Geospatial Landslide Risk Mapping Using AHP and GIS: A Case Study of the Utcubamba River Basin, Peru

Abstract

This study examines the use of a spatial multi-criteria approach based on GIS and AHP techniques to model landslide risk in the Utcubamba river basin, Peru. The methodology consisted of selecting twelve triggering variables: slope angle, geology, precipitation, distance to faults, drainage density, TWI, relative relief, profile curve, land use, elevation, distance to roads, and distance to population centers. These variables were then analyzed using the AHP method and then integrated into a GIS environment, where the weighted linear combination (WLC) method was used to map landslide risk. The risk was categorized into five classes, ranging from very low (1) to very high (5). The main results indicate that 32.81% of the area analyzed in the Utcubamba river basin presents a high and very high risk of landslides. The high-risk areas are mainly located in the southern part of the basin and coincide with areas with steep slopes, high rainfall, and proximity to population centers or communication routes. The model generated was highly accurate (AUC of 0.82), confirming that the integration of the AHP method with GIS allows for the precise identification of critical areas, which is useful for territorial planning, the prioritization of interventions, and emergency management, making it a reliable and replicable methodology in other parts of Peru.

1. Introduction

Natural hazards such as earthquakes, landslides, hurricanes, land erosion/degradation, and tsunamis have caused significant human losses, property damage, loss of agricultural land, and damage to the environment [1]. Globally, the frequency of geological and hydrometeorological disasters has increased in the last two decades, generating highly significant consequences [2]; in addition, changes in the global climate and microclimate, population growth, urbanization, the expansion of communication networks, and other infrastructure developments in areas with fragile slopes have increased landslide-induced risk, affecting both society and the economy [3,4,5].

Natural or human-made disasters are phenomena that can affect large areas and have numerous environmental, social, and economic impacts. Landslides are among the major large-scale disasters that can affect both the natural environment and urban areas [6]. The term landslide is defined as the movement of a mass of rock, debris, or earth down a slope under the influence of gravity [7] and is the result of a combination of several causal factors, such as steep slopes, uneven terrain, the absence of vegetation, loss of forest cover, unconsolidated material, rock erosion, frequent seismic activity, and heavy and prolonged rainfall [8]. In natural areas, landslides can alter landscapes, disrupt ecosystems, and affect wildlife habitats [9]. They occur during the gravitational displacement of soil or rocks destabilized by climatic, geomorphological, or geological natural phenomena or by human activities [10]. These factors are among the geological hazards that cause significant damage, being responsible for approximately 17% of the total number of deaths caused by natural disasters in the world [11,12].

In the context of landslides, physicists define a natural hazard as the probability that a reasonably stable condition will change abruptly or as the probability of occurrence of a potentially damaging phenomenon within a given area and within a given period of time [13,14]. Related to this, landslide risk refers to the potential loss of population and economic activity caused by landslide disasters over a certain period of time [15,16]. On the other hand, landslide susceptibility seeks to identify the areas most vulnerable to future landslides, based on the local physical conditions that determine the propensity to landslide activity [17,18]. Based on these concepts, landslide susceptibility is an important topic mainly because its geospatial analysis provides a useful tool for planning, disaster management, and risk mitigation [18]. However, in Peru, there are still gaps in the generation of risk maps, which limits their effective application in territorial planning and disaster management.

In Peru, landslides and other geological hazards represent a constant threat to the various departments, causing human, material, and economic losses over the years. The difficult geography of this country, characterized by steep slopes, varied geological conditions, and extreme climatic events, places the country in a highly vulnerable situation [19,20]. In Peru, landslides occur mainly on the flanks and steep slopes of coastal valleys and high jungles, as well as on coastal cliffs and road cuts. These events are influenced by specific lithological conditions, intense rainfall, runoff, and terrain inclination, while other geological hazards, such as landslides, rock falls, and slope erosion, occur in most of the country’s 106 watersheds.

Landslide risk assessment is fundamental to identifying and quantifying high-risk areas, vulnerable communities that allow for the planning and implementation of disaster risk reduction strategies and, therefore, the mitigation of landslide disasters [21]. In this sense, remote sensing offers a wide range of techniques and geospatial data, which, integrated with Geographic Information Systems (GIS), facilitate risk assessments at local and regional scales [22,23,24]. For this reason, both remote sensing and GIS have been widely used in various parts of the world to develop studies related to land movement in order to generate valuable information for decision makers.

Using a geographic information system (GIS), several methods have been proposed to assess susceptibility to landslides [25]. A frequently used method for mapping and zoning landslide hazards [26] is the multi-criteria decision analysis technique, or analytical hierarchical process (AHP), developed by Saaty in the 1970s [27]; this model combines qualitative and quantitative methods to decompose a complex decision problem into different levels [28]. AHP is particularly suitable for structuring complex problems and ensuring consistency in decision making [29], as it combines multiple factors or criteria to assess the likelihood of landslides occurring in a particular area.

