Monitoring Slope Movement and Soil Hydrologic Behavior Using IoT and AI Technologies: A Systematic Review

Abstract

Landslides or slope failure pose a significant risk to human lives and infrastructures. The stability of slopes is controlled by various hydrological processes such as rainfall infiltration, soil water dynamics, and unsaturated soil behavior. Accordingly, soil hydrological monitoring and tracking the displacement of slopes become crucial to mitigate such risks by issuing early warnings to the respective authorities. In this context, there have been advancements in monitoring critical soil hydrological parameters and slope movement to ensure potential causative slope failure hazards are identified and mitigated before they escalate into disasters. With the advent of the Internet of Things (IoT), artificial intelligence, and high-speed internet, the potential to use such technologies for remotely monitoring soil hydrological parameters and slope movement is becoming increasingly important. This paper provides an overview of existing hydrological monitoring systems using IoT and AI technologies, including soil sampling, deploying on-site sensors such as capacitance, thermal dissipation, Time-Domain Reflectometers (TDRs), geophysical applications, etc. In addition, we review and compare the traditional slope movement detection systems, including topographic surveys for sophisticated applications such as terrestrial laser scanners, extensometers, tensiometers, inclinometers, GPS, synthetic aperture radar (SAR), LiDAR, and Unmanned Aerial Vehicles (UAVs). Finally, this interdisciplinary research from both Geotechnical Engineering and Computer Science perspectives provides a comprehensive state-of-the-art review of the different methodologies and solutions for monitoring landslides and slope failures, along with key challenges and prospects for potential future study.

1. Introduction

Earthen slopes represent a common geomorphic feature in natural landscapes as well as engineered structures such as embankments, dams, and highway slopes. The movement of soil, rock, and organic materials down these slopes due to gravity is called a landslide or slope failure. The detrimental effects of slope failures, including human casualties, property damage, economic losses, and environmental factors, are widely recognized and extensively documented in scholarly works [1,2,3,4]. Based on data from the Emergency Events Database, retrieved in November 2021 by the Centre for Research on the Epidemiology of Disasters [5], there have been approximately 488,000 fatalities from natural hazards related to landslides or slope failure since 2000. This number also includes ground movement resulting from earthquakes. The database states that since 2000, the total estimated worth of economic losses and damages due to landslides, whether direct or indirect, has exceeded USD 310 billion.

Slope failure or landslide is an external manifestation of multiple factors, such as land type, geological condition, rainfall amount and intensity, melting snow, soil mantle thickness, soil state parameters, groundwater level fluctuation, human activity, erosion, seismic loads, subsurface hydrology, vegetation, and other occurrences [6,7]. It has been demonstrated by several studies that unsaturated soil hydrology is one of these elements that has a major role in the formation and dynamics of slope failure [8,9,10]. Because of the various physicochemical processes influencing slope stability and shear strength, the interactions between unsaturated hydrology and mechanical processes are even more complicated in clay soils. Additional processes that occur in clay soils include weathering, slaking, and soil weakening brought on by cycles of freezing and thawing or wetting and drying. According to refs. [11,12], even a little swelling can cause bonded clays to deteriorate because the buildup of plastic strains causes a reduction in the soil’s shear strength.

The slopes are typically unsaturated before a failure initiates [13]. Volumetric water content (VWC) and related variations in suction significantly impact the behavior of unsaturated slopes [14]. Rainfall, snowmelt, temperature, and many other environmental loadings are examples of triggering events that cause VWC and suction values to rise and fall, respectively. These parameters determine unsaturated shear strength and, ultimately, the stability of the slopes. According to several researchers [8,15], the emergence of positive pressures above the water table and the subsequent drop in matric suction (increase in soil moisture) following precipitation are the primary causes of shallow slope failure. Specifically, as soil matric suction increases, the shear strength of the soil reduces non-linearly. Consequently, as suction becomes less negative and the soil gets closer to saturation, the soil becomes more vulnerable to failure [16,17]. Therefore, moisture (loss of soil suction) is the primary triggering element for most slope destabilizations, among other reasons [18,19]. Water-induced landslides or slope failure is common on slopes and may become slow to fast-moving, making them one of the main risks worldwide. They could be brought on by periods of intense rainfall, snowmelt, or a mix of both.

Hydrological monitoring on slopes was carried out in several studies [13,20,21,22,23,24,25] to elucidate the underlying initiation mechanism of water-induced landslides. According to ref. [26], hydrological monitoring can offer crucial insights into the hydrological processes that take place on similar slopes, such as water penetration and the ensuing variations in suction and VWC [27]. Based on data from hydrological monitoring, landslide occurrence was observed in a naturally occurring slope under partially saturated conditions in the study [22]. The study showed that using the infinite slope stability method for unsaturated conditions, VWC and matric suction data may be used to predict the occurrence of partially saturated shallow landslides [8]. Similarly, the hydrological monitored data on the VWC and suction were overused in the literature to evaluate the stability condition of slopes when combined with the infinite slope stability method for unsaturated conditions [13,23,25,28]. Song et al. [29] also used hydrological monitoring in an early warning system, determining the hazard levels according to the stability evaluation that corresponded with the monitored data. Therefore, the unsaturated zone of slopes, characterized by partial saturation, plays a critical role in regulating water movement, nutrient transport, and slope stability. Thus, it is often necessary to conduct field hydrological monitoring to better understand how the many predisposing complex hydrological processes within the unsaturated zone regulate the mechanism leading to slope failure.

Numerous methodologies have been effectively employed to examine the hydro-logical and mechanical response of slopes subjected to varying triggering circumstances and the mechanisms that initiate slope failure or landslides. Numerical modeling, for instance, has effectively examined how various elements influence the stability of both deep-seated [30,31,32,33,34,35] and shallow [36,37,38,39] slope failures under changing unsaturated soil hydrology due to varying rainfall conditions. Field monitoring has successfully observed slopes’ hydrologic and mechanical performance in various geological and climatic contexts (e.g., [40,41,42]). Several techniques employed to monitor slope mechanical performance or displacement in the field can mainly be classified as manual approaches and remote monitoring [43]. Traditionally, the geotechnical monitoring of slopes has relied on manual techniques such as scheduled inspections by experts, deploying extensometers [44], piezometers [45], and inclinometers [46,47], etc., for both data collection and on-site investigations. However, such methodologies prove intricate and time-consuming [48], especially in hazardous environments or difficult-to-access sites. As a result, throughout the previous two decades, technical developments have prompted the creation of novel instruments that incorporate a variety of sensors [49] for real-time remote monitoring such as GPS systems [50] or interferometry techniques such as Synthetic Aperture Radar (SAR) Satellite Images [51], Light Detection and Ranging (LiDAR) [52,53,54,55,56,57], PSInSAR (Permanent Scatterer Synthetic Aperture Radar Interferometry) [58,59], Differential Interferometric SAR (DIn-SAR) [60,61], SqueeSAR [62], Ground-Based Synthetic Aperture Radar (GBSAR) [63,64,65,66], Terrestrial Laser Scanning (TLS) [67,68], Infrared Thermography (IRT) [69], Wireless Sensor Network (WSN) [70,71], and most recently, Internet of Things (IoT) [72,73]. Recently, artificial intelligence (AI) and machine learning (ML) techniques are increasingly being applied to analyze the vast amounts of data generated by advanced monitoring methods. AI algorithms can identify patterns and correlations within the data, providing predictive insights into slope stability and enabling the development of more accurate and reliable early warning systems. By combining IoT and AI technologies, researchers can enhance their understanding of the complex interactions between soil moisture, matric suction, and slope stability [74,75]. With technological advancement, this research article provides a comprehensive overview of the unsaturated soil hydrology and stability of earthen slopes, focusing on different hydrological monitoring systems and slope movement detection systems from traditional methods to current technologies.

2. Types of Landslides and Influencing Factors

Understanding the various modes of slope movement is essential for effective risk management and mitigation. Landslides result from the interplay of various factors that reduce the stability of a slope. This section highlights the different types of slope movements and the factors influencing their occurrence. As climate change and human activities continue to impact slope stability, ongoing research and innovation in this field will be crucial for safeguarding lives and infrastructure.

2.1. Various Modes of Landslides or Slope Failure

Slope movements are generally categorized based on the type of material involved and the nature of the movement. According to Varnes [76], these categories include falls, topples, slides, spreads, and flows. Each category is distinct in its mechanics and potential impact. Falls involve the free fall, bouncing, or rolling of rock or soil down a slope. These movements are typically rapid and are often triggered by weathering processes or seismic activity. Wieczorek [77] noted that rock falls are common in steep mountainous regions where freeze–thaw cycles weaken rock structures. Another mode of slope movement, toppling, is characterized by the forward rotation and movement of rock or soil around a pivot point. This mode of movement is often associated with steep slopes and can be triggered by undercutting due to erosion or anthropogenic activities. According to Cruden and Varnes [43], toppling failures are prevalent in stratified rock formations where layers dip away from the slope face. Slides down a slope involve the downslope movement of a mass of rock, debris, or earth along a well-defined surface of rupture. Slides can be further classified into rotational and translational slides based on the shape of the rupture surface. Hutchinson [78] describes rotational slides as having a curved rupture surface, while translational slides occur along planar surfaces, often along bedding planes or joints. Spreads of slope soil are characterized by the lateral extension and fracturing of cohesive soil or rock. These movements typically occur on gentle slopes and are often triggered by liquefaction or the presence of weak layers within the soil profile. Terzaghi [79] highlighted that spreads are common in regions with soft, sensitive clay deposits, which can lose strength rapidly when disturbed. Flows involve the movement of soil, rock, and water as a viscous fluid. These movements can range from slow earth flows to rapid debris flows and mudflows. Iverson [80] emphasized that flows are often triggered by intense rainfall, volcanic activity, or the rapid melting of snow and ice, leading to the saturation and mobilization of materials.

