radar backscattering from the clean and oil-covered sea surface. Numerical simulations are compared with the multi-frequency SAR observations and results show that it is possible to estimate the oil thickness at reasonably good accuracy.

This work was supported by the ESA-MOST Dragon 4 cooperation project ID 32235 “Microwave satellite measurements for coastal area and extreme weather monitoring”.

Snow falls frequently over the Tibetan Plateau, and is a vital component of the widespread cryosphere which has vital feedback to climate change. Snowfall and the subsequent evolution of the snowpack have a large effect on surface energy balance and water cycle. Albedo, the main determinant of net radiation flux, is a major driver of land surface processes. However, the current widely used Noah land surface model does not describe snow albedo correctly, although it keeps snow-related variables into account. In the improved albedo scheme, albedo was parameterized as functions of snow depth and age which was developed using remote sensing retrievals of albedo. Numerical experiments were conducted to model a severe snow event in March 2017. The performance of WRF coupled with Noah applying the improved albedo scheme was compared with that of applying the default albedo scheme and with that of WRF coupled with CLM applying CLM’s complex albedo scheme. First, the improved albedo scheme largely reduces the WRF coupled with Noah albedo overestimation in the southeastern Tibetan Plateau, remarkably reducing the large cold bias estimates by 0.7 ℃ air temperature RMSE. Second, the improved albedo scheme gives the highest correlationship between the satellite-derived and the model estimated albedo, contributing to achieve the SWE spatial pattern, heavy snow belt and maximum SWE estimates in eastern Tibetan Plateau. Remarkable underestimation of albedo in WRF coupled with CLM contributes to regional maximum SWE underestimation and failure in heavy snow belt estimates.

Over a decade of OMI observations provide insight into the rapidly changing air quality levels in China. Global documentation of key atmospheric pollutant gases such as nitrogen dioxide (NO₂), sulfur dioxide (SO₂) and formaldehyde (HCHO) allow the study of anthropogenic and natural emissions on different spatial scales.

Based on bottom-up emissions inventories, Chinese SO₂ emissions were the world’s largest, particularly over the North China Plain. SO₂ sources are related to major coal-fired power plants and industrial activities such as oil and gas refining, and metal smelting. North China Plain is also characterized by a strong seasonal footprint of fire emissions, a significant contributor to poor air quality, due to biomass burning from agricultural activities. The highest NO_x emissions are observed over the world’s most populated (increased traffic), highly urbanized and industrialized regions.

Despite the growth of the economy in the rapidly developing China over the past two decades, a substantial overall decrease in the SO₂ and NO_x emissions has been observed with different patterns between the species. Changes are observed also in the human-induced fire emissions over the North China Plain. We identity the origin of these emissions and we investigate the spatial and temporal variability of the source types. Such variability is strongly related to differences in economic and technological activity, and regional environmental policies.

Government efforts to restrain emissions from power plants and industrial sectors (e.g. installation of de-sulfurization devices) have resulted in decreasing SO₂ and NO_x emissions since approximately 2007 and 2011, respectively. We use the SO₂/NO_x ratio to locate and characterize the emissions sources since, to some extent, it reflects the level of the regional modernization and helps us identify the source sector. For instance, the megacity of Shanghai and the areas around it are highly populated with cleaner power plants compared to other regions, therefore a relatively low SO₂/NO_x ratio is observed. We use OMI HCHO observed columns to infer biogenic emissions and fire emissions as an additional human-induced emission source with high impact on the air quality of China. We compare our satellite-derived findings to the MEIC bottom-up anthropogenic emission inventory combined with biogenic emissions from the REAS inventory.

