Conference Agenda

Part 1, Glacier velocity estimation based on Sentinel-2 observations at western High-Mountains Asia.

The Sentinel-2A/B Twin satellites provide 5-day repeat observation to the Earth and capable of deriving glacier velocity with high-temporal resolution. In this study, western High-mountains Asia, including the ‘Karakoram-Pamir anomaly’ region and its surroundings, were taken as the study sites and a data processing procedure was proposed to derive quasi-monthly glacier flow velocity fields. Each acquisition is performed offset-tracking to its next three almost cloud-free acquisitions to increase number of redundant observations. The detector mosaicking errors are eliminated if offset-tracking is performed between two different Sentinel-2 satellites. Flow speed and direction referenced method is taken to remove the wrong matching of offset-tracking. Then an iterative SVD method solves the glacier velocity and removes the observation with large residual. According to the glacier flow velocity time series between Oct 2017 and Sep 2021, it captures plenty of surged glaciers start and/or end their surging phases across this region. Two types of surging glaciers are identified according to the shape of their high temporal resolution flow rates time series. The first types’ surging phase last for only a few years, and shows no seasonal variation. Rimo’s southern tributary is an example of this type, it experienced a full surging phase during our study period and last for about two years, the maximum speed exceeded 10 m/day within the study period. The second type behaves similarly to normal glaciers, albeit with a glacier front that advances and exhibits much higher summer speeds than during stagnation, such as Gando at Pamir. Normal glaciers exhibit annual speed-ups and slowdowns, with acceleration typically beginning in late April or early May and ending before September.

Part 2, Greenland ice sheet melting and re-freezing status monitoring with dual-polarized Sentinel-1 imagery

First this study introduced a method of incidence angle normalization to the backscatter coefficient and polarimetric features of dual-polarized Sentinel-1 images. Additionally, polarization decomposition of SLC data was performed to obtain Alpha and Entropy images. A multiple linear regression model is trained for incidence angle correction for backscatter coefficient and polarimetric features. The precision evaluation of the ascending and descending images suggests better normalization results than the widely-used cosine-square correction method for HH images and little improvement for the HV images.

To assess glacier melting status, MODIS-LST (Land Surface Temperature) and AWS (Automatic Weather Station) data were utilized as ground truth, categorized into different Radar Glacier Zones (RGZ) describing diverse physical properties of glacier surfaces. An ensemble decision tree model is trained using the distinct feature values of both SAR data types to determine glacier melt status. The accuracy of glacier melting detection using backscatter features can reach 80%, while the accuracy using polarization decomposition features is 74%. Despite the lower accuracy compared to the former, polarization decomposition demonstrated greater sensitivity in areas with poor discrimination based on backscatter, particularly in bare ice zones. Combining the features of both backscatter and polarization decomposition achieved an accuracy of 83% and kappa coefficient of 0.61. The spatial distribution of melt detection results closely aligned with recent anomalous melting events. Future research should focus on deepening the understanding and application of various physical parameters, optimizing models to enhance glacier melt detection accuracy and maximizing the utilization of SAR data.

Part 3, Glacier velocity estimation based on both Sentinel-1 and -2 observation at Greenland Ice Sheet

Here we develop different strategies for deriving glacier flow rates for Greenland Ice Sheet. First, for the regions glacier flows at relatively low and or medium rates. A special DD-InSAR method is developed based on Sentinel-1 6 or 12 days interferograms and slant range offset-tracking. The maximum capacity of detecting deformation is ~3.6m for 6-day interferogram than conventional D-InSAR of ~1.4m. The phase jump between adjacent burst is eliminated based on azimuth offset-tracking and azimuth coregistration error estimation. We also developed algorithm to inverse 3D velocity from one direction InSAR observations according to the presumption that ice flow directions are steady. Regions with both ascending and descending Sentinel-1 observation are employed for cross validation to evaluate the precision of 3D velocity inversion. The result show that such method can be applied for most glaciers that distributed along the western and eastern coastal line except for several glaciers with the north-south direction such as Petermann and N79.

Second, for the regions glacier flows at relatively high speed, where coherence of D-InSAR is too low, we combined the offset-tracking results from Sentinel-1 and -2. A two-step least square method is developed for solving the flow rates time series and combine them. This method can provide a full time series of glacier velocity fields. The errors of Sentinel-1 images offset-tracking are ~0.4 m while ~2.5 m for Sentinel-2. We made 6-day glacier flow velocity products for several largest outlet glaciers for Greenland Ice Sheet, including Petermann, N79 (Nioghalvfjerdsfjorden, Zachariae Isstrom, and Storstrommen), Jakobshavn Isbrae, and Kangerlussuaq. The results are more spatial and temporal continuous than most published results such as PROMICE and NSIDC-0766.