Previous studies have used multi-criteria techniques to estimate landslide susceptibility and risk using GIS and remote sensing-based methodologies; e.g., Bhat et al. [30] generated a landslide hazard map along the Mughalen Road in India; Subedi et al. [31] analyzed landslide hazards in the Gandaki province with AHP and GIS; Das et al. [32] mapped the landslide susceptibility (LSZ) in the Kalimpong region of India using AHP. These works combine qualitative approaches based on expert opinion, including geomorphological analysis and inventory methods [7] with quantitative methods that establish mathematical relationships between causal factors and landslide occurrence, such as the studies by Panchal and Shrivastava [21] and Erener and Düzgün [33], who analyzed landslide hazards in Gandaki province using AHP and GIS.

Methods based on landslide inventories have also been implemented that, through classification and weighting techniques such as the Analytic Hierarchy Process (AHP), the Analytic Network Process (ANP), and fuzzy logic, allow for the elaboration of semi-quantitative risk maps [10,30,34]. In particular, the AHP method facilitates the quantification of the opinions of a group of specialists and their transformation into a coherent decision model through consistency measures [35,36,37,38].

In Peru, studies on landslides incorporating advanced technologies are scarce; however, those that have been carried out have undergone significant development, covered various regions of the national territory, and used increasingly innovative methodologies. For example, Aro et al. [39], in a study conducted in Quilahuani (Tacna), analyzed the influence of irrigation water and precipitation on slope stability, identifying how the increase in the degree of saturation reduces the safety factor in areas with active landslides. Anchivilca y Villegas-Lanza [40] used advanced geodetic techniques, including ground-based radar (GBSAR), satellite radar images, and GPS measurements, in Cuenca (Huancavelica) to characterize the surface deformation of a landslide that has affected infrastructure, roads, and agricultural fields for more than a decade. Although landslides are catastrophic events and are frequent in Peru, the lack of studies on the subject is alarming.

In this research, the aim is to use the geospatial database collected from various platforms and integrate it into a semi-quantitative approach to assess the risk of landslides in the Utcubamba river basin in the department of Amazonas, Peru. In this basin, there has been a progressive development of anthropogenic activities that, although necessary for socioeconomic development, have brought with them environmental problems, such as the alteration of the natural stability of slopes, in addition to the rugged topography, unfavorable geological conditions, and intense rainfall regime that characterize the Andean–Non-Amazonian transition and influence the occurrence of landslides.

To date, no comprehensive landslide risk assessment has been conducted in this basin. Despite the increasing frequency of these events and the associated damage, aggravated by climate change, a risk assessment that integrates all the influencing factors has yet to be developed. In this context, it is essential to prepare a precise map that identifies the areas at greatest risk in the Utcubamba river basin. For this purpose, in this study, Geographic Information Systems (GIS) were used together with the Analytical Hierarchy Process (AHP), which allowed for the appropriate weighting and classification of the different factors that influence the occurrence of landslides.

2. Materials and Methods

2.1. Study Area

The Utcubamba river basin is located in the department of Amazonas, in northeastern Peru, and is one of the main freshwater bodies of this department (Figure 1). The Utcubamba River, the main channel of this river basin, has an average length of 253 km [41]. It also connects with the Chinchipe and Marañon rivers at the Pongo de Rentema, located in the province of Bagua, at 360 m above sea level [42]. This basin has a varied relief, characterized by steep mountains at the headwaters and lower hills at its mouth, with a maximum altitude of 4130 m.a.s.l. and a minimum altitude of 360 m.a.s.l. In terms of climate, there is a marked variation according to altitude: in the lower basin, the average annual temperature reaches 29 °C, while in the upper basin, it is reduced to 14 °C. Moreover, the Utcubamba watershed harbors remarkable biodiversity, particularly in its riparian forests, which show great diversity and heterogeneity due to their irregular tree distribution. These riparian ecosystems prove fundamental as biological corridors and natural barriers that protect the integrity of watercourses. There is also a decreasing trend in the quality of the gallery forest, with the highest values recorded in the tributaries, from the upper to the lower section of the basin.

2.2. Variable Selection and Preprocessing

In order to develop the landslide risk assessment model for the Utcubamba river basin, a total of 12 variables that interact in a complex manner were integrated, as they play critical roles in the stability of land slopes. These variables were grouped into five factors according to their nature: topographic, geological, hydrological, climatic, and anthropogenic. All these variables together determine an initial predisposition to soil instability [43] and, in extremely negative conditions, can trigger landslides [44]. The variables were generated and processed in the software QGIS (v. 3.36.3), where they were cropped according to the extent of the study area with the extraction tool and resampled to 30 m with the resampling tool of the QGIS bilinear algorithm.

