Influence of Specific Contributing Area algorithms on slope failure prediction in landslide modeling

Abstract. This study anatomized algorithm effects of specific contributing area (SCA) on soil wetness estimation, consequently landslide prediction, in SHALSTAB. A subtropical mountainous catchment during three typhoon invasions is targeted. The peak 2-day rainfall intensity of the three typhoons: Haitang, Mindulle and Herb are 144, 248 and 327 mm/day, respectively. We use modified success rate (MSR) to retrieve the most satisfying mean condition for model parameters in SHALSTAB at three rainfall intensities and respective pre-typhoon NDVI themes. Simulation indicates that algorithm affects the prediction of landslide susceptibility (i.e. FS, Factor of Safety) significantly.

Based on fixed NDVI and the mean condition, we simulate by using full scale rainfall intensity from 0 to 1200 mm/day. Simulations show that predicted unstable area coverage increases non-linearly as rainfall intensity increases for all algorithms yet with different increasing trends. Compared to Dinf, D8 always gives lower coverage of predicted unstable area during three typhoons. By contrast, FD8 gives higher coverage areas. The absolute difference (compared to Dinf) in predicted unstable area ranges from ~−3% to +4% (percent watershed area). The relative difference (compared to Dinf) ranges from −15% to as high as +40%. The maximum absolute and relative differences in unstable area prediction occur around the condition of 100–300 mm/day, which is common in subtropical mountainous region.

Theoretical relationship among slope, rainfall intensity, SCA and FS value was derived in which FS values are very sensitive to algorithms in the field of slope from 37 to 52degree. Results imply any comparison among SCA-related landslide models or engineering application of rainfall return period analysis must base on the same algorithm to obtain comparable results. This study clarifies the SCA algorithm effect on FS prediction and deepens our understanding on landslide modeling.