---
title: "Iceland Glacier Projections — Water Divides"
subtitle: "OGGM projections on Landsvirkjun's hydrological catchments"
author: "Fabien Maussion & Patrick Schmitt"
date: 2026-10-16
format:
  html:
    output-file: delivery_overview.html
    toc: true
    embed-resources: true
---

## What this is

This delivery complements the first one (July 2026). It contains new OGGM runs for
all Icelandic glaciers in which the ice caps are divided along **your water divides**
(hydrological catchments) instead of the ice-flow divides used so far. These runs
are meant for the analysis of glacier runoff per catchment, reservoir and hydropower
system.

The first delivery remains the reference for Iceland-wide results (and the AMOC
analysis). This one adds the catchment-level view, and shows how much the change of
divides matters.

## What changed, and what did not

**The inventory.** The new inventory (RGI2025-hydro) has 571 entities:

- **59 catchments** from your water-divide file, on Vatnajökull, Hofsjökull,
  Langjökull, Mýrdalsjökull and Drangajökull;
- **20 "undelineated" remnants**: ice on these ice caps that is not covered by any
  catchment (mostly NE and SW Hofsjökull, and small pieces elsewhere). They are
  modelled so that totals stay consistent, but are not assigned to a river;
- **492 unchanged glaciers** from the first delivery (all other ice caps and small
  glaciers, and the southern outlets of Öræfajökull), with the same IDs.

The total glacier area is identical to the first delivery. As agreed with you:
nested catchments are modelled as their smallest units (Halslon10 = Kring + illakv +
jokulsa + jokulkvisl; Jökulsá í Lóni = Skyndidalsá + the rest), Þingvallavatn uses
the Þingvallavatn_2 delineation, and small overlaps and slivers between your polygons
were cleaned up (modelled catchment areas are within 3% of yours, most within 0.3%).

**Everything else is the same as in the first delivery:** CARRA climate,
trapezoidal bed (lambda=14), the same model version, and the same regional
parameters (precipitation factor, temperature bias, ice creep parameter). Only the
quantities that depend on the shape of each entity were recomputed: the melt factor
of each entity (calibrated to the Hugonnet et al. 2000-2020 mass change, with the
same per-ice-cap totals), the ice thickness, and the dynamic spinup. We checked that
re-running this workflow on the original inventory reproduces the first delivery
exactly, so any difference between the two runs comes from the change of divides
only.

**Why the ice volume changes a bit.** Although the glacier area is the same, the
total ice volume is not (about 2% less for Iceland, more for some ice caps). OGGM
computes ice thickness from mass conservation: along each entity, the ice must carry
the mass gained in the accumulation area down to the ablation area, and the thickness
needed for that depends on this mass flux and on the slope and width of the entity.
Dividing an ice cap differently changes these, so the same ice gets a different
thickness. We kept the ice creep parameter (Glen A) of the first delivery, calibrated
there to the ice volume measured by DEM differencing. This keeps all parameters
identical, so that the change of divides is the only difference between the runs, at
the price of a slightly different total volume. Re-calibrating Glen A would have
forced the same volume, but the runs would then differ by a parameter too. Relative
changes (% of 2020) are much less affected than absolute volumes. Notebook `01`
explains this in more detail.

## Contents & how to read this delivery

As last time, each notebook is linked as an **nbviewer** page (read in your
browser), an **HTML** page (same, for offline reading) and a **.py script** (the
plain Python source; the notebook was generated from it).

1. **[01 — Water divides vs ice divides](https://nbviewer.org/urls/cluster.klima.uni-bremen.de/~fmaussion/share/iceland_projections_hydro/01_hydro_vs_ice_divides.ipynb)**
   ([HTML](https://cluster.klima.uni-bremen.de/~fmaussion/share/iceland_projections_hydro/01_hydro_vs_ice_divides.html),
   [.py script](https://cluster.klima.uni-bremen.de/~fmaussion/share/iceland_projections_hydro/01_hydro_vs_ice_divides.py)) —
   volume, area, mass (% of 2020) and runoff from 2000 to 2100 for both runs, for all
   of Iceland and for Vatnajökull, Hofsjökull and Langjökull.

2. **[02 — Catchments](https://nbviewer.org/urls/cluster.klima.uni-bremen.de/~fmaussion/share/iceland_projections_hydro/02_catchments.ipynb)**
   ([HTML](https://cluster.klima.uni-bremen.de/~fmaussion/share/iceland_projections_hydro/02_catchments.html),
   [.py script](https://cluster.klima.uni-bremen.de/~fmaussion/share/iceland_projections_hydro/02_catchments.py)) —
   how to go from your catchments to model entities, a summary table for all
   catchments, and a worked example (Hálslón: mass, runoff, and seasonal cycle of
   runoff). **Start here to build your own catchment analysis.**

## Key results

**For Iceland as a whole, the change of divides hardly matters.** The water-divide
run has about 2% less ice in 2020 and loses nearly the same fraction of it by 2100
(32.8% vs 34.7% of the 2020 mass left under SSP5-8.5; the difference is smaller for
lower emissions). Peak runoff is 2-3% higher. All these differences are much
smaller than the spread across climate models.

