GDM-OPT-ICE: Ground Deformation Monitoring with OPtical image Time series for ICE/glacier analysis¶

Select input data¶

The Geobrowser offers multiple ways to search a large variety of EO-based dataset and the user should refer to the Geobrowser section for a general introduction. For this tutorial we will use a data package which is accessible through the “Data Packages” tab on the upper left of the screen. If you type “Ridgecrest” into the search box you should be able to find a data package named “European_Alps_S2_im”. Alternatively you can access it directly by clicking on the link: https://geohazards-tep.eu/t2api/share?url=https%3A%2F%2Fgeohazards-tep.eu%2Ft2api%2Fdata%2Fpackage%2Fsearch%3Fid%3DEuropean_Alps_S2_im

Please refer to the tutorial of the GDM-OPT-ETQ to learn more on how to manipulate the data on GEP.

Set the processing parameters¶

There are 37 processing parameters that can be adjusted. A short explanation of the parameter is provided when hovering over the parameter fields.

• DEM: Defines the Digital Elevation Model used for filtering the displacement fields. The Merit [8] and the COP-DEM_GLO-30 [9] are available to GEP users. The default DEM is the Merit DEM.
• Sentinel-2 band: Defines the Sentinel-2 band for matching. The option B04 is recommended since the red band is also used for band to band co-registration by the ESA Sentinel-2 production center.

• Matching parameters: These parameters control the network of pairs that will be created. This is of main importance to obtain a measure of the ground deformation. The user should keep in mind that the method is sensitive to 1/10 of pixel. In the case of Sentinel-2, the method is sensitive to displacement of ~1 meter, if there is less than 1 meter of displacement between two dates, the ground motion will not be measured.

  • Matching mode: Defines the unit of the next parameters, it can be “acquisition” or “days”.
  • Minimum matching range: Defines the minimum matching range for creating the image pairs. The matching range is expressed in acquisitions so if a minimum range is set to 1, all the images (N) will be paired with at least the next image in time (N+1). The default value is set to 1.
  • Maximum matching range: Defines the maximum matching range for creating the image pairs. The matching range is expressed in acquisitions so if a maximum range is set to 2, all the images (N) will be paired with at most the next second image in time (N+2). The default value is set to 5.
  • Split date: Is an optional parameter of the form “yyyy-MM-dd” which will split the time series into two subsets. Pairs will only be formed among members of different subsets. This is particularly interesting in the case of quantifying co-seismic displacement. The default value is left empty.
  • Matching direction: Define the time direction for the matching. If Forward is selected, the pairs are only created in the time direction. If Forward+Backward is selected, the pairs will be created in both directions (i.e. time and reverse time direction). The default value is set to Forward.

• Sentinel-2 relative orbit: Defines the relative orbit to filter the acquisitions. Biases exist between acquisitions of different relative orbits, this option allow to ensure only one relative orbit is considered.
• Region Of Interest bounding box: Defines the area the Sentinel-2 acquisitions are cropped to compute the correlation.

• Image Matching parameters: Two different algorithms are proposed for this step: MicMac developped by IGN/ENS and GeFolki developped by ONERA. Micmac is based on the correlation of two images in the spatial domain while GeFolki is optical flow algorithm.

  • MicMac Parameters:

