Nothing Special   »   [go: up one dir, main page]
Next Article in Journal
A Prediction Method for City Traffic Noise Based on Traffic Simulation under a Mixed Distribution Probability
Previous Article in Journal
Optimal Machine Learning Model to Predict Demolition Waste Generation for a Circular Economy
Previous Article in Special Issue
Automatic and Efficient Detection of Loess Landslides Based on Deep Learning
You seem to have javascript disabled. Please note that many of the page functionalities won't work as expected without javascript enabled.
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Sustainability
Volume 16
Issue 16
10.3390/su16167063
sustainability-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Landslide Assessment Classification Using Deep Neural Networks Based on Climate and Geospatial Data

by
Yadviga Tynchenko
1,2,
Vladislav Kukartsev
1,3,
Vadim Tynchenko
1,4,*,
Oksana Kukartseva
1,2,
Tatyana Panfilova
1,5,
Alexey Gladkov
1,
Van Nguyen
6 and
Ivan Malashin
1,*
1
Artificial Intelligence Technology Scientific and Education Center, Bauman Moscow State Technical University, 105005 Moscow, Russia
2
Laboratory of Biofuel Compositions, Siberian Federal University, 660041 Krasnoyarsk, Russia
3
Department of Information Economic Systems, Reshetnev Siberian State University of Science and Technology, 660037 Krasnoyarsk, Russia
4
Information-Control Systems Department, Reshetnev Siberian State University of Science and Technology, 660037 Krasnoyarsk, Russia
5
Department of Technological Machines and Equipment of Oil and Gas Complex, Siberian Federal University, 660041 Krasnoyarsk, Russia
6
Institute of Energy and Mining Mechanical Engineering—Vinacomin, Hanoi 100000, Vietnam
*
Authors to whom correspondence should be addressed.
Sustainability 2024, 16(16), 7063; https://doi.org/10.3390/su16167063
Submission received: 27 June 2024 / Revised: 13 August 2024 / Accepted: 15 August 2024 / Published: 17 August 2024
(This article belongs to the Special Issue Risk Assessment of Landslides Based on Multi-source Data and Machine Learning)
Browse Figures
Figure 1
<p>Overall technical framework for landslide risk assessment.</p> "> Figure 2
<p>Distribution of landslide events by countries.</p> "> Figure 3
<p>Experimental Pipeline.</p> "> Figure 4
<p>Evolution of metrics for landslide size and trigger predictions across individuals.</p> "> Figure 5
<p>Confusion matrix for landslide size classification.</p> "> Figure 6
<p>Confusion matrix for landslide trigger classification.</p> "> Figure 7
<p>Map of landslide risk assessment on Viti Levu Island, Fiji. Red indicates areas with a high probability of landslides according to the model, while green represents areas with low risk.</p> "> Figure 8
<p>Research structure overview.</p> "> Figure A1
<p>Island landslides maps: (<b>a</b>) England and Ireland; (<b>b</b>) Philippines; (<b>c</b>) Hawaii; (<b>d</b>) Fiji.</p> "> Figure A2
<p>Landslides maps: (<b>a</b>) Caribbean Basin; (<b>b</b>) Indonesia and Malaysia; (<b>c</b>) East Australia and New Zealand; (<b>d</b>) Korea and Japan; (<b>e</b>) Columbia and Ecuador; (<b>f</b>) Rio de Janeiro.</p> "> Figure A3
<p>Landslides maps: (<b>a</b>) Alps and Balkans; (<b>b</b>) Caucasus; (<b>c</b>) Lake Victoria Region; (<b>d</b>) West Africa.</p> "> Figure A4
<p>Landslides maps (<b>a</b>) Himalaya, (<b>b</b>) China, (<b>c</b>) Western Ghats, (<b>d</b>) South-East Asia.</p> "> Figure A5
<p>USA landslides maps: (<b>a</b>) Washington; (<b>b</b>) California; (<b>c</b>) the North West states; (<b>d</b>) Utah and Colorado.</p> ">
Review Reports Versions Notes

Abstract

:
This study presents a method for classifying landslide triggers and sizes using climate and geospatial data. The landslide data were sourced from the Global Landslide Catalog (GLC), which identifies rainfall-triggered landslide events globally, regardless of size, impact, or location. Compiled from 2007 to 2018 at NASA Goddard Space Flight Center, the GLC includes various mass movements triggered by rainfall and other events. Climatic data for the 10 years preceding each landslide event, including variables such as rainfall amounts, humidity, pressure, and temperature, were integrated with the landslide data. This dataset was then used to classify landslide triggers and sizes using deep neural networks (DNNs) optimized through genetic algorithm (GA)-driven hyperparameter tuning. The optimized DNN models achieved accuracies of 0.67 and 0.82, respectively, in multiclass classification tasks. This research demonstrates the effectiveness of GA to enhance landslide disaster risk management.

1. Introduction

Landslides are natural hazards that pose risks to human lives and infrastructure worldwide [1,2,3]. Effective disaster management and risk reduction strategies are based on the factors that contribute to landslides. Key climate variables, such as precipitation, temperature, and soil moisture, affect the triggering of landslides, making them essential considerations for predictive models.
Recent advancements in data science, particularly in machine learning (ML) and geospatial analysis, have improved the accuracy of landslide predictions [4,5,6]. By harnessing datasets that integrate climatic and geographical information, researchers are able to develop more precise and reliable models for forecasting and classifying landslides. This progress represents a major focus of current scientific research.
Deep learning (DL), a specialized branch of ML, has impacted various fields, including remote sensing, by enabling the automatic learning and extraction of features from large datasets. In particular, DL techniques have driven advancements in applications such as image classification, semantic segmentation, and the extraction of landslide information [7] from remote sensing imagery. These methods capitalize on the capacity of DL to process and analyze complex visual data, resulting in enhanced accuracy and efficiency in tasks related to detecting and mapping natural hazards.
Given the predictive performance of DL compared to conventional methods, Tehrani et al. [8] provide a review of ML techniques and their effectiveness in landslide detection and mapping and both spatial and temporal forecasting. Their review also identifies key areas for future research to further advance the field. While traditional methods have been widely reviewed, there has been a lack of comprehensive evaluations of modern ML models for landslide susceptibility mapping. Merghadi et al. [9] address this gap by reviewing prominent ML techniques, their operational frameworks, and performance metrics. Their case study from Algeria demonstrates that tree-based ensemble algorithms, such as random forest, can deliver robust performance with minimal preprocessing, thereby enhancing the accuracy of landslide susceptibility mapping.
By building on these advancements, recent studies have further explored the potential of various ML approaches to enhance landslide detection and mapping. Wang et al. [10] developed a method that integrates multiple databases and evaluates algorithms such as logistic regression, SVM, random forest, boosting methods, and convolutional neural networks (CNNs). Their research in Lantau, Hong Kong, demonstrated that CNNs, with their robust feature extraction capabilities, achieved an impressive 92.5% accuracy, surpassing the performance of other methods. Similarly, Ghorbanzadeh et al. [11] investigated different CNN architectures for landslide detection in Nepal, finding that CNNs with small window sizes using spectral data delivered the best results. These studies highlight the growing effectiveness of CNNs in the domain of landslide detection, particularly due to their ability to process complex visual data and capture subtle patterns critical for accurate mapping.
Despite advancements in ML, accurately predicting landslide occurrences remains challenging due to limited real-time data on soil and rock conditions. Korup et al. [12] emphasize the need for simplified models that offer effective predictions despite data quality issues. Future research should focus on refining model selection, improving the predictions of large-scale slope failures, and enhancing forecasting methods to better prepare for landslides.
Several studies compare traditional and modern ML techniques for landslide susceptibility mapping. Marjanovic et al. [13] compare SVM, decision trees, and logistic regression, finding SVM to be particularly effective. Goetz et al. [14] evaluate various models, including random forests and SVM, highlighting the strengths of tree-based methods and the need for careful model selection based on geological and morphological contexts. Micheletti et al. [15] assess SVM, random forests, and AdaBoost, noting that random forests and AdaBoost are more efficient in feature selection for landslide susceptibility mapping.
Hazard engineering underscores the need for innovative approaches in both seismic design and landslide detection. Esteghamati et al. [16] address these challenges by integrating advanced ML techniques for comprehensive hazard assessment. The framework presented enhances seismic design by leveraging ML to identify optimal structural systems and DL for precise seismic risk estimation, as demonstrated in a case study for multistory buildings in Charleston, SC. Simultaneously, Kikuchi et al. [17] applied CNN models to landslide detection, focusing on the Kii Peninsula in Japan. By analyzing 38 landslide and 63 nonlandslide points with a 1 m DEM, the CNN models achieved impressive results, with an ROC of 96.0% and an accuracy of 88.7%, improving over initial models and aligning with local terrain features. These advancements illustrate the potential of DNN in enhancing hazard engineering and disaster risk management.
Table 1 summarizes the key findings from these studies, highlighting the applied methods, results, and limitations encountered in each investigation.
This study proposes an approach that leverages a combination of deep neural networks (DNNs) and genetic algorithms (GAs) for landslide classification. The model’s parameters are optimized through GA, with accuracy maximization as the objective function. While previous research in landslide prediction has predominantly relied on traditional ML techniques, the approach presented here is particularly adept at identifying complex patterns and interactions within climate and geospatial data, enhancing landslide risk assessment. The application of GA to hyperparameter tuning further refines the model’s configuration, ensuring it is precisely tailored to the specific characteristics of the dataset.
This enhanced classification accuracy directly benefits early warning systems and risk management strategies by leveraging comprehensive climate data. By providing more precise predictions based on climatic variables, our approach enables the earlier detection of potential landslide events and more effective risk mitigation measures. The practical implications of this research offer actionable insights for policymakers and disaster management agencies. The effective use of advanced predictive models based on comprehensive climate and geospatial data enables better risk assessment and facilitates timely intervention in vulnerable regions, thereby mitigating potential damage and enhancing disaster preparedness.

2. Materials and Methods

In order to provide an overview of the proposed methodology, the workflow was divided into three main components: Data collection, data processing and integration, and model development and optimization. Each of these step is illustrated in Figure 1.