The radiances measured at top of atmosphere are the light back-scattered from surface and atmosphere, which is of the same order over bright surface such as desert and urban, resulting in the failure of current retrieval algorithm over land. The images from orbit sensor with high spatial resolution provide us very fine quality picture of the cloud and surface, it is very clear to discriminate cloud and its shadows projected on the surface from cloud-free pixels. Over non-transparent cloud’s shadow area, it is assumed that only diffused light are reflected, while over a nearby bright surface, both the solar direct beam and the diffused lights are reflected back to the atmosphere, and then transmitted to the space detector. The contrast of pixels over the shadow and nearby bright surface could be used to derive the aerosol optical properties, this method was first proposed by Duan & Lu in 1999 and simulations are given in 2002. As the release of the high quality data of Landsat 8, this method are revisited, cases studies for different aerosol loading and types, surface albedos show the feasibilities of this algorithm over surface where no dark-pixels exist.

China is suffering from severe air pollution in winter with high load of particulate matters (PM_2.5) and in summer with high ozone (O₃) concentration. A great part of PM_2.5 and O₃ are produced in the polluted atmosphere from their precursors, and nitrogen dioxides (NO_x) are important precursors of both PM_2.5 and O₃. The delivery of the Clean Air Action Plan in 2013 has led to substantial reduction of sulfur dioxide (SO₂) emission and remarkable decrease of PM_2.5 concentrations. NO_x emission control is not as effective as that of SO₂, and nitrate (produced from NO_x) is playing an increasingly important role in PM_2.5 during haze. At the same time, surface O₃ concentrations over China has increased through the years. This study is aimed at constraining the emission of NOx from China in July 2019, and exploring the influence on surface ozone simulation.

Sulfur dioxide (SO2) from volcanic eruption has a significant impact on global climate and aviation. Satellite remote sensing technology provides an unprecedented advantage for continuous, large spatial and short-revisit monitoring for atmospheric SO2. GF-5 Environmental Trace Gas Monitoring Instrument (EMI) with high spatial resolution of 7.5 x 13 km is the China’s first instrument of hyper spectral measurements with wavelength range from 240 nm to 710 nm, and makes daily global observations of key atmospheric constituents, including ozone, nitrogen dioxide, sulfur dioxide. In this study, GF-5/EMI SO2 columns were retrieved by using GF-5 EMI UV-2 band observations and DOAS algorithm, and then the retrieved GF-5/EMI SO2 columns were compared with similar AURA/OMI and S5P/TROPOMI SO2 columns to evaluate the ability of GF-5/EMI on global SO2 monitoring. Results show that, GF-5/EMI can obtain the daily distribution of SO2 from volcanic eruption. The accuracy of GF-5/EMI SO2 columns can meet the needs of application of global volcano monitoring.

We have evaluated TanSat carbon dioxide total column measurements against the ground-based measurements from the Total Carbon Column Observing Network (TCCON) instruments. In the evaluation we determined station-specific biases and precisions. Also the trends and seasonal cycle amplitudes were compared to see how well the longer-term changes can be followed by the TanSat measurements.
High latitude measurements were of special interest in our studies. We determined if the measurements were made above a snow-covered surface using NSIDC's IMS Daily Northern Hemisphere Snow and Ice Analysis data. The availability of high latitude measurements over snow and in general were calculated. The retrieval errors for snow and no-snow cases were compared to assess the errors related to snow-covered scenes.
We utilized the AirCore measurements performed at the Arctic Space Centre of Finnish Meteorological Institute to assess the prior profiles used in the TanSat carbon dioxide retrieval. The AirCore profiles can extend up to 30 kilometers altitude with superior vertical resolution in comparison to remote sensing methods.
We found that the TanSat carbon dioxide total columns agree well with the TCCON measurements. The biases were between -2.9 - 1.6 ppm with precision (1-sigma) varying from 0.5 to 2.5 ppm, depending on the station. There were vast amount of TanSat measurements made over snow surfaces but the average retrieval uncertainty over the snow cover was 3.11 ppm when it was only 1.37 ppm over snow-free land. In general, the prior profiles agreed well with AirCore measurements with some deviations in the upper atmosphere during spring.