    • Window size: Controls the size of the template used for matching. It controls the neighborhood around the central pixel. The minimum value is 1 (3x3 pixels) and the maximum value is 7 (15x15 pixels). The default value is 3 (7x7 pixels). A smaller window size allow better reconstructing small scale variations but can lead to more noise. Vice versa, larger window sizes lead to greater robustness against noise but smooth small scale details. For large scale motion such as co-seismic slip, we recommend to use large window sizes.
    • Decorrelation threshold: Discards the matches with a correlation coefficient below a value expressed in the range [0,1]. The default value is 0.2.
    • Spatial matching range: Defines the search range in pixel for finding matches based on the template. The actual search range is computed from this parameter as round(Spatial matching range/0.8)+2. The parameter has to be adjusted according to the maximum expected displacement taking into account possible coregistration biases of the input images.
    • Regularization parameter: Similar to the window size, controls the smoothness of the expected motion field. Increasing the regularization parameter puts greater emphasis on a smooth motion field where neighboring pixels will have similar displacement values. For large scale features such as co-seismic displacement, large value lead to smoother and less noisy results. The default value is 0.3.
    • Sub-pixel matching: If true, the sub-pixel matching will be computed at different resolution up to sub-pixel level. This multi-resolution approach consists in starting the computation at a coarse resolution and improving the resolution at each matching level. If false, the matching is computed only at the image resolution and no refinement is performed at sub-pixel level.

  • GeFolki parameters:

    • Radius: Define the the size of the window (2R + 1) x (2R + 1) on which the matching between the two images is maximized. The choice of the radius value is a compromise between robustness and the expected level of detail. A large radius makes the algorithm more robust. If the flow is rapidly changing on the image, the radius must be chosen small enough to estimate these variations. The algorithm can be used for several radius sizes in an iterative manner. The radius sequence can be chosen among one or several of the following values: 4, 8, 16, 32, 64, 128.
    • Levels: Define the number of levels (L) in the scale pyramid. The parameter value is conditioned by the maximum size of the displacement Wmax. By default, L=1 so the displacement is assumed to be lower than 2 pixels.
    • Iteration: Defines the number of iterations to reach a minimum.
    • Rank: Define the spatial window of the rank filter. The parameter controls the smoothness of calculated displacement field by averaging the displacement values within the window size. Default value is r=4 (9x9 pixel).

• Masks:

  • Buffer outside the glacier extent: Defines a buffer area around the glacier mask of the GLIMS database _[7]. The unit is meter. By default, a distance of 1000 m is taken around the glacier outline.
  • Glacier mask: If set to True, the correlation is computed only on the pixels located inside the glacier outlines. The glicier oultines are taken from the GLIMS database _[7].
  • Snow mask: If set to True, the areas of the images covered by snow are masked. The default value is set to True.
  • Cloud mask: If set to True, the areas of the images covered by clouds are masked. The default value is set to True.
  • Slope mask range minimum: The pixels located on terrain slopes with a slope angle larger than the value set with the parameter are filtered out in the products. By default, the parameter is set to 80, so pixels located on slopes with angle larger than 80 degrees are filtered.
  • Slope mask range maximum: The pixels located on terrain slopes with a slope angle smaller than the value set with the parameter are filtered out in the products. By default, the parameter is set to 90 degrees, so pixels located on slopes with angle between Slope mask range minimum and 90 degrees are filtered.
  • Topographic shadow: If set to True, the sun illumination is simulated using the position of the sun and the selected DEM. The area in the shadow are then mask out of the acquisitions before computing the correlation.

• Correction and filtering of the displacement fields

  • Apply correction and filtering: If set to True, the geometric corrections (as described in [4] ) and the filtering (as described in [3]) are applied. They are highly recommended for any use case and are applied by default. The user can activate or deactivate each correction.
  • Correction: deramping If set to True, the first geometric correction (as described in [4] ) is applied . It consists in estimating a planar function to correct the ramp commonly present in the displacement fields. It is highly recommended for any use case and is applied by default.
  • Correction: along-track destriping If set to True, the second geometric correction (as described in [4] ) is applied . It consists in estimating a linear shift within each Sentinel-2 sensor stripe to correct the shift present in each stripes of the displacement fields. It is highly recommended for any use case and is applied by default.
  • Correction: along-track destriping value The shift within each stripe can be estimated using the mean or the median of the displacement distribution. By default, the shift is estimated using the mean value.
  • Correction: across-track destriping value This corrects the jitter undulation by filtering out the short wavelength undulation by a wavelet filter [10]. This filter is directional and can affect the results by filtering out part of the signal. In the case of small object like glaciers, it is not recommended use it. By default, it is set to False.
  • Filtering displacement amplitude threshold: Displacement with a magnitude larger than this value will be filtered out in each correlation pair. The unist in in pixel. By default, the threshold is 10 px (i.e. 100 m for Sentinel-2).
  • Filtering: Displacement direction: If set to True, the displacement field is filter by analysing the direction of the displacement with respect to the direction of the slope. By default, it is set to False.
  • Maximum angle deviation for direction filtering: Defines the maximum angle between the displacement direction and the slope direction. If the this angle is larger than this value, the displacement will be removed in the East-West and North-South displacement fields. The unit is in degree and is set to 45° by default.

