Spatial and scale-dependent variability in data quality and their influence on susceptibility maps for gravitational mass movements in soil, modelled by heuristic method - a case study in Lower Austria - Nils Tilch & Leonhard Schwarz Introduction and defining problems Results Conclusions For several years now, different methods have increasingly been used to produce susceptibility maps in regard to spontaneous gravitational mass movements in soil for regions of different size and different landscape areas. The quality of the produced susceptibility maps are highly dependent on the methods employed, particularly on the quality of the available data (process data and spatial data, for describing the spatial conditions). In general, the larger the area, the more (i) heterogeneous the partial data quality will be and (ii) the poorer the average data quality will be. A fundamental question is therefore whether susceptibility maps, modelled for large extended areas, are generally applicable to small-extended areas and large-scaled statements. The resulting susceptibility maps for areas of different size show that: A comparison between SAW-validations of different extended partial areas clearly demonstrates that there can be good SAW-validations of the respective large-extended area, but this is not necessarily the case in the smaller part-areas. In the case of large-extended areas in particular, this is due both to the spatial variability in the quality of the spatial data as well as the random and selective amount and quality of process data, which is not representative for all partial areas. This means that using and making statements based on susceptibility maps produced on a large extended area concerning issues for small-extended areas and large-scales is highly dubious and should probably be reassessed. In any case, the cell size of the modelled susceptibility maps should be adjusted to available spatial-data and process-data quality (for example during map calibration), and such adjustment should be oriented towards the partial areas of poorer data quality. This is highly important, in order to prevent misuses (for example invalid “zooming in”) and misinterpretations (for example, the derivation of quantitative statements). References: CHUNG, C.J. & FABBRI, A.G. (2003): Validation of spatial prediction models for landslide hazard mapping, in: Natural Hazards, 30, pp. 451-472 TILCH, N. & SCHWARZ, L. (2010): Erstellung von Dispositionskarten für Massenbewegungen – Herausforderungen, Methoden, Chancen, Limitierungen.- Vortrag Innsbrucker Hofgespräche 26.05.2010, Innsbruck. [ ] http://bfw.ac.at/050/pdf/IHG260510_Tilch_Schwarz.pdf Lower Austria (area 1) (large regional area extent of very different types of landscape, ca. 20.000 km²) The simple heuristic GBA-Method susceptibility map (cell size: 100 m) Catchment area of the “Klingfurther Bach” (area 3) (local area extent, landscape-specific, ca. 10 km²) “Bucklige Welt - Wechselland” (area 2) (smal regional area extent of different types of landscape, ca. 1000 km²) A simple heuristic Method (developed by GBA, see TILCH & SCHWARZ 2010) was used area-wide in Lower Austria (Area 1: ca. 20.000 km²), and in sub-areas of different size (Area 2: ca. 1000 km²; Area 3: ca. 10 km²) in order to produce susceptibility maps for spontaneous mass movements in soil. The identification of spatially variable susceptibility was done using area-wide data of the highest quality, which was available in each (sub)area. Internationally recognized methods were used for summarised area-wide validation of the maps produced. “Summarised area- wide validation” (SAW-validation) means one validation for the whole area, without distinguishing different partial sub-area validations. Method 0 0,25 0,5 1,0 1,5 2,0 km 0 2,5 5 10 15 km 0 5 10 20 30 40 50 km Legende susceptibility map (cell size: 500 m) susceptibility map (cell size: 20 m) area 1 area 2 area 3 parameter map „Geologie“ Gk200 (1:200,000) Gk50 (1:50,000) GK50 - modified (1:50,000) parameter map „forest “ ÖK50 (1:50,000) landuse map (1:10,000) parameter map „slope“ DHM10 (10 m) DHM10 (10 m) DHM1 (1 m) geology-oriented susceptibility-units slope intervalls process cadastre (cal. and val. data set) high susceptibility low middle Correction-map "forest" (stabilization due to Rooting (+5 °) landuse map (1:10,000) SAW-validation susceptibility class density of landslides [Number / km²] 0 1 2 3 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 percentage of area [%] 0 1 2 3 cumulative percentage of mapped landslides [%] cumulative percentage of study area from high to low susceptibility [%] susceptibility class cell size: 50 m cell size: 100 m cell size: 200 m cell size: 500 m Validation according CHUNG & FABBRI (2003) Validation according CHUNG & FABBRI (2003) 0 1 2 3 0 1 2 3 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 0 1 2 3 SAW-validation cumulative percentage of study area from high to low susceptibility [%] cumulative percentage of mapped landslides [%] density of landslides [Number / km²] percentage of area [%] susceptibility class susceptibility class cell size: 50 m 0 1 2 3 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 0 20 40 60 80 0 200 400 600 800 1000 0 1 2 3 SAW-validation susceptibility class susceptibility class percentage of area [%] density of landslides [Number / km²] Validation according CHUNG & FABBRI (2003) cumulative percentage of mapped landslides [%] cumulative percentage of study area from high to low susceptibility [%] cell size: 20 m Comparison of SAW-validations for the catchment area of the “Klingfurther Bach” (ca. 10 km²) partial susceptibility map for the catchment of the “Klingfurther Bach” (ca. 10 km²) Flächenanteil [%] Flächenanteil [%] 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 0 1 2 3 0 1 2 3 20 40 60 80 0 100 0 200 400 600 Legend Legend mapped landslide mapped landslide very low very low low low middle middle high high Legend mapped landslide very low low middle high 0 0,5 1,0 1,5 2,0 km partial susceptibility map for the catchment of the “Klingfurther Bach” (ca. 10 km²) 0 0,5 1,0 1,5 2,0 km Legend mapped landslide very low low middle high cumulative percentage of study area from high to low susceptibility [%] Validation according CHUNG & FABBRI (2003) cumulative percentage of mapped landslides [%] density of landslides [Number / km²] percentage of area [%] susceptibility class susceptibility class Modeling in area "Klingfurth" and the result for the area "Klingfurth (cell size: 20 m) Modeling in area "BUWELA" and the result for the area "Klingfurth (cell size: 100 m) Modeling in area "Lower Austria" and the result for the area "Klingfurth (cell size: 500 m) 20 40 60 80 0 100 recognition rate (=sensitivity) [%] Geological Survey of Austria - Department of Engineering Geology Neulinggasse 38, A-1030 Vienna Telefon: +43-1-7125674-392 Fax: +43-1-7125674-56 [email protected] www.geologie.ac.at European Geoscience Union General Assembly 2011 Austria,Vienna 04- 08 April 2011 Increasingly better SAW-validation results were achieved with decreasing regional extent and concomitant higher data quality (parameter maps (such as digital elevation model and geological map), completeness and localization accuracy of process data). For example, the recognition rate is 60% for Area 1, 50% for Area 2, and 89% for Area 3. Fundamentally, however, the validation results obtained for all three areas with different spatial extent can be assessed as good. This was only achieved because the cell size of the respective susceptibility maps was adjusted in each case during calibration to the existing quality of the parameter maps and process data. If the susceptibility maps for the larger, more extended areas (Area 1 and 2) are subjected to partial SAW-validation within the scope of the small-extended area (Area 3), the difference in quality between the susceptibility maps becomes very obvious: whereas the susceptibility map produced for the large-extended area provides consistently poor validation results, the results are reasonable for the medium-extended area. However large areas are also declared relatively stable, even though many gravitational mass movements have occurred there. Legend mapped landslide very low low middle high susceptibility maps of different extended and (non)landscape-specific areas