Rainfall-induced landslides are destructive natural hazards that require timely detection and early warning to protect lives and infrastructure. This study presents the development and deployment of an IoT-based, cost-effective, real-time monitoring and early warning system that integrates surface and subsurface sensors to detect slope instability and issue timely warnings for disaster prevention. The monitoring system integrates tilt sensors, volumetric water content sensors, a MEMS-based inclinometer, a rain gauge, and a video camera, all linked to a web-based platform. Field results demonstrated that the tilt sensors effectively detected surface displacement, the volumetric water content sensors responded rapidly to rainfall infiltration, and the MEMS-based inclinometer captured subsurface displacement during rainfall events. Detailed analysis was conducted using multisource monitoring datasets collected during three specific rainfall events. An early warning method for landslides was proposed by combining the tilt rate, horizontal displacement rate derived from the MEMS-based inclinometer, and saturation index. Accordingly, critical threshold values for different warning levels were established based on tilt rate ( Tr), displacement rate ( Dr), and saturation index ( Si). This study provides a robust strategy and guidelines for early warning systems, enabling generation of warning alarms and demonstrating immense potential to reduce the impacts of rainfall-induced shallow landslides and enhance risk management. 1. Introduction Rainfall-induced landslides represent one of the most catastrophic natural hazards, resulting in substantial fatalities and infrastructural devastation, and have a severe impact on local socioeconomic development. Ongoing changes in global weather patterns have led to more frequent and erratic extreme rainfall, causing a rise in rainfall-induced landslides, which account for 90% of all landslide disasters [ 1]. Approximately 17.1% of the Earth is at risk of landslides, with around 8.2% of the world’s population living in landslide-prone areas, thus exposing people, property, and the environment to potential landslide hazards [ 2, 3]. According to statistical reports from the Chinese Institution of Geological Environmental Monitoring, unexpected geological hazards cause 1167 casualties and CNY 6.4 billion in property losses per year [ 4]. Other research reported that more than 230,000 landslides have been reported in China from 2004 to 2019 [ 5]. Hence, there is a clear need to monitor landslides and provide early warnings to enable the evacuation of vulnerable people and the protection of infrastructure. Various methods are currently employed to monitor unstable slopes. However, there is no single technique or instrument that can comprehensively monitor an unstable slope through ground-based monitoring. Extensometers are the most widely used instruments for monitoring the slow displacement slopes, installed across well-defined cracks and other displacement discontinuities. GPS (Global Positioning System) and radar-based remote sensing are also being used for long-term displacement monitoring over wide areas on slope surfaces [ 18, 19]. LIDAR (Light detection and ranging) and InSAR (Interferometric Synthetic Aperture Radar) technologies have been widely applied for monitoring surface deformation, offering advantages such as continuous real-time monitoring and high precision in acquiring detailed ground elevation data [ 20, 21]. In addition, GNSS (Global Navigation Satellite System)-based monitoring systems are capable of achieving millimeter-level positioning accuracy for landslide monitoring [ 22]. However, these technologies are expensive, require skilled personnel, and may lack the accuracy for detecting minimal slope displacement. To address this gap, this study proposes an IoT-based real-time landslide monitoring and early warning system for shallow landslides, which is considered one of the most effective approaches among various mitigation methods. The system incorporates surface tilt sensors, volumetric water content sensors (VWCs), a MEMS-based inclinometer, a rain gauge, and a video camera. By integrating surface and subsurface sensing strategies, the system provides a comprehensive assessment of slope stability. LoRa (Long Range) wireless and low-power communication technology is used for data transmission from sensor units, and the collected data are subsequently transferred to a cloud database via a 5G wireless network. The system also provides a web-based platform for data visualization and alarm dissemination. In addition, data processing and analysis techniques are applied to enhance monitoring accuracy and the effectiveness of early warning performance. The system was deployed on a landslide-prone slope for real-time monitoring. In July 2021, a slope failure induced by heavy rainfall was successfully detected and recorded by the system. This study presents rainfall data, volumetric water content data, and slope surface and subsurface displacement data recorded before and during the failure, highlights the system’s monitoring capabilities, and provides corresponding analysis. Furthermore, quantitative early warning thresholds incorporating tilt rate ( Tr), displacement rate ( Dr), and saturation index ( Si) are established to define different warning levels. This study also provides practical guidance for selecting effective monitoring techniques in the field of slope displacement and landslide prediction. A well-designed landslide monitoring system can deliver accurate and timely warnings, thereby reducing loss of life and property damage associated with landslides. 