2.1. Global Landslide Catalog

The Global Landslide Catalog (GLC) [20] was developed to identify rainfall-triggered landslide events globally without regard for their size, impacts, or location. This catalog includes all types of mass movements triggered by rainfall, as well as some triggered by other events. The data have been compiled since 2007 at NASA Goddard Space Flight Center and are uniquely identified with the “GLC” tag in the landslide editor. Landslides, along with their descriptions, are detailed in Appendix A.
The dataset contains detailed information about each landslide event, including the name and URL of the news entity reporting the event, a unique event identifier, and the date and time of occurrence. The event title and description provide a narrative of what happened, and the location description and accuracy detail the event’s geographic context (Figure 2).
Further, the dataset categorizes the landslide, identifies its trigger and size, and describes the environment in which it occurred. It also includes the number of fatalities and injuries, the name of any associated storm, and a link to a photo of the event. Additional notes offer further context or information.
Administrative details include the country and its ISO code, the name and population of the administrative division, and proximity information to the nearest gazetteer point. Metadata about the data entry process, such as the event import source and ID, timestamps for submission, creation, and last edit, are also provided. The geographical co-ordinates (longitude and latitude) pinpoint the exact location of each event.

2.2. Obtaining Climate Data

In order to improve the understanding of environmental factors influencing landslide susceptibility, climate data were incorporated using geographic co-ordinates. The NASA POWER (Prediction of Worldwide Energy Resource) [21] dataset was utilized to obtain daily weather data, including temperature, precipitation, pressure, and humidity. A Python (version 3.12) script was employed to automate the extraction of this climate data by accepting latitude and longitude co-ordinates, along with the start and end dates for the desired data range. A request is sent to the NASA POWER API, specifying the required meteorological parameters—temperature at 2 m (T2M), precipitation corrected for gauge undercatch (PRECTOTCORR), surface pressure (PS), and specific humidity at 2 m (QV2M). The API response is then parsed to extract relevant weather information for each day within the specified date range. Since GLS contains data on landslides from 2008 to 2018, data from 1998 to 2018 were retrieved for each location. The retrieved data are structured into a list of dictionaries, each containing the latitude, longitude, date, temperature, precipitation, pressure, and humidity for a given day. This format facilitates further integration and analysis within the broader study of landslide susceptibility. The entire experimental pipeline is illustrated in Figure 3.

2.3. Classification Methodology

This study explores the application of GA to optimize DNN architectures for landslide classification assessment. By leveraging the evolutionary principles of GA, the aim was to discover the most effective DNN architectures for accurately predicting plant species. The choice of using DNN combined with GA in this research is driven by the distinct advantages these methods offer for complex predictive tasks such as landslide classification.
DNNs are highly effective at capturing intricate patterns and relationships within large and diverse datasets, such as the climatic and geospatial data used in this study. DNNs can model nonlinear interactions between multiple variables, making them particularly well-suited for understanding the multifaceted nature of landslide triggers and sizes. GAs, on the other hand, are powerful optimization tools that mimic the process of natural selection to find optimal solutions in high-dimensional search spaces. When applied to hyperparameter tuning, GAs systematically explore various model configurations to identify the most effective combination of parameters. This approach is beneficial for enhancing the performance of DNNs, as it ensures that the models are not only accurate but also efficiently configured for the specific characteristics of the dataset. By integrating DNNs with GA-driven hyperparameter optimization, this research leverages the strengths of both methods.
The collected climate data, including temperature, humidity, pressure, and precipitation over a 10-year period, were recorded monthly for each landslide event. This resulted in a dataset with 480 columns—120 for each feature across the 10-year span. Additionally, the dataset contained information about the class of the landslide, as similar types of landslides often recur in the same locations.
Our objective was to predict and classify two key aspects: landslide size and landslide trigger. Accurately predicting the size and trigger of landslides helps in understanding the potential impact and initiating timely preventive measures. The size of a landslide indicates the scale of the disaster, while the trigger provides insight into the immediate cause, such as heavy rainfall, earthquakes, or human activities, enabling better risk management and mitigation strategies.
Given the extensive number of climate features, the dataset was subjected to principal component analysis [22] (PCA) to reduce its dimensionality to 20 components. This dimensionality reduction was necessary because it lowers the computational burden, making the model training process faster and more efficient. Additionally, high-dimensional datasets can lead to overfitting, where the model learns noise and outliers instead of the underlying patterns. PCA helps mitigate this by retaining only the most significant features. By focusing on the principal components that capture the most variance, PCA improves the model’s ability to generalize from the training data to unseen data. Fewer components make it easier to understand and visualize the relationships and patterns within the data, aiding in more effective analysis and decision-making. Despite the transformation of the original features into principal components, the DNN models remain effective in deriving predictions. The principal components are linear combinations of the original features and encapsulate information needed for making forecasts. During both training and prediction, the DNN operates within this reduced feature space, allowing it to learn and predict based on the most informative aspects of the data.
In order to initiate the optimization process, the numerical features were standardized with the dataset to ensure uniform scaling. Then, an extensive hyperparameter search space was defined, encompassing a wide range of DNN architecture [23] configurations. This search space included the number of layers (ranging from 1 to 20), neurons per layer (ranging from 1 to 128), and various activation functions such as relu, sigmoid, tanh, softmax, softplus, softsign, elu, selu, gelu, hard sigmoid, and linear. Additionally, the search space included multiple optimizers, such as adam, sgd, rmsprop, adagrad, adadelta, adamax, and nadam, as well as different learning rates (0.0001, 0.001, 0.01, and 0.1). For loss functions, we focused on categorical cross-entropy.
In order to describe the hyperparameter search space formally, we defined each hyperparameter space individually and then expressed the overall search space as their Cartesian product.
  • Number of layers:
    L = { l ∣ l ∈ Z , 1 ≤ l ≤ 20 }
  • Neurons per layer:
    N = { n ∣ n ∈ Z , 1 ≤ n ≤ 128 }
  • Activation functions:
    A = { relu , sigmoid , tan h , softmax , softplus , softsign , elu , selu , gelu , hard sigmoid , linear }
  • Optimizers:
    O = { adam , sgd , rmsprop , adagrad , adadelta , adamax , nadam }
  • Learning rates:
    R = { 0.0001 , 0.001 , 0.01 , 0.1 }
  • Loss functions:
    F = { categorical crossentropy }
    The overall hyperparameter search space H is given by the Cartesian product of these individual spaces:
    H = L × N × A × O × R × F
    Here, each element of H is a tuple of the form ( l , n , a , o , r , f ) , where
    • l is the number of layers chosen from L ,
    • n is the number of neurons per layer chosen from N ,
    • a is the activation function chosen from A ,
    • o is the optimizer chosen from O ,
    • r is the learning rate chosen from R ,
    • f is the loss function chosen from F .
So, for each DNN layer, the total number of possible combinations is n · a = 128 · 11 = 1408 . The total number of combinations could be calculated as follows:
o · r · f · ∑ l = 1 L ( n · a ) l = 1 · 4 · 7 · ∑ l = 1 20 1408 l ≈ 2.64 × 10 52
This is an extremely large number, illustrating the vast potential variety in neural network designs with these parameters. Given that each architecture took approximately 1 min to evaluate, exploring the entire space of possible architectures would require an impractically long time. Specifically, with the vast number of potential configurations, the total computation time would be prohibitive, underscoring the need for efficient optimization strategies.
By integrating GA from the DEAP library with TensorFlow and Scikit-learn libraries, an iterative optimization framework was developed that evaluated and refined DNN models based on their predictive accuracy. The optimization process was carried out on a computer with an Intel Core i5-6600K CPU operating at 3.50 GHz and 8.00 GB of RAM. Despite the high computational cost associated with GA, the benefits of this application are considerable. GA offers a powerful method for optimizing hyperparameters across an extensive search space of potential DNN architectures. This approach avoids the need for exhaustive manual tuning by employing iterative refinement to discover optimal model configurations. The GA-driven optimization process involved initializing a diverse population of architectures, evaluating their performance, selecting the fittest models as parents, and generating offspring through crossover and mutation operations, which could be itemized as follows:
  • Initialize a population P ( t ) of N candidate solutions (DNN architectures) at generation t = 0 :
    P ( 0 ) = { A 1 , A 2 , … , A N }
    where A i represents the architecture of the i-th individual in the population.
  • Evaluate the fitness of each individual A i in the population using the classification accuracy A i on a validation set:
    f ( A i ) = Accuracy ( A i )
  • Select a subset of individuals based on their fitness scores f ( A i ) to act as parents for the next generation. This can be carried out using methods such as roulette wheel selection or tournament selection:
    P parents ( t ) = Select ( P ( t ) , f ( A i ) )
  • Apply crossover operations to pairs of parent architectures to produce offspring architectures. For instance, if A p and A q are two parents, the offspring A c 1 and A c 2 can be generated as
    A c 1 , A c 2 = Crossover ( A p , A q )
  • Apply mutation operations to the offspring architectures to introduce genetic diversity. For an architecture A c , a mutation might alter one or more hyperparameters:
    A m = Mutation ( A c )
  • Evaluate the fitness of the offspring architectures:
    f ( A m ) = Accuracy ( A m )
  • Replace the least fit individuals in the population with the new offspring, potentially incorporating elitism to retain the best solutions:
    P ( t + 1 ) = Replace ( P ( t ) , A m )
  • Repeat the evaluation, selection, crossover, mutation, and replacement steps for a predefined number of generations G or until convergence criteria are met:
    For t = 1 to G do :
    P ( t ) ← Update Population
    End For
The objective function of the GA is to maximize the classification accuracy of the DNN models, which can be expressed as
max A i ∈ P ( t ) Accuracy ( A i )
The overall optimization problem can, thus, be formulated as
Find A opt such that A opt = arg max A i ∈ P ( G ) Accuracy ( A i )
Here, A opt represents the architecture with the highest accuracy achieved after G generations of optimization.
The dataset was partitioned into training and testing subsets, with 80% allocated for training and 20% for testing. Throughout the optimization process, we monitored the evolution of accuracy scores across generations, gaining valuable insights into the complex interplay between architecture choices and predictive performance. By recording the optimized architecture parameters and corresponding accuracies, we enabled a detailed analysis and comparison of model behaviors under different optimization strategies.

