1. Introduction
Economic impacts of landslides worldwide include damages to infrastructure, buildings, and homes costing hundreds of billions annually, and are forecasted to increase [
1,
2,
3,
4,
5,
6,
7,
8]. Annual landslide fatalities, often associated with large events such as rainstorms or earthquakes, vary significantly due to the vulnerability of elements at risk and hillslope development practices. In the United States, landslides occur in every state, causing billions of dollars in economic losses and estimates of an average of 25–50 fatalities annually [
9]. In the United States, damage from landslides is typically not covered under property insurance policies [
10]. The increasing number of landslides, and the resulting health and economic impacts, are compounding problems that call for not only more comprehensive landslide hazard (or susceptibility) assessment, but also socio-economic risk assessments [
11,
12].
The International Union of Geological Sciences (IUGS) Working Group on Landslides [
13] define landslide risk as the product of a hazard (the likelihood a landslide will occur) and exposure (the health, property, or environmental assets that might be diminished should the landslide occur). However, a range of complex risk modeling equations exist because there are numerous combinations of spatial and temporal inputs for assessment and mapping [
14,
15]. The exposure component can be reformulated to include both vulnerability and consequence (such as cost) so that risk becomes the product of hazard, vulnerability, and consequence [
14,
16,
17]. In reference to landslides, many authors practically define vulnerability as the likelihood of elements at risk of having an adverse result to landslide activity, intensity, and magnitude [
4,
16,
18,
19,
20,
21]. Consequence is the economic or societal loss expected should a landslide affect the asset.
The variability of terms and data inputs prompts differences in units of measure. For example, the hazard input may be a probability with units of 1/time, vulnerability is a probability of an asset being damaged (with no units), and consequence may be in terms of money. However, more importantly, each of the terms in the risk equation carries with it some degree of uncertainty that can arise from incomplete knowledge of landslide processes, triggers, and past occurrences. The uncertainty regarding hillslope soil and rock properties, hydrologic conditions, and landslide triggering mechanisms affects the way hazards are communicated to stakeholders and how stakeholders perceive the communication [
22]. Even when the landslide mechanisms are qualitatively similar, quantifying heterogeneous vulnerability data for different elements at risk makes risk mapping tenuous [
23,
24,
25]. These uncertainties are reflected in differences in the way government and private entities respond to landslides, and landslide mitigation practices that are available and affordable in different areas [
26,
27,
28,
29,
30].
Estimations of vulnerability are equally challenging because, in addition to understanding where a landslide will occur, a risk assessor must also be able to predict how far and fast the landslide will move and the complex behavior of people [
31,
32,
33]. Vulnerability is typically expressed on a scale of 0 (no loss) to 1 (total loss) [
4,
15,
34]. However, a lack of common language and data related to vulnerability poses many challenges because vulnerability is a multi-dimensional, dynamic, scalar, and community-driven concept [
21,
32].
The optimal risk approach finds the most useful combination of risk components and associated data, balanced against what is realistic to accomplish. Depending on data availability and quality, risk assessments fall into quantitative or qualitative approaches [
13]. A quantitative approach may contain extensive and accurate occurrence data, landslide magnitude or kinematics, fatalities, and other vulnerability (of property and people), and consequence data. [
34,
35,
36]. The hazard component may be a probabilistic, deterministic, or scenario-based model that evaluates slope stability, landslide initiation, potential runout, or frequency of occurrence [
28,
37,
38]. Even further, they may have magnitude, velocity, and frequency data associated with dynamic real-time rainfall and population location data, as opposed to static variables. Vulnerability and consequence data may distinguish among building types, market value of buildings, road types, road value, structure strength or resistance, persons in buildings, and loss of life considerations [
39,
40]. Many quantitative risk assessments are time and data-intensive and challenging to implement but may be able to narrowly focus on specific risk types such as societal, individual, financial, and health and safety [
4,
34]. The robust, data-intensive quantitative approaches still require expert experience in communication and risk management.
Qualitative risk assessments, in which vulnerability and consequence data are general or non-existent, equate risk with hazard [
13,
19]. The hazard input may be supported by expert knowledge, landslide inventories, national or global-scale elevation data, and subsequently derived geomorphological or topographic indicators such as slope steepness [
25,
41]. Qualitative assessments may involve a simple frequency analysis of past events or a broad intersection of asset and hazard inputs at a broad scale. Results are often presented as weighted indices, relative ranks of risk, or other qualitative descriptors [
18].
Furthermore, risk is influenced by economic, social, cultural, environmental, climatological, and political factors, that continuously shift perceptions of what is acceptable and tolerable [
8]. Even differences in terms and definitions of landslide types among geologists, engineers, and the public reflect the complexity of landslide processes, and consequently, the ways in which risk is communicated, understood, and managed [
42]. The range of landslide risk assessment approaches can vary based on the quality of available data, which can range from well-established knowledge to broad but geologically plausible assumptions in the absence of data [
4,
22].
Our study focused on evaluating a technical range of model inputs with the need for practical, cost-effective solutions for stakeholders like emergency managers, first responders, local officials, and residents and communities at-risk from landslides. We produced two static socioeconomic risk maps that consider hazard data limitations, as well as limited vulnerability and consequence data, in five eastern Kentucky counties with chronic landslide problems. Both approaches leverage existing remote sensing data and generalized infrastructure and land-use data useful for estimating exposure and consequence. One approach uses a robust landslide inventory, 1.5 m airborne lidar digital elevation models (DEM), and lidar-derived geomorphic datasets to model landslide susceptibility [
43]. High-resolution remote sensing data provides an opportunity to model relevant geomorphic conditions that lead to landslides at a regional scale.
Our second risk approach uses a coarse slope angle map sourced from global 30 m DEM as a hazard input in the risk equation. This model incorporates no landslide susceptibility data, simplified exposure data, but similar vulnerability and economic consequence data. The main purpose of the coarse, slope-based approach is to demonstrate the myriad of results possible at regional scale with limited quality data availability. Given our history of working with regional development groups, we also sought to determine levels on the quantitative risk assessment continuum that met the practical needs of stakeholders. While coarse slope-based models are not computationally intensive and are more readily accessible, we argue that further refinement of risk components creates more practical and uniform risk models while maintaining ease of access.
A practical and useful landslide risk assessment can be developed at a regional scale with limitations regarding hazard behavior and asset data. We reason that somewhere in between quantitative and qualitative risk assessments is a practical level, perhaps a model in which hazard and exposure data are well-established, yet still limited, compared to an advanced quantitative assessment. A static, geomorphic-based landslide hazard input, minimal to no vulnerability data, and consequence data constrained to economic loss will model a true risk assessment that is a useful, realistic combination of inputs. In fact, many approaches have effectively used a hazard input, yet are still limited with vulnerability and exposure data, often over-simplifying results. For example, the Federal Emergency Management Agency (FEMA) National Risk Index for 18 natural hazards, including landslides, leverages available source data for baseline risk for counties and census tracts [
17]. These risk approaches can effectively connect data and modeling needs, computational requirements, and expert knowledge, which makes this approach practical at various scales.
5. Discussion
The primary drawback for both quantitative and qualitative risk assessments is based on relevant degrees of data limitations and complex environmental conditions, and the uncertainties they introduce. Most advanced landslide hazard assessments require landslide frequency and run-out data that does not exist. Similarly, landslide vulnerability and consequence data regarding building types, road infrastructure, and personal safety is virtually non-existent in most places in the world. However, understanding that a range of quantitative risk assessments is a continuum that includes advanced, site-specific models to over-simplified modelling techniques is the foundation of developing a risk assessment that is useful for landslide-prone communities. We demonstrate a variety of methods, from conventional risk estimations to sparse, yet resourceful risk model inputs can effectively address data limitations and produce quality maps.
Our models emphasize a regional scale, limited data approach that provides risk information for imposing risk evaluation and mitigation strategies for communities. The hazard input (landslide susceptibility) to our risk equation highlights the importance of existing deposits that have a moderate to a high probability of subsequent movement and highlight other parts of the slope that do not necessarily show obvious landslide activity but are classified, nonetheless. The hazard input is limited regarding landslide type and behavior (extent or runout), however using static probability serves as a critical risk input. We obtained the economic values from various sources, and all were generalized as total values for the elements in question. Developed and undeveloped land values were determined from a small sample of property values. This analysis lacks data on other highly vulnerable elements, such as powerlines, water lines, and sewerage lines, therefore these elements are not included in the risk assessment. Population considerations did not include where populations would be at any given moment. Vulnerability was assumed at the maximum value (1), which is not likely the case uniformly across the study area. We submit that using a vulnerability value of 1, combats an underestimating of landslide hazard impacts and the often-related reduced awareness and concern [
65].
Our risk calculation and map derived from a coarse slope-based hazard input recognizes similarities in risk distribution at a regional scale, but also highlight the need for further evaluation of over-and under-predicted risk in several areas. These comparisons assume that our susceptibility-based landslide risk map is a strong model primarily because of the landslide susceptibility input, thus better constraining the areas vulnerable to landslides. A slope-based hazard input, limited exposure and consequence data, and limited computing power are a reality for developing countries striving for risk evaluation and implementation. The susceptibility map input is a superior input to slope angle, but minor GIS tasks (such as improving on census block data or how exposure is quantified) can significantly improve the risk calculation and results that make a practical difference in map quality.
Because we developed a socio-economic approach to risk, a recognition of how changing conditions and opportunities could impact community resilience in the long term need to be considered in future assessments. Additional data sets to consider in future risk mapping may include property value administrator data, traffic counts, cell phone locational data, geology, updated land class maps, and time-dependent rainfall. National Oceanic and Atmospheric Administration differenced 30-year averages (1991–2020 minus 1981–2010) which indicated Kentucky has experienced an increase in annual precipitation change across the state that ranges from 3–12% [
66]. Increased precipitation will translate to more landslides and increased risk. Incorporating precipitation data, rainfall triggering thresholds, and other related climate change factors may also improve risk assessments in the future.
Considering the variability in methods for landslide risk, establishing the utility of model and map results, in conversation with local stakeholders, is critical. Regardless of the robustness of data availability and model inputs, landslide risk mapping can provide baseline information for all stakeholders that show economic benefits, improves public safety, and builds trust. Our results contain data to inform mitigation strategies that could support building and infrastructure assessment, land-use planning, event awareness, response, and recovery actions for communities in the region.
6. Conclusions
We evaluated a susceptibility-based landslide risk map and a more limited, slope-based approach in order to emphasize how minor changes in data quality can improve landslide risk assessments. Minor changes in the hazard and vulnerability inputs result in significant changes in the quality and applicability of risk maps. All approaches and inputs of regional-scale landslide risk assessments have limitations and a recognition of data resources and the quality of model inputs allows for comparison of map results that can be used by practitioners and communities to mitigate landslide hazards and reduce risk.
Using different hazard inputs (landslide susceptibility versus slope angle), exposure data, and associated economic value of assets in a risk equation, we generated two landslide risk maps for five counties in a landslide-prone portion of eastern Kentucky. We used a logistic regression-based landslide susceptibility model as the hazard input. The elements at risk included population, road, railroad, and land class inputs, along with associated asset costs (consequence). The vulnerability input was assumed to be (1), modelling a total loss. For the slope-based map, the vulnerability remained (1) and the road, railroad, and land exposure asset raster maps were not included in the consequence component of the risk equation. However, these assets’ cost-per-pixel maps and cost data remained the same.
The susceptibility-based map indicates 64.1 percent of the study area is classified has moderate to high risk, with assets closer to high hazard areas being reliably highlighted as moderate to high risk. The map effectively highlights high-risk road segments, which is helpful to emergency managers, first responders, and local officials who need to communicate the threat of landslides. Broad, wide-open hillslopes and ridges with little to no infrastructure or other elements at risk are classified as low risk. Because of our asset density and hazard input (landslide susceptibility), the model over-predicted risk in some areas (compared to the slope-based map) particularly valley bottoms with dense areas of buildings or roads that are in close proximity to a toe slope or engineered embankment. These areas are mostly flat and have little correlated hazard.
The more data-limited risk assessment used a coarse (30-m) slope input and U.S. Census block group-derived population data, resulting in much less consistent distribution of a risk factor score. The map shows sharp boundaries between areas with moderate and high-risk and large areas of very low risk. These boundaries are coarse renderings of how buildings, roads, and railroads fall within risk classifications. Although identification of risk classes for local scale areas is possible with the slope-based map, the utility is limited for a broad five-county area.