3. Development of Real-Time Monitoring and Early Warning System 3.1. Selection of the Sensors The selection of sensors for the landslide monitoring system is determined by the main triggering factors for the landslides in the studied area. The effectiveness of real-time landslide monitoring and early warning systems depends on the careful selection of sensors that can accurately measure the key parameters affecting slope stability. The reliability of the system is directly linked to the quality and suitability of the sensors deployed in the field. The reason is that the sensors collect massive amounts of monitoring data that are essential for assessing the stability of the slope and issuing early warnings [ 23]. Given that rainfall is the predominant triggering factor for landslides, it is essential to analyze the relationship between rainfall conditions and landslide occurrence using collected measurement data. Under heavy rainfall conditions, water infiltration into the slope leads to an increase in pore water pressure, reduces the effective normal stress and the shear strength of the soil, and ultimately causes slope failure as shown in Figure 3 [ 24]. For rainfall-induced shallow landslides, slope failures are primarily influenced by high-intensity rainfall or prolonged medium-intensity rainfall, as well as the soil properties of the slope. However, shallow failures are generally caused by the progression of the wetting front due to rainfall infiltration, rather than by a rise in groundwater level [ 25]. In recent years, the majority of landslide monitoring has incorporated rainfall observations alongside slope displacement, as well as the hydrological characteristics of soils [ 26, 27]. In the context of geotechnical and hydrological monitoring, slopes are typically equipped with instruments such as tiltmeters, inclinometers, and volumetric water content sensors, which are installed on or within the slope to assess deformation associated with rainfall-triggered landslides [ 28, 29]. Therefore, selection of the sensors relies on the triggering parameters associated with the landslide being considered. In this study, rainfall was considered the main triggering factor of landslides, and the key physical parameters monitored included variations in surface and subsurface movement, changes in volumetric water content, and rainfall intensity and duration. To achieve comprehensive monitoring, tilt sensors, volumetric water content sensors, a MEMS-based inclinometer, a rain gauge, and a video camera were selected and deployed on the surface and within the subsurface to study the deformation behavior of rainfall-induced landslides. Among the selected sensors, one of the most important underlying technologies is the advancement of MEMS (Micro-Electro-Mechanical Systems). With the development of this technology, sensors can be made smaller and lighter, thereby making their installation and maintenance procedures relatively simple. To ensure continuous monitoring in remote and inaccessible areas, these sensors are usually powered by solar energy, which provides a sustainable and maintenance-friendly power source. Furthermore, due to their low cost and low power consumption, MEMS-based sensors are already being widely used in geotechnical instrumentation, especially for real-time landslide monitoring and early warning systems. In the following, a brief description of the selected sensors is provided. The selected sensors are shown in Figure 4, and their specifications are summarized in Table 1. 3.1.1. Tilt Sensor MEMS (Micro-Electro-Mechanical System)-based tilt sensors were installed on the slope surface to monitor subtle angular displacement and to detect small-scale or sudden movement of the soil layer associated with rainfall-induced instability [ 30]. Tilt sensors measure angular displacement in two orthogonal directions: one parallel and one perpendicular to the slope surface. In this system, tilt sensors were mounted on the head of steel rod inserted to a depth of 200 mm, with the rod vertically installed to a depth of 1 m within the unstable slope layer. Tilt sensors provide localized point measurements that represent small areas within the landslide body. The installation and maintenance of tilt sensors are relatively simple and cost-effective. It should be emphasized that this monitoring approach is particularly effective for detecting shallow slope failures. 3.1.2. Volumetric Water Content (VWC) Sensor Volumetric water content sensors were installed to assess soil moisture by measuring the dielectric constant of the surrounding soil, which is highly sensitive to variations in water content [ 31]. Our objective was to obtain data of volumetric water content present in the soil, a key factor governing soil infiltration behavior. In rainfall-induced landslide early warning systems, volumetric water content sensors play a critical role in evaluating slope stability by continuously monitoring rainwater infiltration and the corresponding changes in volumetric water content conditions. Each volumetric water content sensor was placed carefully at a shallow depth of 30 cm. It should be noted that volumetric water content measurements represent conditions at a single point, and because the potential location of a future slip surface is difficult to predict, the practical utility of volumetric water content sensors in forecasting slope failure is limited. 3.1.3. MEMS-Based Inclinometer In this study, we developed a wireless biaxial MEMS-based borehole inclinometer, which was installed within the slope [ 32]. The purpose of incorporating the MEMS inclinometer was to quantify cumulative horizontal displacement at different subsurface depths and to determine the potential sliding surface of the landslides. The inclinometer consisted of biaxial MEMS tilt sensors installed inside a flexible PVC (Polyvinyl chloride) casing pipe with a diameter of 60 mm. The PVC casing was designed to allow deformation within the surrounding soil rather than directly measure displacement. Its primary function was to transfer subsurface deformation to the MEMS tilt sensor installed inside the PVC casing. Each MEMS-based tilt sensor measures local angular displacement in two orthogonal directions at its corresponding installation depth. The lateral displacement profile is subsequently derived from the measured tilt angles and the known vertical spacing between adjacent sensors. In this system, MEMS tilt sensors were installed at 1 m intervals along the inclinometer casing, starting from 0.5 m below the ground surface and extending to a depth of 24.5 m. The inclinometer was installed by pre-drilling a borehole and inserting the PVC casing pipe through the sliding mass, with the lower part extending into the stable bedrock. After installation, the borehole was backfilled with sand to ensure proper contact between the casing and the surrounding soil. In this system, data measured from the inclinometer were transmitted wirelessly through a single cable after being installed in a borehole. 