2.2. Factors Affecting Slope Failure

Geological factors are foundational in determining slope stability. The lithology, structure, and weathering of the slope material significantly influence its strength and stability. Rocks with high fracture density, such as shale and heavily jointed limestone, are more susceptible to slope failure [43,81]. Weathering processes weaken the rock mass over time, reducing its cohesion and increasing the likelihood of failure [76]. Recent studies emphasize the role of detailed geological mapping and remote sensing in assessing geological conditions that predispose slopes to failure. Structural features such as bedding planes, joints, and faults can act as planes of weakness along which failure may occur. For instance, rock layers that dip parallel to the slope face are particularly prone to translational slides [78]. The presence of faults can also facilitate the infiltration of water, further destabilizing the slope [82]. Hydrological factors play a critical role in slope stability by influencing pore water pressures and the moisture content of slope materials. Rainfall infiltration increases the weight of slope materials and pore water pressure, commonly triggering landslides [83]. High pore water pressures reduce the effective stress in the soil or rock, diminishing its shear strength [79]. This can lead to a reduction in the frictional resistance of the material, making it more susceptible to failure [82]. Surface water infiltration from rainfall or irrigation can saturate slope materials, increasing their weight and the driving forces acting on the slope. According to Iverson [84], intense or prolonged rainfall events are a common trigger for landslides, particularly in regions with steep terrain and permeable soils. Groundwater flow can also lead to slope instability by eroding the base of the slope or creating seepage forces that destabilize the material. Recent advancements in hydrological modeling and monitoring have improved the prediction of landslides triggered by rainfall [85]. Climatic factors, including precipitation, temperature, and seasonal variations, significantly affect slope stability. Heavy rainfall can rapidly increase pore water pressures and reduce soil suction, triggering slope failures [77]. Additionally, freeze–thaw cycles contribute to the mechanical weathering of rocks, promoting rock falls and slides in colder climates [76]. Seasonal variations in temperature and moisture content can lead to cyclic loading and unloading of slope materials, inducing stress changes that may culminate in failure [86]. In arid regions, the sudden infiltration of water during rare but intense rainfall events can be particularly destabilizing [87]. Vegetation plays a dual role in slope stability. The root systems of plants and trees can enhance slope stability by reinforcing the soil and providing additional cohesion [88]. However, the removal of vegetation through deforestation, agriculture, or construction can significantly increase the risk of slope failure by reducing root reinforcement and increasing surface erosion [89]. Biological factors, such as the activities of burrowing animals and the decomposition of organic matter, can also affect slope stability. Burrowing can create voids and disrupt the integrity of slope materials, while the decomposition of organic matter can alter soil structure and reduce its shear strength [90]. Seismic activity is a major trigger for slope failures, particularly in tectonically active regions. Earthquakes generate ground shaking that induces dynamic loads on slopes, which can lead to the initiation of landslides [91]. The intensity and duration of shaking, as well as the frequency of seismic events, are critical factors influencing the likelihood of slope failure [92]. Seismic-induced slope failures can occur due to the reduction in shear strength of slope materials, the generation of excess pore water pressures, or the direct mechanical disruption of the slope structure. Historical case studies, such as the 1970 Huascarán debris avalanche in Peru, highlight the devastating impact of earthquake-triggered landslides [93]. Human activities significantly impact slope stability through construction, mining, deforestation, and land use changes. Excavations and cut-and-fill operations can alter the natural slope geometry, reducing stability [94]. Construction activities often involve the addition of loads to slopes, such as buildings or infrastructure, which can increase driving forces and reduce the factor of safety [95]. Mining operations, particularly open-pit mining, can destabilize slopes by removing support and creating steep, unstable faces. The removal of vegetation for agriculture or urban development reduces root reinforcement and increases erosion, further compromising slope stability [89]. Chemical processes, such as the dissolution of soluble minerals or the alteration of clay minerals, can affect the stability of slopes. The dissolution of limestone or gypsum can create cavities and reduce the strength of the rock mass, leading to subsidence or collapse [96]. The alteration of clay minerals through hydration or ion exchange can result in significant volume changes and reductions in shear strength, promoting slope failure [97]. Acidic conditions, often resulting from mining activities, can accelerate the weathering of rocks and the leaching of minerals, further weakening slope materials [98]. Understanding the chemical processes at play is essential for predicting and mitigating slope failures in affected regions.

Several case studies illustrate the complex interplay of factors leading to landslides. The 1963 Vaiont Dam disaster in Italy, caused by the rising water levels in the reservoir, triggered a massive rockslide, highlighting the impact of hydrological changes on slope stability [99]. The 1983 Thistle landslide in Utah, USA, was triggered by prolonged rainfall, underscoring the significance of climatic factors [100]. The 2014 Oso landslide in Washington, USA, demonstrated the combined effects of geological, hydrological, and anthropogenic factors, resulting in a catastrophic failure with significant loss of life and property [101]. These examples emphasize the need for comprehensive risk assessments and the integration of multiple factors in slope stability analysis.

3. Unsaturated Soil Hydrology and Slope Stability

Unsaturated soil hydrology of slopes is a specialized area within the broader field of soil mechanics and hydrology that examines the behavior and movement of water in partially saturated soils on inclined surfaces. In the unsaturated zone of earthen slopes, water movement is governed by a combination of gravity, capillary forces, and hydraulic gradients [102,103], and this movement or water infiltration into the soil matrix is significantly affected by soil properties such as porosity, permeability [104], soil water storage, fissure distribution, initial soil moisture, etc. [105]. The water infiltration process is also significantly influenced by the precipitation’s type, intensity, and duration. Regarding evaporation, the two most important variables are likely to be evaporation intensity and duration [106,107,108,109,110,111]. In slopes, these factors influence the soil water characteristic curve (SWCC) and overall stability. SWCC describes the fundamental relationship between two significant soil hydraulic parameters, pore water pressure or matric suction and water content or degree of saturation [112]. A typical SWCC [113] measured at the laboratory scale is presented in Figure 1. Though a variety of equations can be used to define the SWCC [114], the sigmoidal van Genuchten equation [115] is widely used to analyze the SWCC parametrically. The van Genuchten equation takes the following form (Equation (1)) to describe the SWCC.

$$\theta ={\theta }_r+{\displaystyle \frac{{\theta }_s-{\theta }_r}{{[1+{(\alpha \Psi )}^n]}^m}}$$

(1)

where $\Psi$ is the matric suction, $\theta$ is the volumetric water content, ${\theta }_s$ is the saturated volumetric water content, ${\theta }_r$ is the residual volumetric water content, $\alpha$, n, and m are fitting parameters.

SWCC is essential for comprehending and analyzing the hydrological behaviors of soil. Additionally, SWCC is the fundamental concept of unsaturated soil mechanics that applies to geotechnical and geo-environmental engineering practices [116]. Matric suction, the pressure difference between the pore air and pore water, is a key factor in maintaining slope stability, as it contributes to the apparent cohesion of the soil. Variations in moisture content due to rainfall infiltration, evaporation, and transpiration can lead to changes in matric suction, potentially triggering slope failures. Moreover, vegetation cover and land use practices influence surface runoff and soil moisture dynamics [117,118]. Understanding these factors is essential for predicting the partitioning of rainfall into infiltration, runoff, and evapotranspiration on earthen slopes. Studies have emphasized the importance of incorporating unsaturated soil mechanics into slope stability analyses to provide more accurate and reliable predictions [112,113,114,115,116,117,118,119].

Mathematical equations governing unsaturated flow, such as the Richards equation [120] and the GreenAmpt equation [121], provide the theoretical basis for modeling water movement in unsaturated soils. These equations describe the relationship between soil moisture content, hydraulic conductivity, and pressure head, facilitating the simulation of infiltration processes and moisture redistribution in earthen slopes [115,122]. Incorporating these equations into numerical models allows for the prediction of soil moisture dynamics under varying climatic conditions and land use scenarios [123,124]. In addition, preferential flow pathways in earthen slopes, such as macropores and fractures, significantly influence water infiltration and subsurface flow dynamics [125,126]. These pathways often exhibit higher hydraulic conductivity compared to the surrounding soil matrix, leading to rapid transmission of water and solutes [126,127,128]. Characterizing preferential flow pathways’ spatial distribution and connectivity is essential for accurately predicting water fluxes and contaminant transport in earthen slopes [129]. The following sections describe the hydrologic processes in slope, the governing equations of unsaturated soil hydrology in slope, and the factors affecting the hydrology of unsaturated slope.

3.1. The Infiltration Hydrology in Slope

Three stages can be distinguished in the infiltration process based on water’s force and motion properties. (1) Imbibition stage: Molecular force predominates at this point. Particulate water is created when soil particles absorb infiltration water. This stage is easily visible in the dry soil. This stage eventually disappears as the soil water content exceeds the maximum molecular moisture content. (2) Leakage stage: Capillary force and gravity action are involved in this step. Water unevenly infiltrates soil pores by flowing downward, filling them gradually until they are completely saturated. Both of the stages are generally referred to as the leakage stage. (3) Osmotic stage: Water is moving downward at a constant flow under gravity when all of the soil’s pores are full and saturated with water. Seepage, in general, is the flow of unsaturated water, whereas infiltration is the flow of saturated water. There is not a clear distinction between these phases in the penetration process itself.

The infiltration rate of a landslide, which is the actual amount of water in precipitation through the unit area of the surface per unit of time, controls the infiltration of precipitation into the slope. The infiltration rate is equal to the precipitation intensity if it is less than the topsoil’s minimal infiltrability. In this instance, the infiltration rate will stay constant if it is expected that the precipitation intensity will not change. The moisture content of topsoil progressively rises with ongoing precipitation until it reaches a steady level. A portion of the precipitation turns into runoff when the intensity of the precipitation exceeds the unsaturated topsoil’s capacity to infiltrate. The remainder seeps underground. Water that percolates below the surface has two destinations; some of it is retained in soil pores above the subsurface water level, and any water that remains in the soil’s water retention capacity replenishes the groundwater.

The main factors affecting an unsaturated slope’s infiltration hydrology are the hydraulic properties of unsaturated soil, water storage capacity, evaporation, precipitation intensity, and slope geometry. Slope seepage characteristics, slope gradient, vegetation cover on the slope surface, fissure distribution, and soil capillary water are the primary factors influencing the slope aspect. Darcy’s law often governs water flow through unsaturated soil, much like it does through saturated soil. Nonetheless, two significant distinctions in the water flow in saturated and unsaturated soil are evident. (1) A storage term exists that expresses how the matric suction changes the water content; (2) the permeability coefficient of water is highly dependent on the matric suction [105]. It should be mentioned that during the infiltration process, no change in soil volume is considered. The suction in an unsaturated soil determines the storage period in unsaturated flow, which is determined by the SWCC. It is not a constant. Therefore, for unsaturated soils, the most crucial hydraulic parameters are the water permeability coefficient and SWCC.