• Motion analysis: If set to True, the GDM-OPT-ICE service provides different outputs computed from the stack of correlation pairs.

• Time series Inversion for Optical images parameters

  • Run TIO: If set to True the TIO algorithm computes the displacement time series. By default, it is set to True.
  • Inversion weight: Defines the weight of each displacement pairs. The weight is based on the temporal baseline between the two acquisitions as defined in [5]. The user can choose to give more wait to short baseline pairs (Short-baseline) or long baseline (Long-baseline) or to set no weight (None) in the inversion. By default, it is set to None.
  • Discard pairs: If set to True, pairs can be discarded based on the percentage of masked area in the AOI. This allow to remove the pairs with very few correlated pixels.
  • Discarding threshold: The ratio between masked and non-masked pixel is computed over the AOI. If this ratio is larger than the discarding threshold, the pairs is discarded from the inversion procedure. This parameter is ranging in [0,1] and set to 0.8 by default.
  • Correlation weighting: If set to True, the inversion will take into account the correlation grids to weight the contribution of each pixel for each pair in the inversion.

Results¶

The results are also accessible on this link: https://geohazards-tep.eu/t2api/share?url=https%3A%2F%2Fgeohazards-tep.eu%2Ft2api%2Fjob%2Fwps%2Fsearch%3Fid%3D32259ea3-396c-4ba0-aa67-cc752ee8f9cb%26key%3De490dd3f-d5bf-4c5f-97e7-943f06b60c0f

• The first set of results provides the mean velocity (m/day) computed from the stack of displacement grids. Here, one can see the activity of the European Alpine glaciers with a mean velocity of up to 0.7 m/day.

• The second set of results provides the mean velocity (m/day) estimated from the linear regression of the TIO displacement time series.

Disclaimer¶

The GDM-OPT services are scientific softwares provided at the best CNRS/ForM@Ter (EOST/A2S) knowledge according to state-of-the-art image matching algorithms. No warranty is provided on the processors and results of the services. CNRS/ForM@Ter (EOST/A2S) is not responsible for any software inaccuracies, bugs, errors and misuse. Generated results have a defined accuracy according to the relevant scientific publications available in the literature. Result accuracy is estimated on a statistical basis. Provided results are not validated by CNRS/ForM@Ter and, indeed, it is user responsibility to validate them. CNRS/ForM@Ter is not responsible for the use, quality, accuracy and interpretation of results and products that are generated by using the processors and services provided within the platform. CNRS/ForM@Ter is not responsible for the use, quality, accuracy and interpretation of third party results, products and services derived from the use of the processors and services. CNRS/ForM@Ter is not responsible of possible outages of the provided services. CNRS/ForM@Ter is not responsible of any kind of third party loss derived from service outage, result inaccuracies, software errors of the provided services and products. The maintenance, update and user support are provided by EOST/A2S free of charge and at best effort. EOST/A2S is not responsible for any consequence derived from delays on replies to user requests or support inaccuracies.

• CNRS: Centre National de la Recherche Scientifique / French National Research Council
• ForM@Ter: Pôle Terre Solide / Solid Earth Centre
• EOST: Ecole et Observatoire des Sciences de la Terre / School and Observatory of Earth Sciences
• A2S: Application de Surveillance par Satellite / Application Satellite Survey