# GlaMBIE-2 submissions: context and first overview

Written 2026-10-12 as groundwork for the first GlaMBIE-2 workshop (Wednesday 2026-10-14).
Everything here was read off the files in this folder; numbers were computed directly from the CSVs.
Interpretations that go beyond what the files or methods PDFs state are marked *(inference)*.

## 1. What is in this folder

| Path | Content |
|---|---|
| `model_data_submission_guide.md` | The GlaMBIE-2 model submission guide (retrieved 2026-10-12). Defines the protocol all model runs were asked to follow. |
| `GlaMBIE_DATASETS_INFO.md` | Metadata for the GlaMBIE-1 (2022-24) dataset. Describes `glambie1_results/`, not the GlaMBIE-2 submissions. |
| `glambie2_submitted_data/` | **The GlaMBIE-2 submissions**: 151 CSVs + 9 methods PDFs (files dated 2026-09-22). |
| `glambie1_results/` | GlaMBIE-1 published dataset (DOI 10.5904/wgms-glambie-2024-07): inputs and results, plus the original zip. |
| `.ipynb_checkpoints/` | Empty. |

### 1.1 `glambie2_submitted_data/`

Nine datasets: **seven model-type datasets** (the "6 + 1 additional OGGM") and **two observation-based
"combined" datasets** that serve as the reference.

Files are named `{region}_{model|combined}_{group}.csv`, one file per region and dataset.
These are **not the raw submissions** but files already processed by the GlaMBIE team: dates are ISO
(`YYYY-MM-DD`, not the `DD/MM/YYYY` of the template), the single multi-region CSV has been split per
region, and five columns were appended (`user_group`, `start_date_fractional`, `end_date_fractional`,
`date`, `date_fractional`). `model` files carry the full model template (above/below water level,
frontal ablation); `combined` files carry the shorter observational template without those columns.

Region codes: `ala` 1, `wna` 2, `acn` 3, `acs` 4, `grl` 5, `isl` 6, `sjm` 7, `sca` 8, `rua` 9, `asn` 10,
`ceu` 11, `cau` 12, `asc` 13, `asw` 14, `ase` 15, `trp` 16, `san` 17, `nzl` 18, `ant` 19.

All files: unit `mwe`, no NaN in `glacier_change_observed`, no temporal gaps between consecutive rows.

### 1.2 `glambie1_results/`

- `glambie_results_20240716/calendar_years/`: 19 regions + `0_global.csv`, annual, 2000-2023 (24 rows),
  columns `combined_gt`, `combined_mwe` and errors, `glacier_area`.
- `glambie_results_20240716/hydrological_years/`: 19 regions, with per-data-group columns
  (altimetry, gravimetry, DEM differencing + glaciological).
- `glambie_input_20240716/`: 258 CSVs in 19 region folders, the individual GlaMBIE-1 input datasets
  (fractional-year dates, unit `m` or `Gt`).
- `GlaMBIE_Data_DOI_10.5904_wgms-glambie-2024-07.zip`: the original archive (not opened; presumably the
  same content).

## 2. The nine datasets at a glance

| Dataset (`user_group`) | Type | Regions | Period (start of first row to end of last row) | Resolution | Geometry / area | Uncertainty | Extra fields | Coverage % |
|---|---|---|---|---|---|---|---|---|
| `Fujita et al.` (GLIMB) | model | 19 | 1975-01 to 2026-04 | monthly | fixed, RGI 6.0 area | none | none | 100 everywhere |
| `Rounce_et_al` (PyGEM v1.1.0) | model | 19 | 1975-01 to 2026-04 | monthly | fixed, RGI 6.0 area | none | frontal ablation (constant per month) | 88.1 to 100 |
| `Huss_model` (GloGEM) | model | 19 | 1974-10 to 2025-10 (NH), 1974-04 to 2025-04 (SH) | monthly | **area varies** (50-51 values per region) | all rows | none | 100 everywhere |
| `OGGM_v17a_fixedgeom` | model | 19 | 1975-01 to 2026-01 | monthly | fixed, RGI 6.0 area | none | above-water-level = total | 88.1 to 100 |
| `OGGM_v17a_spinupgeom` | model | 19 | 1975-01 to 2026-01 | monthly | dynamic, area varies annually | none | above-water-level = total | 87.5 to 100 |
| `UIOcryogrid_RGI7` (CryoGrid) | model | 2 (`7 sjm`, `9 rua`) | 1991-01 to 2025-12 (`7 sjm`), to 2024-10 (`9 rua`) | monthly | fixed, RGI 7.0 area | all rows (constant 0.0325/month) | none | 100 (but see 3.6) |
| `WGMS-DUS_monthly` | "model" (downscaled observations) | 15 | varies, see 3.7 | monthly | area varies | all rows | none | 69 to 100 |
| `WGMS-DUS_combined` | combined (obs) | 19 | hydrological years 1975/76 to 2024/25 | **annual** (50 rows) | area varies | all rows | n/a | 69 to 100 |
| `ETHZ extended` | combined (obs + model) | 19 | 1974-10 to 2026-10 (NH), 1974-04 to 2026-04 (SH) | monthly | area varies | all rows | n/a | 100 everywhere |