3. Results

3.1. DNN Results

The optimization process yielded a sorted accuracy growth curve, illustrating the performance of various DNN architectures over successive generations. Initially, many neural networks performed poorly, demonstrating similar low accuracy levels. However, as the GA iteratively refined the architectures, a notable improvement in classification accuracy was observed.
Specifically, the accuracy for predicting the size of the landslide based on climatic data reached a peak of 0.82. This indicates a high level of precision in identifying the potential magnitude of landslides. The accuracy achieved in predicting the cause of the landslide was 0.67. Although lower than the accuracy for size prediction, this still represents an improvement over the initial models. Understanding the triggers of landslides, such as heavy rainfall or seismic activity, may implement effective preventative measures and respond appropriately to landslide threats.
These results (Figure 4) highlight the effectiveness of using a GA to optimize DNN architectures for complex classification tasks. This plot illustrates the progression of key metrics—accuracy, precision, recall, and F1 score—over successive individuals for landslide size and trigger predictions. The curves represent the performance evolution of the predictive models across different generations, with distinct colors and line styles distinguishing between the size and trigger metrics. Vertical shaded regions indicate different generations, providing insights into how the model’s performance evolves over time. The gradual improvement in accuracy shows the importance of a comprehensive hyperparameter search space and iterative refinement in developing robust predictive models for landslide analysis.
Figure 5 illustrates the confusion matrix generated during the classification of landslide sizes based on climatic data. The labels in the confusion matrix correspond to the following categories of landslide size: (0) catastrophic, (1) large, (2) medium, (3) small, (4) unknown, and (5) very large. Each label represents a specific classification used to categorize the size of landslides based on their magnitude and impact. These categories aid in assessing and predicting the potential consequences of landslides in different geographical and environmental contexts.
Figure 6 illustrates the confusion matrix generated during the classification of landslide triggers based on climatic data. The matrix provides a visual representation of the performance of the classification model, where the numbers 0 to 14 correspond to specific trigger labels of landslides based on climatic data: (0) construction, (1) continuous rain, (2) downpour, (3) earthquake, (4) flooding, (5) freeze-thaw, (6) leaking pipe, (7) mining, (8) monsoon, (9) no apparent trigger, (10) other, (11) rain, (12) snowfall/snowmelt, (13) tropical cyclone, and (14) unknown. Each label represents a specific category used in the classification of landslide triggers for assessment and prediction based on environmental conditions.
In Table 2, the optimized configurations of DNN for landslide trigger classification are presented.

3.2. Case Study: Application of the Approach on Viti Levu Island, Fiji

This study presents a method for classifying landslide triggers and sizes using climate and geospatial data. In order to illustrate the effectiveness of this approach, we applied it to analyze landslides on Viti Levu Island in Fiji, where 34 landslides have been documented over an area of 10,388 km2. Below, a description of the approach application is listed:
  • Defining grid parameters: The first step involves setting the parameters for the grid of the study area. This includes specifying the minimum and maximum co-ordinates that define the region of interest. For Viti Levu Island, the boundaries selected were a minimum latitude of −18.5, a maximum latitude of −17.3, a minimum longitude of 177.0, and a maximum longitude of 179.5.
  • Grid division: By using these parameters, the area is divided into a grid with evenly sized sectors. The number of sectors is determined based on the desired division in terms of latitude and longitude, resulting in a grid with specific sector sizes.
  • Filtering out water surfaces: In order to refine the analysis, sectors with a large proportion of water surfaces (such as oceans) are excluded. These sectors are marked as non-informative for further analysis.
  • Integration of climate data: For the remaining sectors, climate data for the 10 years preceding each landslide event are integrated. This includes variables such as rainfall, humidity, pressure, and temperature. These climate data are then reduced using principal component analysis (PCA), which helps to highlight the principal climate components relevant to landslide triggers.
  • Classification and risk assessment: By using geospatial co-ordinates and reduced climate data, the probability of landslide occurrence is assessed. Optimized DNN models, with parameters tuned through GA, are employed to classify the likelihood of landslides. The optimized models achieved accuracies of 0.67 and 0.82 for classifying landslide triggers and sizes, respectively.
The results of the analysis are illustrated in Figure 7, showing a map with a color-coded grid indicating the likelihood of landslides. The map reveals that areas with high landslide risk (marked in red) correspond to known landslide occurrences. However, it also identifies sectors with a high probability of landslides where no actual landslides have been documented. These high-risk sectors are predominantly located in river valleys and inaccessible areas, where the detection of landslides might be challenging.
This approach not only validates existing landslide data but also highlights potential high-risk areas that may require additional monitoring and verification. The results underscore the effectiveness of integrating climate and geospatial data for landslide risk assessment, enhancing early warning systems, and improving disaster risk management strategies.

4. Discussion

This study explored the classification of landslide triggers and sizes based on climatic data using DNN. The results demonstrate the effectiveness of this approach in improving the accuracy of landslide prediction models. The geospatial features in the model were selected based on the co-ordinates of landslide events. Latitude and longitude data were used to map the spatial distribution of landslides and analyze their correlation with environmental factors. This integration allowed the model to account for the geographical context of landslides [24]. For example, Novellino et al. [25] examined the evolution and impact of satellite-based landslide mapping, highlighting an increase in research since 2014 due to Sentinel data, with manual detection methods giving way to artificial intelligence while noting ongoing challenges and future potential with upcoming SAR satellite missions; including these details enhances the model’s spatial accuracy and relevance.
Accurate landslide information extraction influences effective disaster response and risk management, enabling the timely identification and assessment of landslide events despite challenges such as varying scales and obstructive land cover. Xie et al. [26] present a landslide extraction framework that integrates a two-branch multiscale context feature extraction module with self-attention mechanisms and a deeply supervised classifier. Additionally, geological environment information can be used to evaluate vulnerability to landslide debris flows, especially in data-scarce regions such as Yanghe Township. In this context, Guo et al. [27] introduced a method combining SBAS-InSAR and optical remote sensing with NDVI variation to enhance the monitoring and prediction of landslide debris flow. Their approach reveals spatial-temporal patterns and improves vulnerability assessments through the use of multisource remote sensing technology.
The confusion matrix analysis revealed accuracy levels for different triggers, especially rain-confused areas such as rain and downpour, with certain triggers being more accurately predicted than others. Several limitations and challenges should be noted. Gaps or inaccuracies in the landslide and climatic datasets can affect model performance [28]. For instance, missing or incomplete data on certain climatic factors could undermine the reliability of predictions [29]. Additionally, the extensive hyperparameter tuning required in GA can lead to overfitting if not carefully managed [30]. Ensuring that models generalize well to unseen data is vital to avoid this issue.
The grid-based segmentation [31] of the study area introduces its own set of limitations. While sectors with water surfaces were excluded to refine the analysis, this approach may overlook areas where climatic conditions are critical despite high water presence. Furthermore, capturing the complex interactions between climatic variables remains challenging [24,28]. Principal component analysis (PCA) helps in reducing the dimensionality of climate data, but it may not fully capture all relevant interactions affecting landslides.
In order to address these challenges, several strategies can be employed. Improving the quality and granularity of landslide and climatic data could enhance prediction accuracy [32]. Collaboration with local agencies and the utilization of satellite data might provide more comprehensive datasets. Regularization techniques, such as dropout and L1/L2 regularization, can help mitigate overfitting [33] and ensure better model generalization. Exploring adaptive grid-based methods, which dynamically adjust grid size and resolution based on climatic conditions and landslide distribution [34,35], could provide more accurate assessments. Advanced feature engineering and nonlinear interaction exploration may also improve model performance. Lastly, integrating remote sensing data and advanced geospatial analysis could further refine predictions in regions with limited ground-based data [36,37].
While the model’s strength lies in its ability to handle large datasets and provide robust predictions based on static climatic variables, it overlooks the dynamic nature of landslide triggers related to temporal factors. The reduction of climatic data to principal components, while effective for simplifying complex datasets, does not capture seasonal trends or historical patterns that could influence landslide risk [38]. Integrating time-series analysis could address these shortcomings. Techniques, such as seasonal autoregressive integrated moving average (SARIMA) [39] or long short-term memory [40] (LSTM) networks, could be utilized to model and predict seasonality and temporal changes. Additionally, incorporating temporal features such as the month or season could improve the model’s accuracy by reflecting the periodic variations in landslide triggers. Despite its limitations, the model’s ability to process and analyze large-scale data remains a strength, making it a valuable tool for landslide prediction with potential for further refinement.
Further investigations could focus on the following: exploring additional climatic features or incorporating temporal patterns to capture seasonal variations and their impact on landslide triggers [41]; enhancing the interpretability of DNN models to understand the specific climatic conditions that contribute to each trigger category; incorporating additional data sources or augmenting existing datasets to improve model robustness and generalization across diverse geographic regions.
The classification of landslide sizes benefited from the optimized DNN models. The results indicated the precise categorization across catastrophic, large, medium, small, and very large sizes, highlighting the model’s ability to predict the severity of landslide events. Future directions for research may include the following: incorporating geographical information system (GIS) data to analyze terrain characteristics and their influence on landslide [42] size predictions; exploring ensemble techniques to combine predictions from multiple models [43] or algorithms to further enhance accuracy and reliability; investigating the potential for predicting changes in landslide sizes over longer time horizons, considering climate change and land-use patterns.
Figure 8 illustrates the hierarchical structure of the research.
The objective is supported by methodology, including landslide event co-ordinates, confusion matrix analysis, and regularization with an adaptive grid approach. Geographic and climatic factors are examined through geographic context and climatic data analysis, addressing challenges such as data gaps and complexity in climate variable interactions. The results highlight the model’s accuracy in classifying landslide sizes and predicting rainfall-related triggers. The diagram also outlines the limitations and challenges, such as data gaps and hyperparameter tuning. Recommendations and future research suggest improvements in data quality, adaptive segmentation methods, temporal data integration, and the use of additional GIS data and ensemble methods.

5. Conclusions

In conclusion, the integration of DNN with GA-based HPO for classifying landslide triggers and sizes based on climatic data demonstrates the potential to enhance landslide hazard assessment and prediction. Such predictive models may help with proactive disaster preparedness, including the development of early warning systems and the strategic allocation of resources. As these models are further refined and additional data sources are incorporated, the accuracy, robustness, and real-world applicability of these tools in managing and mitigating landslide risks will improve, ultimately helping to minimize the impact of landslides on communities and infrastructure.
Looking ahead, several key areas for future research and development include the continuous refinement of GA and HPO techniques to adapt to evolving datasets and enhance model performance; the development of real-time monitoring systems that integrate meteorological data to provide timely alerts and responses to potential landslide events; and the pursuit of multiscale analyses to assess the models’ applicability across various spatial and temporal scales.