3.1.4. Rain Gauge A rain gauge was installed to monitor rainfall conditions continuously. A wireless tipping-bucket rain gauge was selected due to its high accuracy and capability for real-time data acquisition. This instrument measures rainfall intensity and cumulative rainfall over defined time intervals. The rain gauge was installed above ground at a predetermined height to ensure accurate measurement. 3.1.5. Video Camera A high-resolution camera was installed along the landslide site to visually assess the conditions of the slope. The images and videos captured by the camera provide valuable visual data for experts to analyze, facilitating the early identification of visible indicators of slope instability such as soil movement, displacement, and changes in the surrounding environment. Furthermore, in areas with limited accessibility, cameras enable continuous observation of slope behavior without requiring frequent field visits. 3.1.6. Layout of Monitoring Sensors The layout of the landslide monitoring sensors of early warning system is illustrated in Figure 5. Four tilt sensors (T1–T4) were installed at different locations on the slope in such a way that the X- and Y-directional movements were parallel and perpendicular to the slope, respectively. In addition, four volumetric water content sensors (V1–V4) were installed alongside the tilt sensors. Each of these sensor units consisted of a data logger and a data transmission unit. Furthermore, one MEMS-based inclinometer (IN1) was vertically installed to a depth of 25 m in the slope. A rain gauge was installed in a stable location within the monitoring area and mounted at a height of approximately 3 m above the ground. A central data logger and data transmission unit were also installed alongside the rain gauge on the slope, and the video camera was installed near the toe of the slope. 3.2. System Design We propose an integrated system architecture for a landslide early warning system, as illustrated in Figure 6. Establishing an effective real-time monitoring and early warning system for landslide instability requires the careful selection of suitable monitoring methods and the integration of reliable hardware and software components. Our proposed system includes automated monitoring instruments, efficient and rapid data acquisition and transmission techniques, data processing and analysis capabilities, and timely dissemination of monitoring information. A brief description of the early warning system is provided below. 3.2.1. Monitoring Method Choice To develop a robust and effective early warning system for the Xinyu landslide, it is necessary to identify potential landslides and conduct risk zoning. Therefore, integrating both surface and subsurface monitoring methods is vital for comprehensive slope observation. This requires the selection of suitable monitoring parameters and instruments that fulfill essential requirements, including simplicity, robustness, dependability, and cost-effectiveness. In this context, sensor selection is critical, as monitoring strategies must align with the site-specific conditions and targeted parameters. Geotechnical methods are widely used for various landslides, primarily to detect and provide early warning of ongoing deformation [ 33]. Furthermore, a well-structured monitoring layout enhances the ability to detect deformation patterns associated with imminent slope failure. 3.2.2. Data Acquisition and Transmission We have designed and developed a robust and energy-efficient data acquisition and transmission framework for a real-time landslide monitoring system. Figure 7 presents a schematic representation of the integrated monitoring and early warning system employed in this research. All the sensor units were powered by solar energy and configured to record data at 10-min intervals. In recent years, LPWANs (Low-Power Wide Area Networks) have enabled energy-efficient, long-range, and cost-effective wireless communication for remote monitoring [ 34]. Among them, LoRa (Long Range) has emerged as a leading technology due to its low power consumption, wide coverage, and robust demodulation performance. In this study, each sensor was equipped with a LoRa wireless communication module to facilitate real-time data transmission. It was utilized to transmit the initial sensor data to a centralized data logger installed within the monitoring site. This data logger functioned as a local aggregation node, temporarily storing and organizing data collected from multiple sensors. To ensure stable operation in remote and challenging field environments, the data logger was powered by a combination of solar panels and rechargeable lithium batteries. Subsequently, the data collected by the data logger was transmitted to a cloud-based server using a 5G-network-based DTU (Data Transmission Unit). In this system, the F2A16 LTE IP modem, compliant with the 5G/LTE standard [ 35], was utilized to enable high-speed, reliable data transmission from the monitoring site while balancing data rate and energy efficiency. Upon reaching the cloud-based server, the transmitted data were automatically stored in a structured MySQL database, which served as persistent storage for the sensor measurements. This database was integrated with a web-based platform to support continuous data access, processing, and visualization. 