The water permeability coefficient tending to zero and tending to infinity are two extreme instances of the infiltration problem that might be initially considered. There will not be any water infiltration into the soil layer at the former extreme. At the other end of the spectrum, water can readily seep into the soil layer but will quickly drain out via the borders. Precipitation does, however, partially percolate and does not completely drain away at intermediate water permeability values. This suggests that the maximum amount of rainfall penetration could occur at a critical saturated permeability [130].

Mass landslides or slope failure typically happen during periods of intense rainfall. The permeability coefficient, or soil infiltration capacity, in this instance, determines how much water percolates into the slope [105]. In unsaturated soil, the permeability coefficient varies greatly; typically, there are three to five orders of magnitude. In saturated soil, the permeability coefficient can be thought of as essentially constant. The permeability coefficient of unsaturated soils is determined by the volumetric water content, while the pressure of pore water determines the volumetric water content. Thus, the pore water pressure functions indirectly to determine the water permeability coefficient.

Water storage capacity influences unsaturated soil hydrology and infiltration. The difference between the volumetric water contents of soil that is saturated and residual is known as the water storage capacity of the soil. It is a measurement of the maximum water that capillary action is capable of absorbing or desorbing. It differs slightly from the soil’s capacity to hold water. In general, as pore diameters rise, so does the value of storage capacity. It is somewhat connected to the void ratio as well.

The amount of rainfall that penetrates a certain area is determined by the hydraulic gradient of the topsoil, the water permeability coefficient, and the intensity of the rainfall. Because the soil’s negative pore water pressure varies during the infiltration process, so do the water permeability coefficient and hydraulic gradient. In unsaturated soils, the initial distribution of pore water is a crucial input for transient flow studies. According to earlier research, the antecedent infiltration rate not only indirectly affects the soil’s water permeability coefficient [130], but it also significantly influences the initial pore water distribution and, in turn, the pore water pressure redistribution that follows subsequent rainfall infiltration. Rainfall infiltration in initially unsaturated soils causes an increase in the water permeability coefficient due to an increase in water content, and it simultaneously results in a decrease in the hydraulic gradient due to a decrease in negative pore water pressure. Variations in the subsequent rainfall infiltration rate are frequently limited by the hydraulic properties of the soil and ground conditions. The least amount of rainfall will result in the least change in negative pore water pressure if the antecedent and following rainfall infiltration rates are both optimal. Therefore, it is conceivable that if a lengthy antecedent rainfall is paired with a strong following rainstorm, the negative pore water pressure in the soil may be significantly decreased.

3.2. The Governing Equation of Unsaturated Hydrology of Slope

The governing Richards equation for 1D vertical infiltration can be written in several forms, such as the water content, mixed water content, capillary head form, and head form. In the mixed water content form, the governing 1D equation (Equation (2)) is:

$${\displaystyle \frac{\partial \theta }{\partial \tau }}={\displaystyle \frac{\partial }{\partial z}}\left[k\left(\theta \right){\displaystyle \frac{\partial \Psi \left(\theta \right)}{\partial z}}-1\right]$$

(2)

where z is vertical coordinate (positive downward) $\left[L\right]$; $\tau$ is time $\left[T\right]$; $\theta$ equals $\theta (z,\tau )$, which equals volumetric soil moisture content $[-]$; $\Psi \left(\theta \right)$ is empirical soil hydraulic capillary head function $\left[L\right]$; $K\left(\theta \right)$ is empirical unsaturated hydraulic conductivity function $\left[L{T}^{--1}\right]$.

The effects of capillarity are represented by the first term in parentheses in Equation (1), while the impacts of gravity-driven flux are represented by the second term in parentheses. By introducing the chain rule, Equation (2) is frequently stated exclusively as a function of the water content, as presented in Equation (3):

$${\displaystyle \frac{\partial \theta }{\partial \tau }}={\displaystyle \frac{\partial }{\partial z}}\left[D\left(\theta \right){\displaystyle \frac{\partial \theta }{\partial z}}-k\left(\theta \right)\right]$$

(3)

where z is $D\left(\theta \right)$ is referred to as the soil water diffusivity $\left[L^2{T}^{-1}\right]$ and is presented by the following equation (Equation (4)):

$$D\left(\theta \right)=k\left(\theta \right){\displaystyle \frac{\partial \Psi \left(\theta \right)}{\partial z}}$$

(4)

Since water content is a continuous variable, the water content or combined water content variables of the Richards equation are useful in uniform soils. However, layered soils are common in natural slopes, and soils are rarely uniform across large length scales. Because of the distinct unsaturated capillary head connections in the various soil layers, the water content in layered soils is discontinuous across layer interfaces [131]. Rather, since the capillary head ($\Psi$) is continuous, it is preferable to express the moisture content in terms of $\Psi$, $\theta =\theta (\Psi )$, and use Richards’ equation with the capillary head as the dependent variable. This can be carried out in one of two ways (Equations (5) and (6)):

$${\displaystyle \frac{\partial \theta (\Psi )}{\partial \tau }}-{\displaystyle \frac{\partial }{\partial z}}\left[k(\Psi )\left({\displaystyle \frac{\partial \Psi }{\partial z}}-1\right)\right]=0$$

(5)

$$C(\Psi ){\displaystyle \frac{\partial \Psi }{\partial \tau }}-{\displaystyle \frac{\partial }{\partial z}}\left[k(\Psi )\left({\displaystyle \frac{\partial \Psi }{\partial z}}-1\right)\right]=0$$

(6)

where $C(\Psi )$ is the specific moisture capacity (${\displaystyle \frac{\partial \theta }{\partial \Psi }})\left[L^{-1}\right]$; $K(\Psi )$ equals the hydraulic conductivity function written as a function of $\Psi \left[L{T}^{-1}\right]$. Equations (5) and (6) can both be used to solve saturated and unsaturated slope hydrologic problems.

3.3. Shear Strength of Unsaturated Soil

The shear strength of soil is a crucial parameter in geotechnical engineering, affecting the stability of slopes, foundations, and earth structures. In unsaturated soils, where the pore spaces contain both air and water, the hydrological conditions significantly influence the shear strength. The interplay between soil moisture content, matric suction, and shear strength is complex and essential for understanding and predicting the behavior of unsaturated soils under various environmental conditions. This article explores the effect of unsaturated soil hydrology on the shear strength of soil, integrating theoretical concepts, empirical findings, and advanced modeling techniques.

Unsaturated soil mechanics extends traditional soil mechanics by incorporating the effects of matric suction and moisture content on soil behavior. Matric suction, the difference between pore air pressure ($u_a$) and pore water pressure ($u_w$), plays a pivotal role in enhancing the shear strength of unsaturated soils. The extended Mohr–Coulomb failure criterion for unsaturated soils is expressed as the following Equation (7):

$$\zeta =c^{\prime }(\sigma -u_a)tan{\varphi }^{\prime }+(u_a-u_w)tan{\varphi }_b$$

(7)

where $\zeta$ is the shear strength, $c^{\prime }$ is the effective cohesion, $\sigma$ is the total normal stress, $u_a$ is the pore air pressure, $u_w$ is the pore water pressure, ${\varphi }^{\prime }$ is the effective angle of internal friction, and ${\varphi }_b$ is the angle of friction with respect to matric suction [116].

The key hydrological processes, such as rainfall infiltration and evaporation, influence the matric suction and, consequently, the shear strength of unsaturated soils. During rainfall events, infiltration increases pore water pressure and reduces matric suction, leading to a decrease in shear strength and potential slope instability [132]. Conversely, evaporation can increase matric suction, thereby enhancing soil shear strength [119].

Field studies and laboratory experiments have demonstrated the significant impact of hydrological conditions on the shear strength of unsaturated soils. Techniques such as tensiometers, time domain reflectometry (TDR), and soil moisture sensors provide valuable data on soil suction, moisture content, and pore water pressures [116,133]. These measurements are critical for calibrating and validating numerical models that predict the shear strength of unsaturated soils [134]. The mechanical reinforcement provided by vegetation roots can also influence the shear strength of unsaturated soils. Roots increase soil cohesion and alter hydrological dynamics through transpiration, which affects matric suction [135]. Understanding the interaction between vegetation and soil hydrology, thereby, is vital for managing slope stability, particularly in regions with significant vegetation cover [136]. Climate change poses additional challenges to the shear strength of unsaturated soils. Altered precipitation patterns, increased frequency of extreme weather events, and prolonged droughts can all impact soil hydrology and shear strength [137]. Research by Bogaard and Greco [138] underscores the need for adaptive management strategies to mitigate the effects of climate change on soil stability.

Recent advancements in numerical modeling, such as the finite element method (FEM), have enhanced our understanding of the hydrological and mechanical behavior of unsaturated soils. These models incorporate spatial and temporal variations in soil suction, moisture content, and mechanical properties, providing a comprehensive analysis of shear strength under transient conditions [139]. Numerical simulations, combined with field and laboratory data, offer valuable insights into the mechanisms governing the shear strength of unsaturated soils and guide the design of effective stabilization measures [140]. Therefore, the shear strength of unsaturated soils is intricately linked to hydrological conditions, and the stability of unsaturated slopes is governed by a complex interplay of mechanical, hydraulic, and climatic factors. Advances in soil mechanics, numerical modeling, and field monitoring techniques have significantly enhanced our understanding of these factors and their interactions. Continued research and development in this field with advanced monitoring systems are essential for improving predictive capabilities and developing effective mitigation strategies to ensure the safety and sustainability of slopes in unsaturated soils.

3.4. Early Warning System for Rainfall-Induced Slope Failure

Early warning systems (EWSs) for rainfall-induced slope failures are critical in mitigating these risks by providing timely alerts, enabling preventive measures, and facilitating evacuations. This section explores recent advancements in the development and implementation of early warning systems for rainfall-induced slope failures, highlighting key components, emerging technologies, and case studies that illustrate their effectiveness. A comprehensive early warning system for rainfall-induced slope failures consists of several components, including monitoring, data analysis, prediction, and communication. Monitoring involves the use of sensors to detect changes in environmental and geotechnical conditions that precede slope failures, such as rainfall intensity, soil moisture, and pore water pressure [141]. Recent advancements in sensor technology have improved the accuracy and reliability of these measurements. Data analysis and prediction are critical components, involving the processing and interpretation of monitoring data to predict the likelihood of a slope failure event. Machine learning algorithms and statistical models are increasingly being used to analyze large datasets and identify patterns indicative of imminent slope failures [142]. These analytical tools enhance the ability to issue timely and accurate warnings. Establishing rainfall thresholds is a common approach in early warning systems for slope failures. Rainfall thresholds are defined as the amount of rainfall over a specific period that is likely to trigger slope failures. These thresholds are determined based on historical data and hydrological models that simulate the behavior of slopes under different rainfall conditions [83]. Predictive modeling and machine learning algorithms are increasingly being used to analyze monitoring data and predict slope failures. These models can process large datasets, identify patterns, and generate predictions based on historical and real-time data [142]. Machine learning techniques, such as artificial neural networks and support vector machines, have shown promise in improving the accuracy and reliability of slope failure predictions. Several successful implementations of early warning systems for rainfall-induced slope failures highlight their effectiveness. The Hong Kong landslide early warning system, for example, integrates rainfall monitoring, hydrological modeling, and real-time data analysis to provide timely warnings. This system has significantly reduced landslide-related casualties and property damage in the region. In Brazil, the Cemaden early warning system uses a combination of remote sensing, ground-based monitoring, and predictive modeling to issue warnings for rainfall-induced slope failures. The system’s integration of multiple data sources and technologies has improved its ability to predict and mitigate slope failures. Despite significant advancements, challenges remain in the development and implementation of early warning systems for rainfall-induced slope failures. One major challenge is the variability in rainfall patterns and slope characteristics, which complicates the prediction and modeling of slope failures [143]. The accuracy of early warnings depends on the quality and density of monitoring data, which requires substantial investment in infrastructure and maintenance. Despite the challenges, ongoing research and collaboration continue to enhance the ability to predict and respond to slope failures, ultimately reducing the landslide impact on communities and infrastructure worldwide.

4. Traditional Slope Movement Detection Systems

This section explores the traditional methods utilized in slope stability studies. These methods, rooted in historical practices, provide a comprehensive framework for assessing and addressing the factors contributing to slope instability, ensuring safety and sustainability in infrastructure development and land management [144,145]. Accordingly, continuous monitoring of landslides or slope failure patterns is essential for understanding their dynamics and establishing early warning systems. A study by Petley [146] investigated the distinct patterns of slope movement exhibited by rotational and translational landslides. The study aims to differentiate these types of landslides based on their mechanical behavior and movement characteristics. The authors use a combination of field observations, remote sensing data, and numerical modeling to analyze several case studies of both rotational and translational landslides. They identify key factors influencing the movement patterns, including geological conditions, slope geometry, and external triggers such as rainfall and seismic activity. The study concludes that understanding the distinct movement patterns of rotational and translational landslides is crucial for accurate hazard assessment and mitigation strategies. The findings provide valuable insights into the dynamics of landslide processes and contribute to the development of more effective monitoring and early warning systems. The paper by Park [147] presented an integrated analysis method designed to enhance the stability analysis and maintenance of cut-slopes in urban areas. The proposed method integrates geotechnical analysis, field monitoring, and numerical modeling to assess the stability of cut slopes.

There have been various geophysical techniques employed to probe the subsurface properties contributing to slope instability in the ongoing effort to better understand and predict landslides, including Electrical Resistivity (ER), Self-Potential (SP), and Seismic Monitoring [148]. These methods are used for assessing the internal conditions of landslide-prone areas, providing critical information that supports the development of effective early warning systems. These techniques deliver detailed insights into the hydrological and mechanical processes driving landslides [149].

Table 1. Traditional methods for landslide monitoring [145,149].

Method	Description
Electrical Resistivity (ER)	Involves measuring the electrical resistance of subsurface materials to infer moisture content and structural variations. It effectively delineates moisture variations and other hydrological dynamics within landslide-prone areas.
Self-potential (SP)	SP monitoring detects natural electric potentials in the ground, indicating water flow paths and areas of potential weakness in the subsurface.
Seismic Monitoring	This method records and analyzes ground vibrations to identify movements within landslide bodies. It can differentiate between different landslide movements and detect precursory seismic signals that may precede a landslide event.

compares the traditional methods for landslide monitoring techniques. ER measures how subsurface materials resist electrical currents to reveal moisture content and structural changes. SP detects natural electric fields in the ground, showing water flow paths and potential weak spots. Seismic Monitoring captures ground vibrations to track landslide movements and predict possible events. These methods offer valuable insights for better predicting and managing landslide risks.

Traditional monitoring methods used various approaches that allow for in-depth and continuous monitoring [150], including Inclinometers (shown in Figure 2a), Extensometers (shown in Figure 2b), Piezometers (shown in Figure 2c), and Time Domain Reflectometry systems (TDR) (shown in Figure 2d). These systems are used to effectively detect soil movements and identify shear planes [151]. These methods can identify shear movements at comparable depths. However, TDR systems offer significant cost savings compared to inclinometers for remote soil monitoring, reducing the need for frequent site visits. In addition, TDR requires less installation and data collection times.

Traditional methods for detecting slope movement have relied heavily on in situ measurements and observations, such as inclinometers [152] and extensometers [153], which have been widely used due to their proven reliability and accuracy. These commonly used methods were surveying techniques to track changes in slope position over time and piezometers to measure underground water pressures affecting slope stability [149].

The inclinometer is an effective and cost-efficient tool for landslide monitoring and offers real-time and precise data crucial for risk assessment and emergency management. It is used to transmit real-time data to a central cloud server or to be integrated with other monitoring systems. It provides valuable ground-motion insights essential for proactive disaster management [156]. The inclinometer can also be used to gather data on the subsurface conditions to monitor the soil and rock properties and the landslide movements. The inclinometer readings are integral to modeling and assessing the landslide’s stability, contributing significantly to the overall geotechnical evaluation and enabling targeted and effective mitigation strategies. Using inclinometers highlights their value in providing real-time and precise monitoring critical for managing landslide risks effectively [149].

Inclinometers can effectively measure the tilt or inclination of reference structures, providing precise data on the angular movements related to landslide activities. The inclinometers’ high precision and good linearity allowed for accurate monitoring of the slope’s stability and movements, essential for predicting potential landslide events [157]. On the other hand, extensometers can measure soil displacement, which is crucial for understanding movement patterns and predicting landslide behavior. Implementing a wireless sensor network facilitated the installation in remote areas. It provided reliable, real-time data on soil displacement, enhancing the system’s ability to monitor and analyze landslide dynamics effectively [157].

Table 2. Inclinometers vs. extensometers [152,153].

Sensor Type	Pros	Cons
Inclinometers	Can measure the depth and direction of slope movements. In-place inclinometers provide continuous data, allowing for real-time monitoring. Can be installed in boreholes, making them less vulnerable to surface conditions and vandalism. Suitable for long-term monitoring of slope stability.	Require drilling of boreholes, increasing installation costs. Manual probe inclinometers require site visits for data collection, which may not be timely. Data interpretation can be complex and requires expertise. Vulnerable to mechanical damage.
Extensometers	Can measure displacement magnitude, which helps understand the extent of slope movement. Simple mechanical designs are inexpensive and easy to install. Electronic versions provide continuous data for real-time monitoring.	Surface installations can be prone to vandalism and environmental damage. May not provide information on the depth or direction of movement without multiple installations. Mechanical versions require manual data collection and might miss critical events. More complex electronic systems can be costlier and require power sources.

summarizes the specific conclusions drawn from using each type of sensor in the monitoring system, highlighting their effectiveness and applicability in geotechnical monitoring [157]. Also, the table outlines their pros and cons, capabilities, installation requirements, and limitations to determine their suitability for geotechnical monitoring applications.

The Limit Equilibrium Method (LEM) [158] has been used in geotechnical engineering for slope stability analysis. LEM relies on simplified models and assumptions about soil behavior and failure surface geometry, which can make slope stability predictions inaccurate. In addition, similar to inclinometers and extensometers, LEM often fails to capture the behavior of slopes under varied conditions, such as under dynamic loading or progressive failure mechanisms.

The piezometry tool provides critical data on subsurface hydrological conditions essential for understanding and predicting landslide reactivation. When correlated with geoelectrical resistance measurements and environmental factors such as rainfall, the piezometric data offer a comprehensive view of the dynamic processes affecting slope stability. This data integration allows for more accurate predictions of landslide movements and contributes significantly to developing effective monitoring and early warning systems [159].

Traditional methods for slope stability (e.g., inclinometers and extensometers) have limitations regarding installation difficulty, sensitivity to environmental factors, and cost, which can hinder their broader application in specific contexts. In a nutshell, technological improvements will be needed to enhance the reliability and applicability of these monitoring techniques [160]. In addition, the most significant use and impact of inclinometers and extensometers is observed in middle- to high-income countries, indicating that access to low-cost equipment alone may not be sufficient to overcome all barriers to scientific research without additional support like funding, collaboration, and infrastructure [161].

The existing traditional slope movement detection systems have several limitations, including regular maintenance, susceptibility to environmental conditions, and complex data interpretation [162]:

• Environmental Conditions: Once installed, external factors can influence their readings and affect the accuracy of detecting slope movements.
• Integration with modern technologies: Traditional and contemporary systems often generate data in different formats, making integration complex.

Table 3. Traditional methods for real-time data acquisition.

Method	Pros	Cons
Data Acquisition Units	Records precise measurements from sensors monitoring various environmental and physical parameters of landslides.	Requires setup and maintenance and may be vulnerable to environmental damage.
Remote Telemetry	Enables data transmission over long distances using radio, satellite, or cell phone links.	There are potential interruptions in data transmission depending on signal availability and strength.
Automated Data Processing	Processes large volumes of data quickly to provide timely updates on landslide conditions.	It is a complex setup and requires consistent calibration and validation to ensure accuracy.
Displays of Current Conditions	It offers immediate access to monitored data online, allowing for quick response.	It requires reliable internet connections and a power supply but may not be accessible during power outages.

compares the existing approaches for real-time monitoring of landslides. These systems are instrumental in providing continuous and precise data on landslide activities, which is crucial for timely decision making and implementing safety measures. The table compares the pros and cons of these methods, highlighting their respective advantages and disadvantages in different scenarios. This comparative analysis underscores each method’s importance in improving landslide monitoring and response strategies [71].

Table 4. Traditional and modern methods for assessing and managing landslide risks [162].

