GEDTM30: global ensemble digital terrain model at 30 m and derived multiscale terrain variables

Area of Science:

• Geosciences
• Remote Sensing
• Geomatics

Background:

• Global Digital Terrain Models (DTMs) are crucial for various Earth science applications.
• Existing DTMs often face limitations in accuracy, resolution, and consistency across diverse landscapes.
• Fusion of multiple data sources and advanced modeling techniques are needed to overcome these limitations.

Purpose of the Study:

• To produce and validate an open global ensemble digital terrain model (GEDTM30) at ~30 m spatial resolution.
• To derive standard terrain and hydrological variables at multiple scales.
• To ensure global consistency and local accuracy through a novel transfer learning framework.

Main Methods:

• Data fusion of Copernicus DEM, ALOS World3D, and object height models for DTM generation.
• A global-to-local transfer learning model leveraging extensive lidar datasets (ICESat-2, GEDI) for validation and refinement.
• An optimized tiling workflow using the Equi7 grid system for generating multiscale terrain and hydrological variables.

Main Results:

• GEDTM30 demonstrates significant RMSE reduction compared to Copernicus DEM in built-up and vegetated areas.
• GEDTM30 achieves superior vertical accuracy compared to state-of-the-art DTMs (MERIT DEM, FABDEM, FathomDEM) when validated with GNSS data.
• The tiling-based workflow ensures computational efficiency, absence of boundary artifacts, and preservation of hydrologic connectivity.

Conclusions:

• The GEDTM30 dataset provides a globally consistent and locally accurate DTM with derived variables at ~30 m resolution.
• The developed transfer learning and tiling approach offers a scalable solution for DTM production and analysis.
• Public availability of GEDTM30 and associated code facilitates advancements in Earth surface modeling and geospatial research.