Optimizing rock glaciers activity classification in South Tyrol (North-East Italy): integrating multisource data with statistical modelling
As a consequence of climate warming, high-altitude periglacial and glacial environments exhibit the clearest signs of cryosphere degradation, and the Alps serve as a natural laboratory for studying the primary effects on permafrost-related features. Our research in South Tyrol, North-East Italy, aim...
| Main Authors: | , , , , , |
|---|---|
| Format: | Text |
| Language: | English |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://doi.org/10.5194/egusphere-2024-1511 https://egusphere.copernicus.org/preprints/2024/egusphere-2024-1511/ |
| Summary: | As a consequence of climate warming, high-altitude periglacial and glacial environments exhibit the clearest signs of cryosphere degradation, and the Alps serve as a natural laboratory for studying the primary effects on permafrost-related features. Our research in South Tyrol, North-East Italy, aimed to develop an updated classification system for rock glaciers activity, based on remote sensing data and statistical models, with the aim of categorizing them as active, transitional, or relict according to the recent RGIK guidelines. Since the current regional inventory includes activity attributes based only on morphological observations and differential SAR interferometry (DInSAR) coherence, it lacks a comprehensive definition integrating climatic drivers, displacement rates, and morphometric parameters. To address this, we utilized the Alaska Satellite Facility's InSAR cloud computing, employing small baseline subset (SBAS) approach and MintPy algorithms to extract velocity data for each rock glacier in South Tyrol. Additionally, we analyzed geomorphological and climatic maps derived from in-situ and remote sensing data to obtain descriptive parameters influencing rock glaciers development and activity. From a wide range of potential variables, we selected eight key predictors, representing physical (e.g. temperature), morphological (e.g. roughness), and dynamic (e.g. velocity and coherence indicators) attributes. These predictors were successively integrated in a multiclass generalized additive mixing model (GAM) classifier to categorize the landforms. Applying this model to the entire dataset (achieving an AUC over 0.9) allowed us to address gaps in previous classification methods and provided activity attributes for previously unclassified rock glaciers, along with associated uncertainty values. Our approach improved classification accuracy, leaving only 3.5 % of features unclassified compared to 13 % in morphological classification and 18.5 % in DInSAR-based methods. The results revealed a predominance of ... |
|---|