Landslide hazard zoning using Shannon Entropy and Information Value models in GIS environment - Case study: East Rudbar-e Alamut District-Qazvin Province

Landslide is one of the most important types of natural disasters,which endangers lives and financial security of many people and destroys environment and natural resources.With the present population growth and expansion of urban areas towardsteep areas and hillsides, landslide-related losses can be catastrophic. For an instance, landslides in Badakhshan Province in Afghanistan killed around 2,700 people in 2014, and a landslide in China (Shan’xiprovince)resulted in the disappearance of 64 people in 2015.Therefore, assessingthe possibility of landslides occurrence seems to becrucial. Providing zoning maps is one of the measures which makes identification of areas prone to future landslides possible. Inferences drawn from these maps can be used for land use planning, prevention of unauthorized construction activities, infrastructure development, refurbishment and restoration.

The present research selectsEast Rudbar-e Alamut (a district of Qazvin province), which is affected by landslides and instability of hillsides, as the study area. It takes advantage of Shannon entropy and information value models to develop landslide susceptibility map of the study areain GIS environment.Shannon entropy theory has been used in extensive researcheswith the aim of prioritizinginfluential factors in the probable occurrence of natural disasters such as landslide. Information value (IV) model is one of the statistical models drawn from information theory with a widespread application in the modeling of geological hazards and disaster risk assessment. Information value model aims to find a combination of significant factors anddeterminetheir impacton theoccurrence oflandslide in an area.To implement this model, relevant data and its related criteria maps were prepared. In this regard, the location of previous landslide events in the study area was determined based on the information received from Forests, Range and Watershed Management Organization. 49 landslides were identified in this way. Then, data was randomly divided into 2 categories: training data and validation data. Thus, 70% of data (35 landslides) were used to produce the models and the remaining 30% (14 landslides) were used for validation purposes. In addition to previous landslides, a collection of topographic, environmental and climatic characteristics of the study area including seven criteria of lithology, slope, distance from faults, land use, precipitation, slope-direction and elevation were selected as the most effective independent variablesto produce criteria maps with 30-meter spatial resolution. Basic information used to map these seven influential factors was obtained from Forests, Range and Watershed Management Organization, as well as the SRTM Digital Elevation Model (DEM), and used after some modifications. Considering the capability of ArcGIS in spatial data analysis, thissoftwarewas used to produce information layers and implement the models.

Prioritizing influential factors using Shannon entropy model introducesthree factors (i.e. land use, elevation and precipitation)as the most significant factorsin the occurrence of landslides in the study area. Factors of slope angle, distance from faults (almost equal to slope angle), lithology and slope-direction were in the next influential factors.Also, results of information value model indicate that looking from lithology perspective, the category of marl, calcareous sandstone, sandy limestone and minor conglomerate has an information value of 1 and thus, the highest probability of landslide occurrence. Category of basaltic volcanic rocks, along with category of well bedded green tuff and tuffaceous shale have the lowest probability of landslide occurrence with information values ​​of -2.03 and -1.70, respectively.Only two categories of theslope angle criterionhave a positive-index. The highest information value (0. 93) in this category occurs in the class of 5-12 degrees, followed by the class of 12-20 degrees. The lowest information value occurs in slopes of more than 30 degrees. Based on this observation, it can be clearly concluded that the slope angles of 5 to 20 degrees are most prone to landslides. Distance to faults criterion indicate that the category of500 to 1000-meter distance to faultshave the highest information value (1.67). Regarding land use criterion, three land uses of garden, agriculture and garden-agriculture have the highest information values ​of 2.16 and 1.59 and 1.11, respectively. Regarding precipitation, average annual rainfall of less than 400 millimeters have the highest information value (1.50). Regardingslope-direction criterion, most landslides occur in southwest, south and eastdirections.Northeast, west, and northwest directions have the lowest probability of landslide occurrence, respectively. In terms of elevation, the information value is reduced as the height increases, and the maximum information value is related to the elevations of less than 1200 meters.After assigning a weight to each criterion and related classes, the landslide risk zone map was generated based on Shannon entropy and information valuemodels. The resulting zoning map produced based on natural breaks methods dividesthe area into five classeswith very high, high, moderate, low and very low risk. Resultsof Shannon entropy modelindicate that out of 14 landslides considered as the validation data, 3, 7, 2, 1, 1 landslideshave occurred in very high, high, moderate, low and very low risk zones, respectively. Resultsof the information value modelindicatethat 8, 4, 0, 1, 1 landslideshave occurred in very high, high, moderate, low and very low risk zones, respectively.

Evaluation of results using experimental probability index indicates that with 86% experimental probability,both models of Shannon entropy and information value are effective inidentification of landslide hazard in the East Rudbar-e Alamut region. Also, considering the number of landslides in very high and high risk zones, Shannon entropy and information value modelshave an experimental probability index of 72% and 86%, respectively, which prove higher efficiency of information value model. In Shannon entropy model, total area of very high, high and moderate risk zones covers 34% and 56% of the study area,respectively. In information value model,total area of very high and high risk zones covers 20% and 29% of the study area, respectively. Based on the landslide risk zone map, high and very high risk zones are mainly located in the west of the study area