3.2.3. Web Application A user-friendly, web-based monitoring application was also developed to display real-time monitoring data, as shown in Figure 8. The application serves as a comprehensive, real-time monitoring and data management system, providing users with timely and accessible information from monitoring points anywhere and at any time, thereby supporting proactive decision-making and effective risk mitigation strategies. It also enables interaction with cloud-based services through a UI (User Interface) designed for usability and consistency, ensuring a reliable and efficient experience. To optimize development and reduce code complexity, a JavaScript framework was utilized. The platform provides comprehensive data display options at hourly, daily, weekly, monthly, and yearly intervals via the web platform’s search function. Furthermore, by browsing the web interface, experts and registered users can view each monitoring point and its corresponding monitoring area. The application also allows for the addition of new monitoring points and the deletion of existing ones. It is password-protected to ensure data security and supports simultaneous access by multiple users from any internet-connected device. Additionally, historical data can be downloaded in CSV format for post-processing [ 36]. 3.2.4. Data Processing and Analysis In landslide early warning systems, efficient data processing and analysis are crucial for ensuring timely and reliable risk assessment, as illustrated in Figure 6, step 3. Our developed system is fully automated, enabling the automatic transmission of raw sensor data to a central database for data processing and analysis. However, the acquired raw sensor data cannot be directly used to calculate early warning parameters. Therefore, data processing is responsible for calibrating the collected data to maintain its quality, querying the monitoring data, and revising essential information. Data analysis service provides algorithm services for the entire system, including data denoising using Kalman filtering, regression analysis, early warning, and forecasting. Then the processed monitoring data can be displayed in graphical formats on the website for expert’s review, as discussed in the previous Section 3.2.3. The software developed for this study is capable of achieving all the above requirements. Then the critical parameters such as tilt rate ( Tr), displacement rate ( Dr), and saturation index ( Si) are selected for early warning analysis. Threshold values for these parameters can be derived through heuristic analysis of historical monitoring data, particularly during critical periods of landslide activity. These threshold parameters are subsequently integrated into the system’s algorithm services, which are designed to support automated analysis and generate timely warnings. Based on the defined threshold levels, the proposed risk classification can be divided into distinct warning levels for slope stability assessment. According to the criteria of early warning, when any threshold value is exceeded, the system issues automatic notifications to the designated experts and registered users in real time. However, the final warning is ultimately issued only after expert assessment to avoid false alarms. The system was designed to be flexible, allowing threshold values to be modified as new and more reliable data become available, in accordance with the site-specific situation. However, any modifications to the threshold values must be determined by the designated expert. 3.2.5. Data Release and Early Waning Alarm In the proposed system, tilt rate ( Tr), displacement rate ( Dr), and saturation index ( Si) serve as the main warning parameters. Early warnings will be issued only when at least two of the three warning parameters reach their respective warning levels. Warning level will be further upgraded when at least two warning parameters reach a higher warning level. When the three warning parameters indicate different warning levels, the final warning level will be issued based on the highest warning level that is exceeded by at least two of the three warning parameters. Therefore, if one warning parameter reaches a higher warning level while the other two warning parameters remain at a lower warning level, the lower warning level supported by two warning parameters will be issued. When one warning parameter is unavailable, the warning decision will be determined based on the two remaining available warning parameters. If both warning parameters indicate the same warning level, that warning level will be issued after expert verification. If the two available warning parameters indicate different warning levels, the lower warning level will be adopted, and continuous observation will be maintained until additional data or expert judgment supports further upgrading. When only one parameter is available, the system will not issue any early warning based solely on that single warning parameter. Instead, expert review, field inspection, and supplementary evidence, including rainfall records, video monitoring, and historical deformation trends, are required to support the warning decision. If all warning parameters are unavailable, the situation will be treated as a potential device or communication failure. In this case, equipment inspection together with manual monitoring procedures will be initiated. Once a predefined monitoring threshold is exceeded, the system immediately issues an automatic alert notification to experts and registered users. Before issuing a warning, experts are required to verify the reliability and accuracy of the monitoring data to assess the likelihood of a landslide occurring. Any inaccurate or false warnings can provoke unnecessary public alarm and erode confidence in the early warning system. Furthermore, experts must ensure that all communication infrastructure is fully functional and capable of delivering clear warnings to all stakeholders. In our system, warning messages are usually sent via SMS, WeChat, sirens, and telephone calls. In addition, when the orange or red warning level is exceeded an emergency meeting will be held to determine whether the warning level is back to blue or to continue to issue an orange or red alarm to the public in a formal way. After a warning is issued, local government must implement emergency measures, and the public’s response to landslide hazards is also essential for mitigating potential losses. Figure 9 illustrates the flowchart for issuing early warning alarms [ 37]. 