Since the successful launch of ESA’s Earth Explorer mission Aeolus in August 2018, atmospheric wind profiles from the ground to the lower stratosphere are being acquired on a global scale, deploying the first-ever satellite-borne wind lidar system ALADIN (Atmospheric LAser Doppler INstrument). ALADIN provides one component of the wind vector along the instrument’s line-of-sight (LOS) with a vertical resolution of 0.25 km to 2 km depending on altitude. The wind accuracy is better than 1 m/s, while the random error ranges from 3 to 6 m/s. The near-real-time wind observations contribute to improving the accuracy of numerical weather prediction and advance the understanding of tropical dynamics and processes relevant to climate variability.

Already several years before the launch of the Earth Explorer mission, an airborne prototype of the Aeolus payload – the ALADIN Airborne Demonstrator (A2D) – was developed at the DLR (German Aerospace Center). Like the direct detection Doppler wind lidar on-board Aeolus, the A2D is composed of a frequency-stabilized ultra-violet laser, a Cassegrain telescope and a dual-channel receiver to measure LOS wind speeds by analyzing both molecular and particulate backscatter signals. Thanks to the complementary design of the A2D receiver, broad vertical and horizontal coverage across the troposphere is achieved. In addition to the A2D, DLR’s research aircraft carries a well-established coherent Doppler wind lidar (2-µm DWL). It is equipped with a double-wedge scanner which allows for the determination of the wind vector with accuracy of better than 0.1 m/s and precision of better than 1 m/s. Hence, both wind lidars represent key instruments for the calibration/validation activities during the Aeolus mission.

After the launch of Aeolus, the A2D and 2-µm DWL were deployed during three airborne validation campaigns between November 2018 and September 2019. 20 coordinated flights along the satellite swath were conducted in Central Europe and the North Atlantic region, yielding a large amount of wind data from the troposphere under various atmospheric conditions in terms of cloud cover and dynamics. The high accuracy of the 2-µm DWL allowed to precisely assess the Aeolus systematic and random errors, and thus enabled a comprehensive evaluation of the satellite’s wind data product quality. Due to the high degree of commonality of the A2D with the satellite instrument, the comparative wind results delivered valuable information on potential error sources as well as on the optimization of the Aeolus wind retrieval and related quality-control algorithms. Beyond the airborne campaigns, the A2D has been serving as testbed to explore new measurement strategies and algorithm modifications which cannot be readily implemented in the Aeolus operation modes and processors, respectively.

The poster presentation will provide an overview of DLR’s airborne validation campaigns and show selected highlights both from an instrument and a meteorological point of view.

Humans have transformed many open waters into agricultural/urban land or aquaculture to meet the growing demand for food and living space of the growing population in recent decades. These changes have been putting cumulative impacts on territorial hydrological cycle, the environment and ecology. Though the information on the distribution and changes of water cover types is essential to understand those changes, its timely and accurate acquisition and updating is difficult to be accomplished, mainly due to the lack of effective and practical classification method at various scales.

From water cover perspective, we firstly suggest a comprehensive water cover classification system in terms of water sources, water body shapes and water usage, including rivers, reservoirs, lakes, agricultural ponds, seasonal wetlands and rice fields. Based on this classification system, an automatic classification method based on shape features and flooding frequency that can be carried out on a large scale was developed correspondingly. This method is also based on the time series Sentinel-1 and Sentinel-2 images. Using this method, we develop China Water Cover Map (CWCM) with an unprecedented spatial resolution of 10 meters in the year 2020 based on the Google Earth Engine (GEE) . It has an overall accuracy of 0.86 and kappa coefficient of 0.83. The CWCM is the first effort on water cover mapping for whole China at 10-m spatial resolution, which can provide complete information on water cover types. The CWCM indicates that China’s water cover area is estimated to be 334,550 , of which 47.5% is accounted by artificial water covers (rice fields and agricultural ponds). From the perspective of the geographical distribution of water cover, China’s water cover is mainly distributed in Tibet, Heilongjiang, Xinjiang and Qinghai provinces. The shape-based and inundation frequency-based automatic method highlights the potentials of fast water cover type mapping at various scales using the existing water datasets.