Only Fujita and Rounce cover the requested January 1975 to March 2026 in full. OGGM stops at
December 2025; Huss_model stops at the end of hydrological year 2024/25; UiO starts in 1991.

## 3. Per-dataset notes (from the methods PDFs and the data)

### 3.1 Fujita et al. (`model_dataset_information_Fujita-et-al-.pdf`, 2 pages)

- Koji Fujita, Christina Draeger, Akiko Sakai (Nagoya). **GLIMB**: daily energy and mass balance model
  (Fujita and Ageta, 2000).
- Run on a 0.25 degree ERA5 grid (11146 cells) with glacier hypsometry per cell in 50 m bands, not
  glacier by glacier; results are then summed to individual RGI 6.0 glaciers.
- New ice albedo dataset from Landsat-8 minimum albedo over RGI 7.0 glaciers (Draeger and Fujita,
  in preparation, i.e. not yet peer reviewed).
- Calibration: a precipitation multiplier tuned to match Hugonnet et al. (2021) 2000-2019.
- Fixed geometry. No uncertainties.

### 3.2 Rounce et al. (`model_dataset_information_Rounce-et-al.pdf`, one paragraph)

- **PyGEM** as in Rounce et al. (2023). Fixed geometry.
- Frontal ablation is reported but is a constant per month (annual estimate / 12); non-zero in 8
  regions (`9 rua` 0.047, `19 ant` 0.025, `7 sjm` 0.023, `17 san` 0.012, `1 ala` 0.008, `5 grl` 0.005, `3 acn` 0.004,
  `4 acs` 0.0002 m w.e./month), zero elsewhere.
- CSV remark on every row: "frontal ablation is removed in model calibration and thus is constant
  amount each month".
- Coverage percentages are identical to OGGM fixed geometry to all printed digits (e.g. 88.078427 in
  `12 cau`, 96.5 in `10 asn`), and the reported areas in `10 asn`, `12 cau` are RGI area times that fraction.
  *(inference)* PyGEM runs on OGGM glacier directories, so the same glaciers are missing.

### 3.3 Huss_model (`model_dataset_information_Huss-model.pdf`, 7 pages)

- **The PDF is the ETHZ combined ("hybrid") method description**, prefixed with a note apparently
  written by someone other than the author: *"Method description taken from Huss ETHZ hybrid
  submission. Note: this model run uses the same underlying model but (to my understanding) is only
  constrained by the long-term trend from HUG21. As a result, both monthly and annual values are
  driven by ERA5."* So there is **no dedicated methods description** for this dataset.
- Underlying model: **GloGEM** (Huss and Hock, 2015), ERA5 temperature and precipitation,
  elevation-dependent accumulation and temperature-index melt, calibrated per glacier to
  Hugonnet et al. (2021) 20-year geodetic balances.
- Verified relationship to `ETHZ extended`: over the common period, **8 of every 12 months are
  bit-identical** between the two (408 of 612 months per region), including the uncertainty column.
  The differing months are the four summer months. This matches the ETHZ method statement that the
  model-vs-observation annual bias "is equally distributed over the summer months". So
  `Huss_model` = raw GloGEM, `ETHZ extended` = the same GloGEM series with summer months adjusted so
  annual totals match the observation-based series.
- Areas are identical to `ETHZ extended` (updated annually by volume-area scaling, regional totals
  rescaled to RGI 7.0), so this is **not a fixed-geometry run** in the sense of the guide.
- Uncertainty: per month, the standard deviation of modelled balances for that calendar month
  (described as conservative); not a formal 68/95% interval.
- Does not reproduce the Hugonnet 2000-2019 rate in several regions (see 4.1).

