Skip to content


Folders and files

Last commit message
Last commit date

Latest commit



60 Commits

Repository files navigation

OGGM Mass-Balance sandbox

Next generation of OGGM's mass-balance models. Work in process!

You are welcome to discover some new mass-balance options such as different climate resolutions or surface type distinction. However, we want to make clear that this is work in process and is part of my PhD project:

  • the new mass-balance module from this sandbox is of course less stable, less robust, slower and less documented than the default MBmodel (i.e. the PastMassBalance class in OGGM)
  • it has no pre-mass-balance-calibrated gdirs
  • although I try to include tests as I can, they might be less rigorous as the ones from pure OGGM

The OGGM mass-balance sandbox has been used by the preprint "Glacier projections sensitivity to temperature-index model choices and calibration strategies" (Schuster et al. 2023, in review, submitted to Annals of Glaciology). We uploaded the preprint to EarthArXiv, and it is available under the DOI Most options are described there in details. If you use the OGGM mass-balance sandbox, please cite the preprint. The scripts to reproduce the analysis of the preprint and that use this OGGM mass-balance sandbox are here available:

If you can't wait until the OGGM massbalance-sandbox is integrated into OGGM default and want to use it already now for your specific OGGM application, please contact me first either by:

  • opening an issue
  • writing an e-mail to me (Lilian Schuster)
  • or, the easiest way, discuss it inside of our OGGM slack channel

At the moment these options of climate resolution are available inside TIModel:

  • to compute degree days:
    • using monthly temperatures ('mb_monthly'), default option in OGGM
    • using monthly temperatures and daily temperature standard deviations (monhtly averages) from the past to generate daily temp. assuming normal distributed data ('mb_pseudo_daily_fake')
    • using monthly temperatures and daily temp std to generate daily temp. assuming normal distributed data ('mb_pseudo_daily')
    • using daily temperatures ('mb_real_daily')
  • temperature lapse rates:
    • using a constant calibrated value independent of location and season (-6.5 K/km, grad_type: cte), default option in OGGM
    • using lapse rates from ERA5 that vary throughout the year and inbetween glacier locations, but that are constant inbetween the years (grad_type: 'var_an_cycle')
    • ( this has not been tested: using lapse rates from ERA5 that vary throughout the year and inbetween glacier locations, different for each year (grad_type: 'var') )

In addition, a surface type distinction model is included with a bucket system together with a melt_f that varies with age inside of TIModel_Sfc_Type:

  • there are two options for how often the melt factor should be updated and how many buckets exist

    • melt_f_update=annual
      • If annual, then it uses 1 snow and 5 firn buckets with yearly melt factor updates.
    • melt_f_update=monthly:
      • If monthly, each month the snow is ageing over 6 years (i.e., 72 months -> 72 buckets).
    • the ice bucket is thought as an "infinite" bucket (because we do not know the ice thickness at this model stage)
    • Melt factors are interpolated linearly inbetween the buckets. TODO: include non-linear melt factor change as an option!
  • there are different option of how and how fast the snow/firn melt factor approximates to the ice melt factor.

    • melt_f_change=linear
      • just a linear change assumed
    • melt_f_change='neg_exp'
      • a negative exponential change is assumed via the eq. melt_f = melt_f_ice + (melt_f_snow - melt_f_ice)* np.exp(-time_yr/tau_e_fold_yr)
      • tau_e_fold_yr is per default 1 year (but can be changed to another value)
  • default is to use a spinup of 6 years. So to compute the specific mass balance between 2000 and 2020, with spinup=True, the annual mb is computed since 1994 where at first everything is ice, and then it accumulates over the next years, so that in 2000 there is something in each bucket ...

  • the ratio of snow melt factor to ice melt factor is set to 0.5 (as in GloGEM, PyGEM, ...) but it can be changed via melt_f_ratio_snow_to_ice

    • if we set melt_f_ratio_snow_to_ice=1 the melt factor is equal for all buckets, hence the results are equal to no surface type distinction (as in TIModel)
  • get_annual_mb and get_monthly_mb work as in the OGGM v153 PastMassBalance class, however they only accept the height array that corresponds to the inversion height (so no mass-balance elevation feedback can be included at the moment!)

    • that means the given mass-balance ist the mass-balance over the inversion heights (before doing the inversion and so on)
  • the buckets are automatically updated when using get_annual_mb or get_monthly_mb via the TIModel_Sfc_Type.pd_bucket dataframe

  • to make sure that we do not compute mass-balance twice and to always have a spin-up of 6 years, we save the mass balance under

    • get_annual_mb.pd_mb_annual: for each year
      • when using get_monthly_mb for several years, after computing the December month, the pd_mb_annual dataframe is updated
    • get_annual_mb.pd_mb_monthly: for each month
      • note that this stays empty if we only use get_annual_mb with annual melt_f_update

All options have been tested with the elevation-band flowlines.

How to install!

The OGGM MB sandbox needs OGGMv153 and does currently not work with the latest OGGMv16. It also needs some developments after the OGGM v153 release. Thus, it is best if you install a more recent OGGM development version which is still before OGGM v16 (e.g. "OGGM version: '1.5.4.dev60+g9d17303'": by doing:

$ conda create --name env_mb
$ source activate env_mb
$ pip install --no-deps "git+"
$ git clone
$ cd massbalance-sandbox
$ pip install -e .

Test the installation via pytest while being in the massbalance-sandbox folder, best is if you do :

$ pytest -v -m "not no_w5e5"

(Attention: Just doing pytest ., downloads several climate datasets and does example ensemble projections into the future. If you only want to use W5E5 climate data, the tests run with the line above are sufficient)

If you have issues to install the right package versions, you can install the packages dependent on oggm_v153 by using for example the following .yml file:

The MBsandbox package can be imported in python by

>>> import MBsandbox

code inside of MBsandbox

    • process different climate data (W5E5, WFDE5_CRU, ERA5_daily, W5E5_MSWEP(prcp from MSWEP, temp. from W5E5)),
    • new mass-balance model TIModel_Parent with children TIModel and TIModel_Sfc_Type
  • copies of run_from_climate, run_random_climate that are compatible with TIModel, not yet tested for TIModel_Sfc_Type
  • helper functions to minimize the bias, optimise standard deviation quotient for reference glaciers, to calibrate the melt factor given the precipitation factor and geodetic observations, and to compute performance statistics
  • tests: tests for different functions
  • wip: work in process folder without documentation

How to use!

There is not a lot of documentation, but here are some example notebooks with explanations inside of the docs folder. More example scripts of the OGGM massbalance-sandbox are in the repository.


equilibrium runs

other notebooks not directly related to the MBSandbox:

  • geodetic_mb_calibration_to_volume_changes.ipynb : template how to calibrate a glacier by using the geodetic estimates (for the default PastMassBalanceModel of OGGMv153!!! )

    • also explains docs/calib_log_fit_pf_distribution_change_monthly_cte_melt_f_minus_1.png
  • emulator_bayes_calib :

    • trial notebooks to use Bayesian calibration with PyMC3 -> work in process