General. The albedo of snow and ice is characterized by a large spatial and temporal variability and determines to a large extent the energy and mass balance of the snowpack and glaciers. Spatial variability in snow and ice accumulation and melt leads to mass flow. The dynamic of the atmospheric boundary layer is determined by the energy balance, particularly net radiation, at the land-atmosphere interface. Accurate retrieval of snow and ice albedo using radiometric measurements by space-borne imaging radiometers requires addressing multiple radiative interactions in the land-atmosphere system. Retrieval of atmospheric properties over land requires accurate knowledge of the background reflectance, thus leading to simultaneous retrieval of land surface reflectance and of atmospheric constituents, particularly aerosols. Multiple effects of the complex terrain characteristic of High Mountain Asia need to be taken into account for accurate retrieval of land surface albedo, particularly of the snowpack and glaciers. First, irradiance at a location is largely affected by radiance reflected by the surrounding terrain. Second, the spatial heterogeneity of the latter needs also to be taken into account. Third, the anisotropy of land surface reflectance, particularly of snow and ice has to be characterized and applied both in the estimation of terrain effects and in the retrieval of the target reflectance.

All these aspects are being addressed in an integrated effort to improve the accuracy of albedo retrievals in High Mountain Asia and to understand how and to what extent albedo drives atmospheric boundary layer and land surface processes.

Aerosols + irradiance. Two advanced algorithms to retrieve land surface albedo and aerosol amount and properties were integrated to improve iteratively two strongly related aspects for simultaneous retrievals: a) the characterization of the land surface background to separate the surface and atmospheric signals in aerosol retrievals and b) the accurate separation and estimation of direct and diffuse irradiance in the retrieval of land surface albedo. The correlation of snow and ice albedo with aerosol loading, due to the deposition of fine particles on the surface, was further documented by comparing the evolution of Aerosol Optical Depth (AOD), measured at the ITP Nam Co observatory with glacier albedo in the Western Nyainqentanglha Mountains (WNM) during 2009–2018. It appears that initially AOD was rather low and albedo rather high, both relatively stable. After 2013 a rapid increase in AOD and rapid decrease in albedo was observed.

Terrain contribution to irradiance. Terrain has a two-fold impact on reflectance retrieved from radiometric measurements by a space-borne imaging radiometer: 1) multiple scattering of down-welling radiance on the surrounding terrain modifies the irradiance onto a specific facet; 2) the exposure of the observed facet modifies the inherent anisotropy of the surface. Both effects were evaluated and taken into account in an advanced retrieval algorithm. The evaluation of these radiation-terrain interactions cannot be separated from the accurate description of radiative transfer in the atmosphere. This was documented by analyzing the difference between the reflectance at l = 0.64 mm corrected for terrain effects using either the MODIS data products on atmospheric conditions or the hourly ERA5 and the CALIPSO aerosol extinction coefficients

Anisotropy correction of snow and ice albedo. We calibrated and evaluated four anisotropy correction models for glacier snow and ice, applicable to visible, near-infrared and shortwave-infrared wavelengths using airborne datasets on Bidirectional Reflectance Distribution Function (BRDF). We then tested the ability of the best-performing anisotropy correction model to retrieve albedo from Landsat 5/Thematic Mapper (L5/TM), Landsat 8/Operational Land Imager (L8/OLI) and Moderate Resolution Imaging Spectro-radiometer (MODIS) imagery, and evaluated these results with field measurements collected on eight glaciers around the world.

Parameterization of albedo and PBL processes. We evaluated whether an advanced hydro-meteorological model, i.e. WRF, correctly accounts for the variability in snow and ice albedo and for the response of the latter to atmospheric forcing. In-situ and satellite observations on snow and ice albedo were combined to develop an improved parameterization of ice and snow albedo in relation with snow age and depth. Numerical experiments applying alternate parameterizations and atmospheric forcing were evaluated against in-situ observations of snow water equivalent. The experiments demonstrated the sensitivity of atmospheric processes to ice and snow albedo in the Qinghai-Tibet Plateau and the need to apply realistic parameterizations taking into account ice and snow properties.