Author Contributions

Conceptualization, V.T. and I.M.; Data curation, Y.T., V.K. and I.M.; Formal analysis, V.K.; Funding acquisition, Y.T. and T.P.; Investigation, V.T. and O.K.; Methodology, T.P., A.G., V.N. and I.M.; Project administration, Y.T., V.T., O.K. and V.N.; Resources, V.K., T.P. and I.M.; Software, V.K., V.T., A.G. and V.N.; Supervision, Y.T., O.K. and A.G.; Validation, T.P., A.G., V.N. and I.M.; Visualization, I.M.; Writing—original draft, O.K. and I.M.; Writing—review & editing, V.T. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The interactive landslide map can be accessed via the following link: LandslidesMap (https://github.com/catauggie/landslides) (accessed on 14 August 2024).

Conflicts of Interest

The authors declare no conflict of interest.

Appendix A. Geographical Description of Landslides Worldwide

Islands are particularly vulnerable to landslides due to their unique environmental and geological conditions. In the United Kingdom and Ireland (Figure A1a), high precipitation levels affect the triggering of landslides. These areas receive an average annual rainfall of 1200 to 2000 mm, with some regions experiencing up to 3000 mm during exceptionally wet years. Frequent and intense rainfall saturates the soil, increasing the risk of slope failures, particularly in mountainous and hilly areas. The geological composition of these regions, including clayey soils and stratified rock formations, further exacerbates slope instability. Additionally, human activities, such as construction and agriculture, can destabilize slopes by altering the landscape and reducing the natural forces that maintain slope stability. These combined factors underscore the importance of understanding and mitigating landslide risks in these vulnerable areas.
The Philippines (Figure A1b): The Philippines are frequently subjected to tropical cyclones and monsoons, bringing heavy rainfall of up to 4000 mm annually in some regions [44,45]. This precipitation saturates the soil, increasing landslide risks. The country experiences an average of 20 typhoons per year [46], which often lead to large-scale landslides. The complex topography, with numerous volcanic mountain ranges, further exacerbates the conditions for landslides [47,48]. Additionally, widespread deforestation and unstable land use practices increase the frequency and severity of landslides [49].
The Hawaiian Islands (Figure A1c) experience rainfall, particularly on the windward sides of the mountains, with annual averages ranging from 2500 mm to over 10,000 mm in some locations [50]. This intense rainfall, coupled with a tropical climate, leads to substantial erosion and slope weakening [51,52]. The islands’ geology, dominated by volcanic rock, is prone to weathering and erosion, contributing to landslide vulnerability [53]. Human activities, such as construction and agriculture on slopes, further destabilize the soil [54].
Fiji (Figure A1d) is affected by tropical cyclones and seasonal monsoons, leading to frequent and intense rainfall, with some areas receiving over 3000 mm annually [55]. This increases the risk of landslides, particularly in mountainous regions [56,57]. Fiji’s volcanic activity and seismic instability also create conducive conditions for landslides [58,59]. Deforestation and poor land use practices exacerbate the situation by removing vegetation that stabilizes the soil [60,61].
Central America’s (Figure A2a) countries, such as Guatemala, El Salvador, Nicaragua, Honduras, and Costa Rica, experience a tropical climate with heavy rainfall, especially during the rainy season from May to October, with annual precipitation ranging from 1500 mm to over 4000 mm [62]. These intense rains saturate the soil, increasing landslide likelihood. The region’s volcanic rocks and heterogeneous soils are particularly prone to landslides [63]. Deforestation, uncontrolled land use, and urbanization further exacerbate the risks by reducing natural vegetation that stabilizes the soil.
Sustainability 16 07063 g0a1
Figure A1. Island landslides maps: (a) England and Ireland; (b) Philippines; (c) Hawaii; (d) Fiji.
Figure A1. Island landslides maps: (a) England and Ireland; (b) Philippines; (c) Hawaii; (d) Fiji.
Sustainability 16 07063 g0a1
The Windward Islands (Figure A2a) territories, such as Dominica, Saint Lucia, and Barbados, are susceptible to tropical storms and hurricanes that bring heavy rainfall and strong winds. These islands experience annual precipitation between 2000 and 3500 mm [64]. Periods of intense rain can rapidly saturate the soil, leading to landslides, especially on steep slopes [65]. The islands have diverse terrains with mountainous and hilly areas. The geological features, including volcanic and sedimentary rocks, are prone to erosion and weathering [65]. Human activities, such as deforestation and construction, increase the vulnerability of slopes to landslides.
Haiti (Figure A2a) experiences substantial rainfall, with some regions receiving up to 2000 mm annually during the wet season [66]. The country is frequently hit by hurricanes, which bring intense rains that cause flooding and heighten the risk of landslides [67]. Haiti’s geology is characterized by fragile mountain rocks and erosion-prone soils [68], which are easily destabilized by water. Extreme deforestation, driven by the use of wood as a primary fuel source, significantly increases landslide risks [69]. Uncontrolled urbanization and construction on slopes further contribute to soil destabilization.
Southeast Asia (Indonesia and Malaysia) (Figure A2b) also experiences heavy monsoon rains, with annual precipitation ranging from 2000 mm to over 4500 mm, and frequent volcanic activity, leading to soil saturation and instability [70,71]. The deforestation and poor land management practices in these countries exacerbate the risk by removing vegetation that stabilizes slopes and by constructing infrastructure on unstable land [72].
Oceania (Eastern Australia and New Zealand) (Figure A2c): These regions face landslide risks due to intense seasonal rains, with annual precipitation between 1000 and 2500 mm, and frequent storms. The steep slopes and tectonic activity make these areas particularly vulnerable [73,74]. Human activities such as deforestation, mining, and construction further disturb the natural stability of the land, increasing the likelihood of landslides [75].
East Asia (Korea and Japan) (Figure A2d): This region is prone to landslides due to heavy seasonal rainfall, monsoons, and typhoons, with annual precipitation ranging from 1500 to 3000 mm [76,77]. The mountainous terrain, coupled with volcanic activity and frequent earthquakes, contributes to slope instability [78]. Urban expansion, deforestation, and infrastructure development add to the risk [77].
Sustainability 16 07063 g0a2
Figure A2. Landslides maps: (a) Caribbean Basin; (b) Indonesia and Malaysia; (c) East Australia and New Zealand; (d) Korea and Japan; (e) Columbia and Ecuador; (f) Rio de Janeiro.
Figure A2. Landslides maps: (a) Caribbean Basin; (b) Indonesia and Malaysia; (c) East Australia and New Zealand; (d) Korea and Japan; (e) Columbia and Ecuador; (f) Rio de Janeiro.
Sustainability 16 07063 g0a2
South America (Colombia and Ecuador) (Figure A2e): These countries experience frequent landslides due to heavy rainfall, with some areas in the Andes receiving up to 3000 mm annually [79,80]. Deforestation, mining, and poor land management practices disturb the natural stability of the slopes [79]. Similarly, the Rio de Janeiro area in Brazil (Figure A2f) faces landslide risks due to its tropical climate, steep hills, and urban sprawl into hilly areas, compounded by deforestation and inadequate drainage systems [81].
Alps and Balkans (Figure A3a): These regions are prone to landslides due to their rugged mountainous terrain and seasonal rainfall, with precipitation levels in the Alps reaching up to 2500 mm annually. In the Alps, snowmelt in spring contributes to soil saturation and slope instability [82]. Human activities such as deforestation, construction, and tourism infrastructure development further increase the risk [83]. The Balkans also experience frequent landslides, exacerbated by similar climatic conditions and human-induced land disturbances.
Caucasus (Figure A3b) is highly susceptible to landslides due to its steep slopes and heavy precipitation, with some areas receiving up to 1500 mm annually and seismic activity [84,85]. The combination of snowmelt, intense rainfall, and earthquakes leads to frequent landslides. Deforestation and uncontrolled construction in mountainous areas further destabilize the slopes, increasing the risk of landslides [84].
Sustainability 16 07063 g0a3
Figure A3. Landslides maps: (a) Alps and Balkans; (b) Caucasus; (c) Lake Victoria Region; (d) West Africa.
Figure A3. Landslides maps: (a) Alps and Balkans; (b) Caucasus; (c) Lake Victoria Region; (d) West Africa.
Sustainability 16 07063 g0a3
The Lake Victoria region (Figure A3c) in East Africa is particularly susceptible to landslides due to several interrelated factors. The region experiences heavy seasonal rainfall, with some areas receiving up to 2000 mm annually, particularly during the long rains from March to May and the short rains from October to December [86]. The steep terrain surrounding the highlands exacerbates the landslide risk, especially when combined with deforestation. Agricultural expansion and settlement development reduce the stability of slopes by removing vegetation that normally helps anchor the soil. Poor land management practices, such as overgrazing and improper farming techniques, further increase the vulnerability of the land to landslides.
In West Africa (Figure A3d), landslides are primarily triggered by intense rainfall during the wet season, which lasts from May to October. Regions with hilly or mountainous terrain, such as parts of Cameroon and Nigeria, are particularly at risk. For example, some areas of Cameroon receive annual precipitation of up to 4000 mm, which saturates the soil and increases landslide susceptibility [87,88]. Deforestation for agriculture, logging, and mining activities significantly destabilizes the soil by removing the vegetation cover, which acts as a natural stabilizer. Rapid urbanization without adequate planning or infrastructure development also contributes to the risk by increasing surface runoff and soil erosion, further destabilizing slopes.
The Himalayas (Figure A4a) are one of the most landslide-prone regions in the world. This vulnerability is due to the region’s extreme elevation, with peaks exceeding 8000 m, steep slopes, and heavy monsoon rains that can exceed 5000 mm annually in some areas [89,90,91]. The Himalayas also experience frequent seismic activity, which can trigger landslides, particularly in areas with fractured and weathered rock. Human activities, including deforestation, road construction, and terraced farming, further destabilize slopes by altering the natural drainage patterns and reducing the cohesive strength of the soil. These factors combine to make landslides a common and deadly hazard in the region.
China (Figure A4b) faces significant landslide risks, especially in its mountainous regions, such as the Sichuan and Yunnan provinces. These areas are characterized by rugged terrain and are prone to heavy seasonal rains during the summer monsoon, with some regions receiving up to 2500 mm of precipitation annually [92,93]. The risk is compounded by frequent seismic activity, particularly in Sichuan, which is located near active fault lines. Rapid urbanization and extensive mining operations further exacerbate the risk by disturbing the natural stability of the land. Deforestation for agriculture and construction also contributes to landslide vulnerability by removing the vegetation that stabilizes slopes and reduces erosion. These combined factors make landslides a frequent and serious hazard in China’s mountainous regions.
Western Ghats (Figure A4c) along the western coast of India are susceptible to landslides, primarily due to the region’s intense monsoon rains. The Western Ghats receive an average annual rainfall ranging from 2000 mm to over 7000 mm in some areas, particularly during the southwest monsoon season from June to September [94,95,96]. The combination of steep terrain, with slopes often exceeding 30 degrees, and extensive deforestation for agriculture and urban development significantly increases the likelihood of landslides. The removal of vegetation reduces the soil’s ability to retain water, leading to increased surface runoff and soil erosion. Inadequate drainage systems and poor land management practices, such as slash-and-burn agriculture and unplanned construction, further contribute to slope instability, making the Western Ghats a high-risk area for landslides.
Vietnam (Figure A4d) is prone to landslides, particularly in its northern and central mountainous regions. The country experiences heavy rainfall during the monsoon season, with some regions receiving up to 3000 mm annually [97,98,99]. The risk of landslides is heightened by the frequent occurrence of typhoons, which bring intense, short-duration rainfall that saturates the soil and triggers slope failures. The problem is exacerbated by deforestation, where forests are cleared for agriculture and infrastructure development. This loss of vegetation destabilizes the soil, making it more susceptible to landslides. Additionally, the expansion of agricultural activities and the construction of roads and settlements in hilly areas disturb the natural vegetation cover and soil stability, further increasing the landslide risk in Vietnam.
Washington State (Figure A5a) is highly susceptible to landslides due to its rugged terrain, which includes the Cascade Range and Olympic Mountains, and heavy rainfall, particularly in the western regions. Western Washington receives an average of 1500 to 3800 mm of rain annually, with some areas experiencing even higher amounts, particularly during the wet season from October to March [100,101,102]. The frequent landslides in this region are exacerbated by snowmelt, which adds to soil saturation and volcanic activity, particularly around Mount Rainier and Mount St. Helens. Deforestation for timber, agriculture, and urban development further increases the landslide risk by removing the trees and vegetation that help stabilize the slopes. Mapping landslide-prone areas in Washington is crucial for understanding and mitigating these risks, particularly in areas where human activities have increased the vulnerability of the landscape to landslides.
Sustainability 16 07063 g0a4
Figure A4. Landslides maps (a) Himalaya, (b) China, (c) Western Ghats, (d) South-East Asia.
Figure A4. Landslides maps (a) Himalaya, (b) China, (c) Western Ghats, (d) South-East Asia.
Sustainability 16 07063 g0a4
California (Figure A5b) landslides particularly occur in its steep terrain, such as the Sierra Nevada, coastal ranges, and the San Gabriel Mountains. The region experiences intense seasonal rains, with annual precipitation ranging from 300 mm to over 1200 mm in some areas, depending on location and season [103,104,105]. Earthquakes, such as those from the San Andreas Fault, and frequent wildfires, which remove vegetation and destabilize soil, further contribute to landslide risk.
The Northwestern states (Figure A5c), including Oregon, Idaho, and Montana, face significant landslide risks due to their mountainous terrain and precipitation patterns. This region experiences substantial rainfall, with annual averages ranging from 1000 mm to 1500 mm, combined with significant snowmelt in the spring [106,107,108]. Tectonic activity, particularly along the Cascadia Subduction Zone, also plays a role in triggering landslides. Detailed mapping in these states is essential for identifying vulnerable areas and implementing preventive measures to reduce the risk of landslides.
Utah and Colorado (Figure A5d) experience landslides primarily in their mountainous regions, such as the Wasatch Range in Utah and the Rocky Mountains in Colorado. Snowmelt, heavy rains, and geological instability contribute to landslide risk in these states. Utah’s Wasatch Range receives between 600 mm and 1200 mm of annual precipitation, while Colorado’s Rocky Mountains experience similar levels of precipitation [109,110]. The combination of these factors, along with variable geological conditions, necessitates comprehensive mapping and monitoring to manage and mitigate landslide risks effectively.
Sustainability 16 07063 g0a5
Figure A5. USA landslides maps: (a) Washington; (b) California; (c) the North West states; (d) Utah and Colorado.
Figure A5. USA landslides maps: (a) Washington; (b) California; (c) the North West states; (d) Utah and Colorado.
Sustainability 16 07063 g0a5