2.2.1. Topographic Factor

Topographic variables are of great importance in landslides, as they directly influence the stability of the soil [45,46]. The slope of the terrain influences elevation changes and, consequently, surface and subsurface water movement; slopes with a greater angle favor greater accumulation and runoff and increase erosion and the likelihood of landslides [47,48,49]. Elevation affects local relief, erosion rate, vegetation cover, and rainfall intensity, especially in mountainous areas, as high elevation areas are more prone to landslides [50,51]. Profile curvature directly influences hydrological conditions, erosion, sedimentation rate, and soil properties, factors that affect slope stability [32,52]. Relative relief classifies the relief (flat, undulating, hilly, and mountainous), and the higher the relative relief, the more prone the areas are to landslides [49,53], this variable involved using interpolation IDM [32]. The topographic variables were obtained in QGIS software (v.3.36.3) using a Digital Elevation Model (DEM) from the Shuttle Radar Topography Mission (SRTM) [54] with a 30 m resolution.

2.2.2. Geologic Factor

Two variables were considered: The first is geology, which is widely used in landslide risk research, mainly in mountainous areas [55], as it influences slope stability and landslide risk due to variations in erosion resistance between different rock and soil types [30,31]. This variable was obtained from the geology data available in the Geoserver of the Ministry of Environment—MINAM [56]. The other variable is proximity to fault lines, which increases the risk since cracks, faults, and fissures raise the pore pressure in the rock and soil mass, inducing instability—landslides are observed to occur more frequently along these structures [11,57,58,59]. A spatial database obtained from the GEOCATMIN Portal version 3 (https://geocatmin.ingemmet.gob.pe/geocatmin_v3/, accessed on 15 January 2025) [60] was used to elaborate this variable, which was exported to QGIS for processing. The process was performed in QGIS software (v.3.36.3) using the Euclidean distance function. The Euclidean distance is expressed by the following mathematical equation:

$$d_{E }\left(P_1,P_2\right) = \sqrt{{(X_2 - X_1)}^2 + {(Y_2 - Y_1)}^2}$$

where d_E is the Euclidean distance, P₁ is the starting point, P₂ is the end point, and X₁, X₂, Y₁, and Y₂ are Cartesian coordinates.

2.2.3. Hydrologic Factor

Drainage density is an important variable, as it identifies the distribution of drainage networks (streams and rivers) [31]; high-density areas favor soil erosion, weakening soil stability and increasing the risk of landslides [21,61,62]. This variable was generated from the drainage network of the basin with the Digital Elevation Model (DEM) of the Shuttle Radar Topography Mission (SRTM) with a resolution of 30 m, downloaded from the portal Earth Explorer del USGS [54]. The Topographic Wetness Index (TWI) indicates the accumulation of flow at any point in the study area by determining soil moisture [63]; areas with high TWI indicate higher surface saturation in large areas with gentle slopes, while low values are associated with steep slopes and an efficient drainage system [32]. This increased saturation decreases the shear strength of the materials in the slope, thus increasing the likelihood of landslides [64]. The TWI was processed in QGIS software (v 3.36.3) by applying the following formula:

$$TWI=ln(A/tan\beta )$$

2.2.4. Climatic Factor

Pluviometric precipitation is considered the trigger for landslides, as rainwater infiltration increases the pore pressure in the soil [31,65]. The rainfall variable was generated from raster data obtained from the WorldClim database (v.2.1) (www.worldclim.org/data/bioclim.html, accessed on 29 January 2025) [66], with a spatial resolution of 30 s (approximately 1 km²). Subsequently, in QGIS software (v 3.36.3), resampling was performed with the bilinear algorithm to have a spatial resolution of 30 m.

2.2.5. Anthropogenic Factor

Land use has an indirect influence on landslides since areas with more vegetation have less erosion and slopes are more stable [31,67]. The land use map of the study area was derived from a pre-existing map for the Amazon region, available at the Geoportal of the Peruvian Ministry of Transport and Communications [68]. The distance to roads is a determining factor in the occurrence of landslides since their construction alters the topography and reduces the shear strength at the top of the slope, generating tensile stresses; it also facilitates the infiltration of water into the slopes and adds additional loads due to traffic, compromising the stability of slopes that were originally stable [31,69]. The spatial database of the road network of the Ministry of Transport and Communications (MTC) was used to generate this variable [68] in Peru. The variable distance to population centers is understood as a spatial indicator of geographical exposure to hazards [70]; this variable is understood as the physical proximity of human settlements to areas with high landslide risk [71]. Distance is a quantifiable parameter of high availability and spatial resolution [71,72], widely used in the spatial modeling of landslide risk because of its ability to indirectly reflect the degree of potential interaction between the hazard and inhabited areas [73]. The interaction between natural, anthropogenic, and socioeconomic factors is intensified in urbanized areas, where unplanned urban expansion, deforestation, infrastructure construction in steep slope areas, and the lack of effective public policies significantly increase the risk for the populations closest to landslide zones [74,75]. This variable was generated from the spatial database of population centers of the National Geographic Institute (IGN) of Peru [76]. The variables were generated using Euclidean distance in QGIS software (v 3.36.3).

