An Ensemble Learning Based Framework for Smart Landslide Detection, Monitoring, Prediction and Warning in IoT-Cloud Environment

doi:10.21203/rs.3.rs-3137501/v1

Download PDF

Research Article

An Ensemble Learning Based Framework for Smart Landslide Detection, Monitoring, Prediction and Warning in IoT-Cloud Environment

https://doi.org/10.21203/rs.3.rs-3137501/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 16 Nov, 2023

Read the published version in Environmental Science and Pollution Research →

You are reading this latest preprint version

Landslides occur every year during the Monsoon season in hilly areas. Every year, this natural disaster lead to several fatalities, injuries and property destruction. It is crucial to monitor landslides and promptly alert people to looming disasters in light of these injuries and fatalities. Till date no efficient technique is in practice to predict landslides. The tools that are now available monitor landslides at a very high cost and do not offer early warning or forecasts of soil movement. A innovative, low-cost Internet of Things (IoT)-based system for landslip warning, monitoring, and prediction is the major objective of this research. Its assessment, implementation, and development are described in detail.In this study, an IoT-based smart landslide detection, warning, predicting and monitoring system is proposed. The pre and post measures are considered using sensors and other hardware to deal with landslide disaster. It uses real time monitoring of the environment (landslide site) for any changes and providing appropriate output by comparing the threshold values. The proposed system is put to the test on a prototype model, which performed well in our tests. The database was updated 2.5 seconds after the landslide happened, thanks to a steady internet connection. In less than 5 seconds after the event, the thinkspeak channel is able to display a graphical depiction of the data as well as its position. Multiple readings showed an 80–85% system accuracy rate. Further, the proposed ensemble learning based risk prediction model is applied on static and dynamic data to predict the landslide for future references. The ensemble classifier model has 98.67% recall, 96.56% accuracy, 97.35% F1- value and 96.07% precision. The alert SMS are also sent to concerned authority for medical emergency/PWD department/ District administration.

Earthquake

Casualty prediction

Real Time Monitoring

IOT

Post landslide detection to keep the concerned authorities up to date.
Data collection for future references.
A simplified and reliable technique for practical implementation even in isolated areas.
Instant alerts to the PWD department and rapid evacuation
Prediction of landslide risk using Proposed risk prediction model applied on collected data.

Landslides have always been a part of our planet's history. Mass movements of rock, debris, and soil down a slope of the land are often categorized as landslides (What Are Landslides and How Do They Occur?,n.d.-a). While landslides are a natural environmental danger, they have lately become more common in some regions as a result of human activities. From mountain ranges to coastal cliffs, landslides can occur in a variety of settings with steep or mild slope gradients. A landslide is primarily caused by gravity, but there are additional factors that impact slope stability and generate particular circumstances that render a slope vulnerable to collapse. A specific occurrence (such as heavy rain, an earthquake, or the cutting of a slope to make a road, and many others) often triggers a landslide, but this is not always obvious. Road connection is a big concern during the rainy season. Heavy rain causes total washout of roads up to 50 meters in length, and landslides disrupt traffic movement in many areas. Natural disasters and extreme weather events are also a cause for concern. Small cars and regular traffic flow can only be restored if connectivity is restored in the lowest amount of time feasible. The main issue is when trenches of 50 meters of the steep road are completely washed away, and the only way to get cars moving is to build temporary bridges. These bridges must be a secure, long-lasting, and cost-effective solution.

More intense scientific investigations and engineering procedures focusing on the problem of landslides are needed in Indian states like Himachal Pradesh due to the disastrous landslides. The Himachal Pradesh hills and mountains are vulnerable to landslides during monsoons and high-intensity earthquakes. The first step for developing a landslide prediction model should be to conduct geological and geotechnical investigations to create zoning maps of landslide and rockfall prone zones. Protective measures should be implemented in landslide-prone areas based on professional advice, and these locations should be monitored with continuous data during the monsoons. Landslides systematically impact several lives and cause serious destruction to property and the lifetime of innocent individuals. As a result, in landslide-prone areas, establishing Early Warning Systems capable of the police investigation and notifying compact persons concerning the prevalence before is very important. To meet this need, numerous frameworks are developed such as distant device Networks (RSN)-based systems (What Are Landslides and How Do They Occur?,n.d.-a). The measuring system considers surface morphology and earth science pictures whereas exhibiting the surface and recognizing landslide late action (What Is a Landslide and What Causes One? | U.S. Geological Survey, 2022b). Satellite pictures area unit utilized image threshold ding by Genetic Programming to tell apart land movement within the space Kumar et al. (2017). Complicated computations and a k-NN classifier area unit accustomed classify landslide space and non-landslide neighborhood in Programmed Landslide Detection from distant police investigation symbolism employing a Scene Characterization Strategy derived from BoVW and pLSA Ramesh (2009a). Imaging from a Fixed-Wing UAV for Landslide Mapping Mckean and Roering (2004). Advanced camcorders are wide accustomed monitor mine inclination conditions and warning of sloppy rocks Rosin and Hervás (2002); Cheng et al. (2013a); Rau et al. (2011); Bjorklund (2017); Mchugh and Girard (2002); Mchugh (1900).

Most of these Systems require expensive equipment and upkeep. Therefore, it would be extremely difficult for agricultural nations with limited financial resources to fund such frameworks to screen each landslide-prone zone. This emphasizes the need for a straightforward system that can detect landslides in real-time and notify those who are at danger so they can take the necessary measures. The Internet of Things (IoT) is a group of wireless sensor networks that collaborate to talk to one another and react to their surroundings. Since its conception, the internet of things (IoT) has developed into the next progressive innovation in the process of transforming the internet into a fully coordinated future internet, fueled by a number of recent advancements. The Himalayan mountain range has seen a rise in landslide incidents over the past few years, however it is noted that not enough research is being done to comprehend these landslides and the causes behind them. 2016 Bjorklund. It is physically separated into three zones. Siwalik, Lesser, and Great Himalayan are the three ranges that make up the Shivaliks (the outside Himalayas), Himalayas of the Lower Himalayas (the centre zone), and The Great Himalayas (the northern zone), each having its own litho-tectonic setting. India's Himachal Pradesh is a hilly state. The landslide-prone regions of Himachal Pradesh are depicted in Fig. 1. There are several natural and man-made risks in the state, according to the policy on disaster management. Landslides, blizzard dam collapses, and earthquakes are the greatest hazards. Despite the fact that landslides happen all year round, the monsoon season is when they seem to happen most frequently. The risk of landslides is rising as a result of numerous causes, including an increase in total rainfall, deforestation, and road building. Himachal Pradesh's landslip incidence map is shown in Fig. 2.

