16:15 〜 16:30
[HTT14-04] Sediments sources typology and mapping based on remote sensed data analysis
★Invited Papers
キーワード：Landslide, Sedimentary cascade, (de)coupling, Remote sensing, Machine learning
Before studying the sedimentary cascade from the point of view of connectivity, it is necessary to identify and characterise the functioning of its components and elementary (de)couplings from the source zones. For this purpose, the open-air laboratory represented by the geomorphological disaster of 5-6 July in the Asakura region (Kyushu) allows us to study the sedimentary sources with an estimated sample of more than 1500 mass movements triggered by the heavy rainfall during this High-Magnitude / Low-Frequency (HMLF) event.
This work follows three stages: i) identification and mapping of source zones, ii) characterisation of their nature and iii) estimation of their contribution to the sedimentary cascade.
The mapping of source zones was carried out using multispectral satellite remote sensing combined with a machine-learning method. The images used are from the Pleiades sensor, which provides a very high resolution (VHR) spatial (0.50 m with Pansharpening) and spectral (RGB NIR). The semi-supervised classification was performed by the Random Forest learning algorithm using the Dzetsaka extension in QGIS. Bare soils on slopes are interpreted as instability zones subject to lateral transfer. The typology of the source areas was carried out in two steps. Firstly, the nature of the mass movements and therefore of the source zones was characterised using expert analysis methods associated with morphological criteria (length/width ratio). Secondly, the contribution of the source zones to the sedimentary cascade was estimated using expert analysis methods and geomorphological criteria (e.g. types of contact, obstacles to sedimentary continuity, etc.).
The mapped sedimentary sources represent 6.7 km² and 3.5% of the study area (193 km²). There is a spatial concentration of source areas in the southwest of the study area. The machine error of the classification is about 97% and the operator error is 58%. The nature of the sedimentary sources is characterised by a broad spectrum of hydro-gravity processes that can be grouped mainly between slides/slope failures and flow-like landslide (e.g. debris avalanche, mud flow). These sedimentary sources actively contribute to the sedimentary cascade through generally efficient lateral and longitudinal couplings despite local difficulties of sediment export due to decouplings or anthropogenic / environmental obstacles to sediment continuity.
This work follows three stages: i) identification and mapping of source zones, ii) characterisation of their nature and iii) estimation of their contribution to the sedimentary cascade.
The mapping of source zones was carried out using multispectral satellite remote sensing combined with a machine-learning method. The images used are from the Pleiades sensor, which provides a very high resolution (VHR) spatial (0.50 m with Pansharpening) and spectral (RGB NIR). The semi-supervised classification was performed by the Random Forest learning algorithm using the Dzetsaka extension in QGIS. Bare soils on slopes are interpreted as instability zones subject to lateral transfer. The typology of the source areas was carried out in two steps. Firstly, the nature of the mass movements and therefore of the source zones was characterised using expert analysis methods associated with morphological criteria (length/width ratio). Secondly, the contribution of the source zones to the sedimentary cascade was estimated using expert analysis methods and geomorphological criteria (e.g. types of contact, obstacles to sedimentary continuity, etc.).
The mapped sedimentary sources represent 6.7 km² and 3.5% of the study area (193 km²). There is a spatial concentration of source areas in the southwest of the study area. The machine error of the classification is about 97% and the operator error is 58%. The nature of the sedimentary sources is characterised by a broad spectrum of hydro-gravity processes that can be grouped mainly between slides/slope failures and flow-like landslide (e.g. debris avalanche, mud flow). These sedimentary sources actively contribute to the sedimentary cascade through generally efficient lateral and longitudinal couplings despite local difficulties of sediment export due to decouplings or anthropogenic / environmental obstacles to sediment continuity.