2.3. Assignment of Risk Categories

To carry out the risk assessment model, a classification of each of the study variables was made using the “Reclassify” tool in QGIS (v 3.36.3) (Figure 2). For this purpose, the initial data were reclassified into five risk categories: very low, low, moderate, high, and very high (

Table 1. Risk classes used in the study of the variables.

Variables	Risk Classes
	Very High	High	Moderate	Low	Very Low
Slope angle (P)	>41°	41–29°	29–20°	20–11°	11–0°
Geology (G)	Mitu (M)—River deposits (RD)	River deposits (RD)—Population centers (PC)	Population centers (PC)—Cajamarca formation (CF)	Cajamarca Formation (CF)—Inguilpata Formation (IF)	Inguilpata Formation (IF)—Lavasen Formation (LF)
Precipitation (Pc)	>1261 mm	1261–1094 mm	1094–941 mm	941–774 mm	774–424 mm
Distance to Faults (Df)	8–1774 m	1774–3895 m	3895–6794 m	6794–10,754 m	>10,754 m
Drainage Density (Dd)	>0.001047 m/m2	0.001047–0.000830 m/m2	0.000830–0.000670 m/m2	0.000670–0.000491 m/m2	0.000491–0.000156 m/m2
TWI	>11.85	11.85–10.756	10.76–9.87	9.87–9.06	9.06–6.11
Relative Relief (Rr)	>506 m	506–408 m	408–324 m	324–235 m	235–30 m
Profile Curve (Pc)	Strongly concave	Moderately concave	Flat/Neutral	Moderately convex	Strongly convex
Land use (Lu)	Subsistence Agriculture (SA)—Colan Mountain Range (CMR)	Coffee predominance (CP)—Subsistence agriculture (SA)	Population centers (PC)—Protection lands (PL)	Protection lands (PL)—Predominantly coffee-growing (PCG)	Private conservation (PrC)—Population centers (PC)
Elevation (E)	>3085 m.a.s.l.	3085–2537 m.a.s.l.	2537–1935 m.a.s.l.	1935–1156 m.a.s.l.	1156–325 m.a.s.l.
Distance to Roads (Dr)	3–1477 m	1477–3621 m	3621–6233 m	6233–9448 m	>9448 m
Distance to Population Centers (Dpc)	18–2206 m	2206–3990 m	3990–5890 m	5890–8307 m	8307–14,755 m

). This was performed using the Jenks natural breaks technique, which seeks to maximize the clustering of data by minimizing the variance within each category and maximizing the variance between categories [30,32].

2.4. Analytic Hierarchy Process (AHP)

The AHP is based on paired comparisons between criteria using a numerical scale from 1 to 9 to quantify the relative importance of each (Supplementary Table S1). This scale includes intermediate values (2, 4, 6, 8) to capture nuances in the preferences of the decision-makers [77].

To comparatively evaluate the variables of the model, a 12 × 12 matrix was constructed (Supplementary Table S2), in which each pair of variables was analyzed according to their relevance in the landslide risk assessment. After completing this pairwise comparison (Supplementary Table S2), a normalized matrix was generated that illustrates the relative importance of each variable to the others (

Table 2. Pairwise comparison of normalized AHP matrix.

Variables	P	G	Pc	Df	Dd	TWI	Rr	Pc	Lu	E	Dr	Dpc
Slope angle (P)	0.28	0.40	0.36	0.29	0.21	0.14	0.22	0.25	0.20	0.20	0.19	0.14
Geology (G)	0.09	0.13	0.24	0.19	0.27	0.19	0.09	0.11	0.09	0.11	0.06	0.06
Precipitation (Pc)	0.09	0.07	0.12	0.19	0.21	0.14	0.18	0.11	0.14	0.20	0.19	0.10
Distance to Faults (Df)	0.09	0.07	0.06	0.10	0.14	0.14	0.09	0.15	0.11	0.07	0.15	0.18
Drainage Density (Dd)	0.09	0.03	0.04	0.05	0.07	0.24	0.18	0.11	0.14	0.09	0.08	0.14
TWI	0.09	0.03	0.04	0.03	0.01	0.05	0.13	0.11	0.09	0.11	0.06	0.12
Relative Relief (Rr)	0.06	0.07	0.03	0.05	0.02	0.02	0.04	0.07	0.11	0.07	0.06	0.06
Profile Curve (Pc)	0.04	0.04	0.04	0.02	0.02	0.02	0.02	0.04	0.06	0.07	0.06	0.06
Land Use (Lu)	0.04	0.04	0.02	0.02	0.01	0.02	0.01	0.02	0.03	0.07	0.06	0.06
Elevation (E)	0.03	0.03	0.01	0.03	0.02	0.01	0.01	0.01	0.01	0.02	0.04	0.04
Distance to Roads (Dr)	0.03	0.04	0.01	0.01	0.02	0.02	0.01	0.01	0.01	0.01	0.02	0.04
Distance to Population Centers (Dpc)	0.04	0.04	0.02	0.01	0.01	0.01	0.01	0.01	0.01	0.01	0.01	0.02

).

