Kenfichirou KOSUGI, Atsuhiko KINOSHITA, Masamitsu FUJIMOTO, Takahisa MIZUYAMA and Yoshiki SANDO
Abstract
Distributed hydrological models, which calculate topographically]driven rainwater movement within a catchment, have been combined with mechanical analyses of slope stability, producing spatial and temporal variations in safety factors against slope failures. These methods have been used as practical tools in real]time warning systems for shallow landslide and debris flow hazards. This study examined accuracies and limitations of one of these methods by conducting intensive hydrological observations within a head]water catchmentiNishi'otafuku]Yama catchmentjlocated in the Rokko mountain range of southern Hyogo Prefecture. Observations of a discharge hydrograph, soil mantle groundwater levels, and bedrock groundwater levels indicated that a large amount of rainwater infiltrates into the bedrock and recharges bedrock aquifers. In downstream regions along a main hollow of the catchment, the bedrock groundwater exfiltrates into the soil mantle forming perennial and semi]perennial soil mantle groundwater and a large amount of base flow discharge. Thus, the recharging and discharging processes of the bedrock groundwater dominantly controlled hydrological phenomena of the catchment. The distributed model, which does not consider the recharging and discharging processes of the bedrock groundwater, produced inaccurate estimations of the observed discharge hydrograph and soil mantle groundwater levels. Nevertheless, the model produced reasonable decreases in the safety factor at some past landside locations where topographically]driven rainwater convergences were computed. However, the result did not necessarily guarantee accurate predictions of these landslides, since the model produced unsatisfying predictions of the soil mantle groundwater levels and timings and locations of landslide initiation were greatly controlled by soil internal frictions and cohesions. Thus, unless the hydrological model is validated by observed groundwater levels, computed safety factors should be treated as indices representing relative landslide vulnerabilities. For some past landslides located in downstream regions along the main hollow of the catchment, the model failed to detect decreases in the safety factor. Rainwater convergences by the bedrock groundwater exfiltration were suggested to be a controlling factor for the occurrences of these landslides, which clearly suggested limitations of the distributed hydrological model that calculates topographically]driven rainwater movement.
Key wordsFlandslide prediction, real]time hazard warning system, distributed hydrological model, groundwater