Glacier mass balance and flow. The potential of the high spatial resolution ZiYuan-3 (ZY-3) Three-Line-Array (TLA) stereo images to retrieve glacier mass balance has not been sufficiently explored. We optimized the procedure to extract a Digital Elevation Model (DEM) from ZY-3 TLA stereo images and estimated the geodetic mass balance of representative glaciers in two areas of the Nyainqentanglha Mountains (NM) using ZY-3 DEMs and the C-band Shuttle Radar Topography Mission (SRTM) DEM in three periods, i.e., 2000–2013, 2013–2017 and 2000–2017.

We investigated the glacier surface velocity in the Parlung Zangbo Basin (PZB) by applying the normalized image cross-correlation method to Sentinel 2 (S2) MSI and Landsat-8 (L8) OLI image data acquired from 2013 to 2020. We mapped time-averaged glacier surface velocity and examined four typical glaciers (Yanong, Parlung No.4, Xueyougu and Azha) in the PZB. We explored the driving factors of surface velocity and of its spatial and temporal variability.

Net shortwave radiation is an important forcing of glacier energy balance, contributes more than half of the net radiation and is directly determined by glacier surface albedo. Models of So far, in the glacier-energy balance usually estimate albedo by mapping the snow and free ice components. Snow albedo is estimated by parameterizing snow decay with age, while ice albedo is usually set constant when the snow has melt. This albedo parameterization is likely to be inaccurate, especially where a mixture of snow, ice and debris occurs. Previous studies have shown that glacier surface albedo can be accurately retrieved from remote sensing data, especially the Sentinel-2/MultiSpectral Instrument (S2/MSI) radiometric data of high spatial resolution (10 or 20 m) and high radiometric performance. In this study, we used albedo retrieved from S2 / MSI data to evaluate albedo estimates by a glacier energy balance model and then to improve the albedo parameterization and the model performance. A distributed model named Tethys-Chloris (T&C) was applied to test this approach in the Parlung No.4 glacier in the southeast Tibet Plateau. The results show that T&C performs better when ingesting S2 / MSI albedo retrievals, especially in the spatial patterns of glacier albedo, energy and mass balance.

Glaciers are key components of the mountain water towers of Asia and are vital for downstream domestic, agricultural and industrial uses. The highest rate of glacier mass loss in this region occurs in the Southeastern Tibetan Plateau, where it has accelerated in the past four decades. This has been attributed to a warming climate, but the influence of decadal changes in precipitation and its interplay with recent warming remains unclear. Furthermore, Tibetan glaciers’ changing role in catchment hydrology remains largely unknown. Here, we reconstruct catchment runoff and glacier mass changes since 1975 at Parlung No. 4 glacier, using a fully distributed glacio-hydrological model (TOKPAPI-ETH), to shed light on the drivers of mass losses for the monsoonal, spring-accumulation glaciers of this region. Parlung No.4 Glacier is considered as a benchmark glacier in this region, since its meteorology, surface energy fluxes and mass-balance have been examined since 2006. It is a maritime glacier with a spring (April-May) accumulation regime , which is followed by a period of simultaneous accumulation and ablation during the Indian Summer Monsoon (typically June-September). We force the model with climate reanalysis (ERA5-Land, CMFD) and Zayu’s national meteorological station data bias-corrected to a local weather station. TOPKAPI-ETH is calibrated and validated with automatic weather station data, discharge measurements, geodetic mass balance, stake measurements and snow cover data from MODIS. Our modelling shows that increases in annual mean temperatures (0.23°C dec-1) in the region have simultaneously increased melt and decreased accumulation. An increase of total annual precipitation in the late 1990s initially counteracted the detrimental effect of this warming, though was eventually overcome and dominated by the changing phase of monsoon precipitation, reducing monsoon accumulation, decreasing surface albedo and further intensifying ice-melt and catchment runoff since the start of the 21st Century. Our findings also highlight an augmented spring precipitation during the last two decades which was increasingly important in buffering glacier mass loss by simultaneously providing mass to the glacier and protecting it from melting in the early monsoon period. While spring accumulation was relatively insensitive to recent warming during our study period, its potential to halt more rapid mass losses depends on the complex interplay of climate warming and monsoon dynamics in the near future.