Method	Description
Geological Mapping	Used to investigate all types of landslides, providing detailed data on the geology and structure of the terrain.
Geotechnical Mapping	Includes using inclinometers to measure in-depth displacements and extensometers to monitor cracks and surface movements.
Remote Sensing Techniques	Includes the use of LiDAR and radar for the detection and monitoring of earth flows and other types of movements.
Geodetic Techniques	Employs technologies such as GNSS to monitor earth and rockslides, providing precise data on movements in real time.
Hydrogeological Techniques	Monitors hydrological conditions such as groundwater level and pore water pressure, which can influence slope stability.
Mapping Techniques	It includes geomorphological and engineering geological mapping, which is used for risk map production and studying the morphology and evolution of the terrain.

outlines the critical traditional and modern approaches to assessing and managing landslide risks, including Geological Mapping, Geotechnical Mapping, Geodetic Techniques, Hydrogeological Techniques, and Mapping Techniques [163].

The IoT network is essential for facilitating efficient communication between sensors, data loggers, and databases to enhance the outcomes provided by inclinometers, extensometers, and piezometry [164]. For instance, tilt sensors can also be used for landslide monitoring in four different modes [165]:

• Continuous Monitoring: High-resolution tiltmeters can continuously monitor internal deformations within a landslide body. This capability is crucial for understanding landslides’ dynamics over short and long periods.
• Decomposition of Movement: The tiltmeters allow for the decomposition of landslide movement into different components, such as long-term trends and seasonal variations. This detailed decomposition helps in better understanding the factors influencing landslide movements.
• Influence of Environmental Factors: Tilt measurements can effectively correlate the impact of environmental factors like rainfall, groundwater levels, and temperature changes on landslide behavior. This is vital for developing predictive models and early warning systems.
• Non-linear Responses: While tilt measurements provide valuable data, the landslide response to environmental triggers is not always linear. This suggests the complexity of the landslide’s response mechanisms, necessitating sophisticated analysis techniques to interpret the data accurately.

Using low-cost IoT sensors and devices, such as Arduino and IoT microcontrollers, for building early warning systems can significantly reduce costs by automating data collection and transmission compared to traditional methods, which often involve expensive geotechnical and geodetic equipment [166]. However, despite their lower cost, implementing these systems still requires careful consideration of financial resources, as they need customization and maintenance to operate effectively in diverse and often harsh environments [167].

Yadav et al. [168] presented a review of slope monitoring systems with a vision of unifying Wireless Sensor Networks (WSNs) and IoT. The authors examined slope monitoring technologies, particularly highlighting the integration of WSN and IoT, and explored their effectiveness in detecting and predicting slope failures in mining operations. The paper emphasized the technological advancements that enhance real-time data processing and alert systems. The study concludes that integrating WSN with IoT significantly enhances slope monitoring systems’ efficiency and accuracy in open-cast mines. This integration enables real-time data acquisition and analysis, leading to timely warnings and preventive measures against slope failures, thereby reducing potential risks and costs associated with mining operations, as shown in

Table 5. Examples of the slope monitoring systems used for mining operations.

Method	Description
Automated Total Station	Utilizes automated stations with optical systems to monitor slope movements. It can be hindered by adjacent machinery [1].
LiDAR	Employs Light Detection and Ranging technology to provide broad coverage and rapid update rates [169].
Slope Stability Radar (SSR)	Uses radar technology to continuously scan and monitor slope surface distortions, providing high-resolution measurements even in poor weather conditions [126].
GPS	Uses Global Positioning System technology to monitor discrete points on slopes, providing real-time data [170].
TDR	Time Domain Reflectometry uses electromagnetic signals to detect disturbances within slopes [171].
Photogrammetry	Applies photography in surveying and mapping to measure distances between points on slopes [172].
Microseismic Monitoring	Detects and analyzes microseisms induced by rock fracturing within the slope to anticipate potential failures.
Crack Meters	Monitors the opening of cracks within slopes to detect movements indicative of potential slope failures [128].
OTDR	Optical Time Domain Reflectometry monitors fiber optic cables within the slope to detect deformations that indicate movements.
Shape Accel Array (SAA)	Utilizes arrays of accelerometers to measure acceleration along a slope to determine movements and predict failures.
Ground-based SAR Interferometry	Employs radar signals to detect and monitor subtle slope movements over time, allowing for early detection of potential failures.
Wireless Sensor Networks (WSNs)	Implements networks of sensors that collect data on various physical parameters like moisture, pressure, and movement to continuously monitor and analyze slope stability [173].

.

5. Soil Hydrological Monitoring Systems Using Modern Technologies

Landslide early warning systems have used advanced technologies, including IoT [174], ML [175], WSN [70], UAVs [171], Geographic Information Systems (GIS) [176], and Global Positioning System (GPS) [170], in monitoring and predicting landslide occurrences. The integration of these technologies is vital, as they not only facilitate timely evacuations, thereby safeguarding human lives but also enable the implementation of necessary preventative measures. As such, technology is becoming an indispensable component of modern life, providing essential tools for managing natural disaster risks. This section focuses on recent studies that used such technologies and tools in landslide monitoring systems.

5.1. IoT and WSNs

The importance of IoT and WSN in enhancing landslide monitoring systems is underscored through their roles in real-time data collection and monitoring of critical parameters like soil moisture and geological movements, which are crucial for early landslide detection [173,174,177]. Understanding landslide-prone zones and coordinating multi-agency response operations depend on the continuous transfer of data made possible by integrating varied sensing devices across huge areas, made possible by the Internet of Things. WSN’s scalability and flexibility enable adaptive sensor deployment to satisfy changing monitoring requirements, improving system adaptability under varied circumstances. Additionally, the integration of WSN and IoT provides authorities with helpful information that enhances disaster response and preparedness, may even save lives, and lowers financial losses. WSN considerably strengthens the capabilities of landslide monitoring systems to improve the timeliness and accuracy of data for efficient disaster management and mitigation. This is due to its cost-effectiveness, operational efficiency, and capacity to guarantee data accuracy and reliability [178].

Table 6. WSN vs. IoT solutions for monitoring landslides [174,180,181].

	Characteristics	Advantages	Disadvantages	Applicability
WSN	Real-time monitoring and site-specific monitoring. It works on site-specific Early Warning Systems (EWSs).	High speed and resolution, Continuous technical development in processor, Communication, Low-power, Usage of embedded computing	Security against hackers, Security against vandalism	Directly applicable by programming intelligent algorithms in the sensor node.
IoT	Real-time monitoring, Site-specific monitoring, Integration with cloud computing for data aggregation and analysis. It works on site-specific EWSs, national-level EWSs with integration with broader disaster management systems.	Enables remote monitoring, Scalability, Integration with various data platforms and sensors, Real-time data processing and alerts.	Dependence on network connectivity, Potential issues with data privacy and security, Maintenance of sensors and network infrastructure.	Fully integrated with cloud platforms, data analytics, and machine learning for predictive analysis.

compares the landslide monitoring frameworks depending on WSN and IoT [70]. The type of movement for WSN and IoT involves debris flows, rock falls, rotational and translational slides, complex landslides, and deep-seated landslides [167]. The monitoring parameters for both technologies include rainfall, moisture, pore water pressure, deformation angle, distance, inclination, tilt, angle, displacement, vibrations, speed, sound, acceleration, strain, GPS, etc. [179].

The WSN and IoT frameworks involve several key components and processes that enhance the system’s effectiveness and reliability [182], including:

• Sensor Deployment: As shown in Figure 3, WSN comprises multiple sensors deployed across potentially unstable slopes or areas prone to landslides. These sensors are strategically placed to monitor various environmental and geological parameters, such as soil moisture, pressure changes, and vibrations, critical indicators of potential landslides.
• Data Collection: IoT devices with sensors continuously collect data in real time. This capability is central to the system’s effectiveness, ensuring that the data on critical parameters are updated constantly and available for immediate analysis.
• Connectivity and Communication: The sensors in the WSN are interconnected through a robust network that supports efficient data transmission to a central processing unit or server. IoT technology facilitates this connectivity, ensuring seamless communication across the network. Data from sensors are transmitted wirelessly, minimizing the need for extensive physical infrastructure and enabling easier network scaling.
• Data Processing and Analysis: Once collected, the data are transmitted to centralized systems, where they are processed and analyzed using advanced algorithms. IoT platforms integrate these data and enable sophisticated analytical processes to interpret the information [183], predict potential landslide activities, and determine the immediacy and severity of the risks.
• Scalability and Flexibility: The system is designed to be scalable and flexible, allowing for the addition or reconfiguration of sensors as monitoring needs evolve or as new technologies become available. This adaptability is crucial for accommodating different environmental conditions and expanding the system’s coverage as required.
• Disaster Response Coordination: IoT not only supports data collection and analysis but also plays a crucial role in disaster response coordination. The system can trigger alerts and communicate potential landslide warnings to local authorities and residents, facilitating timely evacuations and mitigations.
• Reliability and Cost-Effectiveness: WSN ensures the reliability of the data collected, which is crucial for the accuracy of landslide predictions and warnings. The cost-effectiveness of the network, coupled with low power consumption, makes it viable for long-term monitoring without requiring frequent maintenance or high operational costs.

Geotechnical applications use multiple techniques to quantify deformation, including fiber optic sensors, resistance-type measuring devices, and vibratory wire gages. Vibrating wire strain gauges are among the most reliable and long-lasting. These gauges work based on the idea that the frequency at which a tensioned wire vibrates is directly proportional to the strain the wire is under. Measuring the wire’s resonance frequency, dependent on its length, tension force, and cross-sectional area, is essential to figure out the strain. With IoT wireless sensor networks, landslide monitoring systems that provide real-time data and alerts on geological changes can successfully use this technology [164].

Figure 4 shows the reference IoT architecture for landslide monitoring systems [173,180], which involves the following components and layers:

• Real-Time Monitoring and Alert Systems: IoT devices continuously monitor geological changes and environmental variables. Data on soil movement, moisture content, and other crucial factors that point to the possibility of landslide activity are gathered by sensors buried in the ground or visible above it. Afterward, real-time alerts and cautions are generated using these data, allowing quick action to reduce risks.
• Predictive Maintenance: IoT systems can forecast possible landslide events ahead of time by evaluating the data gathered. Preventive measures are made more accessible by machine learning algorithms that analyze historical and current data to find patterns or abnormalities before a landslide.
• Environmental Monitoring: IoT sensors track several environmental variables, including changes in terrain, soil water content, and the amount of rainfall that might cause landslides. This thorough monitoring aids in determining an area’s susceptibility to landslides and putting suitable environmental management techniques into place.
• Perception Layer: This is the physical layer where IoT sensors and devices collect environmental and geophysical data. The interface between the physical world and the digital IoT system is crucial in acquiring accurate and timely data.
• Network Layer: The collected data are transmitted through this layer, which involves communication technologies and protocols to ensure that data are sent from sensors to processing units or cloud systems securely and efficiently.
• Middleware Layer: Acts as a bridge between the hardware and application layers. It manages device connectivity, data storage, and processing. It also includes tools and platforms that provide data analytics capabilities, essential for interpreting the vast amounts of data gathered by sensors.
• Application Layer: This layer is where specific applications that utilize the processed data are developed [183]. Landslide monitoring includes visualization tools, alert systems, and decision support systems that help stakeholders make informed decisions.
• Business Layer: Overseeing the entire IoT setup, this layer focuses on integrating IoT solutions into existing business and operational processes. It addresses strategic decision making, monetization, and scalability issues, ensuring that the IoT solutions align with organizational goals and contribute effectively to landslide risk management.

Figure 5 shows a conceptual framework of landslide monitoring techniques using IoT and WSN [180], which involves the following aspects:

• Type of Movement: Landslide movements are categorized into types such as fall, topple, slide, spread, and flow, each characterized by different dynamics and influenced by varying geological and environmental factors. Understanding these types helps in selecting the appropriate monitoring techniques for each scenario. For example, falls involve the sudden detachment and downward movement of rock or earth, while flows are characterized by the rapid movement of debris or mud down a slope.
• Landslide Monitoring Parameters: There is a range of monitoring parameters that are crucial for assessing landslide risks, including meteorological conditions (e.g., rainfall, geological parameters, soil composition, and structure), hydrogeological factors (e.g., water pressure and saturation), physical parameters (e.g., displacement and velocity), and geophysical parameters that help in understanding the subsurface characteristics. Effective monitoring of these parameters can aid in predicting landslide occurrences and mitigating their impacts.
• Monitoring Aspects: This topic is addressed through different monitoring techniques and their spatial and temporal resolution capabilities. It is essential to integrate real-time and near-real-time monitoring systems to enhance the responsiveness of landslide warning and intervention systems.

In ref. [177], the authors developed an IoT-based framework for landslide detection, monitoring, prediction, and warning within an IoT-cloud environment. The proposed approach used real-time environment monitoring to detect changes at landslide sites, enabling immediate data processing and communication of threshold-based alerts. The system utilized ensemble learning for enhanced prediction accuracy and incorporated a model that allowed updates and graphical data representation almost instantaneously after a landslide occurred. This intelligent approach improved responsiveness and precision in landslide risk prediction and management, contributing significantly to disaster readiness and mitigation. The experiments employed an array of sensors integrated through an IoT framework, including Soil Moisture Sensors, Accelerometers, Ultrasonic Sensors, Rain Gauges, Arduino Uno Board, and GSM Modules. This experiment showcased a comprehensive approach to using technology for effective landslide management, highlighting the integration of various sensors and real-time data processing within an innovative IoT framework.

Low-Power Wide-Area Network (LPWAN) [178] is a networking technology designed to connect low-bandwidth, battery-powered devices with low bit rates over long distances. It is specifically recognized for its ability to maintain connectivity over extensive ranges using minimal power. It is suitable for IoT applications where devices must communicate periodic small-scale data over extensive periods without requiring frequent battery replacements. LPWAN is primarily employed in various IoT applications due to its efficient power usage and long-range capabilities. LPWAN is utilized in landslide monitoring systems as it facilitates the deployment of sensor networks across expansive and possibly inaccessible areas. It provides a continuous stream of data from sensors monitoring various parameters like soil movement, moisture levels, and other critical factors that can predict a landslide. The necessity of LPWAN in landslide monitoring stems from its low power consumption and long-range communication capabilities, which are crucial for installing sensors in remote or difficult-to-access locations. These locations often need regular maintenance opportunities and reliable long-term data transmission capabilities to ensure continuous monitoring. LPWAN technologies support the deployment of extensive sensor networks without frequent battery changes or extensive maintenance, which is essential for tracking in unstable and inaccessible terrains where landslides are a risk [184].

5.2. Remote Sensing Technologies (RSTs)

Remote sensing technologies (RSTs) [185] refer to the techniques that involve observing and collecting data about the Earth’s surface without physically contacting the observed area. This is achieved through various sensors and instruments mounted on satellites, aircraft, drones, or ground-based platforms. These technologies capture data across different electromagnetic spectrum wavelengths, including visible, infrared, and microwave bands. RSTs have been recognized as valuable tools for studying natural phenomena, including landslides, because they provide systematic and extensive information on various land surface parameters over different spatial and temporal scales. RSTs have a wide range of applications, including:

• Environmental Monitoring: Tracking changes in land use, vegetation cover, and water bodies.
• Disaster Management: Detecting and monitoring natural disasters such as earthquakes, floods, hurricanes, and landslides.
• Urban Planning: Assisting in city planning and infrastructure development.
• Agriculture: Monitoring crop health, soil moisture, and predicting yields.
• Climate Change Studies: Observing and analyzing changes in climate patterns and their impacts on the environment.
• Geological Mapping: Identifying geological features and mineral exploration.

RST techniques, such as satellite-based Synthetic Aperture Radar (SAR) and multispectral imaging, allow for monitoring ground displacement over large areas. Interferometric SAR (InSAR) can measure ground movements with sub-millimeter accuracy, making it possible to detect even slow-moving landslides. These technologies can identify terrain, vegetation, and soil moisture changes that indicate potential landslide activity. Additionally, ground-based methods like LiDAR and interferometric radar offer high spatial and temporal resolution for detailed monitoring of specific slopes, enabling early warning systems and timely interventions.

We discuss here the most commonly used sensors and devices in remote sensing for landslide detection and monitoring, including:

• Multispectral Satellite Sensors: Instruments that capture data in multiple wavelengths of light, such as Landsat, SPOT, and Sentinel-2.
• Synthetic Aperture Radar (SAR): Active sensors that use microwave signals to measure ground surface deformations, including Sentinel-1 and TerraSAR-X.
• Interferometric Synthetic Aperture Radar (InSAR): A technique that uses SAR data from different time points to detect ground movement.
• Light Detection and Ranging (LiDAR): Ground-based or airborne laser scanning systems that create detailed 3D models of the terrain.
• Ground-Based Interferometric Radar (GBInSAR): Devices that accurately measure ground displacement and deformation and are used for real-time monitoring and early warning systems.
• Doppler Radar: Used for detecting rapid movements such as rockfalls and debris flows.

Table 7. A comparison between the remote sensing technologies (RSTs). ES (Earth Slide); RS (Rock Slide); EF (Earth Flow); LS (Lateral Spread); SL (Shallow Landslide); DF (Debris Flow) [185].

Technique	Detection	Monitoring	Prediction
MTInSAR	ES, RS, EF, LS, SL	ES, RS, EF, LS, SL	ES, RS
Multispectral satellite sensors	ES, RS, EF, LS, SL, DF	ES, RS, EF, LS, SL	ES, RS, EF
Ground-based interferometry	ES, RS, LS	ES, RS, EF, LS, SL	ES, RS, EF
Doppler radar	RF	RF, DF, SL	—
Lidar	ES, RS, EF, LS, RF	ES, RS, EF, LS	—

compares RST techniques for landslide movements in terms of the detection techniques, monitoring system, and prediction (time of failure) using the following parameters:

• ES (Earth Slide): Movement of earth material, typically slow-moving, with velocities ranging from less than 1.6 m per year to 1.8 m per hour.
• RS (Rock Slide): Downward and outward movement of rock material, with velocities from less than 1.6 m per year to more than 1.8 m per hour.
• EF (Earth Flow): Flow of fine-grained earth materials, with velocities from 1.6 m per year to 1.8 m per hour.
• LS (Lateral Spread): Horizontal movement of soil or rock, usually on gentle slopes, with velocities from less than 1.6 m per year to 1.8 m per hour.
• SL (Shallow Landslide): Movement of soil and debris near the surface, with velocities from less than 1.6 m per year to more than 1.8 m per hour.
• DF (Debris Flow): Rapid flow of a mixture of water, soil, and rock, with velocities greater than 1.8 m per hour.

With its extensive landslide-prone regions, China increasingly relies on remote sensing technologies for investigation and monitoring. Significant advancements have been made using Synthetic Aperture Radar (SAR), optical remote sensing, and LiDAR across space-borne, air-borne, and ground-based platforms. Multi-temporal optical images and time series SAR detect active landslides over years and months, while the latest optical and SAR intensity images map fresh, especially coseismic, landslides. LiDAR effectively identifies ancient landslides. To maximize the strengths and mitigate the limitations of each technology, an integrated space–air–ground collaborative strategy has been proposed for early identification and warning. Future comprehensive landslide investigations may integrate remote sensing with geology and geophysical exploration to provide a more complete understanding, as remote sensing alone offers only surface information. In ref. [186], the authors presented a smart approach using remote sensing techniques, which used the following technologies: Synthetic Aperture Radar (SAR) and LiDAR.

Space-borne InSAR and optical remote sensing are used for large-scale and historical landslide identification. At the same time, air-borne UAV photogrammetry and LiDAR provide high-resolution, short-term monitoring of active landslides. Ground-based GBSAR and terrestrial LiDAR offer detailed surveys and real-time tracking [185,187]. This collaborative strategy leverages the strengths of each technology, enabling comprehensive monitoring, cross-validation of data, and accurate risk assessments, ultimately improving landslide management and mitigation efforts.

• Space-Borne Platforms

  –

  InSAR: Utilized for early identification and monitoring of active landslides. The advantage lies in its ability to provide high-accuracy measurements regardless of weather conditions. InSAR has been instrumental in identifying deformation in a wide range of areas.

  –

  Optical Sensors: Used for geological and topographic investigations, boundary mapping, and macro deformation analysis. Space-borne optical sensors provide wide-area coverage with high spatial resolution, making them suitable for mapping large-scale landslides.

• Air-Borne Platforms

  –

  Optical Cameras: Air-borne optical sensors are used for detailed surveys of landslides, particularly in sparsely vegetated areas. UAVs (Unmanned Aerial Vehicles) are commonly employed to capture high-resolution images that can be used to create 3D models of the terrain.

  –

  LiDAR: Air-borne LiDAR systems are effective in densely vegetated areas. They provide detailed 3D information by filtering out vegetation to reveal the actual ground surface.

• Ground-Based Platforms

  –

  Ground-Based Synthetic Aperture Radar (GBSAR): Used for real-time monitoring of specific landslides, providing high-resolution data on ground displacement and deformation.

  –

  Laser Scanners: Ground-based laser scanners (terrestrial LiDAR) are used for detailed surveys of steep slopes and rock faces, offering high-resolution data for small areas.

UAVs have become essential tools in landslide monitoring because they provide high-resolution data quickly and efficiently. They are precious in on-site investigations, where traditional methods may be too time-consuming, costly, or dangerous. UAVs offer a rapid response option, capturing detailed images and data that aid in assessing and monitoring landslide-prone areas, especially in remote or inaccessible regions. Some of the UAV applications are listed here:

• Geological Surveys: UAVs are used for detailed mapping and characterizing landslides, identifying features such as scarps, fissures, and micro-landforms.
• Dynamic Monitoring: UAVs monitor temporal and spatial changes in landslide areas, providing data on surface deformation, volume estimation, and crack detection.
• Emergency Response: UAVs play a crucial role in post-landslide scenarios, aiding search and rescue operations and providing real-time data for decision-making.

As shown in Figure 6, different IoT sensors can be installed on UAVs, including:

• RGB Cameras: Capture high-resolution visible light images, useful for mapping and identifying morphological features shown in Figure 6a.
• Multi-Spectral Cameras: Record data across various spectral bands, enabling detailed spectral analysis for vegetation and geological studies shown in Figure 6b.
• Thermal IR Cameras: Detect emitted thermal radiation, identifying temperature variations associated with geological instability shown in Figure 6c.
• SAR Sensors: Use radar to produce high-resolution images regardless of weather conditions, allowing for precise monitoring of surface changes and deformations shown in Figure 6d.
• LiDAR Sensors: Emit laser pulses to measure distances and create detailed 3D models of the terrain, useful for mapping and monitoring surface deformations shown in Figure 6e.

In ref. [169], UAVs equipped with RGB and multi-spectral cameras capture high-resolution images to map landslide morphological features and create orthophotos and digital elevation models (DEMs). Thermal IR cameras detect thermal anomalies to identify geological instability, aiding in detecting cracks and fissures. LiDAR sensors generate precise 3D terrain models for detailed surface analysis, while SAR sensors monitor surface changes and deformations with high accuracy, supporting early warning systems and risk assessments. Integrating these sensors provides comprehensive data, enhancing the understanding of landslide dynamics, supporting hazard assessment, and enabling timely and effective responses to landslide events.

The typical acquisition and processing workflow for UAV images involves several key steps. First, flight planning is essential, considering factors such as Ground Sampling Distance (GSD) to ensure optimal image resolution and maintaining an image overlap of 60% to 80% for comprehensive coverage. This can be set using flight planning software. Additionally, georeferencing is enhanced by strategically capturing Ground Control Points (GCPs) with known locations. During the flight, a UAV equipped with a camera captures multiple images from varying angles, with camera orientation being either nadir or oblique. These images undergo photogrammetric processing, where features like corners and edges are extracted from each image and matched across images through camera triangulation, which calculates the relative positions and orientations of the cameras. This process leads to dense image matching, comparing every pixel in each image to identify matches, resulting in a dense point cloud that portrays the 3D structure of the imaged terrain or object.

Remote sensing technologies for landslide detection, monitoring, and prediction face several challenges. Weather conditions and atmospheric disturbances, such as cloud cover and persistent snow, can hinder satellite imagery and affect the quality of SAR data. Resolution and accuracy limitations exist for multispectral and SAR satellites, with ground-based sensors providing higher accuracy but limited coverage. Complex data interpretation and processing are required for techniques like InSAR and Doppler radar, necessitating advanced algorithms to filter out false alarms and enhance data reliability. Data integration from various remote sensing sources presents interoperability issues, requiring advanced fusion algorithms and improved standardization and data-sharing practices. High costs for equipment, deployment, and maintenance of high-resolution sensors and ground-based systems limit the extent and duration of monitoring campaigns, particularly in resource-constrained regions. Balancing cost and coverage remains a challenge, necessitating strategic resource allocation and continued technological advancements to maximize the effectiveness of remote sensing technologies in landslide management [187].

Table 8. Satellite vs. aerial vs. ground-based remote sensing [185,187].

Technology	Pros	Cons
Satellite Imagery	Wide area coverage, frequent revisit times, and multi-spectral data.	Lower spatial resolution and potential cloud cover issues.
Aerial Photography	High spatial resolution and flexible deployment.	Limited by weather conditions and can be costly for large areas.
LiDAR	High accuracy can penetrate vegetation and generate detailed 3D models.	Expensive and requires complex data processing.
Synthetic Aperture Radar (SAR)	All-weather capability and can detect ground deformation.	Complex data interpretation and lower spatial resolution compared to optical sensors.
Ground-Based Remote Sensing	High spatial and temporal resolution for specific sites.	Limited coverage area and may require more labor-intensive deployment.

highlights the pros and cons of satellite, aerial (i.e., UAV), and ground-based remote sensing.

5.3. Machine Learning

Machine learning (ML) has emerged as a transformative technology in various domains, including natural hazard management. Its ability to analyze vast amounts of data and identify complex patterns has made it a valuable tool in monitoring and predicting landslides. Landslides pose significant risks to human lives and infrastructure, making efficient monitoring and early warning systems crucial. This introduction explores how ML contributes to landslide monitoring systems and Landslide Early Warning Systems (LEWSs) [164], designed to predict landslides and provide timely warnings to mitigate potential damage. Incorporating ML into these systems enhances their predictive accuracy by analyzing various environmental factors and historical data. Studies have shown that ML models can process and interpret large datasets, improving the reliability of LEWS.

Wang et al. [188] proposed an ML-based approach to identify natural terrain landslides. By integrating geodatabases comprising topographic, geological, and rainfall-related data, the study evaluated various ML algorithms, including logistic regression (LR), Support Vector Machine (SVM), Random Forest (RF), boosting methods, and Convolutional Neural Networks (CNNs). The case study conducted on Lantau Island, Hong Kong, demonstrated that CNN outperformed other algorithms, achieving an identification accuracy of 92.5% on recent landslides due to its superior feature extraction and multidimensional data processing capabilities. Another study presented in ref. [175] explored the application of ML in LEWS, focusing on the integration of real-time monitoring data with predictive models. The research highlighted the effectiveness of ML algorithms such as RF and gradient boosting in processing sensor data and environmental variables to predict landslide occurrences [189,190,191,192]. By continuously learning from new data, these models adapt to changing conditions, enhancing the system’s ability to provide accurate warnings.

LEWS consists of five major components [166]: data collection, data transmission, modeling, analysis and forecasting, warning, and response. However, traditional LEWSs often face challenges such as data transmission failures during extreme weather conditions, leading to gaps in monitoring and delayed warnings. Learning algorithms address these challenges by providing alternate solutions when data collection or transmission components fail. For instance, nowcasting algorithms use historical data to estimate current conditions even when real-time data are unavailable, maintaining the system’s operational reliability [192]. Similarly, forecasting algorithms predict future slope stability conditions, providing additional lead time for early warnings and reducing the risk of last-minute alerts. These advancements ensure continuous monitoring and accurate predictions, making LEWS more robust and dependable during critical periods [193].

Machine learning significantly enhances the capabilities of landslide monitoring and early warning systems. ML contributes to more effective risk management and mitigation strategies by automating the identification process and improving predictive accuracy. The integration of ML in these systems not only streamlines data analysis but also ensures timely and reliable warnings, ultimately protecting lives and infrastructure from the devastating impacts of landslides [190].

Table 9. The pros and cons of using ML in landslide detection [175,188,191].

Pros	Cons
High Accuracy: ML algorithms and deep learning models like CNNs achieve high accuracy in landslide detection by effectively processing complex and high-dimensional data.	Data Dependency: ML models require large amounts of high-quality data for training, which may only sometimes be available.
Automation: Automated detection reduces the need for manual interpretation, saving time and labor.	Overfitting: ML models, especially complex ones, can overfit the training data, reducing their performance on unseen data.
Scalability: ML algorithms can be scaled to process large datasets and cover extensive geographical areas.	Computational Resources: Training and running complex ML models require significant computational resources and infrastructure.
Integration of Multiple Data Sources: ML models can integrate diverse data types such as topographic, geological, and meteorological data for comprehensive analysis.	Complexity: The complexity of ML models can make them difficult to interpret and understand, which can be a barrier to their adoption.
Early Warning Systems: ML-based LEWS can provide timely alerts based on real-time data, potentially saving lives and reducing property damage.	Data Quality Issues: The quality of input data, such as resolution and accuracy, significantly impacts the performance of ML models.
Adaptability: ML models can adapt to different regions and conditions, making them versatile tools for landslide detection.	Maintenance: ML models require ongoing maintenance and updating to incorporate new data and adapt to changing conditions.
Enhanced Feature Extraction: Advanced algorithms can automatically extract and prioritize relevant features from the data, improving model performance.	False Positives/Negatives: Despite high accuracy, ML models can still produce false positives and negatives, undermining system trust.

summarizes the pros and cons of using ML in landslide detection.

5.3.1. ML Approaches for Landslide Detection

• Supervised Learning: Supervised learning algorithms are extensively used in landslide detection due to their ability to classify and predict based on labeled datasets. Logistic regression (LR), Support Vector Machines (SVM), and Random Forests (RF) [190] are prominent examples. Logistic regression, a statistical model, helps in predicting the probability of landslides by analyzing the relationship between dependent and independent variables. SVM, known for its effectiveness in high-dimensional spaces, classifies landslide-prone areas by finding the hyperplane that best separates the data into different classes [193]. Random forests, an ensemble learning method, enhance prediction accuracy by constructing multiple decision trees and merging their results. These algorithms are crucial in analyzing various factors such as topography, geology, and rainfall data to identify and predict landslide occurrences [191,192,194].
• Unsupervised Learning: Unsupervised learning techniques, particularly clustering algorithms, play a significant role in identifying patterns in landslide-prone areas without needing labeled datasets [195]. These algorithms, such as K-means clustering and hierarchical clustering, group similar data points based on their characteristics. In landslide detection, clustering helps recognize natural groupings within the data, such as regions with similar geological features or rainfall patterns. By analyzing these clusters, researchers can identify areas that share common characteristics with previously known landslide-prone zones, thus improving the understanding and prediction of potential landslide sites [188].
• Deep Learning: Deep learning techniques, especially convolutional neural networks (CNNs) [196], have successfully handled large and complex datasets for landslide detection. CNNs can extract features from high-dimensional data, such as satellite images and digital terrain models (DTMs). These networks have multiple layers that automatically learn to identify patterns and features relevant to landslide detection. CNNs excel in processing spatial data, making them ideal for analyzing terrain changes and identifying landslide scars. The ability of deep learning models to handle vast amounts of data and learn intricate patterns makes them powerful tools for enhancing the accuracy and efficiency of landslide detection and prediction.