1.1. Landslide Susceptibility Zonation

Landslide susceptibility is the spatial prediction of a landslide which generally depends on various causative factors. Generally, we tend to predict locations that are more vulnerable to landslides compared to others depending on previous landslides and possible factors. The landslide hazard of an area is determined by multiple factors Mchugh and Girard (2002); Prakasam et al. (2019). Many techniques have been applied to delineate the area into different landslide hazard zones. These techniques are broadly divided into direct and indirect methods. Direct methods include Geomorphological mapping and Landslide distribution analysis whereas there are several indirect ways for landslide Hazard zonation. Indirect methods include Index overlay, Fuzzy logic, analytical Hierarchical process, multivariate technique, Deterministic modeling, etc. The approach relied on a statistical correlation between a Landslide map and several parameter maps. To define the landslide danger zones, all of the weighted maps were combined. In Himachal, slopes, deforestation, high precipitation, and road building are reported to be the primary causes of slope instability, resulting in a variety of landslide motions. There is no widely recognized approach for assessing landslide dangers. Several attempts have been made in recent years to use different LHZ techniques and compare findings to determine the best suitable model. Advanced multivariate methods are highly successful in the geographic prediction of landslides.

1.2. Our Contributions

In this paper, we present an IoT-based system for landslide prediction and watching victimization sensors and electronic boards during this study.

Post landslide detection to keep the concerned authorities up to date.
Dynamic and Static Data collection for future references.
A simplified and reliable technique for practical implementation even in isolated areas.
Instant alerts to the PWD department and rapid evacuation.
Prediction of landslide risk using Proposed risk prediction model applied on collected data.

The remaining sections of the paper are organized as follows: Section 2 presents the detailed overview of related work. Section 3 contains information regarding preliminaries and background. Section 4 presents detailed information regarding the proposed framework. Section 5 conducts an experimental evaluation of the proposed framework. Lastly, Section 6 contains the conclusion and future scope.

This section contains a summary of the numerous approaches for predicting and preventing landslides. Table 1 contains the comparison of existing landslide susceptibility analysis and prediction techniques. The event of continuous watching of basic and crucial applications has benefited from the utilization of remote detector networks. PC researchers' got to study and comprehend more room areas has been fueled by remote detector innovation, permitting them to propose or expand current setups.

In a recent study Lee et al. (2003), the drought forecast and alert system (DFAS) was instructed and developed. This arrangement employs mobile communication to alert shoppers; the proposed framework utilizes continuous information assortment and transmission via distant, satellite organization, detector hubs, Wi-Fi, and the internet. The particular running of data created potential by net access unveils the network to a bigger audience. The reference introduces an associate degree beta soil observant organization that uses a far-off detector network to look at continuous estimates at transient and geographical granularities that were antecedently inconceivable. To build a scene classification approach supported the bag-of-visual-words (BoVW) illustration, Cheng et al. (2013b) proposed merging of the unsupervised probabilistic latent linguistics analysis (pLSA) model with the k-nearest neighbor (KNN) classifier. They divided the sub-images into landslide and non-landslide zones supported item distribution. For affiliation between individual sensors and therefore the knowledge lumberjack, also as from the information lumberjack to online info, Abraham et al. (2020) proposed victimization IoT-based network wireless modules.

Table 1

Comparison of existing Landslide susceptibility analysis and prediction techniques
References	Methodology	Goals	Benchmark Dataset	Performance Parameters
Prakasam et al. (2019)	Artificial neural network (ANN)	Landslide susceptibility analysis	https://link.springer.com/chapter/10.1007/978-3-030-60227-7_37	AUC
Lee et al. (2003)	Using object distribution to divide the sub-images into landslide and non-landslide areas	Landslide detection from satellite images using a scene classification technique	https://www.tandfonline.com/doi/abs/10.1080/01431161.2020.1714774	Accuracy, False Alarm Rate
Abraham et al. (2020)	Unstable Slopes: Geotechnical Monitoring	Early Warning of a Landslide in the Darjeeling Himalayas	https://www.tandfonline.com/doi/full/10.1080/17538947.2020.1860145	Accuracy, Sensitivity
Mohan et al. (2021)	landslide detection with the help of machine and deep learning	A review of landslide detection systems using remote sensing	https://link.springer.com/article/10.1007/s12665-021-09525-6	Accuracy
Sruthy et al. (2020)	Detection of Landslide prone areas using sensors like soil moisture sensor and accelerometer	Landslide Detection and Monitoring System	https://www.researchgate.net/publication/333089233_IoT_Based_Landslide_Detection_and_Monitoring	Accuracy
Susanto et al. (2019)	Vibration Sensors, Slope and Humidity	Landslide Detection and Monitoring System	https://jtsiskom.undip.ac.id/index.php/jtsiskom/article/view/13591	Accuracy
Mezaal et al. (2017)	Dempster–Shafer theory from LiDAR-derived data and orthophotos	Automatic landslide detection	https://www.mdpi.com/2072-4292/12/18/3054	Accuracy Sensitivity
Sameen and Pradhan (2019)	Fusion of Spectral and Topographic Information	Landslide Detection	https://ieeexplore.ieee.org/document/9171238	Accuracy
Ramesh and Vasudevan (2012)	Deep-earth Sensor Probes	Landslide Detection	https://link.springer.com/article/10.1007%2Fs10064-005-0025-y	Accuracy
Liu et al.(2019)	Data Driven Cyber-Physical System	Landslide Detection	https://link.springer.com/article/10.1007%2Fs11036-013-0456-9	Accuracy, Sensitivity
Pathania et al. (2020)	Auto Regression and Sequential Minimal Optimization	Landslide Detection, Prediction, Warning	https://www.researchgate.net/publication/339850328_A_Low_Cost_Sub-Surface_IoT_Framework_for_Landslide_Monitoring_Warning_and_Prediction	RMSE Score
Harsa et al. (2023)	Generalized linear model and Deep Learning Model	Landslide Prediction	https://ijai.iaescore.com/index.php/IJAI/article/view/21152	Log-Loss, AUC Curve
Chang et al. (2023)	Random forest and Multi-layer perceptron	Landslide Prediction	https://www.sciencedirect.com/science/article/pii/S1674775522001615	AUC, Accuracy
Kumar et al. (2022)	GIS-multi-criteria decision analysis Technique	Landslide Prediction	https://jme.shahroodut.ac.ir/article_2535.html	Weighted Rating Scheme