References

  1. Fernández, P.; Ceacero-Moreno, M. Urban sustainability and natural hazards management; designs using simulations. Sustainability 2021, 12, 649. [Google Scholar] [CrossRef]
  2. Shim, J.H.; Kim, C.I. Measuring resilience to natural hazards: Towards sustainable hazard mitigation. Sustainability 2015, 7, 14153–14185. [Google Scholar] [CrossRef]
  3. Porȩbska, A.; Godyń, I.; Radzicki, K.; Nachlik, E.; Rizzi, P. Built heritage, sustainable development, and natural hazards: Flood protection and UNESCO world heritage site protection strategies in Krakow, Poland. Sustainability 2019, 11, 4886. [Google Scholar] [CrossRef]
  4. Nanehkaran, Y.A.; Chen, B.; Cemiloglu, A.; Chen, J. Riverside landslide susceptibility overview: Leveraging artificial neural networks and machine learning in accordance with the United Nations (UN) sustainable development goals. Water 2023, 15, 2707. [Google Scholar] [CrossRef]
  5. Yang, C.; Wang, J.; Li, S.; Xiong, R.; Li, X.; Gao, L.; Guo, X. Landslide Susceptibility Assessment and Future Prediction with Land Use Change and Urbanization towards Sustainable Development: The Case of the Li River. Sustainability 2024, 16, 4416. [Google Scholar] [CrossRef]
  6. Habumugisha, J.M.; Chen, N.; Rahman, M.; Islam, M.M. Landslide susceptibility mapping with deep learning algorithms. Sustainability 2022, 14, 1734. [Google Scholar] [CrossRef]
  7. Wang, S.; Yang, B.; Zhou, Y.; Wang, F.; Zhang, R.; Zhao, Q. Three-dimensional information extraction from GaoFen-1 satellite images for landslide monitoring. Geomorphology 2018, 309, 77–85. [Google Scholar] [CrossRef]
  8. Tehrani, F.S.; Calvello, M.; Liu, Z.; Zhang, L.; Lacasse, S. Machine learning and landslide studies: Recent advances and applications. Nat. Hazards 2022, 114, 1197–1245. [Google Scholar] [CrossRef]
  9. Merghadi, A.; Yunus, A.P.; Dou, J.; Whiteley, J.; ThaiPham, B.; Bui, D.T.; Avtar, R.; Abderrahmane, B. Machine learning methods for landslide susceptibility studies: A comparative overview of algorithm performance. Earth-Sci. Rev. 2020, 207, 103225. [Google Scholar] [CrossRef]
  10. Wang, H.; Zhang, L.; Yin, K.; Luo, H.; Li, J. Landslide identification using machine learning. Geosci. Front. 2021, 12, 351–364. [Google Scholar] [CrossRef]
  11. Ghorbanzadeh, O.; Blaschke, T.; Gholamnia, K.; Meena, S.R.; Tiede, D.; Aryal, J. Evaluation of different machine learning methods and deep-learning convolutional neural networks for landslide detection. Remote Sens. 2019, 11, 196. [Google Scholar] [CrossRef]
  12. Korup, O.; Stolle, A. Landslide prediction from machine learning. Geol. Today 2014, 30, 26–33. [Google Scholar] [CrossRef]
  13. Marjanović, M.; Kovačević, M.; Bajat, B.; Voženílek, V. Landslide susceptibility assessment using SVM machine learning algorithm. Eng. Geol. 2011, 123, 225–234. [Google Scholar] [CrossRef]
  14. Goetz, J.; Brenning, A.; Petschko, H.; Leopold, P. Evaluating machine learning and statistical prediction techniques for landslide susceptibility modeling. Comput. Geosci. 2015, 81, 1–11. [Google Scholar] [CrossRef]
  15. Micheletti, N.; Foresti, L.; Robert, S.; Leuenberger, M.; Pedrazzini, A.; Jaboyedoff, M.; Kanevski, M. Machine learning feature selection methods for landslide susceptibility mapping. Math. Geosci. 2014, 46, 33–57. [Google Scholar] [CrossRef]
  16. Esteghamati, M.Z. Leveraging machine learning techniques to support a holistic performance-based seismic design of civil structures. In Interpretable Machine Learning for the Analysis Design Assessment and Informed Decision Making for Civil Infrastructure; Elsevier: Amsterdam, The Netherlands, 2024; pp. 25–49. [Google Scholar]
  17. Kikuchi, T.; Sakita, K.; Nishiyama, S.; Takahashi, K. Landslide susceptibility mapping using automatically constructed CNN architectures with pre-slide topographic DEM of deep-seated catastrophic landslides caused by Typhoon Talas. Nat. Hazards 2023, 117, 339–364. [Google Scholar] [CrossRef]
  18. Kavzoglu, T.; Colkesen, I.; Sahin, E.K. Machine learning techniques in landslide susceptibility mapping: A survey and a case study. In Landslides: Theory, Practice and Modelling; Springer: Berlin/Heidelberg, Germany, 2019; pp. 283–301. [Google Scholar]
  19. Chen, X.; Chen, W. GIS-based landslide susceptibility assessment using optimized hybrid machine learning methods. Catena 2021, 196, 104833. [Google Scholar] [CrossRef]
  20. Adachi, K. Global Landslide Data. 2019. Available online: https://www.kaggle.com/datasets/kazushiadachi/global-landslide-data (accessed on 26 June 2024).
  21. NASA Langley Research Center (LaRC). NASA POWER API: Daily Data Retrieval. 2024. Available online: https://power.larc.nasa.gov (accessed on 26 June 2024).
  22. Abdi, H.; Williams, L.J. Principal component analysis. Wiley Interdiscip. Rev. Comput. Stat. 2010, 2, 433–459. [Google Scholar] [CrossRef]
  23. Lin, W.W.; Mak, M.W. Wav2Spk: A Simple DNN Architecture for Learning Speaker Embeddings from Waveforms. In Proceedings of the Interspeech 2020, Shanghai, China, 25–29 October 2020; pp. 3211–3215. [Google Scholar]
  24. Fidan, S.; Tanyaş, H.; Akbaş, A.; Lombardo, L.; Petley, D.N.; Görüm, T. Understanding fatal landslides at global scales: A summary of topographic, climatic, and anthropogenic perspectives. Nat. Hazards 2024, 120, 6437–6455. [Google Scholar] [CrossRef]
  25. Novellino, A.; Pennington, C.; Leeming, K.; Taylor, S.; Alvarez, I.G.; McAllister, E.; Arnhardt, C.; Winson, A. Mapping landslides from space: A review. Landslides 2024, 21, 1041–1052. [Google Scholar] [CrossRef]
  26. Xie, Y.; Zhan, N.; Zhu, J.; Xu, B.; Chen, H.; Mao, W.; Luo, X.; Hu, Y. Landslide extraction from aerial imagery considering context association characteristics. Int. J. Appl. Earth Obs. Geoinf. 2024, 131, 103950. [Google Scholar] [CrossRef]
  27. Guo, H.; Martínez-Graña, A. Susceptibility of Landslide Debris Flow in Yanghe Township Based on Multi-Source Remote Sensing Information Extraction Technology (Sichuan, China). Land 2024, 13, 206. [Google Scholar] [CrossRef]
  28. Sharma, A.; Sajjad, H.; Roshani; Rahaman, M.H. A systematic review for assessing the impact of climate change on landslides: Research gaps and directions for future research. Spat. Inf. Res. 2024, 32, 165–185. [Google Scholar] [CrossRef]
  29. Jiang, S.H.; Jie, H.H.; Xie, J.; Huang, J.; Zhou, C.B. Probabilistic back-analysis of rainfall-induced landslides for slope reliability prediction with multi-source information. J. Rock Mech. Geotech. Eng. 2024; in press. [Google Scholar]
  30. El-Hassani, F.Z.; Amri, M.; Joudar, N.E.; Haddouch, K. A new optimization model for MLP hyperparameter tuning: Modeling and resolution by real-coded genetic algorithm. Neural Process. Lett. 2024, 56, 105. [Google Scholar] [CrossRef]
  31. Sun, H.; Yang, S.; Wang, R.; Yang, K. Study on a Landslide Segmentation Algorithm Based on Improved High-Resolution Networks. Appl. Sci. 2024, 14, 6459. [Google Scholar] [CrossRef]
  32. Khalili, M.A.; Palumbo, S.; Madadi, S.; Bausilio, G.; Voosoghi, B.; Calcaterra, D.; Di Martire, D. Enhancing landslide prediction through advanced transformer-based models: Integrating SAR imagery and environmental data. e-J. Nondestruct. Test. 2024, 29, 1–10. [Google Scholar] [CrossRef]
  33. Dzjumajev, S. A Study of Mitigating Advesarial Attacks against Machine Learning Models. Master’s Thesis, University of Agder, Kristiansand, Norway, 2024. [Google Scholar]
  34. Cui, H.; Ji, J.; Hürlimann, M.; Medina, V. Probabilistic and physically-based modelling of rainfall-induced landslide susceptibility using integrated GIS-FORM algorithm. Landslides 2024, 21, 1461–1481. [Google Scholar] [CrossRef]
  35. Zeng, T.; Gong, Q.; Wu, L.; Zhu, Y.; Yin, K.; Peduto, D. Double-index rainfall warning and probabilistic physically based model for fast-moving landslide hazard analysis in subtropical-typhoon area. Landslides 2024, 21, 753–773. [Google Scholar] [CrossRef]