4. Monitoring Result 4.1. Analysis of the Monitoring Result This section contains the results of monitoring data acquired from 1 February to 31 July 2021. This period covers the majority of the rainfall events that occur during June and July, which are typically attributed to the monsoon or rainy season. The data presented here demonstrates the effectiveness of our integrated monitoring system. However, one volumetric water content sensor (V3) was found to be broken immediately after deployment. 4.1.1. Rainfall Data Figure 10 illustrates historical and real-time rainfall data over the six-month monitoring period, including daily rainfall, highest hourly rainfall, and cumulative rainfall. During this period, the total rainfall reached 872.6 mm. On 7 July 2021, the highest hourly and daily rainfall were recorded as 60 mm and 203.4 mm, respectively. According to the rainfall classification specified by the China meteorological department, a daily rainfall of 203.4 mm is classified as heavy rainstorm, which triggered slope failure on 7 July [ 38]. Three significant continuous rainfall events were recorded: a rainfall event 1 from 11 May to 16 May, rainfall event 2 from 18 June to 19 June, and rainfall event 3 from 2 July to 7 July. These three rainfall events produced rainfall amounts of 88.8 mm, 64.8 mm, and 297.2 mm, respectively. Their combined rainfall was 450.8 mm, representing approximately 51.66% of the total rainfall of 872.6 mm recorded during the six-month monitoring period. 4.1.2. Volumetric Water Content (VWC) Data Figure 11 shows the variation in volumetric water content at a soil depth of 30 cm measured along different parts of the slope. Changes in volumetric water content (VWC) are a more reliable indicator of potential slope failure than absolute values [ 39]. Data obtained from sensors V1, V2, and V4 indicate that the volumetric water content responded rapidly to rainfall events and showed a similar trend, exhibiting significant fluctuations ranging from 5.34% to 37.79%. However, after the intense rainfall, the volumetric water content declined rapidly. The most significant variations in volumetric water content occurred during the three continuous rainfall events. During these rainfall periods, average peak volumetric water content values generally exceeded 35%. At the upslope, volumetric water content ranged from 5.34% to 25.92%. Compared with the upslope, higher volumetric water content was observed at the midslope and downslope positions. At the midslope, volumetric water content ranged from 7.65% to 33.28%, while at the downslope, it ranged from 8.42% to 37.79%. The highest recorded volumetric water content of 37.79% was observed on 7 July 2021, during an hourly rainfall of 19.9 mm. 4.1.3. Subsurface Displacement Data To assess the deformation characteristics of the slope, inclinometer data from 1 February to 8 July 2021, along the main sliding direction of the landslide, were analyzed. Figure 12a illustrates the distribution of slope deformation along the depth, recorded by inclinometer IN1. The positive displacement curve indicates that the sliding trend aligns with the main sliding direction of the landslide. Progressive deformation occurred in the upper part of the inclinometer over a shallow depth of approximately 5.5 m. This data indicates that the potential slip surface is located around 5.5 m above the siltstone layer. Borehole excavation revealed that the sliding surface consists of silty clay. Significant displacement occurred during three rainfall events on 15 May, 18 June, and 6–7 July 2021. During these events, the average displacement rates were 1.08 mm/h, 0.81 mm/h, and 3.74 mm/h, respectively, and the maximum displacement reached 104.76 mm at the top of the inclinometer. Overall, the inclinometer observation suggests that rainfall affects the subsurface displacement, and the displacement was significant. Figure 12b illustrates the statistical relationship between daily rainfall and displacement at the slope crest at a depth of 0.5 m. The results show that the first two rainfall events had a limited impact on slope deformation compared with third rainfall event. In contrast, rainfall event 3 resulted in a sharp increase in total displacement of 78.60 mm, due to significantly higher rainfall. Specifically, displacement during rainfall event 3 was 6.59 times greater than that during rainfall event 1 and 7.48 times greater than that during rainfall event 2. 4.1.4. Tilt Sensor Data 4.2. Detailed Analysis of the Tilting Behavior of Sensor T1 4.2.2. Rainfall Event 2 On 18 July, from 2 am to 3 pm, the monitored slope received a total of 37.4 mm of rainfall, with an average hourly rainfall of 2.87 mm. Following the onset of rainfall, a progressive increase in tilt angle was observed in both the X- and Y-direction to the slope, along with cumulative displacement. As shown in Figure 15a, the tilt angle shifted from −0.132° to −0.024° in the X-direction and from −0.078° to −0.058° in the Y-direction. During this time, the tilt rate was 0.009°/h in the X-direction and 0.002°/h in the Y-direction, respectively. The displacement rate was 0.81 mm/h, with the highest displacement rate of 3.7 mm/h as shown in Figure 15b. In addition, during 18–19 June, the average tilt rate was 0.002°/h and 0.0005°/h in the X- and Y-directions, respectively, indicating a slow but steady displacement, and the displacement rate was 0.24 mm/h. Figure 15c shows that the volumetric water content ranged from 16.11% to 33.56% over these two days. On 19 June, a notable volumetric water content of 33.56% was recorded, corresponding to a rainfall intensity of 5.4 mm/h at 12 pm. 4.2.3. Rainfall Event 3 Figure 16a presents the time history of the combined X- and Y-direction tilt angles recorded by sensor T1, along with the horizontal displacement, while Figure 16b shows the corresponding tilt and displacement rates. Between 2 July and 6 July, cumulative rainfall reached 93.8 mm, with an average daily rainfall of 18.76 mm and a maximum hourly rainfall of 9.2 mm recorded on 6 July at 8 am. During this time, the volumetric water content fluctuated between 14.09% and 34.08%. Over the same period, the tilt angles recorded by sensor T1 increased at a slow rate, with an average tilt rate of 0.002°/h in the X-direction and 0.0006°/h in the Y-direction, while the displacement rate was 0.05 mm/h, indicating minimal slope deformation. 