Flash flood is an emergency at the hourly timescale
between a rainfall and the onset of flooding. Researchers assess flash flood hazard using conditioning factors, for instance, slope, soil type, land use type, flow accumulation value and topographic wetness index (TWI) driven by different characteristics of each place. This practice is attributable to dampen complex decision problems. While the former German flash flood studies mainly use inundation maps to illustrate flood hazard, we intended to offset this gap by introducing a new multicriteria decision making (MCDM) method with several conditional factors, applies and further extends frequency ratio (FR) based on standing water records in real events. Four indices are implemented for decision making: slope, land use, soil type and flow accumulation. Five types are classified within each index as statistical layers. We start by choosing 10 typical flash flood events, from 2002 to 2016, for 10 catchments distributed in various directions of Bavaria to represent flash flood in recent decades. Afterwards, these standing water record databases are calculated following frequency ratio method by counting record numbers in each index layer. The frequency ratio results are further applied as the database for three decision methods—weighting index, analytic hierarchy process (AHP) and decision trees. This integration forms three methods as abbreviations: FR-WI, FR-AHP and FR-DT. The weighting orders of indices are examined through each study catchment and the holistic Bavaria.
As a result, flash flood hazard maps are built up for visualization and comparison between conditioning factors, together with sensitivity analysis, calibration and validation. Subsequently, an evaluation framework of flash flood hazard, named as MCDM-FR, has been developed that not only deals with the catchment with recorded events but also hopes to serve flash flood hazard analysis for the holistic Bavaria. This study promotes the interaction between real events as inundation records, and the topographic/human factors as the four applied indices. Moreover, in the calculation of frequency ratio, by applying a much larger database size with two more order of magnitudes compared with former studies, the uncertainties due to biased pixel values could be narrowed.