  36. Zhang, Q.; Wang, T. Deep Learning for Exploring Landslides with Remote Sensing and Geo-Environmental Data: Frameworks, Progress, Challenges, and Opportunities. Remote Sens. 2024, 16, 1344. [Google Scholar] [CrossRef]
  37. Guo, H.; Martínez-Graña, A. Landslide Hazard Prediction Based on Small Baseline Subset–Interferometric Synthetic-Aperture Radar Technology Combined with Land-Use Dynamic Change and Hydrological Conditions (Sichuan, China). Remote Sens. 2024, 16, 2715. [Google Scholar] [CrossRef]
  38. Fang, Z.; Wang, Y.; van Westen, C.; Lombardo, L. Landslide hazard spatiotemporal prediction based on data-driven models: Estimating where, when and how large landslide may be. Int. J. Appl. Earth Obs. Geoinf. 2024, 126, 103631. [Google Scholar] [CrossRef]
  39. Minh, H.V.T.; Van Ty, T.; Nam, N.D.G.; Lien, B.T.B.; Thanh, N.T.; Cong, N.P.; Meraj, G.; Kumar, P.; Van Thinh, L.; Van Duy, D.; et al. Modelling and predicting annual rainfall over the Vietnamese Mekong Delta (VMD) using SARIMA. Discov. Geosci. 2024, 2, 19. [Google Scholar] [CrossRef]
  40. Jin, A.; Yang, S.; Huang, X. Landslide displacement prediction based on time series and long short-term memory networks. Bull. Eng. Geol. Environ. 2024, 83, 264. [Google Scholar] [CrossRef]
  41. Kashyap, R.; Pandey, A.C.; Parida, B.R. Spatio-temporal variability of monsoon precipitation and their effect on precipitation triggered landslides in relation to relief in Himalayas. Spat. Inf. Res. 2021, 29, 857–869. [Google Scholar] [CrossRef]
  42. Dai, F.; Lee, C. Terrain-based mapping of landslide susceptibility using a geographical information system: A case study. Can. Geotech. J. 2001, 38, 911–923. [Google Scholar] [CrossRef]
  43. Umar, I.H.; Lin, H.; Hassan, J.I. Transforming Landslide Prediction: A Novel Approach Combining Numerical Methods and Advanced Correlation Analysis in Slope Stability Investigation. Appl. Sci. 2024, 14, 3685. [Google Scholar] [CrossRef]
  44. Racoma, B.A.B.; Klingaman, N.P.; Holloway, C.E.; Schiemann, R.K.; Bagtasa, G. Tropical cyclone characteristics associated with extreme precipitation in the northern Philippines. Int. J. Climatol. 2022, 42, 3290–3307. [Google Scholar] [CrossRef]
  45. Cayanan, E.O.; Chen, T.C.; Argete, J.C.; Yen, M.C.; D. NILO, P. The effect of tropical cyclones on southwest monsoon rainfall in the Philippines. J. Meteorol. Soc. Jpn. Ser. II 2011, 89, 123–139. [Google Scholar] [CrossRef]
  46. Ribera, P.; García Herrera, R.; Gimeno, L. Historical Deadly Typhoons in the Philippines. 2008. Available online: https://docta.ucm.es/entities/publication/8d7a53b0-0ac1-4adf-b14d-d342235be406 (accessed on 14 August 2024).
  47. Hearn, G.J.; Roy Hart, J. Settlements and slides: A large landslide case study from the Central Cordillera of the Philippines. Q. J. Eng. Geol. Hydrogeol. 2020, 53, 62–73. [Google Scholar] [CrossRef]
  48. Abancó, C.; Bennett, G.L.; Matthews, A.J.; Matera, M.A.M.; Tan, F.J. The role of geomorphology, rainfall and soil moisture in the occurrence of landslides triggered by 2018 Typhoon Mangkhut in the Philippines. Nat. Hazards Earth Syst. Sci. 2021, 21, 1531–1550. [Google Scholar] [CrossRef]
  49. Gabriel, M.J. Dynamics and drivers of deforestation in the Philippines. Ecosyst. Dev. J. 2023, 13, 18–32. [Google Scholar]
  50. Xue, L.; Wang, Y.; Newman, A.J.; Ikeda, K.; Rasmussen, R.M.; Giambelluca, T.W.; Longman, R.J.; Monaghan, A.J.; Clark, M.P.; Arnold, J.R. How will rainfall change over Hawai ‘i in the future? High-resolution regional climate simulation of the Hawaiian Islands. Bull. Atmos. Sci. Technol. 2020, 1, 459–490. [Google Scholar] [CrossRef]
  51. Ward, D.J.; Galewsky, J. Exploring landscape sensitivity to the Pacific Trade Wind Inversion on the subsiding island of Hawaii. J. Geophys. Res. Earth Surf. 2014, 119, 2048–2069. [Google Scholar] [CrossRef]
  52. Kitayama, K.; Mueller-Dombois, D. An altitudinal transect analysis of the windward vegetation on Haleakala, a Hawaiian island mountain:(1) climate and soils. Phytocoenologia 1994, 24, 111–133. [Google Scholar] [CrossRef]
  53. Garcia, M.O.; Frey, F.A.; Grooms, D.G. Petrology of volcanic rocks from Kaula Island, Hawaii: Implications for the origin of Hawaiian phonolites. Contrib. Mineral. Petrol. 1986, 94, 461–471. [Google Scholar] [CrossRef]
  54. Porro, R.; Kim, K.; Spirandelli, D.; Lowry, K. Evaluating erosion management strategies in Waikiki, Hawaii. Ocean. Coast. Manag. 2020, 188, 105113. [Google Scholar] [CrossRef]
  55. Deo, A.; Chand, S.S.; Ramsay, H.; Holbrook, N.J.; McGree, S.; Magee, A.; Bell, S.; Titimaea, M.; Haruhiru, A.; Malsale, P.; et al. Tropical cyclone contribution to extreme rainfall over southwest Pacific Island nations. Clim. Dyn. 2021, 56, 3967–3993. [Google Scholar] [CrossRef]
  56. Chen, J.K.; Taylor, F.W.; Edwards, R.L.; Cheng, H.; Burr, G. Recent emerged reef terraces of the Yenkahe resurgent block, Tanna, Vanuatu: Implications for volcanic, landslide and tsunami hazards. J. Geol. 1995, 103, 577–590. [Google Scholar] [CrossRef]
  57. Kouwenhoven, P. Profile of Risks from Climate Change and Geohazards in Vanuatu: Draft Report; CLIMsystems: Hamilton, New Zealand, 2013; Available online: https://www.nab.vu/sites/default/files/nab/documents/03/04/2014%20-%2012:45/risk_profile_report_draft_1.pdf (accessed on 14 August 2024).
  58. Ioualalen, M.; Pelletier, B.; Gordillo, G.S. Investigating the March 28th 1875 and the September 20th 1920 earthquakes/tsunamis of the Southern Vanuatu arc, offshore Loyalty Islands, New Caledonia. Tectonophysics 2017, 709, 20–38. [Google Scholar] [CrossRef]
  59. Németh, K.; Cronin, S.J. Phreatomagmatic volcanic hazards where rift-systems meet the sea, a study from Ambae Island, Vanuatu. J. Volcanol. Geotherm. Res. 2009, 180, 246–258. [Google Scholar] [CrossRef]
  60. Eckardt, R.; Herold, M.; Sambale, J.; Weaver, S. Monitoring deforestation patterns and processes in the Pacific island state of Vanuatu. In Proceedings of the Geoinformatics Forum, Salzburg, Austria, 1–4 July 2008. [Google Scholar]
  61. Siméoni, P.; Lebot, V. Spatial representation of land use and population density: Integrated layers of data contribute to environmental planning in Vanuatu. Hum. Ecol. 2012, 40, 541–555. [Google Scholar] [CrossRef]
  62. Peña, M.; Douglas, M.W. Characteristics of wet and dry spells over the Pacific side of Central America during the rainy season. Mon. Weather. Rev. 2002, 130, 3054–3073. [Google Scholar] [CrossRef]
  63. Bertsch, F.; Alvarado, A.; Henriquez, C.; Mata, R. Properties, geographic distribution, and management of major soil orders of Costa Rica. In Quantifying Sustainable Development; Elsevier: Amsterdam, The Netherlands, 2000; pp. 265–294. [Google Scholar]
  64. Ferdinand, I.; Parker, E. Hurricane Risk Reduction Strategies in the Windward Islands; Coventry Centre for Disaster Management, Coventry University: Coventry, UK, 2005. [Google Scholar]
  65. Walsh, R. The influence of climate, lithology and time on drainage density and relief development in the tropical volcanic terrain of the Windward Islands. In Environmental Change and Tropical Geomorphology; Routledge: Oxfordshire, UK, 2020; pp. 93–122. [Google Scholar]
  66. Marcelin, L.H.; Cela, T.; Shultz, J.M. Haiti and the politics of governance and community responses to Hurricane Matthew. Disaster Health 2016, 3, 151–161. [Google Scholar] [CrossRef]
  67. Harp, E.L.; Jibson, R.W.; Schmitt, R.G. Map of Landslides Triggered by the January 12, 2010, Haiti Earthquake; Technical Report; US Geological Survey: Reston, VA, USA, 2016. [Google Scholar]