### 3.4 and 3.5 OGGM v17a fixed geometry and spinup geometry (one shared 5-page PDF)

- Our own two submissions. The submitted CSVs are **bit-identical** to the local
  `glambie_submission_{fixed,spinup}_geometry_v17a_SfcTypeModel.csv` (11628 rows each, max diff 0).
- The two PDFs are the same file (`methods_v17a_SfcTypeModel.pdf`).
- `glacier_change_observed_above_water_level` is filled with the same value as the total; below water
  level and frontal ablation are blank (no calving in this setup).
- Details are in the project `CLAUDE.md`.

### 3.6 UIOcryogrid_RGI7 (`model_dataset_information_UIOcryogrid-RGI7.pdf`, 3 pages)

- **CryoGrid** energy-balance model with a subsurface scheme for refreezing (Westermann et al. 2023;
  Schmidt et al. 2023, 2025; Schuler et al. 2025).
- **Not ERA5**: forced by CARRA (Copernicus Arctic Regional Reanalysis, 2.5 km, 3-hourly), available
  from 1991 only.
- **RGI 7.0**, not RGI 6.0.
- Region 9 covers **Franz Josef Land + Novaya Zemlya only** (area 34726 km2 vs ~51600 km2 for the
  whole region; Severnaya Zemlya is missing), yet coverage is reported as 100%. The authors argue the
  hypsometry of the subsample represents the whole region.
- Output is climatic mass balance (no frontal ablation).
- Uncertainty: RMSE 0.39 m w.e./yr (68%) from Schmidt et al. (2023), divided by 12, constant in time.

### 3.7 WGMS-DUS_monthly (`model_dataset_information_WGMS-DUS-monthly.pdf`, 1 page)

- **Not a climate-driven model.** It is the annual `WGMS-DUS_combined` series downscaled to months
  with the sine-wave method of Zemp and Welty (2023), using seasonal (winter/summer) glaciological
  balances from FoG 2026-02-10 to set the annual amplitude (117 glaciers in 15 regions).
- Verified: monthly values summed over each hydrological year reproduce `WGMS-DUS_combined` to
  0.003 m w.e.
- Missing regions: `4 acs`, `9 rua`, `15 ase`, `16 trp` (tropics explicitly omitted).
- Shorter records where seasonal data are lacking: `5 grl` from 2007-10, `14 asw` 2009-10 to 2022-10,
  `18 nzl` from 2004-04, `19 ant` from 2001-04, `6 isl` from 1985-10; `3 acn` ends 2024-10.
- 252 rows in 6 regions are flagged in `remarks` as having no seasonal observations (amplitude taken
  from the surrounding years).
- Monthly uncertainty = annual 1-sigma uncertainty / 12, "a relative rather than absolute measure".

### 3.8 WGMS-DUS_combined (`combined_dataset_information_WGMS-DUS-combined.pdf`, 17 pages)

- Dussaillant et al. (2025) method on FoG 2026-02-10: glaciological anomalies (reference period
  2011-2020) kriged to every RGI 6.0 glacier, calibrated to geodetic trends, aggregated by region.
- Annual, hydrological years (start 1 Oct in the north, 1 Apr in the south, 1 Jan in `16 trp`).
- 1-sigma uncertainties (median 0.34 m w.e./yr).
- Low coverage in `5 grl` (69%), `9 rua` (90%), `18 nzl` (91%).
- Some deviations from Dussaillant et al. (2025) in the choice of glaciological series (their
  Table 1), most notably in `10 asn`.
- Most of the PDF is the contributor and agency listing.

### 3.9 ETHZ extended (`combined_dataset_information_ETHZ-extended.pdf`, 5 pages)

- Matthias Huss, Lander van Tricht. Hugonnet et al. (2021) geodetic balances per glacier, plus
  glaciological anomalies (reference 2010-2020) extrapolated by inverse-distance weighting, plus
  GloGEM for the monthly distribution (van Tiel et al., 2026).
- By construction the annual totals are meant to be independent of meteorological data.
- Areas from volume-area scaling, regional totals set to RGI 7.0; balances are still RGI 6.0 based.
- **Extends to September 2026** (NH), i.e. a year beyond Huss_model and beyond the date of submission.
  The PDF says the product covers "1990-2025" and does not explain the extension or the pre-1990
  part. The months after 2025-10 are stored with a different numeric precision (5 decimals vs 4),
  which suggests they were produced differently *(inference: possibly a climatology or forecast)*.

