Landslide model evaluation

Introduction Conclusions References Tables Figures Back Close Full Screen / Esc Papers published in Hydrology and Earth System Sciences Discussions are under open-access review for the journal Hydrology and Earth System Sciences Abstract Introduction Conclusions References Tables Figures Back Close Full Screen / Esc Abstract The often-used success rate (SR) in measuring cell-based landslide model efficiency is based on the ratio of successfully predicted unstable cells over total actual landslide sites without considering the performance in predicting stable cells. We proposed a modified SR (MSR), in which we include the performance of stable cell prediction. The 5 goal and virtue of MSR is to avoid over-prediction while upholding stable sensitivity throughout all simulated cases. Landslide susceptibility maps (a total of 3969 cases) with full range of performance (from worse to perfect) in stable and unstable cell predictions are created and used to probe how estimators respond to model results in calculating efficiency. The kappa method used for satellite image analysis is drawn for 10 comparison. Results indicate that kappa is too stern for landslide modeling giving very low efficiency values in 90% simulated cases. The old SR tends to give high model efficiency under certain conditions yet with significant over-prediction. To examine the capability of MSR and the differences between SR and MSR as performance indicator, we applied the SHALSTAB model onto a mountainous watershed in Taiwan. Despite 15 the fact the best model result deduced by SR projects 120 hits over 131 actual landslide sites, this high efficiency is only obtained when unstable cells cover an incredibly high percentage (75%) of the entire watershed. By contrast, the best simulation indicated by MSR projects 83 hits over 131 actual landslide sites while unstable cells only cover 16% of the studied watershed.