Mohan et al. (2021) reviewed remote sensing strategies used for landslide detection with the assistance of machine learning and deep learning. They mentioned however trendy geospatial techniques like aerial photogrammetry, Terrestrial optical device scanning, satellite remote sensing pictures (i.e., panchromatic, multispectral, microwave radar images), and high-resolution Digital Elevation Model (HR-DEM) are often accustomed to update ancient approaches for characteristic landslides and predicting future risk related to slope failures. They recommend deep learning as a reliable approach for remote sensing problems as well as imaging system trade-offs that end in low-quality studies, likewise as for dashing up activities like image identification, object detection, classification, and so onSruthy et al. (2020) proposed the usage of soil wetness and measuring system sensors to look at the conditions that result in its incidence and apprize well before. Susanto et al. (2019) proposed slope, humidity, and vibration were the 3 sensors whose performance for soil movement watching was evaluated. Information from sensors reveals that every one parameter has the potential to be given for landslide watching. Device detection was used to build the system, and information was processed on the online server victimization cloud services. Mezaal et al. (2017) suggests utilizing the Dempster Shafer theory (DST) to mix high-resolution measuring instrument information products with the Greenness index calculated from orthophoto imagination. Using spectral (RGB bands) and geographic data, Sameen and Pradhan (2019) proposed victimization residual networks for landslide detection (altitude, slope, aspect, curvature). Ramesh and Vasudevan (2012) provides the newest wireless device network (WSN) with deep-earth probes (DEPs) that has been placed to observe Associate in Nursing in progress landslides within the Western Ghats of South Bharat. Liu et al. (2019) proposed a system consisting of Wi-Sun acceleration sensors which will discover the acceleration of the encircling surroundings in 3D area, with the sensors connected to the router (serving as a sink node) through Wi-Sun transmission, lowering energy usage while not sacrificing detection capabilitiesPathania et al. (2020) created and installed a novel subsurface system that can provide SMS warnings of large subsurface movements. Auto regression and sequential minimal optimization machine-learning models are used for predictive analytics on system data. The suggested technology is cheap and uses the same movement detecting mechanism as existing solutions. This low-cost technology can be deployed widely to monitor landslides and alert people. Harsa et al. (2023) used machine learning and AI systems to forecast landslides. The algorithms employed landslip event occurrence data from the Indonesian National Board for Disaster Management and precipitation data from global satellite mapping of precipitation satellite observation. The first technique makes use of the generalized linear model, and the second makes use of deep learning. The top models for each strategy had an AUC curve of 0.828 and 0.836 as well as 0.156 and 0.154 log-losses. Chang et al. (2023) used multi-scale segmentation slope units for landslide prediction. Mean, range, and standard deviation data is adopted to calculate the factor heterogeneity of conditions. The results showed that slope-machine learning models exhibit lower accuracy than Grid models.Kumar et al. (2022) used GIS-MCDA to zone landslide hazards in Kullu, Himachal Pradesh. This study evaluated the analytical hierarchy method (AHP) for landslip hazard mapping. The geospatial information system prepared databases, analyses, models, and outcomes. ArcGIS 10.0 weights input layers for integration is adopted. Six landslide causation components employed geographical database parameters.

Before discussing our proposed framework, some basic terms, background, and preliminaries are discussed in this section. We discussed the sensors that are used, how they work, and the algorithm that underpins their code in the proposed framework. Additionally, block diagrams are used to illustrate the system architecture.

3.1. Sensor Network Design and Implementation

IoT (Internet of Things) describes actuating and communicating wireless sensor networks that appear to fit in with their surroundings. A Common Operating Picture (COP) is created by the sharing of this data across numerous platforms. IoT has emerged from its infancy and is now the next breakthrough technology in changing the Internet into a fully integrated Future Internet, spurred by the adoption of a number of contemporary technologies. By offering practical communication, trustworthy inspection, and performing applications, WSNs have grown to be an essential component of a variety of applications, including environmental monitoring Suryadevara et al. (2014), military surveillance urii et al. (2012), and medical Furtado and Trobec (2011). WSNs are made up of several sensor nodes that are widely dispersed and wirelessly connected in order to transmit and receive environmental data. Each sensor node has a radio transceiver, a CPU, a power supply section, and at least one or more sensors. Due to the complexity of these systems, developing WSNs becomes an extremely difficult process. The power consumption requirement, which serves as the primary key, is one of several significant criteria that must be met throughout the design of WSN. Because of this, a lot of current research focuses on surveying WSNs. Because of this, a lot of current research focuses on surveying WSNs. The notion of WSNs and their properties were discussed by the writers in Akyildiz et al. (2002) and Mottola and Picco (2006). The generations, routing, architecture, and storage management of WSN were all covered. Other research An overview of the current routing protocols for sensor networks is provided by Adeel et al. (2019) and Jeong et al. (2019). The authors of Ramesh (2009b) offered a variety of existing middleware for sensor networks. In Gui et al. (2011), the authors discussed several hierarchical taxonomy and sensor localisation algorithms as well as their applicability in various scenarios. They discussed brand-new sensor localisation techniques and how IoT infrastructure may use them.

The detail of the proposed system is presented below: -

4.1. System Architecture

In this section, how the flow of data takes place in proposed system is presented. Figure 3 describes the work flow diagram for our proposed system of smart landslide detection and monitoring. There will be two different setups that will respectively monitor Pre-Landslide and Post-landslide.

Pre-Landslide: - In this, the live monitoring system (Fig. 3) is used which will collect live data from its surroundings, and when a particular threshold is met the flow is transferred with a suitable message to the Arduino Board. Then the GSM module will be triggered and hence an appropriate message shall be sent to the PWD department. This alert will help the instantaneous arrival of rescue vehicles and reduce deaths that are caused due to delays caused due to traffic jams and delays in arrival.

Post-Landslide: - In this, setup consisting of an ultrasonic sensor and WIFI module will be placed. In case a landslide occurs and a threshold value for the ultrasonic sensor is met it will send a signal to the Arduino board. Then the Arduino board is responsible to hold the data and send it to the thingspeak channel with the help of a WIFI module where it will be stored and can be accessed anytime for the risk prediction, we desire to study the conditions at those sites. Further, the risk is prediction by applying the proposed risk prediction model.

4.2. System’s Logic

The system operates according to the straightforward logic of interrupts generated by accelerometers as illustrated in Fig. 4. The system first verifies that all of the sensors are accessible. The system will only move forward if all the sensors are available. If not, the system rotates indefinitely until every sensor is available. The microcontroller then monitors the accelerometers while putting the remaining sensors to sleep. When the accelerometers are interrupted by movement or when ten minutes have passed, the microcontroller wakes up and reads the values from every sensor. After reading values, data are transmitted to a cloud server. Once data is received in the cloud, alerts are generated based on the accelerometer-recorded motions. There are three levels of alert generation: large, minor, and moderate movements. In addition, the alert notifications are generated in two ways: 1) via SMS on registered users' phones 2) via blinkers and hooters for road traffic. The issuance of alerts is determined by the magnitude of movement and predefined thresholds. Further the risk of landslide is also predicted using the proposed risk prediction model using machine learning classifiers.