  68. White, T.A. Policy Lessons from History and Natural Resource Projects in Rural Haiti; Department of Forest Resources, University of Minnesota: St. Paul, MN, USA, 1994. Available online: https://ageconsearch.umn.edu/record/11892/?v=pdf (accessed on 14 August 2024).
  69. Mompremier, R.; Her, Y.; Hoogenboom, G.; Song, J. Effects of deforestation and afforestation on water availability for dry bean production in Haiti. Agric. Ecosyst. Environ. 2022, 325, 107721. [Google Scholar] [CrossRef]
  70. Lee, H.S. General rainfall patterns in Indonesia and the potential impacts of local seas on rainfall intensity. Water 2015, 7, 1751–1768. [Google Scholar] [CrossRef]
  71. Wangwongchai, A.; Sixiong, Z.; Qingcun, Z. A case study on a strong tropical disturbance and record heavy rainfall in Hat Yai, Thailand during the winter monsoon. Adv. Atmos. Sci. 2005, 22, 436–450. [Google Scholar] [CrossRef]
  72. Wicke, B.; Sikkema, R.; Dornburg, V.; Faaij, A. Exploring land use changes and the role of palm oil production in Indonesia and Malaysia. Land Use Policy 2011, 28, 193–206. [Google Scholar] [CrossRef]
  73. Kingma, J. The tectonic history of New Zealand. N. Zealand J. Geol. Geophys. 1959, 2, 1–55. [Google Scholar] [CrossRef]
  74. Jessop, K.; Daczko, N.; Piazolo, S. Tectonic cycles of the New England Orogen, eastern Australia: A review. Aust. J. Earth Sci. 2019, 66, 459–496. [Google Scholar] [CrossRef]
  75. Allen, S.K.; Cox, S.C.; Owens, I.F. Rock avalanches and other landslides in the central Southern Alps of New Zealand: A regional study considering possible climate change impacts. Landslides 2011, 8, 33–48. [Google Scholar] [CrossRef]
  76. Kim, S.W.; Chun, K.W.; Otsuki, K.; Shinohara, Y.; Kim, M.I.; Kim, M.S.; Lee, D.K.; Seo, J.I.; Choi, B.K. Heavy rain types for triggering shallow landslides in South Korea 2015, 60, 243–249. J. Fac. Agric. Kyushu Univ. 2015, 60, 243–249. [Google Scholar]
  77. Park, D.S.R.; Ho, C.H.; Kim, J.H.; Kim, H.S. Strong landfall typhoons in Korea and Japan in a recent decade. J. Geophys. Res. Atmos. 2011, 116. [Google Scholar] [CrossRef]
  78. Obara, K. Characteristic activities of slow earthquakes in Japan. Proc. Jpn. Acad. Ser. B 2020, 96, 297–315. [Google Scholar] [CrossRef] [PubMed]
  79. Garcia-Delgado, H.; Petley, D.N.; Bermúdez, M.A.; Sepúlveda, S.A. Fatal landslides in Colombia (from historical times to 2020) and their socio-economic impacts. Landslides 2022, 19, 1689–1716. [Google Scholar] [CrossRef]
  80. Montero-Olarte, J. The physical environment and landslides in the Colombian Andean tropics. In Landslides and Engineered Slopes. Experience, Theory and Practice; CRC Press: Boca Raton, FL, USA, 2018; pp. 1445–1451. [Google Scholar]
  81. Fernandes, N.F.e.a. Topographic controls of landslides in Rio de Janeiro: Field evidence and modeling. Catena 2004, 55, 163–181. [Google Scholar] [CrossRef]
  82. Savi, S.; Comiti, F.; Strecker, M.R. Pronounced increase in slope instability linked to global warming: A case study from the eastern European Alps. Earth Surf. Process. Landforms 2021, 46, 1328–1347. [Google Scholar] [CrossRef]
  83. Schlögl, M.; Avian, M.; Richter, G.; Thaler, T.; Heiss, G.; Fuchs, S.; Lenz, G. On the nexus between landslide susceptibility and transport infrastructure–agent-based vulnerability assessment of rural road networks in the Eastern European Alps. Nat. Hazards Earth Syst. Sci. 2018, 19, 201–219. [Google Scholar] [CrossRef]
  84. Matossian, A.O.; Baghdasaryan, H.; Avagyan, A.; Igityan, H.; Gevorgyan, M.; Havenith, H.B. A new landslide inventory for the Armenian Lesser Caucasus: Slope failure morphologies and seismotectonic influences on large landslides. Geosciences 2020, 10, 111. [Google Scholar] [CrossRef]
  85. Tibaldi, A.; Oppizzi, P.; Gierke, J.; Oommen, T.; Tsereteli, N.; Gogoladze, Z. Landslides near Enguri dam (Caucasus, Georgia) and possible seismotectonic effects. Nat. Hazards Earth Syst. Sci. 2019, 19, 71–91. [Google Scholar] [CrossRef]
  86. Ngecu, W.M.; Nyamai, C.; Erima, G. The extent and significance of mass-movements in Eastern Africa: Case studies of some major landslides in Uganda and Kenya. Environ. Geol. 2004, 46, 1123–1133. [Google Scholar] [CrossRef]
  87. Igwe, O. The geotechnical characteristics of landslides on the sedimentary and metamorphic terrains of South-East Nigeria, West Africa. Geoenviron. Disasters 2015, 2, 1. [Google Scholar] [CrossRef]
  88. Igwe, O.; Mode, A.W.; Nnebedum, O.; Okonkwo, I.; Oha, I. The mechanisms and characteristics of a complex rock-debris avalanche at the Nigeria–Cameroon border, West Africa. Geomorphology 2015, 234, 1–10. [Google Scholar] [CrossRef]
  89. Sarkar, S.; Kanungo, D.; Mehrotra, G. Landslide hazard zonation: A case study in Garhwal Himalaya, India. Mt. Res. Dev. 1995, 15, 301–309. [Google Scholar] [CrossRef]
  90. Dortch, J.; Owen, L.; Haneberg, W.; Caffee, M. Nature and timing of large landslides in the Himalaya and Transhimalaya of northern India. Quat. Sci. Rev. 2009, 28, 1037–1054. [Google Scholar] [CrossRef]
  91. Du, J.; Glade, T.; Woldai, T.; Chai, B.; Zeng, B. Landslide susceptibility assessment based on an incomplete landslide inventory in the Jilong Valley, Tibet, Chinese Himalayas. Eng. Geol. 2020, 270, 105572. [Google Scholar] [CrossRef]
  92. Wang, D.; Hao, M.; Chen, S.; Meng, Z.; Jiang, D.; Ding, F. Assessment of landslide susceptibility and risk factors in China. Nat. Hazards 2021, 108, 3045–3059. [Google Scholar] [CrossRef]
  93. Liu, X.; Miao, C. Large-scale assessment of landslide hazard, vulnerability and risk in China. Geomat. Nat. Hazards Risk 2018, 9, 1037–1052. [Google Scholar] [CrossRef]
  94. Kuriakose, S.; Sankar, G.; Muraleedharan, C. History of landslide susceptibility and a chorology of landslide-prone areas in the Western Ghats of Kerala, India. Environ. Geol. 2009, 57, 1553–1568. [Google Scholar] [CrossRef]
  95. Sajinkumar, K.; Anbazhagan, S. Geomorphic appraisal of landslides on the windward slope of Western Ghats, southern India. Nat. Hazards 2015, 75, 953–973. [Google Scholar] [CrossRef]
  96. Martha, T.; Roy, P.; Khanna, K.; Mrinalni, K.; Kumar, K. Landslides mapped using satellite data in the Western Ghats of India after excess rainfall during August 2018. Curr. Sci. 2019, 117, 804–812. [Google Scholar] [CrossRef]
  97. Lee, S.; Dan, N. Probabilistic landslide susceptibility mapping in the Lai Chau province of Vietnam: Focus on the relationship between tectonic fractures and landslides. Environ. Geol. 2005, 48, 778–787. [Google Scholar] [CrossRef]
  98. Van Tien, P.; Luong, L.; Duc, D.; Trinh, P.; Quynh, D. Rainfall-induced catastrophic landslide in Quang Tri Province: The deadliest single landslide event in Vietnam in 2020. Landslides 2021, 18, 2323–2327. [Google Scholar] [CrossRef]
  99. Tien Bui, D.; Pradhan, B.; Lofman, O.; Revhaug, I.; Dick, Ø.B. Regional prediction of landslide hazard using probability analysis of intense rainfall in the Hoa Binh province, Vietnam. Nat. Hazards 2013, 66, 707–730. [Google Scholar] [CrossRef]
  100. Harp, E. Landslides and Landslide Hazards in Washington State Due to February 5–9, 1996 Storm; USGS: Reston, VA, USA, 1997.
  101. Pierson, T.; Evarts, R.; Bard, J. Landslides in the Western Columbia Gorge, Skamania County, Washington; US Geological Survey: Reston, VA, USA, 2016.
  102. Schuster, R.; Chleborad, A. Landslides in Washington and Oregon—An Overview. Open-File Rep. 1989, 89, 86. [Google Scholar]
  103. Jibson, R. The 2005 La Conchita, California, landslide. Landslides 2006, 3, 73–78. [Google Scholar] [CrossRef]
  104. Handwerger, A.L.; Fielding, E.J.; Huang, M.H.; Bennett, G.L.; Liang, C.; Schulz, W.H. Widespread initiation, reactivation, and acceleration of landslides in the northern California Coast Ranges due to extreme rainfall. J. Geophys. Res. Earth Surface 2019, 124, 1782–1797. [Google Scholar] [CrossRef]
