**5. Results**

For the validation of results, rainfall and landslide data of 2016 and 2017 have been used. For each day of the validation period, the level of warning predicted by the decisional algorithm was compared with the reported landslide events according to the classical scheme of a confusion matrix, as shown in Figure 7. A confusion matrix is used to describe the performance of a decisional algorithm on a validation dataset for which the true values are known. The performance of an algorithm can be visualized using this matrix.

**Figure 7.** Confusion matrix for quantitative comparison.

Correct predictions can be both true positives (days in which the model forecasted correctly the occurrence of at least a landslide) and true negatives (days in which the model forecasted correctly that no landslides occurred). False negatives are those days in which the algorithm missed the alarms, and false positives are days in which the model issued false alarms. During 2016, eight shallow landslide events are reported in the region. The events were rapid in nature and happened to occur on a single day. Among the eight events, six were correctly predicted by the SIGMA model. Ground displacements were reported at two locations in the Chibo–Pashyor area during seven days in 2017: on 28th–29th July 2017 and 13th–17th August 2017 [2]. In all seven days, ordinary criticality was well-predicted in the present analysis. 55 false alarms were forecasted in a span of two years. An overview of the quantitative validation of the model for Kalimpong is tabulated in Table 2.