4.3. Methodology

Four pins are included in the FC-28 soil moisture sensor. VCC is the power pin, A0 is the analog output, D0 is the digital output, and GND is the ground pin. This module also has a potentiometer for setting the threshold value, which the comparator-LM393 may then assess. The threshold value will trigger an alert.

The LDO transformer / Low dropout transformer in Associate in Nursing integrated three.3 V within the MMA845X measuring device module allows these modules to be supercharged from a three.6 V six|to six} V DC power supply. The microcontroller/system (Arduino, Raspberry-Pi, etc.) could communicate with the device module through the Associate in Nursing I2C interface capable of hastens to 25. This module is extremely appropriate to be used in moveable electronic devices as a result of it frees the microcontroller to interrupt to operate on your time wasted grouping knowledge (continuous polling data). Associate in the Nursing interrupt is set to get Associate in Nursing interrupt trigger signal (wakeup interrupt signal) of 1 or a mix of events (event-driven interrupt) that allows these sensors to observe the condition whereas in power saving mode.

The second is a microcontroller called an Arduino Uno R3. A microcontroller board supported by the ATmega328 chip may be the Arduino Uno R3. The Arduino Uno has a button, a USB connection, an influence jack, six PWM output pins, six analogue input pins, and fourteen digital output pins. It also contains an influence jack and an associate ICSP header. A sixteen MHz quartz oscillator, a USB connection, a control interface, a connected ICSP header, and a button are also included. The Arduino just has to be connected to your laptop through USB to be put to use, however the adapter and battery should provide enough power. Specifications Atmega 328 microcontroller, 5V power supply, input voltage (recommended), 7–12V, input voltage (limited) in the Arduino Uno R3 Six-20V Digital I/O Pins 14 (of which six output PWM) DC Current per I/O Pin 40mA, when 3.3V Pin 50mA DC, non-volatile storage, analogue input pin 6 boo loader zero uses 32 ATmega328 computer memory units.5 KB, 2 SRAM (ATmega328) and 1 KB EEPROM (ATmega328) memory units, with a clock speed of 16 MHz.

Ultrasonic sensors operate by emitting an undulation that's higher than the human hearing varies. The sensor's electrical device functions as a mike, receiving and sending inaudible sound. To deliver a pulse and receive the echo, our inaudible sensors, like several others, use one electrical device. The sensing element measures the time between delivering and receiving an inaudible pulse to estimate the space to a target. This module's operation is easy. It emits a 40 kHz inaudible pulse that travels through the air and bounces back to the sensing element if it encounters an obstruction or item.

The distance is also determined by multiplying the transit time by the sound speed. The gap will probably be calculated by victimizing the TRD (time/rate/distance) measure formula. The calculated distance is the distance traveled from the unbearable electrical device to the thing and back to the electrical device, so it is a two-way journey. Divide this distance by 2 to urge the particular distance between the electrical device and therefore the object. Unbearable waves travel at an identical rate as sound waves (343 meters per second at twenty degrees Celsius). The gap between the thing and therefore the sensing element is 0.5 the gap traveled by the wave. The gap to associate object place before associate unbearable sensing element is calculated using the below equation:

$$distance=\frac{time taken x speed of sound}{2}$$

Increased soil wet content and movement of rocks and dirt with the layer square measure to components that cause landslides. During this case, a soil wet sensing element and a measuring instrument square measure were used. A measuring instrument is an electrical sensing element that detects the acceleration forces working on an item to ascertain the object's location in the house and track its movement, whereas a soil wet sensing element can observe the wet content therein a specific region. This device (in Fig. 5) is going to be capable of providing enough of a time interval to permit officers to warn residents and motorists shortly before a possible landslide. With appropriate algorithms, we will predict such events within twenty-four hours or maybe every week before. Both the soil moisture sensor and the accelerometer sensor will be placed underneath the soil in a proper container. These sites have been selected based on the geographical study of the terrain and landslide-prone areas will be capped.

The data are going to be compared with the edge which can be set by the researchers and acceptable alerts are going to be sent to the exploitation GSM module to the allotted authority. These alerts may be simply distinguished exploitation; the serial no and Landslide location will simply be determined. The collection may be sent to a server wherever it'll be held on and monitored. With the assistance of a wireless fidelity module, the information is going to be recorded and continuous chase can happen.

A second setup (shown in Fig. 6) will be placed at the roadside for post landslide detection. The ultrasonic sensor will be placed at a certain height set with a threshold value. When an obstacle comes within the periphery the alert is sent to the PWD department and thus helps in the Instantaneous arrival of clearance vehicles.

4.3.1. Phase 1: Component description

A. Soil Moisture Sensor:

The volumetric water content within the soil is measured by a soil wet detector. as a result of direct measurement measure of free wet necessitates the removal, drying, and consideration of a sample, soil wet sensors indirectly live the volumetric water content by exploiting another property of the soil as a proxy for the wet content, like electric resistance, nonconductor constant, or nucleon interaction.

The fork-shaped probe, which has two exposed conductors, functions as a rheostat (similar to a potentiometer) whose resistance changes with the number of water within the soil. The soil wet has an associate inverse relationship with this resistance:

The higher the water content within the soil, the larger the conduction and therefore the lower the resistance.
The lower the water contents within the soil, the lower the conduction, and also the larger the resistance. The sensing element provides an output voltage, which can be used to discover the wetness level.

B. Accelerometer:

An associate in nursing measurement instrument is a gismo that tracks proper acceleration. Contrary to coordinate acceleration, which is defined as acceleration during a mounted reference system, correct acceleration is defined as a body's acceleration (or rate of modification of velocity) in its immediate rest frame. A three-axis measurement device with little to no noise and little power consumption may be the ADXL335. It measures acceleration along the x, y, and z axes, one of the three g fluctuates. This module's output signals square analogue volts proportional to acceleration. It has a polysilicon sensing element with a surface-micro milled surface and a proof process circuit. Polysilicon springs that are located on site regulate this structure. When an exact axis receives the acceleration, the structure can deflect. The capacitance between mounted plates and plates linked to the hanging structure changes as a function of deflection. The change in capacitance directly relates to the acceleration along its axis. The sensor element converts this fluctuation in capacitance into an analogue output voltage through processing. In tilt-sensing applications, it will detect both the static acceleration of gravity and the dynamic acceleration brought on by motion, shock, or vibration.

C. Arduino uno:

The Arduino Nano could be a reasonably priced microcontroller board created by Arduino. cc. It’s going to be made employing a microcontroller like the Atmega328. Supported by the ATmega328, the Arduino Nano could be a little, comprehensive, and breadboard-friendly board (Arduino Nano three. x). It offers a lot of similar options because of the Arduino Duemilanove, however, it comes in a completely different packaging. It simply includes a DC power connection and uses a Mini-B USB cable instead of a standard one.

