Mapping Alluvial Terraces in Watersheds Using Gaussian Mixture Model on Relative Height.

Abstract:
Alluvial terraces in watersheds are geomorphic features formed by river incision and sediment deposit, representing former floodplain levels. They serve as valuable records of fluvial dynamics, climatic changes, and tectonic activity. Mapping methods for terraces that rely on field-acquired data, often involving physical or chemical analyses, are not feasible for large-scale applications. When aiming to map at the national scale, the development of a methodology that eliminates the need for such detailed information enhances scalability and broadens applicability.

We proposed a semi-automatic predictive mapping method for watershed terraces using 25m Digital Earth Model (DEM) provided by the IGN (French geographical service) and derived variables such as curvature, slope, and the difference from a base level (Raingeard et al., 2019). This method demonstrated meaningful results in the Pyrenean Piedmont for the Baïse and Ousse rivers, with the predicted map showing strong alignment with the geological reality.

In this study, we propose an automated approach for identifying alluvial terraces using relative height. Relative height is defined as the difference between the elevation derived from a DEM and the base level. Our methodology is based on the hypothesis that terraces are represented as flat areas in the relative height, where pixels exhibit similar statistical distributions. To capture these patterns, we employ a Gaussian Mixture Model, a probabilistic framework that approximates data as a combination of multiple Gaussian distributions. In this context, each Gaussian distribution corresponds to a specific alluvial terrace.

We conducted experiments on the study areas used by Raingeard et al. (2019), and the results are consistent with both the semi-automatic method and the geological reality. These outcomes provide promising prospects for the predictive mapping of superficial deposits

Reference:

Raingeard A., Tourlière B., Lacquement. F, Baptiste. J, Tissoux. H. Semi-automatic quaternary alluvial deposits mapping - Methodology for the predictive mapping of flat terrains within a watershed, by semi-automatic analysis of the Digital Elevation Model. INQUA 2019, Jul 2019, Dublin, Ireland. 2019.