## 4. First look at the numbers

Mean specific mass-change rates in m w.e./yr, computed as sum of `glacier_change_observed` over the
rows falling inside the window, divided by the time those rows span. Calendar-year windows; the
annual `WGMS-DUS_combined` rows are hydrological years, so its windows are shifted by a few months.
`G1` = GlaMBIE-1 calendar-year consensus from `glambie1_results/`.

### 4.1 2000-2019 (the calibration window for most models)

| Region | G1 | DUS comb. | ETHZ ext. | Fujita | Rounce | Huss | OGGM fixed | OGGM spinup | UiO | DUS monthly |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 ala | -0.726 | -0.900 | -0.798 | -0.748 | -0.725 | **-0.527** | -0.769 | -0.743 | | -0.842 |
| 2 wna | -0.526 | -0.709 | -0.554 | -0.524 | -0.466 | -0.425 | -0.520 | -0.518 | | -0.646 |
| 3 acn | -0.285 | -0.304 | -0.291 | -0.294 | -0.261 | -0.211 | -0.291 | -0.293 | | -0.309 |
| 4 acs | -0.578 | -0.543 | -0.665 | -0.653 | -0.664 | -0.630 | -0.648 | -0.651 | | |
| 5 grl | -0.382 | -0.341 | -0.410 | -0.398 | -0.382 | **-0.228** | -0.396 | -0.406 | | (partial) |
| 6 isl | -0.793 | -0.855 | -0.851 | -0.852 | -0.857 | -0.799 | -0.846 | -0.850 | | -0.871 |
| 7 sjm | -0.326 | -0.343 | -0.314 | -0.307 | **-0.111** | **-0.049** | -0.303 | -0.300 | **-0.096** | -0.327 |
| 8 sca | -0.581 | -0.635 | -0.596 | -0.568 | -0.451 | -0.465 | -0.565 | -0.556 | | -0.565 |
| 9 rua | -0.241 | -0.223 | -0.205 | -0.205 | **+0.037** | **-0.032** | -0.202 | -0.206 | **+0.141** | |
| 10 asn | -0.541 | -0.573 | -0.515 | -0.526 | -0.488 | -0.483 | -0.520 | -0.508 | | -0.541 |
| 11 ceu | -0.879 | -0.858 | -0.874 | -0.812 | -0.763 | -0.727 | -0.805 | -0.801 | | -0.833 |
| 12 cau | -0.587 | -0.473 | -0.602 | -0.507 | -0.489 | -0.484 | -0.507 | -0.493 | | -0.497 |
| 13 asc | -0.169 | -0.181 | -0.198 | -0.185 | -0.198 | -0.160 | -0.195 | -0.189 | | -0.168 |
| 14 asw | -0.135 | -0.140 | -0.130 | -0.132 | -0.148 | -0.134 | -0.136 | -0.133 | | (partial) |
| 15 ase | -0.471 | -0.476 | -0.483 | -0.524 | -0.500 | -0.419 | -0.468 | -0.461 | | |
| 16 trp | -0.425 | -0.403 | -0.462 | -0.396 | -0.457 | -0.332 | -0.399 | -0.393 | | |
| 17 san | -0.848 | -0.732 | -0.737 | -0.706 | -0.682 | **-0.314** | -0.702 | -0.660 | | -0.750 |
| 18 nzl | -0.867 | -0.578 | **-0.032** | -0.570 | -0.538 | -0.534 | -0.554 | -0.477 | | (partial) |
| 19 ant | -0.084 | -0.066 | -0.133 | -0.156 | **-0.004** | **+0.267** | -0.161 | -0.192 | | (partial) |

Readings:

- Fujita and both OGGM runs sit on the Hugonnet-calibrated values in every region, as expected.
- **Huss_model and Rounce are much less negative in the marine-terminating regions** (`7 sjm`, `9 rua`,
  `19 ant`, and for Huss also `1 ala`, `5 grl`, `17 san`), as is UiO in `7 sjm` and `9 rua`. *(inference)* These
  totals are climatic mass balance without frontal ablation. For Rounce, adding 12 x the reported
  frontal ablation moves `7 sjm` to about -0.39, but overshoots in `9 rua` (about -0.53) and `19 ant`
  (about -0.30), so the accounting is not obvious and should be asked.
- `ETHZ extended` in `18 nzl` is -0.032, far from every other dataset (see 5).

