This article presents a geographic information system (GIS)-based artificial neural network (GANN) model for flood susceptibility assessment of Keelung City, Taiwan. Various factors, including elevation, slope angle, slope aspect, flow accumulation, flow direction, topographic wetness index (TWI), drainage density, rainfall, and normalized difference vegetation index, were generated using a digital elevation model and LANDSAT 8 imagery. Historical flood data from 2015 to 2019, including 307 flood events, were adopted for a comparison of flood susceptibility. Using these factors, the GANN model, based on the back-propagation neural network (BPNN), was employed to provide flood susceptibility. The validation results indicate that a satisfactory result, with a correlation coefficient of 0.814, was obtained. A comparison of the GANN model with those from the SOBEK model was conducted. The comparative results demonstrated that the proposed method can provide good accuracy in predicting flood susceptibility. The results of flood susceptibility are categorized into five classes: Very low, low, moderate, high, and very high, with coverage areas of 60.5%, 27.4%, 8.6%, 2.5%, and 1%, respectively. The results demonstrate that nearly 3.5% of the study area, including the core district of the city and an exceedingly populated area including the financial center of the city, can be categorized as high to very high flood susceptibility zones.

Debris flows pose severe hazards to communities in mountainous areas, often resulting in the loss of life and property. Helping debris-flow-prone communities delineate potential hazard zones provides local authorities with useful information for developing emergency plans and disaster management policies. In 2003, the Soil and Water Conservation Bureau of Taiwan proposed an empirical model to delineate hazard zones for all creeks (1420 in total) with potential of debris flows and utilized the model to help establish a hazard prevention system. However, the model does not fully consider hydrologic and physiographical conditions for a given creek in simulation. The objective of this study is to propose new approaches that can improve hazard zone delineation accuracy and simulate hazard zones in response to different rainfall intensity. In this study, a two-dimensional commercial model FLO-2D, physically based and taking into account the momentum and energy conservation of flow, was used to simulate debris-flow inundated areas. Sensitivity analysis with the model was conducted to determine the main influence parameters which affect debris flow simulation. Results indicate that the roughness coefficient, yield stress and volumetric sediment concentration dominate the computed results. To improve accuracy of the model, the study examined the performance of the rainfall-runoff model of FLO-2D as compared with that of the HSPF (Hydrological Simulation Program Fortran) model, and then the proper values of the significant parameters were evaluated through the calibration process. Results reveal that the HSPF model has a better performance than the FLO-2D model at peak flow and flow recession period, and the volumetric sediment concentration and yield stress can be estimated by the channel slope. The validation of the model for simulating debris-flow hazard zones has been confirmed by a comparison of field evidence from historical debris-flow disaster data. The model can successfully replicate the influence zone of the debris-flow disaster event with an acceptable error and demonstrate a better result than the empirical model adopted by the Soil and Water Conservation Bureau of Taiwan.

In this study, the land subsidence in Yunlin County, Taiwan, was modeled using an artificial neural network (ANN). Maps of the fine-grained soil percentage, average maximum drainage path length, agricultural land use percentage, electricity consumption of wells, and accumulated land subsidence depth were produced through geographic information system spatial analysis for 5607 cells in the study area. An ANN model based on a backpropagation neural network was developed to predict the accumulated land subsidence depth. A comparison of the model predictions with ground-truth leveling survey data indicated that the developed model had high accuracy. Moreover, the developed model was used to investigate the relationship of electricity consumption reduction with reductions in the total area of land with severe subsidence (> 4 cm per year); the relationship was approximately linear. In particular, the optimal results were obtained when decreasing the electricity consumption from 80 to 70% of the current value, with the area of severe land subsidence decreasing by 13.66%.