Work

Unsupervised Patch‐Based Ecoregion Classification in Austria Using DEM ‐Derived Features, Deep Learning, and Climate Data

Year: 2026

Type: article

Abstract: ABSTRACT Digital elevation models (DEMs) are widely used to characterize terrain and support applications ranging from geomorphology to environmental management. While traditional analyses rely on handcrafted geomorphometric features, advances in machine learning enable richer, data‐driven representations of topographic surfaces. In this study, we present a multi‐source feature fusion approach that integrates statistical descriptors of elevation, derived terrain attributes, deep features extracted via a ResNet encoder, and climate variables from weather datasets. Using 5175 randomly sampled patches, we generated a comprehensive feature set and applied unsupervised clustering to identify ecoregional units. Results indicate that combining DEM‐derived features with climate data produces clusters that are both quantitatively robust, as indicated by internal validity indices, and exhibit meaningful spatial distribution when compared with known ecoregions.

Source: Transactions in GIS

DOI: 10.1111/tgis.70421

Authors: Thomas Schneidergruber, Johannes Scholz

Institutions: University of Salzburg, Fachhochschule Salzburg

Language: English

Topic: Geography and Environmental Studies

Subfield: Water Science and Technology

Field: Environmental Science

Domain: Physical Sciences

SDGs: Climate action, Life on land

Open Access status: hybrid

Estimated APC paid: $3,800