Next Article in Journal
Seismic Activity Along the Periadriatic and Sava Faults in the Past Two Millennia—An Archaeoseismological Assessment
Previous Article in Journal
Flood Hazard and Risk in Urban Areas
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Geosciences
Volume 14
Issue 12
10.3390/geosciences14120330
geosciences-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 thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

High Resolution Precipitation and Soil Moisture Data Integration for Landslide Susceptibility Mapping

by
Yaser Peiro
1,2,
Evelina Volpe
1,
Luca Ciabatta
2 and
Elisabetta Cattoni
1,*
1
Department of Theoretical and Applied Sciences, eCampus University, 22060 Novedrate, Italy
2
Research Institute for Geo-Hydrological Protection, National Research Council, 06128 Perugia, Italy
*
Author to whom correspondence should be addressed.
Geosciences 2024, 14(12), 330; https://doi.org/10.3390/geosciences14120330
Submission received: 18 October 2024 / Revised: 18 November 2024 / Accepted: 23 November 2024 / Published: 5 December 2024
(This article belongs to the Section Natural Hazards)
Browse Figures
Versions Notes

Abstract

:
Satellite-derived high-resolution soil moisture and precipitation data have become widely adopted in natural hazard and climate change research. Landslide susceptibility mapping, which often relies on static predisposing factors, faces challenges in accounting for temporal changes, limiting its efficacy in accurately identifying potential locations for landslide occurrences. A key challenge is the lack of sufficient ground-based monitoring networks for soil moisture and precipitation, especially in remote areas with limited access to rain gauge data. This study addresses these limitations by integrating static landslide conditioning factors—such as topography, geology, and landscape features—with high-resolution dynamic satellite data, including soil moisture and precipitation. Using machine learning techniques, particularly the random forest (RF) algorithm, the approach enables the generation of dynamic landslide susceptibility maps that incorporate both spatial and temporal variations. To validate the proposed method, two significant rainfall events that occurred in Italy in October and November 2019—each triggering more than 40 landslides—were analyzed. High-resolution satellite rainfall and soil moisture data were integrated with statistical conditioning factors to identify high-probability landslide areas successfully. A differential susceptibility map was generated for these events to compare the results between them, illustrating how susceptibility variations within the study area are influenced by hydrological factors. The distinct susceptibility patterns associated with different hydrological conditions were accurately captured. It is suggested that future research focus on leveraging time-series high-resolution satellite data to enhance landslide susceptibility assessments further.

