Abstract: Different landslide prediction models have the problem of predicting the same landslide in advance or delay, and the prediction accuracy is quite different. At present, the fusion multi-model early-warning method fails to fully reflect the characteristics of individual model prediction, and the fusion early-warning accuracy is not high. On the basis of analyzing the reliability of these models, two landslide occurrence membership functions of delay and advance were designed respectively. According to the principle of minimum risk, the membership function of landslide warning was determined, and multi-model fusion of landslide early warning was realized by fuzzy integral. Sixteen known landslides were used to evaluate the reliability of the prediction models, and the other four known landslides were used to carry out the fusion early warning experiment of the landslide three days before and one day before the landslide occurrence, and the false alarm rate of multi-model fusion early warning was reduced by 16.6% and 25% respectively, compared with the average false alarm rate of multiple single models. The experiment shows that using early warning membership function for multi-model landslide early warning can improve the early warning accuracy for about 20%.

Key words: landslide early warning, early warning membership function, fuzzy integral, multi landslide prediction model fusion