GitHub

Muninn.jl

Muninn.jl is the surface mass balance (SMB) module of the ODINN ecosystem. It computes the net ice accumulation and ablation at the glacier surface — the key atmospheric forcing that drives glacier volume change over time.

The default model is a distributed temperature-index (TI) model, which approximates ablation from positive degree-days (PDD) and accumulation from solid precipitation — see Hock (2003) [2] for a review of temperature-index melt modelling in mountain areas. Two variants are provided: TImodel1 (a single degree-day factor) and TImodel2 (separate snow and ice degree-day factors). The climate forcing and glacier geometry driving these models are preprocessed with OGGM, Maussion et al. (2019) [3], through Gungnir.

Neural-network-based mass balance is supported via MassBalanceMachine.jl, which ports pre-trained PyTorch models exported as JSON into Lux.jl as a CustomMLP <: MBmodel. See the MassBalanceMachine package page and the Models page for details.

Muninn also handles the calibration of temperature-index models against geodetic mass balance observations from Hugonnet et al. (2021) [4]. calibrate_MB_model tunes DDF, prcp_fac and temp_bias per glacier so that the modelled mean annual mass balance matches the observed one, and compute_mean_annual_MB evaluates the result. See the SMB calibration tutorial for a full worked example.

Muninn re-exports all of Sleipnir, so importing Muninn gives access to the full Sleipnir API without a separate using Sleipnir statement.

Use directly vs. use ODINN.jl

Use Muninn directly when you want to:

  • Compute mass balance independently of ice flow (e.g. sensitivity studies, regional mass balance assessments).
  • Calibrate a temperature-index model against geodetic observations, without running any ice flow simulation.
  • Plug in a custom SMB model into a downstream tool that accepts a MBmodel object.
  • Experiment with a new mass balance parameterization before integrating it into a full Huginn or ODINN simulation.

Use Huginn when you need the coupled ice dynamics + mass balance forward simulation, and ODINN.jl when you additionally need gradients through that coupling — for instance when training a UDE that involves the mass balance component.

Minimal usage example

using Muninn

params = Parameters(
    simulation = SimulationParameters(
    tspan = (2010.0, 2015.0),
    multiprocessing = false,
    use_MB = true
),
)

glaciers = initialize_glaciers(["RGI60-11.00897"], params)

# Temperature-index mass balance model
glacier = glaciers[1]
mb_model = TImodel1(params; DDF = 6.0 / 1000.0, prcp_fac = 1.2)

# Compute the mass balance for one monthly step, independently of ice flow
step = params.simulation.step_MB  # monthly step (1/12 yr)
get_cumulative_climate!(glacier.climate, 2010.5, step)
climate_2D = downscale_2D_climate(glacier.climate.climate_step, glacier.S, glacier.Coords)
MB = compute_MB(mb_model, climate_2D, step)
@show size(MB)   # (nx, ny) mass balance grid in m w.e.

Extending Muninn

To add a new mass balance model (new MBmodel subtype), see the Extending ODINN guide.

API reference

See Muninn API for the full list of exported types and functions.