### 4.2 1976-1999 (before the calibration window; the part the experiment is really about)

| Region | DUS comb. | ETHZ ext. | Fujita | Rounce | Huss | OGGM fixed | OGGM spinup | UiO (1991+) |
|---|---|---|---|---|---|---|---|---|
| 1 ala | -0.450 | -0.297 | -0.258 | -0.275 | +0.012 | -0.155 | -0.067 | |
| 2 wna | -0.369 | -0.414 | -0.267 | -0.030 | -0.022 | -0.062 | -0.048 | |
| 3 acn | -0.054 | -0.064 | +0.343 | -0.024 | +0.090 | +0.094 | +0.085 | |
| 4 acs | -0.291 | -0.203 | +0.191 | -0.237 | -0.122 | -0.105 | -0.092 | |
| 5 grl | -0.048 | -0.066 | +0.459 | -0.107 | +0.153 | +0.053 | +0.038 | |
| 6 isl | -0.334 | -0.072 | -0.110 | -0.270 | +0.113 | -0.001 | +0.089 | |
| 7 sjm | -0.124 | -0.116 | -0.320 | +0.056 | +0.203 | -0.013 | +0.034 | +0.014 |
| 8 sca | -0.002 | -0.083 | +0.157 | +0.401 | +0.317 | +0.218 | +0.224 | |
| 9 rua | -0.013 | -0.091 | -0.116 | +0.045 | +0.014 | -0.135 | -0.117 | +0.194 |
| 10 asn | -0.297 | -0.122 | +0.114 | +0.064 | +0.026 | +0.082 | +0.045 | |
| 11 ceu | -0.149 | -0.260 | +0.090 | +0.113 | +0.349 | +0.368 | +0.470 | |
| 12 cau | +0.043 | -0.134 | +0.126 | +0.056 | +0.127 | +0.164 | +0.256 | |
| 13 asc | -0.191 | -0.160 | -0.081 | -0.034 | +0.056 | +0.083 | +0.082 | |
| 14 asw | -0.160 | +0.199 | +0.010 | -0.002 | +0.201 | +0.245 | +0.278 | |
| 15 ase | -0.471 | -0.091 | -0.337 | -0.212 | -0.036 | -0.028 | -0.093 | |
| 16 trp | -0.141 | -0.398 | +0.224 | +0.181 | +0.505 | +0.826 | +0.190 | |
| 17 san | -0.484 | -0.009 | -1.038 | -0.536 | -0.233 | -0.536 | -0.605 | |
| 18 nzl | +0.008 | -0.520 | +1.029 | -0.165 | +0.009 | -0.068 | -0.012 | |
| 19 ant | +0.249 | +0.046 | -0.431 | -0.014 | +0.308 | -0.016 | -0.043 | |

Readings:

- Outside the calibration window the spread is large, and **the ERA5-driven models are
  systematically more positive than the two observation-based datasets** before 2000 in most
  mid-latitude regions (`2 wna`, `11 ceu`, `8 sca`, `10 asn`, `13 asc`, `14 asw`, `16 trp`): models show balanced or
  positive budgets where the observation-based series show loss. This is the central result to
  discuss.
- The two observation-based datasets also disagree with each other in places (`6 isl`, `15 ase`, `17 san`,
  `18 nzl`, `14 asw`, `16 trp`), so "the observations" are not a single reference before 2000.
- Fujita is an outlier in `3 acn`, `5 grl` (strongly positive), `17 san` (-1.04) and `18 nzl` (+1.03).
- OGGM fixed vs spinup in `16 trp` (+0.83 vs +0.19) is the known fixed-geometry artefact documented in
  `CLAUDE.md`.

