A New Method to Estimate Changes in Glacier Surface Elevation Based on Polynomial Fitting of Sparse ICESat - GLAS Footprints

Tianjin Huang, Li Jia, Massimo Menenti, Jing Lu, Jie Zhou, Guangcheng Hu

TL;DR

This paper introduces a novel polynomial fitting method to more accurately estimate glacier surface elevation changes from sparse ICESat-GLAS altimetry data, particularly in complex mountainous terrain. By moving beyond simple planar fitting, this approach accounts for local topographic variations, leading to more reliable glacier thickness change estimates and a better understanding of glacier mass balance.

What Problem Does This Paper Address?

Accurately measuring glacier thickness change is crucial for understanding their contribution to sea-level rise and for climate change monitoring. Traditional methods like in-situ measurements are labor-intensive, while differencing digital elevation models (DEMs) can be challenging due to DEM accuracy. Satellite radar altimetry offers a broader coverage but can be complicated by terrain. The Geoscience Laser Altimeter System (GLAS) on ICESat provided valuable elevation data, but its sparse, along-track footprints posed a challenge for estimating glacier thickness change, especially in complex mountainous regions. Existing methods, like planar fitting, assumed a flat glacier surface within a sampled facet, which is often not the case in these environments, leading to significant errors and biases in the estimated thickness changes.

Methods

The core of this paper is the development of an improved polynomial fitting method to overcome the limitations of the traditional planar fitting method.

  • Planar Fitting Method: This standard approach assumes that a small segment of a glacier surface can be approximated by a plane. It fits a plane to observed GLAS footprints within a defined "facet" and estimates elevation changes based on the difference in these fitted planes over time. However, this method struggles with complex topography where the glacier surface deviates significantly from a plane.
  • Improved Polynomial Fitting Method: The authors propose replacing the planar assumption with a binary polynomial function. This allows for a more flexible representation of the glacier surface, better capturing local topographic variations. They determined that a fourth-order (p=4) binary polynomial function provided the best balance between fitting accuracy and the complexity of the resulting equations, especially when compared to a first-order polynomial (planar fitting).
  • Data Integration: To establish a reference elevation and account for potential co-registration errors between datasets, the study co-registered GLAS data with the Shuttle Radar Topography Mission (SRTM) DEM. This ensured a more consistent baseline for elevation change calculations.

Datasets

  • ICESat-GLAS Global Land Surface Altimetry Data (GLA14): This dataset provided the primary source of glacier surface elevation measurements. Data from February/March of various years between 2003 and 2009 were used to minimize seasonal effects.
  • Shuttle Radar Topography Mission (SRTM) DEM: This dataset provided a near-global, spatially continuous elevation reference (90m resolution) for the year 2000. It was crucial for co-registration and as a baseline for elevation change calculations.

The study did not introduce new benchmark datasets but rather utilized existing, well-established datasets for its evaluation.

Key Insights and Ideas

  • The "Flat Earth" Assumption is Problematic: The fundamental insight is that assuming a glacier surface is flat within a sampled facet (the planar fitting assumption) introduces significant errors, especially in mountainous regions. The improved method directly addresses this by allowing for more complex surface shapes.
  • Polynomial Order Matters: The choice of polynomial order (p) is critical. While a higher order can better fit complex surfaces, it also requires more data and can lead to overfitting. The study found that p=4 offered a good compromise, significantly outperforming p=1 (planar fitting) in terms of RMSE, particularly on complex terrain.
  • SRTM DEM is a Valuable Reference: Using the SRTM DEM as a reference for co-registration and as a baseline for elevation change calculations proved effective in improving the accuracy and extending the temporal coverage of the glacier thickness change estimates.
  • Non-Rigid Facets Introduce Uncertainty: The study acknowledges that glacier facets are not perfectly rigid and can change shape over time. The polynomial fitting method was found to be less susceptible to these changes compared to planar fitting, as indicated by smaller standard deviations in thickness change estimates.

Key Results

  • Improved Accuracy: The improved polynomial fitting method (p=4) consistently yielded smaller Root Mean Square Errors (RMSE) compared to the planar fitting method (p=1), especially on complex glacier surfaces. For example, on a complex glacier surface, RMSE decreased from 42.73 m (planar) to 9.66 m (polynomial).
  • More Reliable Estimates: For the Naimona'nyi glacier, the improved method's results (-0.46 to -0.97 m/year) showed much smaller variations with facet length and were more consistent with in-situ DGPS measurements (-0.45 to -0.67 m/year) than the planar fitting results (-0.55 to -3.28 m/year).
  • Extended Temporal Coverage: By using the 2000 SRTM DEM as a reference, the improved method could estimate glacier thickness changes over a longer period (2000-2008/2009) compared to methods relying solely on GLAS data (e.g., 2004-2008/2009).
  • Dealing with Complexity: The method demonstrated its ability to handle both large, relatively flat glaciers (Guliya) and glaciers with very complex topography (Chasku Muba).

Limitations and Future Work

  • Data Sparsity: Despite the improvements, the inherent sparsity of GLAS footprints remains a limitation, especially in areas with very complex topography where even polynomial fitting might struggle to fully capture the surface.
  • SRTM DEM Errors: The study acknowledges that the SRTM DEM itself may have errors, which could influence the accuracy of the co-registration and the baseline elevation.
  • Future Work: The authors suggest that future missions like ICESat-2, with denser footprints, will further enhance the potential of this polynomial fitting approach for retrieving glacial thickness change on larger mountain glaciers without the need for additional DEMs. They also highlight the need for further investigation into the impact of non-rigid facets on thickness change estimations.

Summary updated

BibTeX
@inproceedings{huang2017new,
  author = {Tianjin Huang and Li Jia and Massimo Menenti and Jing Lu and Jie Zhou and Guangcheng Hu},
  title = {{A New Method to Estimate Changes in Glacier Surface Elevation Based on Polynomial Fitting of Sparse ICESat - GLAS Footprints}},
  booktitle = {Sensors},
  year = {2017},
  url = {https://openreview.net/forum?id=7Aicc9ZKBf},
}

Loading related content...

Figure