1. Introduction

Climate variables and their changes affect landslide activity [1,2,3], leading to an increase in rainfall-induced landslides in many areas of the world. The threat of shallow landslides to people, the environment, structures, and infrastructures is becoming more severe. Landslide susceptibility assessment allows us to predict and demonstrate the probability and distribution of possible landslides in a certain area [4,5,6,7]. The spatial distribution of landslides occurrence at the regional scale [8,9] provides a scientific basis and represents a fundamental step for landslide disaster prevention and regional development planning [10].
To predict where landslides are likely to occur, several methods have been proposed [11]. Geomorphological mapping, analysis of landslide inventories, heuristic approaches to terrain analysis and susceptibility zoning, physically based numerical modeling, and statistically based classification methods are just some of the proposed approaches [12,13,14,15,16,17,18,19].
In addition to the traditional models, multifarious machine learning models have been developed and used to evaluate landslide susceptibility [20,21,22,23].
In recent years, due to the rapid advancement of computer technology, there has been a notable shift towards exploring advanced modeling architectures capable of delivering superior predictive performance. Techniques such as deep learning [24,25,26], which leverages artificial neural networks with multiple layers to extract complex patterns from large and complex datasets, have demonstrated significant potential in landslide susceptibility modeling. By automatically learning hierarchical representations of data, deep learning models can capture intricate spatial and temporal dependencies, leading to more accurate predictions. Ensemble modeling [27], which combines multiple models, such as decision trees, support vector machines, or neural networks, has also gained traction. By aggregating the predictions of diverse models, ensemble methods can reduce overfitting and improve generalization performance. Hybrid approaches that integrate multiple techniques, such as statistical and machine learning methods, have shown promise in capturing the intricate factors influencing landslide occurrence. By combining the strengths of different methods, hybrid approaches can provide more comprehensive and accurate predictions [28,29]. Among these techniques, the random forest (RF) machine learning algorithm is particularly promising. This model operates by constructing an ensemble of decision trees from randomly selected subsets of the input data, combining their predictions to derive a final output. RF demonstrated effectiveness in addressing both classification and regression tasks [30,31]. The strength of the RF model lies in its ability to use the combined knowledge of individual trees while mitigating the risk of overfitting [32].
In addition, the RF model is particularly advantageous in landslide susceptibility assessment, due to its capacity to include different types of data [33]. Both numerical and categorical variables, covering factors such as slope angle, lithology, and land use, can be seamlessly integrated without forcing strict distributional assumptions. This flexibility makes the RF model well-suited to analyzing the complex interaction of factors contributing to landslides occurrence.
The main factors that induce shallow landslides are divided into two categories: static and dynamic. Static factors, also known as conditioning factors, include topographic features, geological conditions, and land cover. Dynamic factors, also known as triggering factors, are the rainfall and soil moisture conditions.
The assessment of triggering factors is important for the reliability of susceptibility maps. Landslide susceptibility modeling has typically focused on static information [34,35,36,37,38], while the dynamic influence of rainfall and soil moisture have often been ignored [11,39]. Recent advancements in data-driven solutions allowed for the incorporation of spatiotemporal variations of these variables, leading to more accurate and comprehensive landslide hazard assessments [40]. By incorporating explanatory variables that reflect spatiotemporal variations, such as rainfall and soil moisture parameters, the models can capture the complex interactions between environmental factors and landslide triggering mechanisms [41,42].
The evolution of data-driven landslide susceptibility models has been marked by an increasing sophistication in methodological approaches and a growing recognition of the importance of spatiotemporal dynamics [33,37,41,43,44,45,46].
The space-time solution approaches explicitly account for the temporal dimension of landslide occurrences, incorporating dynamic variables like rainfall and antecedent hydrological conditions to capture the evolving nature of landslide hazards [47]. These methodologies allow for the simultaneous assessment of landslide risk across both geographic space and time, enabling a more comprehensive understanding of the dynamic nature of landslide occurrences [48].
High-resolution soil moisture and precipitation data have become essential in understanding and predicting landslide events, especially in regions prone to rainfall-induced landslides [49,50,51,52].
Soil moisture and precipitation data can be derived from in situ measurements, but they have limitations. For example, the accuracy of both rainfall intensity duration and cumulative rainfall duration (E-D) models is influenced by various factors, such as the resolution of rainfall and geohazard data, the spatial relationship between rain gauges and landslide locations, and the criteria used to define threshold limits [53,54]. These approaches typically involve calculating rainfall parameters for each landslide event based on data from relevant rain gauges [55]. The selection of appropriate rain gauges remains a challenge, with no universally accepted methodology [56]. In addition, the spatial limitations of rain gauges hinder the accurate assessment of rainfall and soil moisture conditions in close proximity to landslide sites.
A promising solution to overcome these limitations is offered by satellite-derived data. By leveraging remote sensing technologies, researchers can obtain comprehensive, high-resolution rainfall and soil moisture data over large areas, providing a more accurate and spatially continuous representation of the environmental conditions that influence landslide occurrence [57,58,59,60]. The integration of these satellite-derived data into landslide prediction models has the potential to improve the accuracy and reliability of landslide forecasts, thereby enhancing the ability to mitigate the risks associated with these natural hazards.
In this study, the limitations of raingauge-based susceptibility models are addressed by integrating high-resolution satellite-derived soil moisture and precipitation data with advanced machine learning techniques. By combining these powerful tools, a more sophisticated model that captures the intricate relationships between the various factors is developed, leading to a deeper understanding of the critical conditions under which landslides are triggered.
The document is organized as follows. Section 2 provides a detailed description of the study area, defining the main geological and climatic features. An overview of the static and dynamic factors considered in the study is reported in Section 3. Section 4 describes the machine learning method used, and Section 5 presents the results of the application of the model to a selected study area and the discussion related to the 2019 severe weather events which triggered a significant number of landslides in the area.

2. Study Area

2.1. Geological and Geomorphological Setting

The studied area covers 9500 km2 between the Piemonte, Lombardia, Liguria, and Emilia Romagna regions (Figure 1A). The area has experienced numerous landslide events in the past, indicating a high susceptibility to landslides. The red dots in Figure 1B represent the approximate center locations of these landslides. For landslide events, a detailed inventory is available and the date and location of each event occurrence are detailed. From January 2016 to December 2021, 410 phenomena were recorded in the area.
The geology of the area consists of the alluvial, lacustrine, and marine deposits in the central part of the area, while an extensive area of bedrock is located in the northern zone [61] (see Figure 2). The northwestern part is primarily covered by unconsolidated clastic rock, comprising loosely arranged and uncemented materials of all grain sizes with a heterogeneous origin, such as clay soil, sand, and conglomerate.
The southern sectors are dominated by hilly reliefs composed of carbonate-rich sedimentary rocks (limestone, dolomite, and marl) and chaotic terrains with a clay matrix. In contrast, the southwest part is characterized by marlstone and schistose metamorphic rocks, encompassing a wider variety of rock types.
The digital elevation model with a 10 m resolution reveals that the territory shows different morphology, ranging from flat areas such as Pianura Padana by the Po River, to peaks surpassing 1700 m (the highest peak being Falterona Mount at 1655 m a.s.l.).
Geosciences 14 00330 g001
Figure 1. (A) Localization of the study area (in orange) within Italy; (B) spatial distribution of landslides (red dots) in the study area (2017–2022) [62,63].
Figure 1. (A) Localization of the study area (in orange) within Italy; (B) spatial distribution of landslides (red dots) in the study area (2017–2022) [62,63].
Geosciences 14 00330 g001
Geosciences 14 00330 g002
Figure 2. Lithology distribution in the study area [60].
Figure 2. Lithology distribution in the study area [60].
Geosciences 14 00330 g002