The consistency of the AHP method and the robustness of the results were determined using the consistency ratio (CR) to assess the quality of the pairwise comparisons performed [31]. If the CR values are less than 10%, consequently, the matrix is consistent [21]. The consistency of the relative importance weights assigned during pairwise comparison can be verified using the equation shown below:

$$Consistency ratio \left(CR\right) = {\displaystyle \frac{CI}{RI}}$$

where CI is the consistency index while RI is the random consistency index.

The CI was calculated as follows:

$$Consistency index \left(CI\right) = {\displaystyle \frac{{\mathsf{\lambda }}_{max} - \mathrm{n}}{n - 1}}$$

where λ_max = higher eigenvalue and n = matrix order.

The values of the random consistency index (RI) are proposed and standardized by Saaty (Table S3) and the value of RI is established according to the number of variables used in the AHP process [78,79]. In this study, we used 12 variables; therefore, the value of RI is 1.54.

2.5. Weighted Linear Combination (WLC) for Landslide Risk Map

Finally, using QGIS software (v3.36.3), the landslide risk map (LRM) for the Utcubamba river basin was generated using the weighted linear combination (WLC), which is widely used in studies with the AHP method [25,80,81]. Weighted linear combination (WLC) combined with the AHP method is widely used in landslide susceptibility mapping, as it can incorporate several conditioning variables by assigning weights obtained from expert judgment [82,83]. Moreover, it is easily developed in GIS environments [84,85]. Recent studies in Ethiopia and China yielded evidence that when integrating between 11 and 12 variables and validating with AUC–ROC, AHP integrated to WLC performs well (AUC ≈ 0.75–0.82) and is comparable to more complex techniques [80,83,86]. Because of this, the integration of these two methods is useful for disaster risk planning and management [84], and their validation with the AUC–ROC provides confidence in the results obtained [87], although it is recommended that these statistical metrics be complemented with in situ validations [85]. AHP and WLC is widely used in LRM studies due to its good results, providing highly accurate and reliable maps [88,89,90]. WLC consists of weighting and combining all variables in a map, where the map grades are grouped as landslide risk zones [83,91].

$$LRM=(\sum {i=1}^n{\omega }_{ij}{W}_j)/(\sum {i=1}^n{\omega }_{ij}{W}_j)max$$

where ${\omega }_{ij}$ is the weight factor of class i in the conditioning factor j, and $W_j$ is the weight of the landslide conditioning factor j.

Subsequently, the landslide risk map was divided into uniform intervals by means of natural breaks using the Jenks method (Vergara et al., 2024 [92], Bhat et al., 2023 [30]; Shah et al., 2023 [12]), yielding a total of five risk categories: very high, high, moderate, low, and very low.

The Natural Jenks method, used in spatial modeling studies integrating GIS, is capable of minimizing intraclass variance and maximizing interclass variance [93]. This method allows for the identification and establishment of optimal thresholds within the probability values generated by the spatial model [94]. The Jenks method differs from other categorization methods, such as equal interval methods that divide into ranges of values with uniform segments and that do not consider statistical distribution [95], or quantiles that force the same number of observations per class that mix heterogeneous ranges [96]. On the other hand, other methods such as standard deviation or geometric interval do not have the same capacity to detect natural breaks since they are based on assumptions of normality or mathematical progressions that do not always correspond to the intrinsic variability of the phenomenon [64,97]. This is why the method used in this developed study is better adapted to the real structure of the data, preserving significant groupings and avoiding the loss of data in relevant information [96] since it is possible to obtain susceptibility categories that better represent the spatial and statistical patterns of the probability index and better relate to the historical landslide inventory obtained [98].

2.6. Model Validation

The Receiver-operating Characteristic Curve (ROC), a method widely used for the validation of landslide risk maps [51], was used. This method diagnoses the accuracy of a developed model and quantifies its predictive capability in terms of the area under the curve (AUC) considering the specificity and sensitivity results. The AUC can take values ranging from 0 to 1, where values closer to 1 indicate an excellent goodness of the model and values below 0.5 indicate that the model fails to measure the level of accuracy. The landslide and non-slide inventory map was used for validation.