### 4.3 2020 to 2025-10 (after the calibration window)

| Region | DUS comb. | ETHZ ext. | Fujita | Rounce | Huss | OGGM fixed | OGGM spinup | UiO |
|---|---|---|---|---|---|---|---|---|
| 1 ala | -1.050 | -1.031 | -0.878 | -1.043 | -0.849 | -1.165 | -1.197 | |
| 2 wna | -1.621 | -1.862 | -1.858 | -1.438 | -1.352 | -1.567 | -1.584 | |
| 3 acn | -0.273 | -0.321 | -0.373 | -0.341 | -0.301 | -0.381 | -0.396 | |
| 4 acs | -0.530 | -0.678 | -0.688 | -0.631 | -0.588 | -0.578 | -0.612 | |
| 5 grl | -0.570 | -0.898 | -0.470 | -0.470 | -0.333 | -0.515 | -0.535 | |
| 6 isl | -0.989 | -1.074 | -2.141 | -0.825 | -0.880 | -0.858 | -0.933 | |
| 7 sjm | -1.014 | -1.143 | -2.448 | -0.554 | -0.588 | -1.002 | -1.035 | -0.753 |
| 8 sca | -1.002 | -0.864 | -1.470 | -0.728 | -0.556 | -0.908 | -0.942 | |
| 9 rua | -0.913 | -0.851 | -2.515 | -0.353 | -0.618 | -0.898 | -0.906 | -0.506 |
| 10 asn | -0.876 | -0.724 | -0.922 | -0.728 | -0.719 | -0.774 | -0.756 | |
| 11 ceu | -1.610 | -1.708 | -2.041 | -1.635 | -1.759 | -1.873 | -1.815 | |
| 12 cau | -0.533 | -0.735 | -1.386 | -0.807 | -0.801 | -0.855 | -0.750 | |
| 13 asc | -0.752 | -0.836 | -0.677 | -0.637 | -0.771 | -0.949 | -0.941 | |
| 14 asw | -0.682 | -1.108 | -0.539 | -0.720 | -1.154 | -1.251 | -1.278 | |
| 15 ase | -0.876 | -0.830 | -0.957 | -1.101 | -1.132 | -1.253 | -1.147 | |
| 16 trp | -0.589 | -0.180 | -1.422 | -1.355 | -1.633 | -2.024 | -1.508 | |
| 17 san | -1.121 | -1.200 | -1.433 | -1.353 | -1.731 | -1.660 | -1.428 | |
| 18 nzl | -0.983 | **+1.116** | -2.896 | -1.568 | -2.448 | -1.757 | -1.535 | |
| 19 ant | -0.340 | -0.482 | -0.077 | -0.135 | +0.039 | -0.488 | -0.501 | |

Readings:

- All datasets agree on strong acceleration after 2020.
- Fujita loses two to three times more than everyone else in the Arctic maritime regions (`6 isl`,
  `7 sjm`, `9 rua`) and in `18 nzl`.
- The models lose markedly more than the observation-based series in `16 trp` and, less so, `14 asw`,
  `15 ase`, `17 san`.

### 4.4 Reported areas

- Fujita, Rounce and OGGM fixed report the same RGI 6.0 regional areas (e.g. `1 ala` 86725 km2),
  constant in time. Rounce is slightly lower where glaciers are missing (`10 asn`, `12 cau`, `3 acn`, `9 rua`,
  `19 ant`).
- Huss_model and ETHZ extended share areas that start larger than RGI 6.0 in 1975 and shrink
  (e.g. `1 ala` 91881 to 78040; `11 ceu` 2337 to 1442; `18 nzl` 959 to 587).
- WGMS-DUS starts larger still in many regions (`1 ala` 101161, `5 grl` 101511, `16 trp` 3158) and shrinks
  (`5 grl` to 73026).
- OGGM spinup varies little by comparison (`1 ala` 85228 to 84155; `11 ceu` 1861 to 1806).
- Because everything is in m w.e., these area differences matter as soon as anything is converted to
  Gt: 1975 areas differ by up to ~15-30% between datasets in several regions.