4.3. Relationship Between Surface Tilt and Horizontal Displacement Figure 18 presents the relationship between tilt rates recorded by tilt sensors T1 to T4 and displacement rates measured by the inclinometer IN1 at a depth of 0.5 m during three rainfall events. Data from three rainfall events were analyzed using linear regression, with both variables plotted on a logarithmic scale. As shown in Figure 18a, the high coefficient of determination (R 2 = 0.95) for tilt sensor T1 indicates a strong linear relationship between tilt rates and subsurface displacement rates, reflecting close agreement between the measured values and the fitted trend. Data from all three rainfall events demonstrated a consistent trend, with increasing displacement rates corresponding to increasing tilt rates, although minor scatters are observed at lower displacement rates. Moreover, the slope parameter (m) plays a key role in interpreting this relationship, with a value of approximately 1/100, indicating that tilt rates are proportional to displacement rates scaled by a factor of 1/100. This relationship allows the development of early-warning criteria based on monitored tilt rates and displacement rates. As shown in Figure 18a, tilt rates below 0.005°/h correspond to displacement rates of approximately 0.5 mm/h or less. Under all three rainfall conditions, tilt rates ranging from 0.005 to 0.5°/h are associated with displacement rates between 0.5 and 50 mm/h. When the tilt rate exceeds 0.5°/h, slope failure occurs, and the corresponding displacement rate also exceeds 50 mm/h. However, the other sensors showed no measurable response during the three rainfall events; therefore, regression analysis revealed no significant correlation, and their coefficient of determination was limited to R 2 = 0.64. In this study, inclinometer data were considered as an indicator of the overall horizontal displacement of the landslide [ 18]. Based on the relationship between tilt rates and displacement rates, four distinct types of slope movement were classified. Tilt rates lower than 0.005°/h are classified as very slow movement. Tilt rates between 0.005 and 0.05°/h, corresponding to displacement rates of 0.5 mm/h to 5 mm/h, reflect slow movement. Similarly, moderate movement is observed when tilt rates range from 0.05 to 0.2°/h, accompanied by displacement rates of 5 mm/h to 20 mm/h. Furthermore, tilt rates exceeding 0.2°/h are defined as rapid movement, with displacement rates greater than 20 mm/h. The corresponding early-warning thresholds for tilt rate ( Tr) and displacement rate ( Dr) are presented in Table 2, together with the saturation index ( Si) established in the following section. 4.4. Saturation Index-Based Threshold Determination for Volumetric Water Content Due to the considerable heterogeneity of the landslide deposit, volumetric water content could not be used to define a unified early-warning threshold. Therefore, in this study, the saturation index ( S i ) , introduced by [ 41], was adopted as a uniform parameter to represent the real-time volumetric water content by using the following equations: S i = θ − θ r θ p − θ r (1) where θ θ r are the real-time volumetric water content and residual water content, respectively. The value of θ r is derived from the soil–water characteristic curve. θ p is the peak volumetric water content recorded over the monitored historical period. As a result, when the measured volumetric water content θ exceeds the historical peak value θ p , the value of S i becomes greater than 1, and the newly recorded maximum volumetric water content should be adopted in practice as the new peak value θ p . Analysis of Figure 11 shows that volumetric water content sensors responded uniformly to all rainfall events, with volumetric water content consistently exceeding 20% in each instance. This consistent response enabled the normalization of raw volumetric water content measurements into a relative soil saturation index ranging from unsaturated to near-saturated states. Threshold levels for the saturation index ( S i ) were determined using a percentile-based approach. The distribution of saturation index ( S i ) values during rainfall periods was used to compute the 50th, 90th, 95th, and 99th percentiles, representing progressively extreme soil saturation conditions. Figure 19 illustrates the temporal variation in hourly rainfall and saturation index ( S i ), showing that landslides were associated with higher saturation index ( S i ) , indicating fully saturated soil conditions at the time of failure. 5. Discussion This study developed an IoT-based real-time monitoring and early warning system for rainfall-induced shallow landslides. The integrated sensor unit, consisting of tilt sensors, volumetric water content sensors (VWCs), a MEMS-based inclinometer, a rain gauge, and a video camera, was deployed on a slope within a residential area. The system demonstrated the potential to provide early warnings of rainfall-induced landslides by capturing surface and subsurface displacement, volumetric water content variation and rainfall in real time. All the sensor units were connected through solar-powered, long-range wireless data transmission and linked to a web-based platform for real-time data visualization and analysis. Change in tilt angle was recorded using two-axis tilt sensors. Data collected from all the tilt sensors revealed that all sensors remained stable except for sensor T1, which exhibited significant displacement during all three rainfall events in the unstable section of the slope and moved faster during the slope failure. This behavior indicates that the local displacement of the slope was not associated with the entire landslide body. The volumetric water content sensors responded rapidly and increased simultaneously during each rainfall event, and then declined after the rainfall ceased. A MEMS-based inclinometer was used to monitor subsurface displacement and to determine the potential sliding surface of the landslides. Landslide warning levels and corresponding thresholds are suggested based on the relationship among tilting