![Iceland: ice divides (solid) vs water divides (dashed). Rows: volume, area, mass (% of 2020), runoff; columns: SSPs (notebook 01)](figures/comparison_iceland.png)

**Per ice cap, the differences are larger, but remain moderate:**

- **Vatnajökull**: about 3% less ice, nearly identical relative mass loss, runoff and
  peak-water timing.
- **Hofsjökull**: slightly faster relative mass loss, and a runoff peak that is
  slightly higher and equal or a few years later.
- **Langjökull**: the largest change. The water-divide run has about 14% more ice
  (it is divided into fewer, larger catchments, which come out thicker in the
  inversion, see above), loses it faster after about 2050, and has a 10-13% higher
  runoff peak, 3-10 years later.

![Langjökull: the ice cap where the change of divides matters most (notebook 01)](figures/comparison_langjokull.png)

The figures for [Vatnajökull](figures/comparison_vatnajokull.png) and
[Hofsjökull](figures/comparison_hofsjokull.png) are in the `figures/` folder.

**At the catchment level**, notebook `02` gives the numbers for each catchment. As an
example, the Hálslón catchment keeps about two thirds of its 2020 mass by 2100 under
SSP2-4.5, and its runoff keeps increasing until late in the century:

![Hálslón catchment: mass and runoff (notebook 02)](figures/halslon_mass_runoff.png)

![Hálslón catchment: seasonal cycle of runoff for three decades, SSP2-4.5 (notebook 02)](figures/halslon_seasonal_cycle.png)

## Caveats

- **Catchments in the accumulation area.** OGGM treats each catchment as a glacier
  with its own flowline, while in reality ice flows across water divides towards the
  outlet glaciers. Catchments that lie mostly in the interior of an ice cap therefore
  lose their ice too slowly in the model. The clearest case is **grimsvotn**, which
  even gains mass. The large northern Vatnajökull catchments (including those of
  Hálslón, Kreppa and Jökulsá á Fjöllum) are probably also affected to some degree:
  for them, the late peak water should be taken with caution. Runoff over the next
  decades is more robust, because it is driven by surface melt and rain. Notebook
  `02` lists these catchments.
- **Undelineated remnants** are only there to keep the totals consistent. They are
  not hydrologically meaningful and can be left out of a catchment analysis.
- **Runoff** is computed as in the first delivery: melt and rain over a fixed area
  (the glacier outline at the start of the projections). It does not include the rest
  of the river catchment.
- **Four CMIP6 runs** (CESM2-WACCM and MRI-ESM2-0, ssp126 and ssp534-over) lose many
  small glaciers that regrow beyond their model domain, a known OGGM limitation. They
  are excluded from the comparisons in both runs; they are still in the data files.

## Downloading this delivery

Everything is hosted at:

<https://cluster.klima.uni-bremen.de/~fmaussion/share/iceland_projections_hydro/>

or as a single archive:

```bash
wget https://cluster.klima.uni-bremen.de/~fmaussion/share/iceland_projections_hydro.tar.gz
tar -xzf iceland_projections_hydro.tar.gz
```

The uncompressed data is about 9 GB, mostly the projection files.

## Data

```
data/
  inventory/      — the RGI2025-hydro inventory (shapefile, RGI7 format, with the
                    extra columns ent_type, ws_id, catchment, icecap), the lookup table
                    from your catchments to model entities, the ice cap lists, and the
                    per-entity geodetic mass balance used for calibration
  historical/     — historical runs (1991-2026) for the water-divide run (hydro_v2)
                    and, for comparison, the first delivery (rgi2025)
  projections/
    hydro_v2/     — merged CMIP5/CMIP6 projections, 2100 and 2300, water divides
    rgi2025/      — CMIP6 to 2100 from the first delivery, for the comparison
figures/
  inventory_maps/ — maps of the new entities per ice cap, with your water divides
tables/
  catchment_summary.csv — the summary table of notebook 02
```

The projection files have the same structure as in the first delivery, so the
notebooks of the first delivery also work with them (change the data path).

## Running this delivery

The notebooks use relative paths only and must be run from this folder. They need
`xarray`, `pandas`, `numpy`, `matplotlib` and, for notebook `02`, `geopandas`.

## About this delivery

As for the first delivery, these notebooks were built with the assistance of Claude
(Anthropic's AI coding assistant), under our direction and review. The data and
notebooks are self-contained, so they can be extended directly, by you or with the
help of Claude Code pointed at this folder.