For validation with AUC–ROC, a data set of landslide occurrence and non-occurrence was used. The landslide occurrence inventory was built with 314 records that occurred during the period from 2020 to 2023, which were obtained from the Disaster Risk Management Information System portal SIGRID (https://sigrid.cenepred.gob.pe/sigridv3/mapa, accessed on 17 January 2025). Data cleaning consisted of cleaning duplicate and/or empty data and ensuring geographic accuracy and thus avoiding bias among them [55,99]. The inventory of non-occurrence points was generated randomly and is representative of areas with no evidence of landslides to ensure a balanced spatial distribution and avoid bias due to autocorrelation [73,100]. Both data sets were integrated into a single data set and coded with 0 (no landslide) and 1 (landslide) [101]. The data set was then randomly divided into two subsets, 70% for model training and 30% for validation. From this process, the risk of overfitting is significantly reduced [102,103,104].

The methodology developed in this research can be seen in the methodological flowchart (Figure 3).

3. Results

3.1. Statistical Metrics of Model Performance

The AHP method applied to map the landslide risk areas in the Utcubamba river basin, Peru, obtained positive results. Since the consistency (CR) of the matrices generated was 0.085, which is less than 10%, the stable threshold for the consistency of the results is in accord with Saaty’s methodology; in addition, the most important prioritization vector (%) was the slope variable (P) with 23.96%, followed by the precipitation variable (P_c) with 14.446% (

Table 3. Weighted weights of each variable in the study area.

Variables	Prioritization Vector	Prioritization Vector (%)
Slope angle (P)	0.240	23.96
Geology (G)	0.137	13.65
Precipitation (Pc)	0.144	14.44
Distance to Faults (Df)	0.111	11.12
Drainage Density (Dd)	0.105	10.51
TWI	0.073	7.37
Relative Relief (Rr)	0.054	5.43
Profile Curve (Pc)	0.041	4.08
Land Use (Lu)	0.034	3.40
Elevation (E)	0.022	2.25
Distance to Roads (Dr)	0.020	2.03
Distance to Population Centers (Dpc)	0.018	1.80
λmax	13.44
n	12
IR	1.54
CI	0.131
CR	0.085

), which are in accordance with reality since both precipitation and slope are variables with high importance in triggering landslides in any part of the world.

The current landslide risk model for the Utcubamba river basin in Peru, developed using the Analytical Hierarchy Process (AHP) method, demonstrated significant accuracy, reaching an area under the curve (AUC) of 0.82 (Figure 4)—the non-discrimination line (or random reference line).

3.2. Landslide Risk Mapping of the Utcubamba River Basin, Peru

The mapping of landslide risk in the Utcubamba river basin in Peru (Figure 5) indicates that the very high risk class obtained an area of 731.56 km², representing 11.06% of the basin area, and the high risk class obtained an area of 1438.30 km², representing 21.75% of the total basin area; the high risk class occurs with greater incidence in the southern part of the basin. In addition, during the years 2020–2023, 314 landslides were recorded in the study area, 56 of which were in the very high risk class and 29 in the high risk class (

Table 4. Areas of landslide risk classes in the Utcubamba river basin, Peru.

Risk Classes	Landslides	Area (km2)	Area (%)
Very low	64	964.25	14.58%
Low	94	1657.14	25.06%
Moderate	71	1821.76	27.55%
High	56	1438.29	21.75%
Very high	29	731.56	11.06%
Total	314	6613.0207	100.00%

).

4. Discussions

Landslides are events that bring with them many ecological and social problems. In different parts of the world, different methods are used for mapping this event, from methodologies that include statistical probability models [105,106] to bivariate models [107,108]; however, apart from selecting the method for mapping, it is necessary to know the triggering factors of landslides and their weighted relationship to obtain the product sought [109]. This research integrated 12 variables belonging to the following factors: topographical, geological, hydrological, climatic, and anthropogenic, which have been used previously in other studies, obtaining high-quality and reliable results [110].

This research used GIS techniques, remote sensing data, and the hierarchical AHP method to map landslide risk zones in the Utcubamba river basin based on topographic, geological, hydrological, climatic, and anthropogenic triggers, all of which are directly related to each other. The pairwise comparison matrix was used to obtain the weightings by factors and the variables for each factor and subsequently evaluate the impact on the occurrence of landslides. Of all the variables integrated in the AHP model, the slope variable was assigned the highest weighting with a value of 0.240 since, in various international contexts, experts indicate that topographic factors, especially slope, have been shown to be determinant in the occurrence of landslides [109,111], and other studies indicate that topographical factors are integrated with climatic factors such as precipitation, which facilitates mass movement and increases the risk of landslides in an area [112].