D. Ultrasonic Sensor:

An inaudible sensing element is a device that uses inaudible sound waves to leave the space between a target item and transforms the mirrored sound into an electrical signal. Inaudible waves move faster than sonic sound waves (i.e., the sound that humans will hear). Inaudible sensors are a unit created of two primary components: a transmitter (which uses electricity crystals to make sound) and a receiver (which encounters the sound once it's cosmopolitan to and from the target). Inaudible sensor area units are usually used as proximity detectors. They are employed in self-parking technologies and anti-collision safety systems in cars.

E. GSM Module:

GSM stands for global system for mobile communication. It is a mobile communication modem and it is a widely used mobile communication system in the world. The GSM was developed at Bell Laboratories in 1970. It is an open and digital cellular technology; it is used for transmitting mobile voice and data services and operates at the 850MHz, 900MHz, 1800MHz, and 1900MHz frequency bands.

4.3.2. Phase 2: Experimental Setup and Circuit Diagram

A. Experimental Setup

Landslide theory has not however advanced sufficiently to form the correct predictions of landslides or to model the phenomena in a lot of depth the manner we'd like them. We tend to design associates in nursing and develop an experimental setup to simulate and take a look at the operating of the system. The experimental setup provides work for developing, testing, and calibrating the devices and subsystems of the wireless sensor network. The soil is packed into the setup in conjunction with the device column for the experiment. Water is then further within the variety of flow, till the slope fails.

B. Circuital Diagram

The pictorial representation of the sensors and boards is captured using the Circuito.io interface and depicts accurate connections among them. The Fig. 7 shown below describes the circuital connection for setup 1 (w.r.t Fig. 4 of section 4). It consists of a Soil Moisture Sensor, WIFI module, Accelerometer Sensor, and Arduino Board.

The Fig. 8 shown below describes the circuital connection for setup 2 (w.r.t Fig. 6 of section 4). It consists of an Ultrasonic Sensor, GSM module, and Arduino board.

4.3.3. Proposed Risk Prediction Model

In this section the risk of landslide is predicted using the static and dynamic data of various geotechnical in the cloud. These parameters are the slope of the terrain, rain severity, soil shear strength, soil moisture, acceleration values etc. as shown in Table 2. The landslide is predicted using an ensemble prediction model in the cloud computing layer. As illustrated in Fig. 3, the ensemble model has two levels.

Table 2

Data Categories Of Landslide Monitoring
Data Type	Sub Categories	Description
Static Data	Geographical data	Geologically fundamental arrow data, comprising huge scale administrative division, lake and water systems, residential areas, and lattice spatial data for the digital elevation mode
	Potential landslide sites Survey Data	Rock ,Plane graph, soil mass characteristics data, and profile, location map of landslides,
	Documentary Information	Relevant statistical materials recorded about historical disasters and technical reports.
Dynamic Data	Real-time Sensor monitoring data	Ground surface displacement, Rain Severity, depth displacement, Soil moisture, groundwater level, inclination, acceleration values (2,924 values measured every 5 minutes from each sensor node) and slope of terrain

In Level-1, two ML algorithms are utilized: Weighted Random Forest (WRF) and Extreme Gradient Boosting (EGB) which are able to concurrently process the features with high processing speed Sarangi et al. (2022); Reddy et al. (2022). EGB is chosen because they have both boosting and Gradient Descent (GD) capabilities. By simultaneously updating the value in each direction of the gradient, GD is an effective and well-formed optimization algorithm that seeks to find the best solution to a loss function. Boosting increases, the accuracy of GD by giving more weight to the high error (worst case) and retraining the model. To make up for the shortcomings of the preceding weak learners, a new weak learner is introduced at each gradient boosting level. EGB also uses a more standardized model formalization to avoid overfitting, which improves performance. The first tree in the EGB building process uses the whole training set, but the second tree uses the training set based on residuals different from those in the first tree. In other words, the disadvantages of the first tree are evaluated throughout the subsequent training phase. Training is repeated until the termination requirement is reached. The final result may be calculated by adding the prediction of each tree. Eq. 2 depicts the objective function of EGB at step t.

Where, ${f}_{k}$ represents the k^th independent tree in F, ${x}_{i}^{t}$ is prediction result of i-sample at t in Eq. 3, ${x}_{i}$ is expected prediction result of i- sample, Ω is the regularization in Eq. 4 and $l$ indicates the loss function.

Where, ${y}_{i}$ represents the input variable, L is the number of leaves, $\omega$ indicates the leaf weight magnitude, $\gamma and \lambda$ are the controlling parameters.

WRF is one of the most precise machine learning algorithms now in use, and it can process a large number of input variables without variable elimination. In this study, all selected features give information to predict the landslide risk. RF attempts to utilize all of them for predicting. The working steps of the proposed WRF classifier is given in Algorithm 1. The findings of this layer serve as input for the second layer.

Algorithm 3 : WRF Algorithm
Input: Landslide dataset (D), number of tree (NT), random subset of features (FS) Output: Random Forest (RF) tree with severity prediction
	Step 1:	Initially, the random forest tree is empty (RF = = Null).
	Step 2:	For each i = 1 to NT, do following
	Step 3:	Apply the bootstrap algorithm on training dataset (D) such as${D}_{i}=bootstrap \left(D\right).$
	Step 4:	Apply the decision tree (DT) algorithm${DT}_{i}=random decision tree \left(D, FS\right).$
	Step 5:	Build the random forest (RF) tree as $RF=RF\cup {DT}_{i}$.
	Step 6:	End for
	Step 7:	For each i = 1 to n, do following
	Step 8:	Determine the weight (${w}_{i}^{t}$) of i^th sample using Eq. 1. ${w}_{i}^{t}=\frac{1}{OB}{\sum }_{jϵOB}^{}\left\|{Y\_predicted}_{i,j}-{Y\_actual}_{i,j}\right\|\left(5\right)$
	Step 9:	End for
	Step 10:	For each i = I to NT, do following
	Step 11:	${\varnothing }_{i}=f\left(AUC{\left({w}^{t}\right)}_{{IB}_{i}},AUC{\left({w}^{t}\right)}_{{oB}_{i}}\right)\left(6\right)$
	Step 12:	End for
	Step 13:	For each i = 1 to NT, do following
	Step 14:	Determine the weight (${w}_{i}$) using Eq. 3. ${w}_{i}=\frac{{\left(NT-{p}_{i}+1\right)}^{k}}{{\sum }_{k=1}^{NT}{\left(NT-{p}_{i}+1\right)}^{{k}^{{\prime }}}}\left(7\right)$
	Step 15:	For each i = 1 to n, do following
	Step 16:	Determine the final prediction using Eq. 4. ${Y\_predicted}_{i}=\frac{1}{NT}{\sum }_{j=1}^{NT}{Y\_predicted}_{i,j}\times {w}_{j} \left(8\right)$
	Step 17:		End for
	Step 18:		Return RF tree.