## 5. Things that look wrong or need a question at the workshop

1. **`ETHZ extended`, `18 nzl`**: annual totals are positive in recent years (2021: +1.84, 2024: +2.01,
   2025: +1.50 m w.e.) while the same group's raw model gives -1.3 to -3.0 and WGMS-DUS gives about
   -1; 2000-2019 mean is -0.03 against -0.5 to -0.9 everywhere else. Looks like a processing problem
   in that region. `16 trp` (2020-25: -0.18) is also far from the others.
2. **`ETHZ extended` runs to September 2026**, past the submission date, with no explanation.
3. **`Huss_model` has no methods description of its own**, and the organiser's note that it is "only
   constrained by the long-term trend from HUG21" does not match the data in calving regions
   (`19 ant` +0.27, `7 sjm` -0.05, `9 rua` -0.03 over 2000-2019). Needs clarification of whether the total
   excludes frontal ablation. It is also not a fixed-geometry run.
4. **Rounce: is frontal ablation inside or outside `glacier_change_observed`?** The totals in
   `7 sjm`, `9 rua`, `19 ant` suggest outside, but subtracting it does not consistently recover Hugonnet.
5. **UiO `9 rua`**: reported as 100% coverage of an area that is two thirds of the region; forcing is
   CARRA and inventory is RGI 7.0, both outside the protocol; starts 1991.
6. **`WGMS-DUS_monthly` is filed as a model** but is a statistical downscaling of the combined
   observations, so it is not independent of `WGMS-DUS_combined` and carries no climate forcing.
7. **Protocol compliance is uneven**: only 2 of 7 model datasets reach March 2026; uncertainties are
   given by 3 of 7 and are not comparable (monthly model spread, annual sigma / 12, constant RMSE / 12);
   nobody provides a real above/below water level split (OGGM fills above = total).
8. **Pre-2000 model vs observation offset** (4.2) is the main scientific point. It bears on the
   first stated purpose of the experiment (comparison outside the calibration period); whether it
   also matters for the second (hydrological-to-calendar-year correction, which depends on the
   seasonal cycle rather than the long-term mean) has not been looked at here.

## 6. Open questions to raise at the workshop

Added 2026-10-12 after the first plots (`plots/`, made by `plot_overview.py`). Each entry says who
the question is for, what the files show, and what is still a guess.

### 6.1 WGMS-DUS_monthly: filed as a model by mistake? (for the organisers / WGMS)

- It is filed as `model` but is the annual `WGMS-DUS_combined` series downscaled with a sine wave
  from seasonal glaciological balances. It has no climate forcing and reproduces the combined annual
  values to 0.003 m w.e.
- It covers 15 of 19 regions (no `4 acs`, `9 rua`, `15 ase`, `16 trp`) and has short records in `5 grl`
  (from 2007), `14 asw` (2009-2022), `18 nzl` (from 2004), `19 ant` (from 2001) and `6 isl` (from 1985).
- Fabien's reading: its presence among the model submissions is probably a mistake.
- Question: should it be in the model intercomparison at all, or treated as a monthly version of the
  observational reference? If it stays, it cannot enter global totals.
- Side observation if it stays: its seasonal cycle runs one to two months ahead of every model in
  the northern mid-latitudes (turns negative in April, models in May or June), a consequence of the
  assumed 1 October to 1 May winter and the sine shape.

### 6.2 New Zealand in the GloGEM-based products (for Matthias Huss)

- `ETHZ extended` in `18 nzl` gives -0.03 m w.e./yr over 2000-2019 and is positive after 2020
  (2021: +1.84, 2024: +2.01, 2025: +1.50 m w.e.). Every other dataset shows loss of about 0.5 to
  0.9 m w.e./yr for 2000-2019 and 1 to 3 m w.e./yr after 2020. In the cumulative plots it drifts
  upward after 2015 while everything else falls.
- The raw GloGEM run (`Huss_model`) is **not** affected: -0.53 m w.e./yr for 2000-2019, in line with
  the other models. The two differ only in the four summer months (December to March here), where
  `ETHZ extended` absorbs the offset between model and observation-based annual totals.
- So the problem sits in the observation-based annual totals or in how the correction is applied for
  this region, not in the model *(inference)*. In the mean annual cycle `ETHZ extended` is flat from
  January to March where all others melt strongly.
- Question: what annual series is `ETHZ extended` tied to in `18 nzl`, and is the sign or the
  hydrological-year assignment of the correction right for the southern hemisphere? `16 trp`
  (2020-25: -0.18 against -0.6 to -2.0 elsewhere) may be worth the same check.