Monitoring glacier surface velocity is important to understand the response of mountain glaciers to environmental forcing under climate change. The variability of surface velocity in the temperate glaciers of the Parlung Zangbo Basin (PZB) has attracted wide attention. This study investigated in detail the spatial pattern and the temporal variability of surface velocity of glaciers in the PZB. Thanks to the Sentinel-2 MSI instrument, observations of the same region every 5 to10 days are feasible. In this study, we monitored the glacier surface velocity by using the normalized cross-correlation method. We used the Sentinel-2A/B data form October 2016 to December 2019 and Landsat-8 data from June, 2013 to June 2020 to generate velocity maps. Based on the satellite images acquired from 2013 to 2020, we present a map of the annual and of time-averaged glacier surface velocity and examined four typical glaciers (Yanong, Parlung No.4, Xueyougu, and Azha) in the PZB. The results show that the glacier centerline velocity increased slightly from 2017 to 2020. Results of the L8 / OLI and S2 / MSI datasets were compared to evaluate their accuracy, and assess the reliability of cross correlation for ice flow monitoring. The analysis of meteorological data at two weather stations on the outskirts of the glacier area provided some indications of increased precipitation during winter-spring. Besides, there was a clear winter-spring glacier velocity speedup of 40% in the upper glacier region, while a summer speedup occurred at the glacier tongue. The seasonal and interannual variability of surface velocity was captured by the transverse velocity profiles in the four selected glaciers. The observed spatial pattern and seasonal variability in glacier surface velocity suggests that the winter-spring snow might be a driver of glacier flow in the central and upper portions of glaciers. The findings on glacier velocity suggest that the transfer of winter-spring accumulated ice triggered by mass conservation seems to be the main driver of changes in glacier velocity. The interannual and seasonal glacier velocity result would contribute to modelling glacier dynamics in the future.

Ice cliffs play a key role in the mass balance of debris-covered glaciers, dramatically enhancing melt, but their importance has only been assessed for select glaciers due to a lack of datasets on their distribution and evolution. This is partly attributable to the lack of robust automatic methods that can replace the subjective and time-consuming manual delineation approaches used to date. Existing methods have used digital elevation models or multispectral images, but the varying slope and mixed spectral signal of these dynamic features makes the transferability of these approaches particularly challenging. We developed two automated and objective new approaches, based on the Spectral Curvature (SC) and Linear Spectral Unmixing (LSU) of multispectral images, to map these features at glacier to regional scales. These methods are i) robust and transferable to sites with varying lithologic, glaciological and climatic settings, and encompassing different periods of the melt season; and ii) applicable to atmospherically corrected Pléiades (2 m resolution) and Sentinel-2 (10 m resolution) images. We find that the SC method works best for the high spatial resolution, four band Pléaides images, while a modification of the LSU using the scaling factor of the unmixing method is best for the Sentinel-2 products with their coarser spatial resolution but greater number of bands. Comparison of the results with other existing methods shows that our approaches are more accurate and robust than any other existing automated or semi-automated approaches. We then use the LSU-based method to measure the distribution of ice cliffs across selected areas of High Mountain Asia and assess the main drivers of ice cliff density using a statistical analysis. This approach also outlines a high number of small, sometimes shallow-sloping and thinly debris-covered ice patches that differ from our traditional understanding of ice cliffs as isolated, prominent features, and which may have a non-negligible impact on the mass balance of debris-covered glaciers. Overall, our results give insights into the regional distribution of ice cliffs across High Mountain Asia, provide a database for the assessment of their role on the mass balance of glaciers regionally, and pave the way for representation of these features in debris-covered glacier melt models.