In Level- 2, The prediction of landslide is computed using an Artificial Neural Network (ANN) model. The three input components of the neural network are weighted equally, which further simplifies the model and has the potential to produce excellent prediction accuracy. A neural network's three parts are the input layer, hidden layers, and output layer. Three nodes that reflect the outputs of WRF and EGB make up the input layer in this investigation. Unit severity prediction is the single node in the output layer. Identifying the structure of the hidden layer(s), including the number of layers and hidden units in each layer, is a challenge. It is known that there will only be one hidden layer because there are only three input nodes. By testing various node counts, cross-validation is utilised to determine the ideal number of concealed nodes. This study makes six attempts at training the hidden layer with 1, 3, 5, 7, 9, and 11 units. Nine is found to be the optimal number through iteration since it gives the neural network the least Root Mean Square Error (RMSE).

This section presents the experimental results of proposed IoT and cloud computing based smart landslide detection and monitoring systems. The experiments are divided into five subparts that are illustrated below: -

Performance Parameters
Result of Proposed Smart Landslide Detection and Monitoring System Using Threshold Value.
Evolution of Alert Generation
Result of Proposed Risk Prediction Model.

5.1. Performance Parameters

A) The amount of water within the soil is remarked as soil wetness. A soil wetness detector, that consists of 2 conducting probes that perform as a groundwork, is also accustomed to living this. The variation in resistance between the 2 conducting plates is also accustomed to verifying the wetness content of the soil. The amount of wetness within the soil has an associate degree inverse relationship with the resistance between the 2 conducting plates. The ADC is employed to method the analog output of the soil wetness detector. On the serial monitor, the wetness content is shown as a proportion. The soil wetness sensor's output varies across the ADC price from zero to 1023. Exploitation the strategy below, this might be expressed as a wetness price in percentages.

$$AnalogOutput=\frac{ADCValue}{1023}$$

Percentage of Moisture = 100 – (Analog output * 100)

When we execute the sketch, we should see something similar to the following in the serial monitor:

When the soil was all wet (400) once the soil was fully dry (850)

The software package below establishes the subsequent ranges to assess the soil standing supported the standardization values.

the quantity five hundred is overly dampish (alert condition) 500 to 750 (to be closely monitored)

It's dry enough at 750.

When the manufactory standardization formulae square measure is applied, the sensor's accuracy ranges between 0. 03 cm³ and 0.04 cm³. The exactitude of soil-specific standardization, on the other hand, may improve greatly, reaching 0.01 cm³ or maybe higher.

B) Accelerometers are unit devices that measure acceleration that is usually made by the movement. The sole acceleration the measuring system detects after they are standing unmoving is thanks to gravity pressing down on that. The quantity of acceleration that the sensor's signal necessities will sustain, typically expressed in g. {this is often this will be} the most quantity of acceleration that the part can properly associate degreed categorical as an output. For America on planet Earth, one gee equals nine.8 m/s2, but this varies somewhat with elevation (and are completely different/special/a unique/a distinct} worth on different planets thanks to variations in attraction pull). Accelerometers are accustomed to observe vibrations in systems or to help with orientation.

The measuring device measures acceleration:

Positive values

It indicates a rise in speed.

Negative values

It indicates a decrease in speed.

Zero values

It indicates constant speed (which may not be zero).

C) The distinction between the output price obtained by the inaudible sensing element and therefore the actual target distance is thought of as accuracy/absolute accuracy. Absolute accuracies of 1 Chronicle to three are realistic in industrial applications for inaudible sensors operating within the temperature vary of -25° C to + 70° C. The foremost usually utilized vary for inaudible sensing is forty to seventy kilocycle per second. Vary and backbone is determined by frequency; the lower the frequency, the bigger the sensing varies. The measuring resolution is one metric linear unit (cm) at fifty-eight kilocycles per second, a widely used frequency, and therefore the range is up to eleven meters. Inaudible sensors will observe the space between a range of objects, freelance of their type, color, or texture. They will conjointly confirm if AN item is returning or receding.

The performance of the proposed risk prediction Model is evaluation using precision, recall and F1-value Kukkar et al. (2022); Singh et al. (2021). Precision (P) measures how precise models are at detecting landslides, while Recall (R) indicates the number of correctly detected landslide samples. The F1 measure is a combination of the Precision and Recall measures. The metrics are derived from equations (10) through (12), and the accuracy evaluation results are shown in Table 2.

$$P=\frac{TP}{TP+FP}$$

$$R=\frac{TP}{TP+FN}$$

$$F1-value=2*\frac{P*R}{P+R}$$

Where, TP, FP, FN, and TN represent the True Positive, False Positive, False Negative and True Negative respectively.

5.2. Result of Proposed Smart Landslide Detection and Monitoring System Using Threshold Value.

The landslides occurring due to rainfall are the major natural disasters in India and even across the globe. Despite having various technologies for landslide monitoring, we are still lacking in the matter of real-time monitoring in geophysical research. This system helps to warn the natives of the particular region and also tells them risk assessments about the landslides that occurred. Furthermore, using a WIFI Module in the Arduino controller, the data is sent to the Thingspeak channel. Therefore, whenever a vibration/movement occurs at a landslide-prone area, the sensors detect them and send all the frequencies to a Thingspeak channel which acts as a data storing mediator and stores all the data in the form of an excel sheet which further represents all this data in the form of graphs. On gathering data through multiple readings, the average accuracy rate of the system is found between 80–85%. Figures 9–10 show the graphs for landslide data reading.

5.3. Evolution of Alert Generation

Through the GSM module, data is transmitted to the cloud server database where it is registered against a station-id unique to each station. The soil moisture, force, and movement readings are given varying threshold levels based on the movement observed and the site's geological characteristics. The task for alert generating begins after the threshold is pre-defined. By comparing the sensor readings with the preset threshold values, warnings are created. Alerts are generated in this system at two different levels: local and global. To alert the local populace and traffic, blinkers and hooters are placed on the scene. At the global level, SMS notifications are delivered to various registered cell numbers together with the position of the movement through GSM Module as shown in Fig. 11. Additionally, this technology allows users to pre-register mobile phone numbers to get SMS warnings. After the prediction analysis using dynamic and static dataset, the alert SMS are also sent to concerned authority for medical emergency/PWD department/ District administration.

