*2.2. Database for Analysis*

Building a chronology of landslides based on the historical records is the first stage of any landslide hazard study [33]. A landslide database for the research has been developed taking inputs from the Geological Survey of India [22], newspapers, state government reports [1,34], and from interactions with the people of the area. The dates of initiation of landslides were collected with a weekly accuracy, and the locations were collected with a spatial accuracy of nearest mentioned site from the reports. The database consists of the spatial (Figure 3) and temporal distribution of landslides and the typology.

**Figure 3.** Digital Elevation Model [35] of the Idukki district along with the spatial distribution of landslide locations and rain gauge stations (2010–2018).

The rainfall data of daily resolution from the year 2010 was collected from four rain gauge stations in the Idukki district, maintained by the India Meteorological Department (IMD) [36], for the analysis. The locations of rain gauge stations are given in Table 1. The monthly distribution of effective rainfall in the Idukki district from 2010 to 2018 is shown in the box plot shown in Figure 4.

The distribution of rainfall is not uniform throughout the district. In a long term rainfall analysis conducted by GSI, it was found that the average annual rainfall varies from less than 1000 mm in the northeast parts of Anamudi peak to around 5000 mm near Peermedu [18]. The four rain gauges from which we collected data are located at Thodupuzha, Peermedu, Idukki, and Munnar (Figure 3). The variation of annual rainfall from the four rain gauges and the district average is plotted in Figure 5. The differences in rainfall conditions will lead to over-estimation or under-estimation of the intensity and duration values if we consider the average rainfall. Hence the rainfall event associated with each landslide was found out based on the spatial distribution of the four rain gauges [37].


**Table 1.** Location of rain gauge stations.

**Figure 4.** Box and whisker plot with monthly distributions of rainfall in the Idukki district (2010–2018). The bottom and top lines indicate minimum and maximum values respectively and the line inside the box represents the median.

**Figure 5.** Variation of annual rainfall measured in four rain gauges during the study period.

Identifying a reference rain gauge is a challenging task as explained by many practitioners [14,38], especially when the number of available rain gauges is limited. One of the most common practices is to choose the rain gauge based on its proximity to the landslide location. Hence in this study, the district was divided into four Thiessen polygons, based on the location of rain gauges (Figure 6). P1 Polygon is occupied by a flat and plain territory, P2 is located in the eastern hilly sector of the study area, P3 represents the central hilly sector, and P4 contains the flanks of the mountain and the hills immediately at the foot of the mountainside, thus separating this area with peculiar physiographic characteristics from the other three. As a consequence, splitting up the area in four sectors by means of Thiessen polygons is better than operating considering the entire area as a whole.

**Figure 6.** Conceptual sketch showing development of dataset: P1, P2, P3, and P4 represent the four Polygons and R1, R2, R3, and R4 are the reference rain gauges in each polygon. D = Duration of rainfall (hours); I = Intensity of rainfall (mmh<sup>−</sup>1); L = Occurrence of landslide (Modified after [37]).

Each polygon defines a space, which is closest to the rain gauge in it (reference gauge). Each point inside a polygon is closer to the reference gauge, than the other three rain gauges. The division of polygons and the selection of reference gauge is constrained by spatial distribution only. Each polygon is assumed to be an area of similar rainfall conditions with a reference rain gauge.

The method of developing a dataset is illustrated in Figure 6 [37] using a sample dataset, i.e, the values (I,D) and the locations of landslides are not from the actual dataset, but are arbitrarily chosen for demonstrating the methodology. For all landslide events that happened in Thiessen polygon P1, the readings from R1 are considered. The procedure was same for all landslide events.

The readings corresponding to landslide events, recorded by individual reference rain gauges, were then merged to a single database. The exact number of triggered landslides and sites were not available from the reports and therefore multiple landslides on the same date within the same polygon are considered as a single landslide event. A threshold defines the possibility of occurrence of a minimum of one landslide event in the region. Thus, a total of 225 landslide events are considered in the present analysis, which happened during the time period of 2010–2018.