The application of the WLC method made it possible to draw up a landslide risk map with five categories. It was identified that 11.06% and 21.75% of the area have very high and high risk levels, respectively. These values are consistent with the history of landslides in the area and warn of the need to prioritize mitigation actions. The developed model showed an acceptable level of consistency (CR = 0.085) and a high predictive capacity (AUC–ROC = 0.82), exceeding the results of comparable studies, such as that of Vakhshoori and Zare et al. [113], which obtained an AUC = 0.77, and that of Mengstie et al. [83], who obtained an AUC = 0.74. With this, it is assumed that the weights assigned to each variable are reliable, which confirms that both slope and geological faults are variables that have a greater incidence on landslide risk [114,115]. Also, it should be noted that the landslide risk map is highly reliable and can be used by society, as the generation of landslide risk maps is a necessary prerequisite for landslide hazard and risk assessment [116].

Although the validation of the AHP–GIS model using the AUC–ROC statistical metric is widely accepted in several scientific studies on landslide risk [117], several authors recommend completing the validations with in situ verification. These may include the direct inspection of geotechnical conditions, such as shear surfaces, water tables, or other indicators of potential instability [118], as well as field validation campaigns with the installation of specialized sensors (inclinometers, piezometers, etc.) and the performance of geotechnical tests, which allow for correlating areas classified as highly susceptible with directly measured physical parameters [119]. However, these validation methods have the disadvantage that they are highly expensive and logistically complex, which gives greater value to validations with statistical parameters such as AUC–ROC, whose usefulness and reliability for models developed in GIS environments have been widely demonstrated. However, it is recognized that the integration of field measurements and validations could significantly improve the accuracy and predictive capacity of the models.

The landslide inventory used reveals that most landslides in the Utcubamba river basin occur in areas with slopes between 20 and 25%. These findings corroborate the trends described in previous studies, such as those found by Khan et al. [120], who showed that 89.76% of landslides occurred on slopes greater than 20°, and Bhagya et al. [121], who identified that the highly susceptible zone has steeper slopes (22.48–74.47°). However, despite the fact that in the Utcubamba river basin, landslide records are mainly found in areas with low slopes, these areas have the highest amount of rainfall, which makes them more prone to landslides since rainfall controls the flow of water and the transport of mud and rock particles; therefore, soil moisture conditions influence the occurrence of landslides [122].

On the other hand, the topographic wetness index (TWI) obtained a prioritization value of 7, 37%, placing it as one of the most important factors in landslide risk, as it increases slope instability due to moisture accumulation [123]. This high TWI score is similar to those in other studies that point out TWI as one of the most influential factors in AHP-based susceptibility models [11,124]. However, TWI may overestimate risk in areas with high flow accumulation but geotechnically stable materials. In this research, this overestimation problem was addressed by integrating variables such as vegetation cover, land use, and slope [125].

For geology and distance to faults, variables that integrate the geological factor were considered, obtaining an importance of 13.65% and 11.12%, respectively. However, there are other necessary and important variables to determine landslide risk, such as shear strength [126], porosity, and soil saturation thresholds [127]. In this research, such variables were not included due to the limited availability of updated and spatially continuous data for the whole analysis area [128]. Further, the fact that each of these variables requires direct in situ measurements that are costly and time-consuming to perform, in addition to being unfeasible in many cases for large areas, makes them unfeasible in large-scale studies or in areas of difficult access [129].

Therefore, considering these limitations, we chose to use proxy variables that roughly represent the effects of such properties, such as slope, geology, vegetation cover, and distance to the drainage network, which are fully available and present adequate resolution for spatial analysis [130,131]. This approach is consistent with practices described in the literature for landslide susceptibility modeling. However, it is recognized that the future incorporation of data such as shear strength, porosity, and saturation thresholds, provided they are available in geospatial format, could significantly improve the accuracy and predictive capability of subsequent investigations.

In this research, the land use variable had a low prioritization value (3.40%), which reflects the theoretical approach adopted, the reality of the study area considering the soil management practices, the levels of technification, and the social conditions of the producers [132,133]. In the analysis, this variable included categories such as subsistence agriculture and coffee cultivation [134]. In the Peruvian context, agricultural areas such as coffee crops are usually established in areas near roads and on slopes; depending on the degree of technification, these agricultural areas can contribute to soil stabilization [135,136] and reduce erosive processes and landslide control [137]. However, the positive contribution of this type of crop is not uniformly manifested throughout the territory since many farmers have poor management practices and a low level of technification [138], generating a structural weakening of the soil and an increase in surface runoff, increasing the risk of landslides [139,140]. Therefore, when hierarchizing the land use variable with AHP, it is necessary to know the reality and to know the technical context of agronomic management so as not to assume that all agricultural areas contribute to soil stabilization [141].