  105. Slosson, J.; Krohn, J. Southern California landslides of 1978 and 1980. In Debris Flows in Southern California and Their Relation to Landslide and Flood Hazards; Caltech: Pasadena, CA, USA, 1978. [Google Scholar]
  106. Lu, Z.; Kim, J. A framework for studying hydrology-driven landslide hazards in northwestern US using satellite InSAR, precipitation and soil moisture observations: Early results and future directions. GeoHazards 2021, 2, 17–40. [Google Scholar] [CrossRef]
  107. Mirus, B.B.; Jones, E.S.; Baum, R.L.; Godt, J.W.; Slaughter, S.; Crawford, M.M.; Lancaster, J.; Stanley, T.; Kirschbaum, D.B.; Burns, W.J.; et al. Landslides across the USA: Occurrence, susceptibility, and data limitations. Landslides 2020, 17, 2271–2285. [Google Scholar] [CrossRef]
  108. Xu, Y.; Schulz, W.; Lu, Z.; Kim, J.; Baxstrom, K. Geologic controls of slow-moving landslides near the US West Coast. Landslides 2021, 18, 3353–3365. [Google Scholar] [CrossRef]
  109. Regmi, N.; Giardino, J.; Vitek, J. Characteristics of landslides in western Colorado, USA. Landslides 2014, 11, 589–603. [Google Scholar] [CrossRef]
  110. Jibson, R.; Harp, E. The Springdale, Utah, landslide: An extraordinary event. Environ. Eng. Geosci. 1996, 2, 137–150. [Google Scholar] [CrossRef]
Sustainability 16 07063 g001
Figure 1. Overall technical framework for landslide risk assessment.
Figure 1. Overall technical framework for landslide risk assessment.
Sustainability 16 07063 g001
Sustainability 16 07063 g002
Figure 2. Distribution of landslide events by countries.
Figure 2. Distribution of landslide events by countries.
Sustainability 16 07063 g002
Sustainability 16 07063 g003
Figure 3. Experimental Pipeline.
Figure 3. Experimental Pipeline.
Sustainability 16 07063 g003
Sustainability 16 07063 g004
Figure 4. Evolution of metrics for landslide size and trigger predictions across individuals.
Figure 4. Evolution of metrics for landslide size and trigger predictions across individuals.
Sustainability 16 07063 g004
Sustainability 16 07063 g005
Figure 5. Confusion matrix for landslide size classification.
Figure 5. Confusion matrix for landslide size classification.
Sustainability 16 07063 g005
Sustainability 16 07063 g006
Figure 6. Confusion matrix for landslide trigger classification.
Figure 6. Confusion matrix for landslide trigger classification.
Sustainability 16 07063 g006
Sustainability 16 07063 g007
Figure 7. Map of landslide risk assessment on Viti Levu Island, Fiji. Red indicates areas with a high probability of landslides according to the model, while green represents areas with low risk.
Figure 7. Map of landslide risk assessment on Viti Levu Island, Fiji. Red indicates areas with a high probability of landslides according to the model, while green represents areas with low risk.
Sustainability 16 07063 g007
Sustainability 16 07063 g008
Figure 8. Research structure overview.
Figure 8. Research structure overview.
Sustainability 16 07063 g008
Table 1. Summary of studies on ML in landslide susceptibility mapping.
Table 1. Summary of studies on ML in landslide susceptibility mapping.
ReferenceFocusApplied MethodResultsLimitations
Merghadi et al. [9]Overview of ML techniques in landslide susceptibility mappingSVM, decision trees, logistic regressionTree-based ensemble algorithms, random forest standout; robust performance in AlgeriaLimited data on landslide types and sizes; potential overfitting and data quality issues
Tehrani et al. [8]ML applications in landslide detection; spatial and temporal forecastingVarious ML techniquesSignificant advancements in complex landslide modeling; challenges remain in temporal forecastsLack of standardized model evaluation metrics; reliance on empirical data for temporal forecasts
Wang et al. [10]Landslide identification using ML and DLCNN, SVM, random forestCNN achieves 92.5% accuracy in landslide identification in Hong Kong; RF and SVM also effectiveData preprocessing challenges; limited improvement with additional data from DEM
Korup et al. [12]ML for landslide prediction based on historical dataML, Data mining techniquesSuccess rates of 75–95% in predicting landslides; challenges with data quality and model selectionDifficulty in predicting specific landslide types and sizes
Marjanovic et al. [13]SVM in landslide susceptibility mappingSVM, decision trees, logistic regressionSVM outperforms AHP in mapping landslide susceptibility in Fruška Gora Mountain (Serbia)Challenges in integrating complex geological and morphological data into models
Goetz et al. [14]Comparison of ML and statistical models for regional susceptibility mappingLogistic regression, GAM, SVM, random forestRandom forest and bundling show superior predictive performance; challenges with spatial artifactsVariability in model performance across different geological settings
Kavzoglu et al. [18]ML versus traditional statistical methods in landslide susceptibility mappingVarious ML algorithmsML techniques show promise in areas with limited geotechnical data; challenges in model selectionLimited studies on the scalability of ML models for large-scale landslide mapping
Ghorbanzadeh [11]ML methods for landslide detection using remote sensing dataANN, SVM, RF, CNNCNN achieves 78.26% mIOU in landslide detection; variability in CNN effectiveness based on designNeed for better understanding of CNN parameter effects; limitations in training data and augmentation
Chen et al. [19]KLR models for landslide susceptibility mappingKernel logistic regression (PLKLR, PUKLR, and RBFKLR)PUKLR model outperforms others in Zichang City, China; valuable insights for hazard preventionDependence on historical data for model construction; challenges in integrating diverse datasets
Micheletti et al. [15]Adaptive ML techniques for landslide susceptibility mappingSVM, random forest, AdaBoostRandom forest and AdaBoost effective in feature selection; challenges in deep-seated landslide characterizationEfficiency of adaptive SVM in landslide mapping; challenges with adaptive scaling in deep-seated landslides
Table 2. Optimized DNN architectures with GA.
Table 2. Optimized DNN architectures with GA.
#Num LayersNeurons per LayerActivation FunctionsOptimizerAlphaAccuracy
12[71, 126][‘tanh’, ‘tanh’]adagrad0.1000.677789
22[64, 28][‘gelu’, ‘softsign’]adam0.0010.680290
39[72, 94, 60, 64, 55, 106, 37, 74, 9][‘linear’, ‘gelu’, ‘linear’, ‘softsign’, ‘softsign’, ‘sigmoid’, ‘gelu’, ‘softsign’, ‘hard_sigmoid’]nadam0.00010.813857
44[59, 114, 84, 76][‘gelu’, ‘softmax’, ‘softplus’, ‘relu’]rmsprop0.01000.818859
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Tynchenko, Y.; Kukartsev, V.; Tynchenko, V.; Kukartseva, O.; Panfilova, T.; Gladkov, A.; Nguyen, V.; Malashin, I. Landslide Assessment Classification Using Deep Neural Networks Based on Climate and Geospatial Data. Sustainability 2024, 16, 7063. https://doi.org/10.3390/su16167063

AMA Style

Tynchenko Y, Kukartsev V, Tynchenko V, Kukartseva O, Panfilova T, Gladkov A, Nguyen V, Malashin I. Landslide Assessment Classification Using Deep Neural Networks Based on Climate and Geospatial Data. Sustainability. 2024; 16(16):7063. https://doi.org/10.3390/su16167063

Chicago/Turabian Style

Tynchenko, Yadviga, Vladislav Kukartsev, Vadim Tynchenko, Oksana Kukartseva, Tatyana Panfilova, Alexey Gladkov, Van Nguyen, and Ivan Malashin. 2024. "Landslide Assessment Classification Using Deep Neural Networks Based on Climate and Geospatial Data" Sustainability 16, no. 16: 7063. https://doi.org/10.3390/su16167063

APA Style

Tynchenko, Y., Kukartsev, V., Tynchenko, V., Kukartseva, O., Panfilova, T., Gladkov, A., Nguyen, V., & Malashin, I. (2024). Landslide Assessment Classification Using Deep Neural Networks Based on Climate and Geospatial Data. Sustainability, 16(16), 7063. https://doi.org/10.3390/su16167063

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop