There is no standard method to create a landslide inventory dataset. Landslide inventory preparation aims to gather as much data as possible. A dataset includes various attributes, but some attributes may not be available for all landslides [14
]. Most landslides are compiled from different sources for different purposes. We adopted three methods in our study: the visual interpretation of Google Earth images, a literature search, and field mapping where each method has advantages and disadvantages. For example, we could detect the landslide location, type, dimension, and date in Google Earth mapping, while for the literature search, we could only obtain data that were recorded in the literature. In contrast, most attributes related to landslide inventories can be gathered in field mapping. The combination of these three methods can help to generate a landslide inventory of events that have occurred recently, in the past, and in inaccessible and remote areas.
In Google Earth, we adopted four criteria to detect landslides: change of vegetation in historical images; the slope and elevation of the area; morphological changes in the images; and the presence of debris [15
]. We considered the change in vegetation as the first indicator of landslides in historical images of Google Earth. Landslides remove vegetation, and this can be detected in pre-event and post-event Google Earth images. We checked the slope and elevation of the area in Google Earth using the Add Path tool. Next, we checked the morphological changes and presence of debris. The removal of vegetation can also occur in plain lands, but landslides cannot, which is why we included the slope and elevation in the criteria of detection. When all four criteria were met for a suspected area, we considered it to be a landslide. As above-mentioned, we had previously mapped the landslides from January 2001 to March 2017 and recorded the location, date, type, and dimensions of the landslide. The details of the landslide mapping in Google Earth are given in [5
]. We searched the existing literature and newspaper reports before the field mapping. CDMP-II (2012) [12
] and Rahman et al. (2016) [16
] provided the landslide locations, date, type, causalities, and triggers of landslides for the Chittagong Metropolitan Area (CMP), Cox’s Bazar, and Teknaf municipalities. CDMP-II (2012) [12
] did not provide the size of landslides, while Rahman et al. (2016) [16
] provided the size of the landslides. Newspaper reports (1980–2017), records of the Disaster Management Department of the People’s Republic of Bangladesh, and the Roads and Highways Department provided the landslide data of landslides that caused casualties and damage to roads. Based on these reports, we selected Rangamati, Bandarban, Khagrachari, and part of the Chittagong district for field mapping. We adopted participatory field mapping with the help of four field investigators to record the landslide locations and various attributes of landslides (Table 1
) including causalities, damages, and economic losses [5
]. From the literature search, we found the location of landslides, but we did not know the exact locations. Participatory field mapping helped us in this regard since local people knew the exact location of the landslides. The trigger of landslides and financial losses were detected by interviewing local people, government officials, and local political leaders. The damage intensities were identified based on the qualitative judgment of the field investigators (classified into three categories: high, medium, and low). We used measuring tapes and Global Positioning System (GPS) to measure the area of each landslide. Four well-trained field investigators were hired to collect the type, distribution, state, and water content of landslides through visual investigation using classification schemes outlined in [14
The final dataset was the compilation of the data gathered from field mapping, Google Earth mapping, and the literature search. We also combined the same landslide locations mapped by Google Earth and field mapping.