5.4. Result of Proposed Risk Prediction Model

Between May 2021 and June 2021, data is collected from the deployed system over a period of sixteen days. For the objectives of predictive analysis, the generated dataset is used as shown in Table 2. 40% of the data is used to test proposed and existing machine learning prediction models, while 60% of the data is used to train the models. The aim of the proposed risk prediction model to predict the movement in test data by learning the pattern of movement in training data. The recall, precision and F1-value is used to evaluate the performance. The proposed ensemble model is compared with existing models such as WRF, EGB, ANN, Auto-Regression (AR), Logic Regression (LR), Support Vector Machine (SVM) and Sequential Minimal Optimization (SMO). The comparison between the new model and the existing model assessment results were shown in Table 3 and Fig. 12. The findings demonstrate that, in comparison to other models, the proposed model has a higher F1-value rate. The average recall, accuracy, and precision of the proposed model with ensemble classifier is 98.67%, 96.56%, 96.07% respectively. By achieving higher values, the proposed system minimizes error rate. Furthermore, the proposed ensemble classifier generates less prediction errors due to its overall better F1- value of 97.35%. As a result, it is safe to conclude that the level of performance derived from the aforementioned parameters justifies the use of ensemble learning in the proposed risk prediction model for IOT-based smart landslide detection and monitoring systems.

Table 3

Performance Comparison of Proposed Ensemble Model and Existing Models
Models	Precision	Recall	F1-value	Accuracy
WRF	90.07	92.17	91.11	90.23
EGB	87.90	86.34	87.11	88.45
ANN	94.89	96.17	95.53	94.78
AR	78.67	80.96	79.80	80.55
SMO	81.33	82.89	82.10	81.56
LR	85.86	87.78	86.81	86.23
SVM	86.56	82.30	84.38	85.56
Proposed Ensemble Model	96.07	98.67	97.35	96.54

In this study, a low-cost IoT platform for lab use is suggested for monitoring landslides. This system measures soil moisture and soil movement using MEMS-based sensors. The experiment-derived sensor threshold values are compared to real-time sensor data, and when the sensor readings go beyond the critical threshold levels, SMS notifications are issued to the registered mobile phones. Landslides brought on by rainfall are now the most catastrophic natural catastrophes occurring, and we are yet unable to fully excel in their real-time monitoring. The data gathered from a certain area is examined using Matlab software. The probability of landslides and the pattern in variation of climate change can be observed through proposed risk prediction models based on ensemble learning. Through this, we can embark on a better change in landslide prediction technologies. This network can be used in the future for the advancement of wireless sensor networks in real-time situations and also in landslide prediction technologies. We further hope that the proposed approach will widen the upcoming capabilities of wireless sensors projects in other emergencies or other real-time projects. So many methods can be used for landslide detection using various algorithms like Distributed algorithms, geophone networks, geodetic monitoring, etc. Therefore, we aim at building a complete hardware and software part for this project and want to extend this project for building the hardware system and making it available for receivers to update their location information and the data for most prone landslide occurrence areas.

Competing interests

There are no competing interests, according to all the authors.

Ethical approval

Each author has reviewed and approved the publication of this work into ESPR.

Consent to participate

Animals were not employed in this investigation; instead, real-time and stored data were used.

Consent to publish

The authors are fully aware and certain that the data utilised in this work has never been published.

Funding

Not applicable

Authors’ Contribution

AshimaKukkar was responsible for the study's idea and design; RajiniMohana collected the data; Aman Sharma, Varun Chodha, and Pranjal Bansal analysed and interpreted the findings; and Y. Author and Z. Author prepared the draught text. The final copy of the paper was approved by all authors after they had evaluated the findings.

