A Critical Integrative Review of Slope Stability Prediction in Open-Pit Mining: Geotechnical Indicators, Modelling Paradigms, Uncertainty-Aware Intelligence, and Emerging Monitoring Systems

2.2. Review Scope and Evidence Mapping

The reviewed studies are primarily from the last 6 years (from 2020 to 2025) due to the recent development of the artificial intelligence, remote sensing, probabilistic modelling, real-time monitoring and hybrid decision-support systems for open-pit slope stability prediction. Earlier approaches that are still relevant to recent developments, such as limit equilibrium analysis or finite element modelling or rock mass classification, are discussed.

The evidence base focuses on open pit and open cast mine slopes. Relevant studies were those that dealt with predicting the stability of slopes, assessment of slope failure, slope deformation monitoring, probabilistic risk assessment, geotechnical modelling, environmental degradation, remote sensing, machine learning and/or decision support systems in the context of mine slopes. Research on natural slopes, tunnels, embankments, dams or underground mines was not considered as core evidence unless there was a clear connection between the methods used and open-pit slope prediction.

The selected evidence found its way into five general families: numerical modelling and geomechanical modelling, probabilistic modelling and uncertainty-aware modelling, remote sensing and monitoring, machine learning and artificial intelligence, hybrid/integrated decision-support systems. This mapping was not intended to be a classification of approaches, but rather a tool for organizing the thinking that occurred during this review as this was the primary goal of this review, to get a sense of how these approaches work together in terms of slope stability intelligence.

2.3. Literature Identification and Evidence Orientation

A structured but flexible search of academic literature, which contains indexes on research related to mining, geotechnical, remote sensing, and computational intelligence, was conducted to identify relevant literature. The primary academic sources used were ScienceDirect, Springer Nature Link, IEEE Xplore and Taylor & Francis Online while Google Scholar served as a secondary source to locate interdisciplinary or recently indexed research. The search was not designed to be a complete systematic review protocol, but rather it was employed as a tool to build an even evidence base on the major stability prediction technical traditions used in open-pit slope design.

The search terms included open-pit mining, open-cast mining, slope stability, slope failure, slope monitoring, prediction, risk assessment, deformation, numerical modelling, finite element modelling, limit equilibrium analysis, probabilistic analysis, random fields, Bayesian networks, machine learning, artificial intelligence, InSAR, LiDAR, radar, UAV, IoT, rainfall, groundwater, blasting, weathering, freeze-thaw degradation, cohesion, friction angle, RMR and GSI. The following terms were employed to assure that the primary methodological families and indicator groups for slope prediction in the review were identified.

The search process was not only employed as a screening process, but also as a means to facilitate thematic evidence mapping. Studies were read in the context of their conceptual and practical contribution – either explaining the mechanisms of slope failure, quantifying uncertainty, detecting deformation, predicting patterns of non-linearity, or early warning – or contributing to integrated decision support. This enabled the review to extend beyond the story of finding a suitable model in a data base to a more holistic approach that developed into an understanding of the contribution of the various modelling traditions to the management of the stability of open-pit slopes. Database searches were re-run and checked on 14 May 2026to ensure the last evidence base was up-to-date and the chosen studies sufficiently covered the major methodological categories in slope stability prediction of 2020-25 (Table 1).

Table 1. Search transparency matrix for structured evidence identification.

2.4. Evidence Selection and Relevance Criteria

The studies were included only if they made a clear technical or conceptual contribution to open-pit or open-cast mine slope stability prediction, monitoring, risk assessment or decision support. Journal articles and full conference papers that were peer-reviewed and reported on geotechnical indicators, modelling techniques, monitoring technologies, empirical results or the implications for the use of open-pit slopes were given priority.

