The ITMIX results showed that averaging the results of an ensemble of independent models is strongly beneficial for increasing the accuracy of glacier ice thickness estimates. The goal of G2TI is to provide the first ensemble-based ice thickness estimate for all glaciers included in the Randolph Glacier Inventory (RGI). The resulting data set is anticipated to become an integral part of RGI starting from RGI7.0, and thus become the standard reference for glacier ice thickness information across the globe.
G2TI shall generate the largest possible number of independent ice thickness estimates for the largest possible number of RGI glaciers. The final results (i.e. spatially distributed ice thickness estimates for individual glaciers) will be provided to the community as a gridded RGI dataset. The results shall be updated with every new RGI release.
For every glacier, the RGI dataset shall include (a) a "best guess", i.e. the ice thickness distribution as obtained by a pixel-by-pixel average of any model result, and (b) an "uncertainty", i.e. a pixel-by-pixel estimate for the uncertainty of the best guess. The number of models that considered a particular glacier will be provided as a "normal" (non-gridded) RGI attribute. It is envisioned to have the "best guess" computed from a weighted average, where the weights will reflect the average model performance. The "uncertainty" will take into account both the spread between and the accuracy of individual models.
Model accuracy will be evaluated against measured ice thickness data. All data included in GlaThiDa v2.0 will be used for that. The accuracy will be estimated in a cross-validation experiment in which 1/3 of the measurements will be used for calibration, and 2/3 for validation. This procedure will be repeated three times, i.e. "cycling" through the three possible 1/3-combinations. The data to be used in each of the steps was selected at random but ensuring that (a) data for individual glaciers always "stay together" (a glacier will be used for calibration or not, but not have only half of the available measurements used, for example), and (b) the selected validation data maintain coverage of different RGI regions.
For participating to G2TI, do the following:
3.0) Make sure to have a model capable of estimating the ice thickness distribution for a large set of glaciers (say >10^3) if only a glacier mask/outline and a digital elevation model of the surface are given.
3.1) Decide for which of the RGI region you are going to run the model. We encourage you to tackle as many regions as possible, ideally all of them.
3.2) Retrieve the necessary data from the data repository (if you don't have the corresponding link, request it here). Point "5)" below provides information on the repository's content.
3.3) Run your estimation model THREE times. Only consider glaciers for which ice thickness measurements are available (the "INDEX_geometry"-file described at point 4.2 tells which ones those are). For these glaciers, produce three estimates of the ice thickness distribution. For every of the three estimates, only 1/3 of the ice thickness data shall be used for calibration. The "INDEX_geometry"-file defines which data shall be used for the 1st, 2nd, and 3th estimate.
3.4) Run your estimation model a fourth time. In this run, apply your model to the largest possible number of glaciers and use all available ice thickness data for calibration.
3.5) Upload your results to the link provided by the organizers (if you don't have it, request it here). Please provide the data respecting the following instructions:
(a) organize all data in a repository called "results_yourname" (e.g. "results_farinotti")
(b) the repository shall contain four subfolders called "cali1", "cali2", "cali3", and "final".
(c) the "cali*"-subfolders shall directly contain the individual, gridded estimates. Please call the individual files "thickness_RGI60-RR.NNNNN_cX.tif", where "RR" is the RGI region ID, "NNNNN" is the RGI glacier number, and "X" is the calibration subset (e.g. "thickness_RGI60-03.00108_c1.tif").
(d) the "final"-subfolder shall contain one subfolder for every region (as the "./data/geometry/" repository; see point 4.3). Individual results shall be named "thickness_RGI60-RR.NNNNN.tif" (e.g. "thickness_RGI60-03.00108.tif").
(e) Once the results are uploaded, inform the Working Group chairs.
IMPORTANT: Make sure that the results have the same format, coordinate system, projection, and resolution as the originally provided DEMs (cf. point 4.3a here below).
The data repository contains the following (again, if you don't have access to that but are committed to participate, request an access link here):
An index file (plain ASCII text) for the data contained in the folder "data/geometry". The file lists every glacier ID for RGI6.0 and provides basic information including glacier area, approximate coordinate, UTM zone, and cell size of the provided DEMs. Refer to the file header for details.
An index file (plain ASCII text) for the data contained in the folder "data/thickness". The file lists all glacier IDs for RGI6.0-glaciers that have measured ice thickness. The latter may either be point data (provided as separate files; see below) or information about the average thickness (provided in the corresponding column). The file additionally defines which glaciers should be used for calibration, and which for validation, in the different calibration runs. A total of three different calibration-subsets is defined. For details, refer to the file header.
A repository containing all geometry files required for the computations. The data are organized in individual sub-folders named with the RGI glacier ID (e.g. "RGI60-01.10689"). Individual glaciers are grouped into folders corresponding to the RGI first-order regions (e.g. "RGI60-01").
For every glacier, the following is provided:
(a) gridded DEM of the glacier surface and its surroundings (file "dem.tif")
(b) gridded mask ("mask.tif") indicating whether a given pixel of the DEM is glacierized (1) or not (0)
(c) the glacier outlines of the glacier ("outlines.*", where "*" is "dbf", "prj", "shp", and "shx").
i) For some few cases, no DEM or mask could be provided. These cases are flagged in the "INDEX_geometry.txt"-file (point 4.1).
ii) For RGI region "05" (Greenland Periphery), glaciers with "connectivity level 2" (strong connection to the ice sheet) are excluded. This is to avoid creating estimates conflicting with "BedMachine v3" (Greenland's reference dataset) with methods that were mainly developed for mountain glaciers.
A repository containing all available point thickness measurements. The repository contains one file per glacier (plain ASCII text; only glaciers with point data are included). The files are column-organized and contain the following columns
ptID: ID of the point thickness measurement
xUTM: point easting as UTM coordinate (m)
yUTM: point northing as UTM coordinate (m)
UTMz: UTM zone of the given point. Negative values refer to the southern hemisphere.
surf: surface elevation (m)
bed: bedrock elevation (m)
thick: ice thickness (m)
Note that the three latter entries are related through "thick=surf-bed". "Surf" was extracted from the DEM of the corresponding glacier (see point 4.3). File nomenclature is "RGI60-xxxx_point_thickness.txt"
The "GLIMS Technical Report" for RGI6.0. This file is for reference only. The file's Section 2 defines the RGI regions.
The main goal is to provide results in time for the IPCC Special Report on Ocean and Cryosphere in a Changing Climate (IPCC SROCC). The paper-submission cut-off for SROCC is October 2018. The following timeline is thus envisaged:
Dec. 2017: Launch of G2TI
Apr. 2017: Results from individual models ready; discussion in a splinter meeting at EGU 2018 (Vienna, 08-13 Apr 2018). Strategy for "meta-analysis" and paper-outline defined.
Jun. 2018: Consolidated results ready; meta-analysis performed. Discussion in a side meeting at POLAR 2018 (Davos, 19-23 June 2018).
Aug. 2018: Paper draft ready and circulated amongst authors.
Sep. 2018: Paper submitted.
Jul. 2019: Final results presented at IUGG 2019 (Montreal, 08-19 July 2019).
If you have any additional question related to G2TI, please contact us here.