the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
A minimal machine learning glacier mass balance model
Abstract. Glacier retreat presents significant environmental and social challenges. Understanding the local impacts of climatic drivers on glacier evolution is crucial, with mass balance being a central concept. This study introduces miniML-MB, a new minimal machine learning model designed to estimate annual point surface mass balance (PMB) for very small datasets. Based on an XGBoost architecture, miniML-MB is applied to model PMB at individual sites in the Swiss Alps, emphasizing the need for an appropriate training framework and dimensionality reduction techniques. A substantial added value of miniML-MB is its data-driven identification of key climatic drivers of local mass balance. The best PMB prediction performance was achieved with two predictors: mean air temperature (May–August) and total precipitation (October–February). miniML-MB models PMB accurately from 1961 to 2021, with a mean absolute error (MAE) of 0.417 m w.e. across all sites. Notably, miniML-MB demonstrates similar and, in most cases, superior predictive capabilities compared to a simple positive degree-day (PDD) model (MAE of 0.541 m w.e.). Compared to the PDD model, miniML-MB is less effective at reproducing extreme mass balance values (e.g., 2022) that fall outside its training range. As such, miniML-MB shows promise as a gap-filling tool for sites with incomplete PMB measurements, as long as the missing year's climate conditions are within the training range. This study underscores potential ways for further refinement and broader applications of data-driven approaches in glaciology.
- Preprint
(3051 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 07 Nov 2024)
-
RC1: 'Comment on egusphere-2024-2378', Signe Hillerup Larsen, 01 Oct 2024
reply
Review of van der Meer et al.: A minimal machine learning glacier mass balance model
The main objective of the study is to create a Machine Learning model to estimate annual point surface mass balance measurements. This could serve as a tool to fill gaps in in-situ measurements, which are crucial for the calibration and validation of models of global glacier change. The usefulness of a machine learning approach is discussed in comparison to the traditional Positive Degree Day (PDD) approach, and given the large amount of observational data points in the Swiss Alps, this method could potentially provide a better constraint on annual point mass balance modeling.
Main comments:
Thanks to the authors for a well-written manuscript! I think using a machine learning approach instead of a PDD approach to fill gaps in point mass balance measurements is an excellent idea, and the choice of method is appropriate. However, I do have one major concern:
The authors took a very open-ended approach to determining the predictors for the miniML_MB model, neglecting the physical background knowledge that is well-established in glaciology. The best predictors of annual point mass balance, which can be found within temperature and precipitation data, are typically the annual snow accumulation and the annual heat content (total number of positive degree days) over the full hydrological year. This knowledge forms the basis of the PDD model, but it is not incorporated into the miniML_MB model in the same way. As a result, I don’t believe the comparison between the two methods is entirely fair. To clarify my point, I’ve identified a few specific lines that illustrate the issue:
Line 171: It is not clear whether the annual predictors are based on the hydrological year or the calendar year.
Line 239: It is stated that temperature [April–August] and precipitation [October–February] are the best-suited predictors. This is unsurprising, given the physics behind glacier mass balance, but is written as if it could have been any of the combinations. I also believe that using data from the full hydrological year would likely yield even better predictors.
Lines 391–396: Snow cover on ice is also crucial in determining glacier mass balance, as glaciers cannot melt when snow is present.
While I am not an expert in Machine Learning, I believe the manuscript requires revision, particularly in how the predictors are defined. I suggest basing them more strongly on established physical knowledge. Predictors could include annual precipitation sum and annual PDD (based on the hydrological year), but could also be expanded to include annual snow, rain, and PDD or something similar? I think that the knowledge we already have on what drives glacier balance should be made more clear throughout the manuscript but mainly in the description of the model and the discussion.
I am happy to discuss this further if you think it would be helpful.
With best regards,
Signe Hillerup Larsen
Geological Survey of Denmark and Greenland (GEUS)
shl@geus.dkCitation: https://doi.org/10.5194/egusphere-2024-2378-RC1
Interactive computing environment
miniML-MB: Release v.1.1 Marijn van der Meer https://doi.org/10.5281/zenodo.12905503
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
245 | 55 | 27 | 327 | 2 | 3 |
- HTML: 245
- PDF: 55
- XML: 27
- Total: 327
- BibTeX: 2
- EndNote: 3
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1