rate ( Tr), displacement rate ( Dr), and saturation index ( Si). Analysis of tilt rate and displacement rate identified four types of slope movement (very slow, slow, moderate, and rapid). During rainfall events, volumetric water content (VWC) consistently exceeded 20%, corresponding to a saturation index ( Si) of approximately 0.5, indicating unsaturated to near-saturated conditions. Therefore, adopting Si < 0.5 as a blue warning level is reasonable. In addition, the saturation index with higher values reflects more critical slope instability conditions, serving as a reliable hydrological early-warning indicator. According to the Emergency Response Law of the People’s Republic of China, natural disasters should be categorized into four levels. In accordance with this framework, the early warning was divided into four levels (blue, yellow, orange, and red) based on the level of urgency, development stage, and effectiveness of risk management. However, there is no standard reference for the specific division of the four levels of geological disaster warning. The threshold proposed in this study is limited to the site-specific scenario. However, continuous monitoring is essential, as threshold values associated with slope instability may vary over time in response to changing environmental and geological conditions. The proposed monitoring system allows threshold values to be modified after installation. In addition, one volumetric water content sensor was found to be broken immediately after installation. Therefore, regular inspections are required to ensure that the local early-warning system operates reliably. Furthermore, rainfall data were not considered as threshold parameters in this study; future research will focus on developing and validating rainfall-based thresholds. Tilt sensors demonstrated effectiveness in monitoring the displacement behavior of shallow landslides, and signs of slope failure can be evaluated using tilt rates and horizontal displacement rates obtained from MEMS-based inclinometers, thereby demonstrating their effectiveness in landslide monitoring. Tilt sensors can be installed at a large scale to enhance early warning efficiency. In addition, rainfall events can be identified by observing temporal changes in volumetric water content using a volumetric water content sensor. Overall, the findings suggest that the developed low-cost, real-time monitoring and early warning system can be deployed on slopes prone to rainfall-induced shallow landslides and can support timely risk mitigation by providing early warnings. 6. Conclusions This study successfully developed and deployed a low-cost, simple, and reliable IoT- based real-time landslide monitoring and early warning system that combines surface and subsurface sensors to monitor rainfall-induced shallow landslides. The key findings can be summarized as follows: (1) The proposed real-time monitoring system integrates surface tilt sensors, volumetric water content sensors (VWC), a MEMS-based inclinometer, a rain gauge, and a video camera. The system provides reliable real-time monitoring with low power consumption. The system automatically acquires and transmits data via LoRa-based, solar-powered sensors, and the collected data are subsequently visualized through a real-time web interface, facilitating effective and efficient remote monitoring. (2) Based on field observations from the monitoring system, tilt sensors successfully recorded surface displacement, while increases in volumetric water content reflected rainfall infiltration. The MEMS-based inclinometer effectively identified horizontal displacement and potential slip surfaces, providing valuable insights into the stability of rainfall-induced shallow landslides. The system demonstrated reliable and consistent performance during the monitoring period. Compared with traditional warning approaches, the system enables the simultaneous evaluation of slope stability across multiple slope locations. (3) A strong relationship (R 2 = 0.95) was observed between tilt rate ( Tr) and displacement rate ( Dr). Based on this relationship, four types of slope movement were identified, enabling the development of a four-level early warning system, comprising blue, yellow, orange, and red. Saturation index ( Si) was further incorporated to characterize slope saturation conditions and define a hydrological threshold for landslide early warning. (4) In the proposed system, tilt rate ( Tr), displacement rate ( Dr), and saturation index ( Si) are used as the primary warning parameters. An early warning will be issued only when at least two of these three parameters reach their corresponding warning thresholds. Similarly, the warning level is upgraded only when at least two parameters exceed the threshold of a higher warning level. This multiparameter warning level reduces the possibility of false alarms, enhances the reliability of warning issuance, and supports timely public notification and emergency response. (5) Further analysis is suggested, as the developed system offers considerable potential for validating rainfall threshold models and to develop site-specific empirical equations. In future research, we aim to develop rainfall-based thresholds. It is concluded that the proposed monitoring approach serves as a good reference for real-time monitoring and early warning systems and can be further applied to other landslides. Author Contributions Conceptualization, A.U.M. and B.L.; methodology, B.L.; software, A.U.M.; validation, A.U.M.; formal analysis, A.U.M. and B.L.; investigation, X.L.; resources, X.L.; data curation, A.U.M.; writing—original draft preparation, A.U.M.; writing—review and editing, A.U.M. and B.L.; visualization, A.U.M.; supervision, X.L.; project administration, X.L. All authors have read and agreed to the published version of the manuscript. Funding This study was funded by the Key Research and Development Program of Yunnan Provincial Science and Technology Department [Grant No. 202503AA080030]. Institutional Review Board Statement Not applicable. Informed Consent Statement Not applicable. Data Availability Statement All data used in this study are available from the corresponding author upon reasonable request. Conflicts of Interest The authors declare no conflicts of interest. References Mondini, A.C.; Guzzetti, F.; Melillo, M. Deep learning forecast of rainfall-induced shallow landslides. Nat. Commun. 