With respect to seismicity, it is known that earthquakes generate landslides [142], as they are a destabilizing force, mainly on mountain slopes. The inclusion of this variable depends mainly on the historical seismic activity in the study area since, in areas with high seismic activity, it is necessary to integrate this variable to avoid underestimates in the modeling [143,144]. In this context, this variable was not integrated in this study because the study area has low seismic activity and the historical behavior of landslides in the Utcubamba river basin, as in other parts of the world [145], indicates that the main causes of landslides are climatic, geological, topographical, and hydrological factors. For this reason, the risk model generated in this study was based on the aforementioned variables.

The geological variable obtained one of the highest prioritization values, at 13.65%. This variable was classified into five risk levels, and the Mitú Group was considered to be at very high landslide risk. This is due to the lithological competition of the local conditions observed, such as intense fracturing, deep weathering, and the presence of planes of weakness that generate residual soils with low cohesion and high permeability [146], reducing shear strength and favoring water infiltration during heavy rains. Considering these conditions, combined with the steep relief, significant instability is generated [147,148]. To this end, geological evaluations conducted in part of the Utcubamba river basin confirm that the Mitú Group formations have weak surficial materials that are very susceptible to mass movements [149,150]. This is reinforced by Kumar et al. [49], who indicate that initially competent lithologies can show high susceptibility when degraded by fracturing and alteration, which occurs especially in tropical and Andean regions where steep slopes favor infiltration and water runoff, which, added to steep reliefs, increase the risk of landslides [151].

On the other hand, each of the lithologic properties of the geologic units in the study area differently influences the susceptibility of the terrain to landslides [152]. Among the most relevant properties identified are cohesion, permeability, and the degree of fracturing, which vary significantly depending on the nature and alteration state of each unit [153]. Cohesion is the force of internal attraction between particles or crystals that confers resistance to the material, where low values, such as those recalled in residual soils resulting from the weathering of the Mitú Group, favor the mobilization of the masses under saturation conditions [154,155]. Permeability, which is the capacity of the material to allow for the passage of water, directly affects infiltration and the increase in pore pressure; therefore, materials with high permeability facilitate the accumulation of water in planes of weakness and reduce shear strength [156]. In turn, the degree of fracturing and the presence of diaclases decrease the structural integrity of the rock mass and generate preferential infiltration pathways. Consequently, the combination of high permeability, low cohesion, and intense fracturing, especially on steep slopes, generates a scenario of very high susceptibility to landslides [157,158].

In this study, the Euclidean distance to geological faults was used as a spatial metric to represent the structural influence on different geodynamic and geotechnical processes in a continuous and quantifiable way [159] since areas near faults usually present a higher density of fractures, mineralogical alteration, and lower mechanical strength of the rock mass, conditions that favor the occurrence of instability processes, stress concentration, and, therefore, the preferential circulation of fluids [16,160]. This variable is an indicator of rupture when detailed information on parameters such as fault type, kinematics, dip, slip velocity, or local seismic history are not available [161]. Euclidean distance is computationally efficient, as it is calculated using simple geometric operations by measuring the shortest linear distance between each cell or analysis point and the fault trace, avoiding complex iterative processes or the need to model detailed structural parameters [162]. Since it requires a small number of mathematical operations and does not rely on large volumes of input data, its processing in GIS environments is fast even for large areas and high spatial resolutions [163].

5. Conclusions

The objective of this study was to map landslide risk in the Utcubamba river basin, Peru, using a geospatial approach based on Geographic Information Systems (GIS) and the Analytic Hierarchical Process Method (AHP). Twelve triggering variables were integrated and grouped into topographic, geological, hydrological, climatic, and anthropogenic factors, which were weighted and combined using the weighted linear combination (WLC) technique in a GIS environment.

The methodology used identified that 32.81% of the surface area analyzed in the Utcubamba river basin is at high and very high risk of landslides. These high-risk zones are located mainly in the south of the basin and coincide with areas with steep slopes, high precipitation, and proximity to population centers or communication routes.

The validation of the model by means of the ROC curve showed an AUC value of 0.82, which confirms that the integration of the AHP method with Geographic Information Systems (GIS) tools allows for the precise identification of critical areas, useful for territorial planning, the prioritization of interventions, and emergency management.

The methodology applied in this study has a high potential for replicability in other Peruvian watersheds and departments with similar geographic and climatic characteristics. The incorporation of more advanced approaches, such as machine learning, fuzzy logic, and the integration of socioeconomic and demographic variables, would strengthen future assessments and provide a more complete view of risk, encompassing both environmental and social aspects. However, limitations related to the resolution of geospatial data and the lack of historical landslide inventories should be considered and identified.

Based on these findings, it is recommended to incorporate in future research the use of complementary approaches such as machine learning algorithms, fuzzy logic, multi-temporal analysis, and more advanced data-driven methods or their combination with multi-criteria techniques.