The selection process was based on the relevance and conceptual contribution of the applicants, and not on the numerical score. This is appropriate as the studies are reviewing different types of evidence and aims. In fact, there is no single quality score that can be used to provide a fair comparison of a finite element model, an InSAR monitoring study, a random forest classifier, a Bayesian risk model, and a hybrid AI–physics framework. The value is associated with their various roles in slope stability prediction. A finite element model, an InSAR monitoring study, a Random Forest classifier, a Bayesian risk model, and a hybrid AI-physics framework would be unfairly compared using a numerical score since the objectives, data configuration, validation process and use of these studies are different (Table 2).

Table 2. Study selection criteria for the critical review.

2.5. Methodological Interpretation of Evidence

The selected studies were discussed in terms of their methodological approach, their sources of data, their geotechnical markers, their models and their use, their methods of validation and their usefulness. The numerical modelling studies were discussed for their capability to explain the mechanical behavior, stress redistribution, deformation, hydro-mechanical coupling, response to rainfall, blasting effects, hydro–mechanical degradation and weathering. The interpretation of the probabilistic studies was based on the methodology used to represents the uncertainty of the parameters, spatial variability and reliability logic and probability of failure.

The use of remote sensing and monitoring studies in detecting deformation, capturing the spatial-temporal movement of slopes, providing early warning and monitoring inaccessible or high-risk slopes was considered. The studies of machine learning and artificial intelligence were translated based on the following input variables, predictive performance, feature importance, explainability, data quality, overfitting risk and cross-site transferability. The studies were split into hybrid and integrated studies, with the latter being evaluated based on their capability to link together physical modelling, monitoring data, probabilistic reasoning and decision-support functionality.

This interpretive approach supports the integrative purpose of the review. It does not attempt to place all the evidence on a single evaluation scale, but rather it identifies the contribution of each paradigm to a comprehensive intelligence framework of slope stability (Table 3).

Table 3. Methodological appraisal framework.

2.6. Cross-Paradigm Thematic Synthesis

Cross-paradigm thematic synthesis was used to integrate the evidence. This was done as the reviewed studies do not attempt to try to solve the same type of prediction problem wholly. There are no particular models for numerical explanation that are available for the main mechanical behavior and stress redistribution. Probabilistic models numerically represent uncertainty and the likelihood of failure. The remote sensing techniques were used to detect the deformation and to obtain spatial-temporal monitoring. Machine learning models are used to discover the predictive nonlinear patterns from the data. Hybrid systems try to integrate physical modelling, monitoring measurement data, uncertainty reasoning, and AI-based decision making.

The five lenses used in the review to integrate the various types of evidence were geotechnical indicators, modelling logic, uncertainty treatment, monitoring capability and decision-support functionality. For each study, the interpretation was not only based on the reported performance but also on the added value of the study to the overall body of knowledge of open pit slope stability prediction. Examples include a finite element study that provides the mechanistic explanation, an InSAR study that provides information on deformation observability, a Bayesian model that provides risk interpretation, and a machine learning model that provides nonlinear pattern recognition.

The integral contribution of the review is to gather together these contributions. Taken together, these findings indicate a transition from a single factor-of-safety analysis of open-pit slope stability to a dynamic and evolving process of geotechnical intelligence based on uncertainty and monitoring. This synthesis logic is also the foundation for the Adaptive Slope Stability Intelligence Framework presented in Fig. (1) and elaborated upon in the discussion. The synthesis hence underpins the key thesis of the paper: open-pit slope stability forecasting is evolving into a non-isolated deterministic evaluation to a system that is uncertain, monitored, climate-resilient geotechnical wisdom (Table 4).

Table 4. Evidence mapping of reviewed studies by methodological category.

2.7. Methodological Limitations

This review does not purport to be a comprehensive review. Despite a structured search and transparent selection process, the intent was critical synthesis and evidence mapping as opposed to full bibliometric coverage. The results of Google Scholar were taken as an additional source of discovery due to the inability to reproduce all the results of this source fully. The type of data, modelling purpose, method of validation, and conditions of the site also vary greatly in the reviewed studies, which makes it impossible to directly compare the performance values. These constraints were resolved through grouping of the evidence based on methodological groups and the interpretation of each study based on its modelling purpose and practical contribution.

3. GEOTECHNICAL INDICATORS AS INPUT LAYERS IN OPEN-PIT SLOPE STABILITY PREDICTION

The evidence reviewed in this section shows that slope stability indicators should not be interpreted as isolated variables. Rock mass quality, shear strength, groundwater pressure, rainfall infiltration, blasting vibration, deformation history and environmental degradation operate together as interacting control layers. Their predictive importance depends on the modelling paradigm being used. Numerical models use these indicators to explain stress redistribution and failure mechanisms, probabilistic models use them to represent uncertainty and parameter variability, remote sensing systems detect their surface expression through deformation, and machine learning models use them as predictive features [14-17]. The integrative value of the review therefore lies in showing how the same indicator can have different analytical roles across different modelling traditions.

3.1. Rock Mass Rating (RMR) and Geological Strength Index (GSI)

The most commonly used indices in the reviewed studies are rock mass classification indices, and especially Rock Mass Rating (RMR) and Geological Strength Index (GSI). These indices are an integrated measure of the quality of rock masses, since they include intact rock strength, joint spacing and condition, effects of groundwater, and structural orientation.

In addition, Stolecki et al., [17] directly incorporated GSI into an artificial neural network model with the generalized Hoek-Brown failure criterion, which is able to prove that GSI was one of the strongest predictors of slope stability. They found that the predictions of ANN were highly similar to the output of numerical simulations, and GSI is an effective condensed set of rock mass behavior. Similarly, Mahmoodzadeh et al., [18] in their case study of Dexing Copper Mine applied GSI alongside joint roughness coefficient (JRC) and joint wall compressive strength (JCS) to assess deterministic safety factors as well as probabilistic risk of failure.

The degradation of the GSI over time in unfavorable environmental conditions is also mentioned in several studies. According to [19], freeze-thaw cycling in open-pit mines with a high altitude resulted in a gradual decrease in GSI, which directly corresponded to decreasing slope safety factors. This effect, which was further quantified by [20], indicated that repeated freeze-thaw cycles led to cohesion losses that are more than 70% thus indirectly lowering rock mass quality indices based on strength parameters.

These observations show that RMR and GSI remain central indicators in slope stability assessment, particularly when they are dynamically interpreted alongside environmental degradation, groundwater variation and time-dependent weakening processes.

3.2. Shear Strength Parameters (Cohesion and Friction Angle)

The parameters of shear strength, namely, cohesion (c), and internal friction angle (φ), are always found to be the most important geotechnical parameters that prescribe slope stability as failure forecasting in the studies reviewed. Empirical modelling and machine-learning-based studies both confirm the high sensitivity of stability results to changes in these parameters.

Gao et al., [21] compared several machine-learning models to predict rock slope failures and discovered that cohesion and friction angle have always been the most effective predictors in the three models (random forest, support vector machine, and artificial neural networks). Similarly, Sahoo et al., [8] showed that factor-of-safety prediction through variation in c and φ exhibited the largest variance in their machine-learning-based stability classification framework, with random forest models having a prediction accuracy of about 95%. Field-scale and laboratory-based studies also emphasize the process of shear strength degradation under environmental load. As demonstrated by Zhang et al., [22], freeze-thaw cycling led to a loss in cohesion exceeding 70% a loss in friction angle up to 50% and exponential losses in slope stability.

3.3. Hydrogeological Indicators

The hydrogeological indicators, such as the pore-water pressure, the groundwater level, the rainfall infiltration, and the seepage forces, are determined as the prevailing factors that initiate slope instability in the open-pit mines. Several case studies indicate that the hydro-geological conditions may change rapidly, leading to increased deformation and decreased effective stress. In a study, Lu et al., [23] observed the interactive impacts between mining blasting and rainwater infiltration and discovered that blasting damage was grossly enhanced by the accumulation of pore-pressure due to rainfall, resulting in more than 30% displacement. Slope stability has also been demonstrated to be influenced by post-mining hydrogeological evolution.

3.4. Blasting-Induced Stress and Mining Operations

Blasting and excavation activities leading to dynamic loading are another negative geotechnical indicator in active open-pit mines. Redistribution of the stress caused by blasting may reopen the discontinuities, lower the mass integrity of the rock, and cause slope deformation to increase. Li et al., [24] quantitatively measured the impact of blast on the ground vibrations and concluded that the maximum particle velocities of over 10.9 mm/s led to a 13% decrease in the safety factor. Xu et al., [25] also showed that cumulative damage caused by repeated blasting together with rainfall infiltration had a significant effect on the increment in the rate of slope deformation.

3.5. Case Studies Demonstrating Indicator Interactions

A number of studies clearly indicate that slope instability occurs as a result of interaction between various geotechnical indicators in relation to each other and not a single factor. Ismail et al., [26] demonstrated that the quality of the rock mass and the degradation of shear strength, as well as the excavation stage, collectively determined the probability of failures at the Dexing Copper Mine. Heddallikar et al., [27] pointed to the compound impact of freeze-thaw degradation on slope stability coupled with rock mass classification in cold mines.

To demonstrate that all the three features, namely, InSAR, infrared thermography, and numerical modelling, provided the same contribution to the slope failure processes, [28] combined them to demonstrate the deformation patterns, thermal anomalies, and mechanical indicators. These case studies highlight the increasing change towards multi-indicator and integrated evaluation framework of slope stability prediction.

4. MODELLING PARADIGMS AS ANALYTICAL LAYERS IN SLOPE STABILITY PREDICTION

The modelling approaches reviewed in this section should be understood as complementary rather than competing paradigms. Numerical modelling explains why slopes fail, probabilistic modelling estimates how uncertain that failure risk is, remote sensing identifies where and when deformation is occurring, machine learning detects nonlinear patterns in multi-source data, and hybrid systems attempt to connect these functions into operational decision support. When considered together, these approaches show that slope stability prediction is shifting from a single-output design calculation toward a multi-layered intelligence architecture. Although they are still common, especially in the field of engineering design, the constraints of such methods with regard to uncertainty and complex geological conditions have spurred the use of more sophisticated modelling techniques [4, 29].

However, caution should be used when comparing the different modelling paradigms since the methods reviewed are used to address different analytical objectives for slope stability management systems. Numerical methods like FEM and LEM are mainly used for modelling of mechanical behavior and stress redistribution, while machine learning models are used for classification or prediction problems with the help of historical data. The primary applications of remote sensing methods are deformation monitoring and early warning, and the uncertainty quantification and risk assessment are the main applications of Bayesian and probabilistic methods. The methodological superiority cannot therefore be generalized, as dependency of modelling suitability is significant regarding the operational objectives, data availability, geological complexity and monitoring requirements.

4.1. Numerical Modelling Approaches

In open-pit mine slope stability prediction, the numerical modelling is of paramount importance.The most frequently used methods in the reviewed studies are Finite Element Method (FEM), Finite Difference Method (FDM) and, to a minor degree, Discrete Element Method (DEM) to model stress-strain behavior, deformation development, and failure processes.

A number of case studies show that FEM-based methods can be effective in modeling complicated slope behavior. Fengyan et al., [28] combined field observations, laboratory testing and numerical modelling to examine the effect of weathering on open-pit slope stability. They reported that progressive weathering reduced slope stability through degradation of mechanical properties.

Yadav et al. [29] should be discussed under machine learning or hybrid numerical–ANN modelling rather than as a freeze–thaw FEM study. Their work used generalized Hoek–Brown-based stability information and artificial neural networks to predict the stability factor of rock slopes. It is therefore relevant to this review because it demonstrates how conventional geomechanical criteria can be translated into data-driven prediction tools, but it should not be used as evidence for freeze–thaw degradation. Fu et al., [30] also measured freeze-thaw-degradation by laboratory experiment and FEM and found that the loss in cohesion was more than 70%.

4.2. Probabilistic and Reliability-Based Modelling

In order to overcome uncertainty of geotechnical parameters, a number of studies embrace probabilistic and reliability-based modelling methodology. Ahour et al., [31] applied a hybrid deterministic-probabilistic model (Monte Carlo simulation with LEM) in order to assess slope stability in the Dexing Copper Mine. Their findings showed that the probability of failure of slopes that seemed to be stable during the deterministic analysis was up to 38%. Similarly, Mirmazloumi et al., [32] have factored in soil parameter uncertainty with Latin Hypercube Sampling, which has shown failure rates as high as 46.7%, an indication on the inevitable drawback of deterministic assessments.

Another significant direction in probabilistic modelling is based on Bayesian. Gao et al., [33] created a Bayesian Network model to evaluate the slope risk in cold areas and found freeze-thaw effects to be the key risk factor. All these studies show that probabilistic modelling presents a more realistic model of slope stability in case of uncertainty.

4.3. Remote Sensing-Based Modelling and Monitoring

The significance of remote sensing technologies in slope stability prediction has been on the rise especially in deformation monitoring and early warning. InSAR, ground-based SAR (GBSAR), LiDAR, and UAV-based photogrammetry are also extensively used in the reviewed studies to trace the trends of surface movement and deformation.

Dai et al., [34] also applied multi-temporal InSAR to slope stability evaluation of the Gevra Coal Mine and observed up to -30 mm/year of deformation of slope and mountainous open-pit mining regions deformation from −232 to +81 mm/year. These experiments show that InSAR is effective in the recording of long-term deformation patterns at large spatial scales.

On-ground monitoring techniques have better temporal resolution. Barkhordari et al., [35] implemented GBSAR in the opencast coal mine and obtained a spatial resolution of 0.6 m, which made it possible to detect the early-stage deformation. The study by Luo et al., [36] used radar to forecast slope failure days earlier along with an enhanced inverse velocity model and demonstrated the possible potential of radar-based forecasts.

4.4. Machine Learning and Artificial Intelligence Approaches

The most rapidly developing paradigm of modelling in slope stability prediction consists of machine learning (ML) and artificial intelligence (AI) solutions. The techniques exploit the huge datasets to establish complex and non-linear correlations between the geotechnical indicators and slope performance.

Zhang et al., [37] compared the use of ML algorithms in a detailed manner and discovered that the accuracy of the prediction by random forest models was approximately 94%. Likewise, Yadav et al., [38] used random forest models in an open-pit mine in Malaysia with a prediction of the stability accuracy in the order of 92%. An et al., [39] used Light Gradient Boosting Machine models in seismic situations and achieved an AUC of about 0.95.

Deformation forecasting is becoming more a time-series based approach. Li et al., [40] created an entire distribution optimization fuzzy time series model with lower prediction errors compared to the baseline models. Lyu et al., [41] used InSAR time series to predict the LSTM networks with RMSE that reached less than 5 mm. Du et al., [42] also improved the time-series prediction with the recurrent generative adversarial networks that minimized the mean absolute error by about 33%.

Although they perform highly, ML models have issues that are associated with data quality, interpretability, and generalization. It has been reported in numerous studies that, when site-specific data is used to train the models, the reliability of the results can be low when applied in other geological settings. Many machine learning studies have reported strong predictive performance, though this should be taken with a grain of salt. Many models have been trained with synthetic, simulated or very site-specific data sets, which may not be transferable to other geologic settings or mines. Moreover, machine learning models are often lacking in interpretability, have data imbalance issues, limited failure data, and potential overfitting when failure data is scarce. Therefore, there is no guarantee of operational robustness and generalisability under real mining conditions with high prediction accuracy.

4.5. Hybrid and Integrated Modelling Frameworks

The recent trend in the literature reviewed is the creation of hybrid modelling frameworks which incorporate numerical models, probabilistic techniques, remote sensing and machine learning. The strategies are supposed to capitalize on the strengths of both paradigms and reduce their unique drawbacks.

Jin et al., [43] also integrated in-situ InSAR, infrared thermography, and FEM in order to determine slope stability in an iron ore mine, and critical instability areas occur where the factor of safety decreased below unity due to seismic loading. Likewise, Wu et al., [44] incorporated deterministic, probabilistic, and geotechnical classification techniques to enhance the reliability of decision making.

There is also monitoring-integrated systems that are also an important direction. Dong et al., [45] established a Fog-IoT-based slope deformation monitoring system using LoRa communication to provide real-time results, which allows acquiring data with low latency. Although these systems do not actually compute safety factors, they provide streams of data that can be used in predictive modeling.

5. EMERGING TRENDS AND TECHNOLOGIES IN SLOPE STABILITY PREDICTION

5.1. Hybrid Physics-Data-Driven Modelling Frameworks

A key emerging trend is the creation of hybrid modelling frameworks that merge conventional geomechanical principles with machine learning or probabilistic intelligence. Such methods in contrast to purely data-driven models incorporate the physical knowledge in the predictive structures enhancing robustness and interpretability.

Li et al., [46] were the first to combine both deterministic and probabilistic models to study the probabilities of failures up to 38% in slopes that were considered stable using deterministic models only. The study took the concept a step further to combine InSAR deformation measurements, infrared thermography and finite element modeling, enabling thermal and surface deformations to directly drive the numerical stability models. Their outcomes showed that the factors of safety decreased to values below unity in areas that were determined using multi-sensor data fusion.

Similarly, Dong et al., [45] came up with an InSAR-based workflow that directly inputs the deformation-derived digital elevation models to slope risk prediction to close the remote sensing and mechanical analysis gap. These works suggest a change to physics-informed intelligence, with the meaning of geotechnical restricting data-based learning and not substituting it.

5.2. Deep Learning and Time-Series Forecasting

The second significant upcoming trend is the use of deep learning architectures, especially on deformation prediction and early warning. Deep learning models can, in contrast to traditional machine learning models, model temporal dependencies and progressive failure behavior, since they use dynamic input features.

Du et al., [42] found that long short-term memory (LSTM) networks with InSAR time-series data could be effectively used with root mean square errors <5mm in deformations forecasting. The study used the fuzzy time series models that are optimized by the optimization of the entire distribution and found that the prediction error is significantly less than that of the traditional statistical forecasting.

Recently more advanced architecture has been developed. To enhance the predictability of slope instability, [43] presented a recurrent generative adversarial network (RGAN) which improves the predictive error by about 33% when augmenting geotechnical time-series data. These methodologies deal with the key weakness in slope surveillance and the lack of asymmetry in failure information.

The growing popularity of deep learning represents the movement of binary stability classification to continuous deformation prediction, which allows the accelerating failure trends earlier.

5.3. Advanced Remote Sensing Integration

Remote sensing technologies have not only been developed but they have become part of the predictive modelling systems rather than just a tool used in isolation to monitor. It is now a trend in slope stability forecasting that data fusion is done across multi-platforms.

This evolution still revolves around Multi-temporal InSAR. To measure deformation between −232 mm/year and +81 mm/year in open-pit mines in the mountains, Wu et al., [44] assembled ascending and descending orbit InSAR data. SAR technologies on the ground supplement satellite-based systems in that they offer high temporal resolution. The LiDAR-based surface modelling is also becoming a key enabling technology. According to [45], an enhanced algorithm of segmenting open-pit terrain using LiDAR is applied to stepped terrain and gives an F1-score of about 75%. Although LiDAR does not specifically estimate stability, it drastically increases geometrical accuracy on the numerical and data-driven models.

Taken together, these works point to the shift in the mode of remote sensing monitoring to predictive, model-based sensing systems.

5.4. Uncertainty-Aware and Probabilistic Intelligence

The other trend that is emerging and is critical is the explicit description of uncertainty using probabilistic intelligence. Conventional slope stability tests typically minimize risk by basing them on individual deterministic safety factors. Recent research incorporates uncertainty more and more into predictive models.

Lyu et al., [41] combined soil parameters with uncertainty in the form of Latin Hypercube Sampling and probabilistic limit equilibrium analysis, and found the failure probability to reach 46.7%. Bayesian intelligence has likewise become popular. Li et al., [46] came up with a Bayesian Network model to determine slope risk in cold regions, where the freeze thaw processes are the most common instability driver. These methods are a step in the direction of risk-based slope management, in which probability of failure and confidence intervals are decision-making scales, as opposed to determinism thresholds.

5.5. Digital Monitoring Systems and IoT-Based Technologies

One of the most revolutionary emerging trends is the integration of the real-time sensing, communication, and computation. Instead of using periodical surveys, a number of studies recommend ongoing digital surveillance ecosystems.

Xu et al., [25] proposed a Fog-IoT-based slope surveillance system on the basis of LoRa communication and allowed the transmission of deformation data in open-cast mines in real time at low latency. Though their system was not the one that specifically calculated safety factors, it nonetheless created a structure of real-time data pipelines that could supply predictive models. By conducting continuous inclinometric monitoring in the Polish copper mines, the study proved the effectiveness of such monitoring in predicting millimeter-scale trends in deformation related to underground excavation. These systems allow real-time-close response and are the basis of the development of digital twins of mine slopes, a new idea that is implicitly mentioned in various studies.

5.6. Climate-Resilient and Environment-Aware Modelling

The slope instability related to climate change has become one of the most important areas of research, especially in cold and high-precipitation areas. The degradation by freezes, thaws and rain infiltration is also becoming a part of predictive frameworks. [42] and [44] measured extreme changes in rock strength that occurred because of freeze-thaw cycling, while [9] used machine learning to forecast geotechnical parameters with high precipitation conditions, with R² over 0.90. [43] also showed that the formation of pit lakes following the closure of the mines has a major impact on pore-water pressure regime, which impacts long-term slope stability. According to these studies, there is an increased focus on predicting climate-resilient slope stability, especially under changing environmental conditions.

6. DISCUSSION

The discussion interprets the reviewed evidence through the Adaptive Slope Stability Intelligence Framework introduced in Fig. (1). The key finding was that none of the modelling paradigms could be used to reliably model slopes in a complex open-pit environment. Numerical models are used to give mechanical explanation; probabilistic models allow uncertainty quantification; remote sensing systems allow deformation observability; machine learning models allow nonlinear prediction; and hybrid systems offer a means towards operational decision support. The classification of the methods is thus just a part of the contribution of the review; the other part is that the methods are integrated within a multi-layered understanding of the intelligence of slope stability. A distinct transition can be seen when the reviewed studies are looked upon together. The prediction of open pit slope stability is shifting from static design evaluation to adaptive management of slope stability risks. However, this requires more and more interaction between geotechnical indicators, uncertainty-aware modelling, real-time or periodic monitoring, climate-sensitive degradation analysis and decision-support logic in order to manage the slope. This integration of two paradigms gives the conceptual grounding for the framework proposed in this review.

6.1. Comparative Analysis of Modelling Approaches and Types of Prediction

Fig. (2) demonstrates that machine learning and numerical modelling are the most prevalent methods of recent research, and most of the reviewed studies belong to them. Random forest, support vector machines, and gradient boosting models are machine learning models that show high predictive accuracy, which is usually more than 90% [8, 18, 22]. These methods are particularly successful in the establishment of nonlinear relationships between geotechnical indicators and slope failure products.