### 6.3 Calving vs surface mass balance: what does the reported total contain? (for all modellers)

- The guide asks for total mass change, with frontal ablation as a separate positive field. The
  submissions do not treat this consistently, and it dominates the spread in marine-terminating
  regions and in the global sum.
- 2000-2019 means in m w.e./yr, against roughly -0.31 (`7 sjm`), -0.21 (`9 rua`), -0.13 (`19 ant`) for the
  Hugonnet-calibrated datasets:
  - `Huss_model`: `7 sjm` -0.05, `9 rua` -0.03, `19 ant` +0.27; also low in `1 ala` (-0.53), `5 grl` (-0.23),
    `17 san` (-0.31). Globally it loses 3685 Gt over 2000-2024 against 6542 +- 387 for GlaMBIE-1,
    mostly because of the Antarctic periphery.
  - `Rounce_et_al`: `7 sjm` -0.11, `9 rua` +0.04, `19 ant` -0.00. Reports a constant monthly frontal
    ablation in 8 regions with the remark "frontal ablation is removed in model calibration".
    Adding 12 x that value brings `7 sjm` to about -0.39 but overshoots in `9 rua` (about -0.53) and
    `19 ant` (about -0.30).
  - `UIOcryogrid_RGI7`: `7 sjm` -0.10, `9 rua` +0.14. Stated to be climatic mass balance.
  - OGGM (both runs) and Fujita have no calving and are calibrated so that the climatic balance
    alone matches the geodetic total; they therefore match Hugonnet everywhere but for the wrong
    reason in calving regions.
- *(inference)* `Huss_model`, `Rounce_et_al` and UiO report climatic/surface balance, with frontal
  ablation outside the total. This is not stated explicitly for Huss or Rounce.
- Not explained by calving: `Huss_model` is also 0.13 to 0.15 m w.e./yr less negative than
  `ETHZ extended` in regions without calving (`11 ceu`, `8 sca`, `2 wna`), although GloGEM is described as
  calibrated per glacier to Hugonnet.
- Questions:
  1. Rounce, Huss: is `glacier_change_observed` the total, or climatic balance only? If the latter,
     should total = reported value minus frontal ablation, and why does that not recover the
     calibration target in `9 rua` and `19 ant`?
  2. Huss: where does the remaining offset in non-calving regions come from, and can a methods note
     specific to this run be provided (the current PDF is the ETHZ combined description with an
     organiser's caveat)?
  3. Organisers: which quantity is compared against observations, total or climatic balance? Models
     without calving that are tuned to the geodetic total (OGGM, Fujita) and models that separate
     frontal ablation are not reporting the same thing under the same column.

### 6.4 Smaller points

- `ETHZ extended` runs to September 2026, past the submission date, with no explanation (see 3.9).
- UiO `9 rua` reports 100% coverage for Franz Josef Land + Novaya Zemlya only, uses CARRA and RGI 7.0,
  and starts in 1991 (see 3.6).
- Fujita has about twice the seasonal amplitude of the other models in most regions, large pre-2000
  gains in `3 acn`, `5 grl`, `18 nzl`, and abrupt single-summer losses in `7 sjm` and `9 rua` after 2020.
- OGGM has a much stronger seasonal cycle than the others in `19 ant` (January about -0.57 m w.e.
  against about -0.2); both runs, so it is the mass-balance model. One for us to look at.
- Which area should be used to convert to Gt? Reported 1975 areas differ by 15 to 30% between
  datasets in several regions (see 4.4); the plots use each dataset's own area.

## 7. How the numbers here were produced

- Read every CSV with pandas; window means as described at the top of section 4.
- Cross-checks done: Huss_model vs ETHZ extended month by month (5 regions); WGMS-DUS monthly summed
  to hydrological years vs WGMS-DUS combined (4 regions); submitted OGGM files vs local submission
  CSVs (all 19 regions, both runs).
- Methods PDFs read in full via `pdftotext`, except the contributor listing in the WGMS-DUS combined
  PDF (pages 5-17) and its Table 1 of special cases, which were only skimmed. Figures in the PDFs
  were not looked at.
- Not done yet: Gt conversion and global totals, any plots, comparison against Hugonnet et al. (2021)
  directly (GlaMBIE-1 was used as the 2000-2019 reference instead), seasonal-cycle comparison, the
  GlaMBIE-1 input and hydrological-year files, and the contents of the zip.