2023, 14, 2466. [] [ CrossRef] Qiu, H.; Su, L.; Tang, B.; Yang, D.; Ullah, M.; Zhu, Y.; Kamp, U. The effect of location and geometric properties of landslides caused by rainstorms and earthquakes. Earth Surf. Process. Landf. 2024, 49, 2067–2079. [] [ CrossRef] Wei, Y.; Qiu, H.; Liu, Z.; Huangfu, W.; Zhu, Y.; Liu, Y.; Yang, D.; Kamp, U. Refined and dynamic susceptibility assessment of landslides using InSAR and machine learning models. Geosci. Front. 2024, 15, 101890. [] [ CrossRef] Zhang, Y.; Zhang, F.; Zhang, J.; Guo, E.; Liu, X.; Tong, Z. Research on the Geological Disaster Forecast and Early Warning Model Based on the Optimal Combination Weighing Law and Extension Method: A Case Study in China. Pol. J. Environ. Stud. 2017, 26, 2385–2395. [] [ CrossRef] Gong, W.; Zhang, S.; Juang, C.H.; Tang, H.; Pudasaini, S.P. Displacement prediction of landslides at slope-scale: Review of physics-based and data-driven approaches. Earth-Sci. Rev. 2024, 258, 104948. [] [ CrossRef] Zhao, B.; Zhang, L.; Gu, X.; Luo, W.; Yu, Z.; Yuan, L. How is the occurrence of rainfall-triggered landslides related to extreme rainfall? Geomorphology 2025, 475, 109666. [] [ CrossRef] Guo, Z.; Ferrer, J.V.; Hürlimann, M.; Medina, V.; Puig-Polo, C.; Yin, K.; Huang, D. Shallow landslide susceptibility assessment under future climate and land cover changes: A case study from southwest China. Geosci. Front. 2023, 14, 101542. [] [ CrossRef] Dijkstra, T.A.; Dixon, N. Climate change and slope stability in the UK: Challenges and approaches. Q. J. Eng. Geol. Hydrogeol. 2010, 43, 371–385. [] [ CrossRef] García, A.; Hördt, A.; Fabian, M. Landslide monitoring with high resolution tilt measurements at the Dollendorfer Hardt landslide, Germany. Geomorphology 2010, 120, 16–25. [] [ CrossRef] Kumar, V.; Cauchie, L.; Mreyen, A.S.; Micu, M.; Havenith, H.B. Evaluating landslide response in a seismic and rainfall regime: A case study from the SE Carpathians, Romania. Nat. Hazards Earth Syst. Sci. 2021, 21, 3767–3788. [] [ CrossRef] Auflič, M.J.; Herrera, G.; Mateos, R.M.; Poyiadji, E.; Quental, L.; Severine, B.; Peternel, T.; Podolszki, L.; Calcaterra, S.; Kociu, A.; et al. Landslide monitoring techniques in the Geological Surveys of Europe. Landslides 2023, 20, 951–965. [] [ CrossRef] Hara, T.; Tatta, N.; Yashima, A. Assessment of ground-anchored slope stability based on variation in residual tensile forces. Soils Found. 2023, 63, 101353. [] [ CrossRef] Dai, B.; Wang, J.; Gu, X.; Xu, C.; Yu, X.; Zhang, H.; Yuan, C.; Nie, W. Development of Modified LSTM Model for Reservoir Capacity Prediction in Huanggang Reservoir, Fujian, China. Geofluids 2022, 2022, 2891029. [] [ CrossRef] Fan, X.; Yang, F.; Siva Subramanian, S.; Xu, Q.; Feng, Z.; Mavrouli, O.; Peng, M.; Ouyang, C.; Jansen, J.D.; Huang, R. Prediction of a multi-hazard chain by an integrated numerical simulation approach: The Baige landslide, Jinsha River, China. Landslides 2020, 17, 147–164. [] [ CrossRef] Wang, D.; Huang, G.; Zhang, Q.; Gao, Y.; Du, Y.; Liu, X. A multi-source landslide early warning model based on dynamic monitoring data and the SAAHP-FCE method. Geomat. Nat. Hazards Risk 2025, 16, 2513536. [] [ CrossRef] Yalçınkaya, M.; Bayrak, T. Dynamic model for monitoring landslides with emphasis on underground water in Trabzon Province, Northeastern Turkey. J. Surv. Eng. 2003, 129, 115–124. [] [ CrossRef] Rosi, A.; Segoni, S.; Canavesi, V.; Monni, A.; Gallucci, A.; Casagli, N. Definition of 3D rainfall thresholds to increase operative landslide early warning system performances. Landslides 2021, 18, 1045–1057. [] [ CrossRef] Liu, J.; Tang, H.; Li, Q.; Su, A.; Liu, Q.; Zhong, C. Multi-sensor fusion of data for monitoring of Huangtupo landslide in the three Gorges Reservoir (China). Geomat. Nat. Hazards Risk 2018, 9, 881–891. [] [ CrossRef] Xu, Q.; Zhao, B.; Dai, K.; Dong, X.; Li, W.; Zhu, X.; Yang, Y.; Xiao, X.; Wang, X.; Huang, J.; et al. Remote sensing for landslide investigations: A progress report from China. Eng. Geol. 2023, 321, 107156. [] [ CrossRef] He, Y.; Wang, W.; Zhang, L.; Chen, Y.; Chen, Y.; Chen, B.; He, X.; Zhao, Z. An identification method of potential landslide zones using InSAR data and landslide susceptibility. Geomat. Nat. Hazards Risk 2023, 14, 2185120. [] [ CrossRef] Sestras, P.; Badea, G.; Badea, A.C.; Salagean, T.; Oniga, V.-E.; Roșca, S.; Bilașco, Ș.; Bruma, S.; Spalević, V.; Kader, S.; et al. A novel method for landslide deformation monitoring by fusing UAV photogrammetry and LiDAR data based on each sensor’s mapping advantage in regards to terrain feature. Eng. Geol. 2025, 346, 107890. [] [ CrossRef] Huang, G.; Du, S.; Wang, D. GNSS techniques for real-time monitoring of landslides: A review. Satell. Navig. 2023, 4, 5. [] [ CrossRef] Gian, Q.A.; Tran, D.-T.; Nguyen, D.C.; Nhu, V.H.; Tien Bui, D. Design and implementation of site-specific rainfall-induced landslide early warning and monitoring system: A case study at Nam Dan landslide (Vietnam). Geomat. Nat. Hazards Risk 2017, 8, 1978–1996. [] [ CrossRef] Yu, X.; Zhao, T.; Gong, B.; Tang, C.a. The water weakening effect on the progressive slope failure under excavation and rainfall conditions. Bull. Eng. Geol. Environ. 2024, 83, 316. [] [ CrossRef] Qian, C.; Li, Y.; Vardon, P.J.; Shao, W.; Song, J.; Zhang, B.; Xu, N. Temporal stability and risk analysis of soil slopes subject to rainfall: The influence of heterogeneity. Eng. Geol. 2025, 347, 107895. [] [ CrossRef] Pajalić, S.; Peranić, J.; Maksimović, S.; Čeh, N.; Jagodnik, V.; Arbanas, Ž. Monitoring and Data Analysis in Small-Scale Landslide Physical Model. Appl. Sci. 2021, 11, 5040. [] [ CrossRef] Palis, E.; Lebourg, T.; Tric, E.; Malet, J.-P.; Vidal, M. Long-term monitoring of a large deep-seated landslide (La Clapiere, South-East French Alps): Initial study. Landslides 2017, 14, 155–170. [] [ CrossRef] Alam, M.J.; Manzano, L.S.; Debnath, R.; Ahmed, A.A. Monitoring Slope Movement and Soil Hydrologic Behavior Using IoT and AI Technologies: A
