Glacier Monitoring Using GEDI Data in Google Earth Engine: Outlier Removal and Accuracy Assessment

Climate change has notably altered the elevation of mountain glaciers, particularly in alpine regions. Alpine glaciers play a pivotal role not only as indicators of climate change but also as crucial elements for human and wildlife well-being, regulating freshwater supply and providing vital habitats in Europe. Consequently, continuous monitoring of these glaciers offers valuable insights into their changing structure and surface dynamics [1].

 

While Unmanned Aerial Vehicles (UAV) offer the most precise method for tracking glacier surface changes, their practicality is often hindered by cost limitations and challenging in-situ measurements in extreme weather or remote areas. Therefore, remote sensing and satellite altimetry emerge as a feasible alternative in such scenarios.

 

Numerous LiDAR and RADAR altimetry sensors, such as Jason-2 and 3, CryoSat, and ICESat-1 and 2, have been employed. However, the Global Ecosystem Dynamics Investigation (GEDI), a reliable source of altimetry data, has been overlooked due to its restricted latitude range of 51.6 and -51.6 [2]. GEDI has proven its efficacy in measuring forest and canopy top height, monitoring lakes and water resources and generating Digital Surface Models (DSM).

 

Google Earth Engine (GEE), a cloud-based platform renowned for its ability to integrate diverse datasets and potent analytical tools, has recently incorporated GEDI into its extensive repository [3].

Our initial analysis aims to assess the accuracy of GEDI data for glacier monitoring. Firstly, we focus on detecting and eliminating outliers. Secondly, we compare the glacier levels obtained from GEDI with reference ground truth. Thus, we've chosen the Rutor and Belvedere glaciers in Northern Italy, where we have access to reference-level measurements from UAV DEMs.

 

The proposed outlier detection consists of two steps for each GEDI passage over the glacier surface.
The first step relies on quality surface flags available within GEDI bands, In the subsequent phase, the outlier removal process was refined by employing the x-means algorithm, an unsupervised classifier available within GEE. This approach facilitated the identification and elimination of outliers within the GEDI data set, contributing to refining the dataset's accuracy for comparative analysis with the reference ground truth.

After the above-mentioned outlier removals, we obtained a median difference of -0.27m and NMAD of 4.9 m for Rutor Glacier in 2021 from more than 500 footprints, whereas for Belvedere a median difference of -0.43 and NMAD of 3.7m were obtained. These underestimated values might be due to the nearly 2-month difference between the DEM and the GEDI acquisitions.

 

[1] Belloni, V., et al. (2023). High-resolution high-accuracy orthophoto map and digital surface model of Forni Glacier tongue (Central Italian Alps) from UAV photogrammetry. Journal of Maps, 19(1), 2217508.

[2] Hamoudzadeh, A., et al.: Gedi Data Within Google Earth Engine: Potentials And Analysis For Inland Surface Water Monitoring, EGU General Assembly 2023, Vienna, Austria, EGU23-15083

 

[3] Hamoudzadeh, A., et al. (2023). GEDI data within google earth engine: preliminary analysis of a resource for inland surface water monitoring. In The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences.