Remote Sensing Techniques for Landslide Prediction, Monitoring, and Early Warning

With the rapid increase in construction projects, the demand for effective landslide prevention has become increasingly urgent. Landslides pose threats to infrastructure and can lead to casualties and economic losses. Therefore, timely warnings and effective prevention measures are crucial. In recent decades, remote sensing technology, with its wide coverage, all-weather capability, and ability to dynamically record spatiotemporal changes, has shown great potential in investigating, assessing, predicting, monitoring, and providing warnings about landslide occurrences. These technologies can provide real-time information on surface changes, helping researchers and decision-makers to better understand the risks of landslide occurrences, thus enabling the development of scientific prevention strategies and enhancing the efficiency and safety of disaster management. Current research on landslide prediction, monitoring, and early warnings demonstrates the dual characteristics of accelerated technological iteration and deepening theoretical frameworks.

In landslide prediction, studies focus on intelligent modeling and multi-source data fusion [1,2]. Machine learning-based hybrid models have significantly improved prediction accuracy [3,4,5,6]. For example, in reference [7], a hybrid model combining the Particle Swarm Optimized Support Vector Machine (PSO-SVM) and Convolutional Gated Recurrent Attention Mechanism (CNN-GRU-Attention) achieved high-precision prediction by decomposing trend displacement and periodic displacement. The fuzzy deep learning (FuDL) model was first introduced to improve the generalization capability of earthquake-triggered landslide prediction [8]. Research is increasingly shifting toward dynamic multi-scale analysis, such as models integrating Long Short-Term Memory (LSTM) and Temporal Convolutional Networks (TCNs) [9], which capture both long-term evolution and localized abrupt changes in displacement. Additionally, transfer learning methods interfere with traditional dependencies on continuous monitoring data, filling data gaps through knowledge transfer to enable real-time prediction under combined rainfall and reservoir water level effects [10]. Climate change-driven mechanisms have been further explored, such as the work in reference [11], which analyzed the sensitivity differences among landslide types to reveal dynamic interactions between environmental factors and landslide dynamics while establishing climate-adaptive prediction frameworks.

In landslide monitoring, technologies exhibit trends toward integrated spatial collaboration and cost-effective universalization. Hu and colleagues [12] integrated multi-parameter sensors with high-speed data transmission networks for second-level data acquisition and processing, while the authors in reference [13] systematically evaluated the feasibility of large-scale landslide deformation monitoring using Synthetic Aperture Radar (SAR)-based techniques. Breakthroughs in deep monitoring include reference [14], where acoustic emission (AE) arrays were both experimentally and field-validated to precisely capture shear displacement acoustic signatures at sliding surfaces; alongside this, the work in reference [15] developed an intelligent infrasound signal recognition system to distinguish landslide failure signals from environmental noise via time–frequency domain analysis and classification algorithms. To reduce costs, lightweight solutions were explored, such as validating the economic feasibility of single-frequency differential GNSS devices for slow-moving landslides [16] and optimizing combinations of displacement meters and rain gauges to extend low-cost, wide-coverage monitoring [17].

In landslide early warning, systems evolve toward dynamic threshold optimization and intelligent decision chains [18,19]. The authors in [20] integrated statistical and physical methods via multi-level decision algorithms to enhance regional warning accuracy by combining rainfall thresholds and geological attributes. Zhan et al. [21] revealed abrupt changes in pore water pressure and volumetric moisture content through model experiments, providing theoretical support for warning indicator selection. Increasingly, intelligent technologies are deeply embedded in early warning research. The authors in reference [22] combined deep learning with time-series decomposition to dynamically analyze the evolution of displacement in loose slopes, while Du et al. [23] proposed a dual-speed ratio (DSRM) criterion to reduce false alarms in step-like deformation landslides by fusing trend and displacement rate features. To refine the model’s precision, multi-stage hierarchical warnings are becoming mainstream. In reference [24], a four-stage model spanning long-term trend analysis for sudden disaster determination was constructed, enabling full-chain hazard tracking. Meanwhile, the authors in reference [25] established a spatiotemporal early warning framework based on micro-seismic event frequency, providing preemptive judgment through rockfall activity pattern recognition.

The existing body of landslide research demonstrates the insufficient utilization of remote sensing technologies. This Special Issue primarily presents the latest advancements and trends in landslide prediction, monitoring, and early warning using modern remote sensing technologies (ground-based, airborne, and spaceborne sensors). With the continuous advancement of technology, the application of remote sensing in landslide research has become increasingly widespread, providing more precise and real-time data support. Through these research developments, we can gain a better understanding of the mechanisms behind landslide occurrences, thus providing a scientific basis for disaster management and emergency response. A more detailed description of the included works is given as follows:

In reference [26], the authors propose a novel heterogeneous ensemble framework based on Bayesian optimization (BO), known as stratified weighted averaging (SWA), to enhance the classification performance of individual machine learning (ML) models in landslide susceptibility mapping (LSM). After conducting field investigations and collecting historical data in the Yanbian region of China, a spatial database was established that included 16 triggering factors. The results showed that SWA outperformed single ML models in terms of accuracy, AUC, and model robustness.

Bahti and colleagues [27] conducted a parametric test on the Sentinel 1A Persistent Scatterer (PS)- and Small Baseline Subset (SBAS)-InSAR using the Stanford Method for Persistent Scatterers (StaMPSs) to optimize landslide monitoring methods. The study compared the performance of StaMPS with Global Navigation Satellite Systems (GNSSs) in landslide cases, verifying the feasibility of SBAS in practical landslide monitoring.

The authors in reference [28] focused on the 1991 rock avalanche in Touzhai, Zhaotong, Yunnan, China. The study combined field investigations, indoor and outdoor experiments, and numerical simulations to invert the occurrence and spreading range of rock avalanche–debris flow hazards. The study revealed the dynamic characteristics and the crushing process of the rock avalanche accumulation body.

In reference [29], Wang and colleagues provide a review article on the research advances and prospects of underwater terrain-aided navigation (TAN). TAN can achieve high-precision positioning independently and autonomously under communication-rejected space conditions, which is crucial for the autonomous and refined operation of deep-sea autonomous underwater vehicles near the seabed. This article reviewed the relevant algorithms involved in the two main modules of underwater TAN and summarized other cutting-edge issues in the field.

The work in reference [30] proposes a slow-moving landslide displacement prediction method based on the Informer deep learning model combined with InSAR technology. The study used the Dawanzi landslide in the Baihetan reservoir area of China as a case study, training the model with 50-time series deformation data points to predict the displacement results of 12-time series deformation data points.

These research articles cover various fields, including landslide susceptibility mapping, monitoring and prediction, debris flow movement and accumulation characteristics, and underwater topography-assisted navigation. They demonstrate the critical role of advanced remote sensing technologies in geological disaster research. These technologies not only improve the efficiency of landslide detection, prediction, monitoring, and hazard mapping but also provide essential support for disaster prevention and emergency response, underscoring the broad application potential of remote sensing in geological studies. The current development of landslide research demonstrates a synergistic advancement in technological convergence and innovation alongside the deepening of theoretical frameworks [31,32,33]. However, challenges persist in overcoming setbacks, such as data heterogeneity and model generalization limitations. Through interdisciplinary collaboration and technological innovation, landslide prevention and control systems are being comprehensively transformed toward intelligent pre-disaster mitigation.