Mapping geomorphology based on the information from existing geomorphological maps with a multiple-point geostatistics technique

Geomorphological maps are valuable tools for studying land surfaceprocesses. Information obtained from the field geomorphological maps canbe, in turn, used in mapping geomorphology at another area. In thisstudy, we present an application of the multiple-point geostatistics(MPG) technique in geomorphological mapping. This technique makes use ofthe field geomorphological maps, together with the topographical dataobtained from the Digital Elevation Model (DEM), to derive the knowledgeon the formation of different geomorphological units in the landscape(e.g. river terrace, alluvial fan, badlands) as a basis in mapping areaswith unknown geomorphology. This approach starts from characterizing theoccurrence of each geomorphological class as a function of land surfaceattributes (i.e. attribute pattern), which consists of DEM derivativesdiscretized into classes (e.g. slope class) observed at that celllocation, and geomorphology at multiple neighboring locations. Thelatter gives information on the spatial relation betweengeomorphological units. Number of cell occurrences per geomorphologicalclass per attribute pattern is counted and stored in the frequencydatabase, which will be subsequently used in the mapping. In the mappingstage, the algorithm assigns a realization of a geomorphological classto the target mapping cell, based on the probability function ofgeomorphological occurrences conditioned to the observed attributepattern at the target mapping cell, as retrieved from the frequencydatabase. The approach is tested to map the geomorphology in the 280 km2Buëch catchment, French Alps. We use 4 morphometric attributes,extracted from a 37.5-m DEM resolution (i.e. height above the nearestdrainage, slope gradient, profile curvature, and slope variability); and2 non-morphometric attributes (i.e. geomorphology observed at 1-cell and10-cells downstream from a cell of interest). Mapping is done usingdifferent frequency databases created from different training areas withsizes ranging between 7-28 km2 (2.5-10% of the mapping area). The MPGtechnique yields the geomorphological map with the highest cell accuracyof 51.2% evaluated against the field geomorphological map, using thetraining image size with 10% of the mapping area. The unit mapped withthe highest accuracy is the debris slope, while hogback and glacis weremapped with the lowest accuracy. The mapping accuracy is highest fortraining areas with a size of 7.5-10% of the total area. Reducing thesize of the training images resulted in a decreased mapping quality, asthe frequency database only represents local characteristics of thegeomorphology that are not representative for other areas. Increasingthe size of training images beyond this range may not considerablyincrease the mapping quality. This will, instead, result in a redundancyof information and more variations in geomorphological class occurrencesper attribute patterns in the frequency database, reducing thecapability to discriminate between geomorphological units.