5.3.2. Data Collection and Preprocessing

• Data Types: For machine learning (ML) applications in landslide detection, various data types are crucial. Topographic data include digital elevation models (DEMs) and slope gradient maps [175,197], which provide information about the terrain’s shape and steepness. Geological data comprise soil types, rock formations, and fault lines, essential for understanding the subsurface conditions. Meteorological data encompasses rainfall, temperature, and humidity, key factors influencing landslide triggers. These data types collectively enable comprehensive analysis and accurate landslide prediction [188].
• Data Integration: Integrating various data sources into a cohesive geodatabase is essential for effective analysis [177,198]. This involves compiling data from multiple sensors and surveys into a unified system. For example, topographic data from DEMs can be combined with geological maps and meteorological records to form an integrated geodatabase. This geodatabase allows for efficient querying and analysis, enabling the identification of correlations and patterns relevant to landslide occurrences [189,193].
• Data Cleaning: Data cleaning involves removing duplicates, correcting errors, and handling missing values to ensure the dataset’s integrity [195].

5.3.3. Feature Engineering

• Key Features: Identifying significant features is crucial for the performance of landslide detection models. Slope gradient is a primary feature, as steeper slopes are more prone to landslides. Aspect, which indicates the slope’s orientation, affects moisture retention and sunlight exposure, influencing landslide risk. Curvature helps identify concave or convex surfaces, which indicate potential accumulation or dispersion of materials. The Topographic Wetness Index (TWI) measures the potential water accumulation in the terrain, a critical factor in landslide initiation. These features are fundamental in capturing the physical and environmental conditions leading to landslides [191].
• Feature Importance and Selection Methods: Assessing and selecting important features involves several methods. Correlation analysis identifies the relationships between variables, helping select features that significantly impact landslide occurrences. Principal Component Analysis (PCA) reduces the dataset’s dimensionality by transforming it into uncorrelated variables, retaining most of the variance. Feature importance rankings from tree-based models, such as random forests, provide insights into which features contribute most to the model’s predictions. These methods ensure the most relevant features are used, enhancing the model’s efficiency and accuracy [199].

5.3.4. Model Training and Evaluation

• Training Techniques: Training ML models for landslide detection involves various techniques. Cross-validation splits the data into training and validation sets multiple times to ensure the model’s robustness. Hyperparameter tuning optimizes the model’s parameters to improve performance. Ensemble methods, such as bagging and boosting, combine multiple models to enhance prediction accuracy and reduce overfitting. These approaches ensure that the ML models are well-trained and capable of generalizing to new data [175].
• Evaluation Metrics: Model performance is evaluated using several metrics. Accuracy measures the proportion of correctly predicted instances. Precision indicates the proportion of true positive predictions among all positive predictions. Recall measures the ability to identify all actual positives. The F1 score balances precision and recall, providing a single metric for model performance. The Area Under the ROC Curve (AUC-ROC) evaluates the model’s ability to distinguish between classes, with higher values indicating better performance. These metrics comprehensively evaluate the model’s effectiveness in landslide detection [174].

5.4. Geographic Information Systems (GIS) and Global Positioning Systems (GPS)

Geographic Information Systems (GIS) and Global Positioning Systems (GPS) have become integral tools in the study and management of landslides. GIS is a framework for gathering, managing, and analyzing spatial and geographic data. It enables the visualization, interpretation, and understanding of data patterns, relationships, and trends through maps, globes, reports, and charts [184]. GPS, on the other hand, is a satellite-based navigation system that allows users to determine their precise location anywhere on Earth. GPS technology is essential for tracking and monitoring movements on the Earth’s surface, making it particularly useful in landslide monitoring and management [170].

Monitoring landslides involves collecting and analyzing data on land movements and the factors that trigger these movements. GIS and GPS technologies are crucial in this regard. GIS facilitates the integration and analysis of various types of spatial data, such as topography, geology, and hydrology, which are critical for understanding landslide dynamics [200]. By overlaying different data layers, GIS helps identify landslide-prone areas and assess the potential impact of landslides on these regions.

GPS, with its ability to provide precise location data, is used to monitor the displacement and deformation of the ground surface. For example, a study implemented a monitoring network using single-frequency GPS sensors to monitor an active landslide in the Carnic Alps, northeastern Italy, comprising 12 single-frequency GPS stations and other sensors that provided daily reports of landslide motion, proving to be a valuable, flexible, and cost-effective tool for quick landslide characterization and early warning [201].

LEWS relies heavily on the integration of GIS and GPS technologies. LEWS aims to provide timely warnings of potential landslide occurrences, allowing for preventative measures to mitigate damage and save lives. The GPS sensors’ ability to detect minute ground movements in near-real time is critical for early warning. These sensors can detect the initial slow movements of a landslide, which can indicate a potentially more significant movement, providing essential data for LEWS [144].

Table 10. GPS vs. GIS for landslide monitoring [176,200].

Technology	Pros	Cons
GPS	High-precision monitoring of ground movements Integral to Landslide Early Warning Systems (LEWSs) Cost-effective solutions with single-frequency sensors	Requires continuous power supply and maintenance Limited by environmental conditions (e.g., dense vegetation) Initial setup costs can be high for comprehensive systems
GIS	Integration of various spatial data types Effective in landslide susceptibility mapping Facilitates temporal analysis of changes over time	Data integration can be complex and time-consuming Accuracy depends on the quality of input data Requires skilled personnel for effective use

summarizes the pros and cons of GPS and GIS. Below, we discuss the major GIS and GPS techniques for landslide monitoring.

• GIS Techniques for Landslide Monitoring

  –

  Integration of Spatial Data: GIS is instrumental in integrating various spatial data types, such as topography, geology, hydrology, and vegetation cover. This integration helps understand landslides’ spatial patterns and relationships, enabling better risk assessment and management [202].

  –

  Landslide Susceptibility Mapping: GIS is widely used in landslide susceptibility mapping [170,202], which involves identifying areas prone to landslides by analyzing conditioning factors like slope, aspect, and soil type. These maps are crucial for planning and mitigating landslide risks [202].

  –

  Temporal Analysis: GIS facilitates temporal analysis, allowing researchers to study changes over time in landslide-prone areas. This capability is essential for understanding the progression of landslides and the effectiveness of mitigation measures [176].

• GPS Techniques for Landslide Monitoring

  –

  High-Precision Monitoring: GPS technology provides high-precision data on ground movements. Single-frequency GPS sensors, in particular, offer a cost-effective solution for continuously monitoring landslide-prone areas, enabling near real-time data acquisition and analysis [150].

  –

  LEWS: GPS is integral to Landslide Early Warning Systems (LEWSs), providing critical data that can predict potential landslide events. Detecting minute displacements helps issue timely warnings to prevent damage and save lives [164].

  –

  Case Study—Carnic Alps: A practical example of GPS application is the monitoring system implemented in the Carnic Alps, northeastern Italy. This system comprises 12 single-frequency GPS stations and provides daily reports of landslide movements, demonstrating its effectiveness in real-world scenarios [69].

  –

  3D Displacement Monitoring: GPS systems can monitor three-dimensional displacements, providing comprehensive data on the movement patterns of landslides. This capability is crucial for understanding the kinematics of landslides and implementing appropriate mitigation strategies [171].

6. Conclusions

In geotechnical engineering, detecting and monitoring slope movements are critical in understanding and mitigating landslide hazards. The impact of unsaturated soil hydrology on slope stability is a multifaceted issue that requires an integrated approach combining theoretical knowledge, empirical research, field implementation, and advanced modeling. The interplay between unsaturated soil hydrology and slope stability is one of the critical areas in geotechnical engineering, with significant implications for the durability, resiliency, and safety of infrastructure. This article has explored the intricate relationships between soil moisture content, matric suction, and shear strength, highlighting how these factors collectively influence slope stability in unsaturated soils. The article provides a comprehensive understanding of how hydrological processes impact slope stability by delving into the theoretical frameworks, relevant studies, and field implications.

One of the key findings of the article is the pivotal role of matric suction in enhancing the shear strength of unsaturated soils. Matric suction, arising from the presence of both air and water in soil pores, contributes significantly to the soil’s ability to resist shear forces. This enhanced shear strength is a crucial factor in maintaining slope stability, particularly in regions where soils are frequently unsaturated due to climatic conditions or fluctuating water tables. The article also emphasized the dynamic nature of slope stability in response to hydrological changes such as rainfall infiltration. Rainfall events can reduce matric suction and decrease soil shear strength, increasing the likelihood of slope failures. These findings underscore the importance of continuous monitoring and adaptive management strategies to account for the transient nature of hydrological influences on slope stability.

Landslide identification is a critical component of risk assessment and mitigation. Traditional techniques such as inclinometer, extensometer, piezometer, TDR, and geophysical system are highlighted as essential tools for real-time monitoring of soil moisture and matric suction. However, these traditional methods of landslide detection, which rely on visual interpretation of remote sensing images and topographic surfaces, are often time-consuming and labor-intensive.

With technological advancement, the integration of the IoT and AI in monitoring soil hydrologic behavior and its mechanical responses represents a transformative approach to landslide monitoring. The advent of ML has revolutionized this process by enabling automated or semi-automated methods for landslide identification using remote sensing techniques. In addition, integrating GIS and GPS technologies has revolutionized the field of landslide monitoring and management. GIS provides a powerful data integration and analysis platform, while GPS offers precise location tracking essential for monitoring ground movements. Together, they form the backbone of practical Landslide Early Warning Systems, enhancing our ability to mitigate the impacts of landslides through timely and informed decision making [184].