Abraham, M. T., Satyam, N., Pradhan, B., &Alamri, A. M. (2020). IoT-based geotechnical monitoring of unstable slopes for landslide early warning in the Darjeeling Himalayas. Sensors, 20(9), 2611.
Adeel, A., Gogate, M., Farooq, S., Ieracitano, C., Dashtipour, K., Larijani, H., & Hussain, A. (2019). A survey on the role of wireless sensor networks and IoT in disaster management. Geological disaster monitoring based on sensor networks, 57-66.
Akyildiz, I. F., Su, W., Sankarasubramaniam, Y., &Cayirci, E. (2002). Wireless sensor networks: a survey. Computer networks, 38(4), 393-422.
Bjorklund, M. (2017). Automatic spot detection in large-scale fluorescence microscopy image dataset.
Chang, Z., Catani, F., Huang, F., Liu, G., Meena, S. R., Huang, J., & Zhou, C. (2023). Landslide susceptibility prediction using slope unit-based machine learning models considering the heterogeneity of conditioning factors. Journal of Rock Mechanics and Geotechnical Engineering, 15(5), 1127-1143.
Cheng, G., Guo, L., Zhao, T., Han, J., Li, H., & Fang, J. (2013). Automatic landslide detection from remote-sensing imagery using a scene classification method based on BoVW and pLSA. International Journal of Remote Sensing, 34(1), 45-59.
Đurišić, M. P., Tafa, Z., Dimić, G., &Milutinović, V. (2012, June). A survey of military applications of wireless sensor networks. In 2012 Mediterranean conference on embedded computing (MECO) (pp. 196-199). IEEE.
Furtado, H., &Trobec, R. (2011, May). Applications of wireless sensors in medicine. In 2011 Proceedings of the 34th International Convention MIPRO (pp. 257-261). IEEE.
Gui, Y., Tao, Z. G., Wang, C. J., &Xie, X. (2011). Study on remote monitoring system for landslide hazard based on wireless sensor network and its application. Journal of Coal Science and Engineering (China), 17(4), 464-468.
Harsa, H., Anistia, M. H., Mulsandi, A., Suprihadi, B., Kurniawan, R., Habibie, M. N., ... &Praja, A. S. (2023). Machine learning and artificial intelligence models development in rainfall-induced landslide prediction. IAES International Journal of Artificial Intelligence, 12(1), 262.
Jeong, S., Ko, J., & Kim, J. (2019). The effectiveness of a wireless sensor network system for landslide monitoring. IEEE Access, 8, 8073-8086.
Kukkar, A., Gupta, D., Beram, S. M., Soni, M., Singh, N. K., Sharma, A., ... &Rizwan, A. (2022). Optimizing deep learning model parameters using socially implemented IoMT systems for diabetic retinopathy classification problem. IEEE Transactions on Computational Social Systems.
Kumar, A., Asthana, A. K. L., Priyanka, R. S., Jayangondaperumal, R., Gupta, A. K., &Bhakuni, S. S. (2017). Assessment of landslide hazards induced by extreme rainfall event in Jammu and Kashmir Himalaya, northwest India. Geomorphology, 284, 72-87.
Kumar, A., Sharma, R. K., & Bansal, V. K. (2022). Spatial Prediction of Landslide Hazard using GIS-multi-criteria Decision Analysis in Kullu District of Himachal Pradesh, India. Journal of Mining and Environment, 13(4), 943-956.
Lee, S., Ryu, J. H., Min, K., & Won, J. S. (2003). Landslide susceptibility analysis using GIS and artificial neural network. Earth Surface Processes and Landforms: The Journal of the British Geomorphological Research Group, 28(12), 1361-1376.
Liu, Z., Tsuda, T., Watanabe, H., Ryuo, S., &Iwasawa, N. (2019). Data driven cyber-physical system for landslide detection. Mobile Networks and Applications, 24, 991-1002.
McHugh, E. L. (1900). Video motion detection for real-time hazard warnings in surface mines.
McHugh, E. L., & Girard, J. M. (2002). Evaluating techniques for monitoring rock falls and slope stability.
McKean, J., &Roering, J. (2004). Objective landslide detection and surface morphology mapping using high-resolution airborne laser altimetry. Geomorphology, 57(3-4), 331-351.
Mezaal, M. R., Pradhan, B., Shafri, H. Z. M., &Yusoff, Z. M. (2017). Automatic landslide detection using Dempster–Shafer theory from LiDAR-derived data and orthophotos. Geomatics, natural hazards and risk, 8(2), 1935-1954.
Mohan, A., Singh, A. K., Kumar, B., &Dwivedi, R. (2021). Review on remote sensing methods for landslide detection using machine and deep learning. Transactions on Emerging Telecommunications Technologies, 32(7), e3998.
Mottola, L., &Picco, G. P. (2006). Logical neighborhoods: A programming abstraction for wireless sensor networks. In Distributed Computing in Sensor Systems: Second IEEE International Conference, DCOSS 2006, San Francisco, CA, USA, June 18-20, 2006. Proceedings 2 (pp. 150-168). Springer Berlin Heidelberg.
Pathania, A., Kumar, P., Sihag, P., Chaturvedi, P., Singh, R., Uday, K. V., &Dutt, V. (2020). A lowcost, sub-surface iot framework for landslide monitoring, warning, and prediction. In Proceedings of 2020 International conference on advances in computing, communication, embedded and secure systems.
Prakasam, C., Aravinth, R., &Saravanan, R. (2022). Report on NRDMS: DST Sponsored 21-day (Level 2) Summer School on Mountain Disaster Management: Landslide from 9 to 29th May 2019. National Academy Science Letters, 45(1), 95-103.
Ramesh, M. V. (2009, June). Real-time wireless sensor network for landslide detection. In 2009 Third International Conference on Sensor Technologies and Applications (pp. 405-409). IEEE.
Ramesh, M.V, &Vasudevan, N. (2012). The deployment of deep-earth sensor probes for landslide detection. Landslides, 9, 457-474.
Rau, J. Y., Jhan, J. P., Lo, C. F., & Lin, Y. S. (2011). Landslide mapping using imagery acquired by a fixed-wing UAV. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci, 38(1/C22), 195-200.
Reddy, K., Reddy, M. A., Kaur, V., & Kaur, G. (2022). Career guidance system using ensemble learning. Akhila and Kaur, Veerpal and Kaur, Gagandeep, Career Guidance System Using Ensemble Learning (July 14, 2022).
Rosin, P. L., &Hervás, J. (2002). Image thresholding for landslide detection by genetic programming. In Analysis of multi-temporal remote sensing images (pp. 67-74).
Sameen, M. I., & Pradhan, B. (2019). Landslide detection using residual networks and the fusion of spectral and topographic information. IEEE Access, 7, 114363-114373.
Sarangi, P. K., Sahoo, A. K., Kaur, G., Nayak, S. R., &Bhoi, A. K. (2022). Gurmukhi numerals recognition using ann. In Cognitive Informatics and Soft Computing: Proceeding of CISC 2021 (pp. 377-386). Singapore: Springer Nature Singapore.
Singh, S., Ramkumar, K. R., &Kukkar, A. (2021, October). Machine Learning Techniques and Implementation of Different ML Algorithms. In 2021 2nd Global Conference for Advancement in Technology (GCAT) (pp. 1-6). IEEE.
Sruthy, M. R., Anjana, R., Archana, R., Dhanya, V., &Hridya, A. H. (2020). IoT based landslide detection and monitoring system. Int J Res EngSciManag, 596-9.
Suryadevara, N. K., Mukhopadhyay, S. C., Kelly, S. D. T., & Gill, S. P. S. (2014). WSN-based smart sensors and actuator for power management in intelligent buildings. IEEE/ASME transactions on mechatronics, 20(2), 564-571.
Susanto, E., Budiman, F., Mukhtar, D. P. H., &Latief, M. H. (2019, November). Slope, humidity and vibration sensors performance for landslide monitoring system. In 2019 IEEE Asia Pacific Conference on Wireless and Mobile (APWiMob) (pp. 139-142). IEEE.
What are landslides and how do they occur? (n.d.) https://people.uwec.edu/jolhm/eh2/rogge/causes.htm
What is a landslide and what causes one? | U.S. Geological Survey. (2022, September 28). https://www.usgs.gov/faqs/what-a-landslide-and-what-causes-one?qt-news_science_products=0#qt-news_science_products

Download PDF

Journal Publication

published 16 Nov, 2023

Read the published version in Environmental Science and Pollution Research →

Editorial decision: Major Revision
31 Aug, 2023
Reviewers agreed at journal
05 Aug, 2023
Reviewers invited by journal
04 Aug, 2023
Editor assigned by journal
27 Jul, 2023
First submitted to journal
23 Jul, 2023

You are reading this latest preprint version

An Ensemble Learning Based Framework for Smart Landslide Detection, Monitoring, Prediction and Warning in IoT-Cloud Environment

Status:

Journal Publication

Version 1

Abstract

Figures

Highlights

1. Introduction

1.1. Landslide Susceptibility Zonation

1.2. Our Contributions

2. Related Work

3. Preliminaries and Background

3.1. Sensor Network Design and Implementation

4. Proposed Smart Landslide Detection and Monitoring System

4.1. System Architecture

4.2. System’s Logic

4.3. Methodology

4.3.1. Phase 1: Component description

4.3.2. Phase 2: Experimental Setup and Circuit Diagram

4.3.3. Proposed Risk Prediction Model

5. Experimental Evaluation and Results

5.1. Performance Parameters

5.2. Result of Proposed Smart Landslide Detection and Monitoring System Using Threshold Value.

5.3. Evolution of Alert Generation

5.4. Result of Proposed Risk Prediction Model

6. Conclusion and Future Scope

Declarations

References

Status:

Journal Publication

Version 1