2.2. Climatic Features

The study area is located in the Po River basin. It is influenced by the Alps, which protect the Po Valley from cold winds from north Europe and by the Apennines limit, which mitigates the action of the sea [64]. The lowest temperatures are recorded in January, followed by a rise until July when the peak is reached, and then a decline from September to December, with values similar to those in January. Winter (December to February) is the coldest season, while summer (June to August) is the warmest. Autumn (September to November) tends to have slightly higher temperatures compared to spring (March to May) [64].
According to the Po River Basin Authority [64], the average annual precipitation over the Po River basin is approximately 1200 mm at the Pontelagoscuro closure section. Of this total, about two-thirds flows to the Adriatic Sea as discharge, while the remaining one-third is believed to be lost due to evaporation and/or vegetation interception.
Analyzing rainfall since 1980 shows that rainfall events have been more intense but less frequent, resulting in a 20% reduction in total annual rainfall [65].
Seasonally, rainfall events have been reduced by up to 50% in spring and summer, while they have varied little in autumn.

3. Materials

Data acquisition for landslide susceptibility analysis began with the collection of raw data from various sources, including digital terrain models (DTMs), geological maps, Sentinel-2 satellite data (Sentinel-2, a key part of the Copernicus Programme, provides high-resolution images for monitoring Earth’s surface changes), and satellite-based rainfall and soil moisture products. The collected data are detailed in Table 1 where the resolution and sources of each dataset are provided. Ten static conditioning factors were selected and derived from the collected data: elevation, slope angle, aspect, plan curvature, profile curvature, geology, land cover, normalized difference vegetation index (NDVI), distance to road, and distance to river. These static factors have been commonly used by previous researchers in studying landslides (see e.g., [9,66]).
In addition to the static conditioning factors, daily rainfall and daily soil moisture data, along with two antecedent cumulative rainfall parameters (7 days and 15 days before the events to capture respectively short-term rainfall effects and longer moisture build-up), were selected to characterize soil moisture and rainfall conditions that trigger landslides during rainstorms.

3.1. Static Conditioning Factors

Landslide occurrence is directly impacted by the angle of the slope, a crucial factor contributing to shear stress and subsequent terrain instability [67]. The digital terrain model (DTM) provides insights into regional topography and geomorphology, reflecting the influence of slope’s geographical features on landslide evolution. For this reason, five widely used slope characteristics—elevation, slope angle, aspect, plan curvature, and profile curvature—are calculated correspondingly. For instance, aspect indicates the direction a slope faces (representing the compass direction 0° to 360°), influencing sunlight exposure, moisture content, vegetation growth, and erosion rates. Slopes with certain aspects may retain more moisture, have less vegetation, or experience rapid snowmelt and freeze–thaw cycles. Slopes with specific aspects, such as north-facing slopes in temperate regions, may retain more moisture due to reduced solar radiation, leading to increased soil saturation and potential landslide initiation. Conversely, south-facing slopes may experience more rapid snowmelt and freeze–thaw cycles, which can weaken soil structure and contribute to slope instability. Additionally, the aspect can influence vegetation cover, with certain aspects supporting more vegetation growth, which can help stabilize slopes through root reinforcement [68].
Plan curvature and profile curvature represent the effects of topographic gradient on flow rate and water flow patterns, respectively. Curvature, generally, describes a surface’s deviation from flatness [69].
For this analysis, we used QGIS version 3.34.2-Prizren [70], a free and open-source geographic information system (GIS) software. QGIS was employed to calculate and map topographic and geomorphological factors. Its user-friendly interface and robust tools for spatial data analysis, visualization, and modeling make it an ideal choice for landslide susceptibility assessments. Figure 3 illustrates the distribution of elevation (Figure 3A) and slope angle (Figure 3B) across the study area.
To define the lithology distribution, the classification proposed by Bucci et al. 2022 [61] was adopted (Figure 2). The distance to the nearest river map (Figure 4) was derived using the Google Earth Engine (GEE) platform. A digital elevation model (DEM) and the HydroSHEDS [71,72] flow accumulation dataset were utilized to calculate the distance to the nearest river for each point within the study area. The flow accumulation dataset, representing the accumulated flow over a given area, was used as a proxy for river networks. By taking the square root of the flow accumulation and dividing by a scaling factor, we obtained a distance measure. This derived distance map was then incorporated into the landslide susceptibility analysis.
Land cover, particularly vegetation, plays a crucial role in influencing slope stability. Vegetation intercepts rainfall, reducing surface runoff and infiltration rates, and its roots bind soil particles, enhancing soil cohesion and shear strength. These factors significantly mitigate the risk of landslides, especially those triggered by rainfall [73,74]. To assess vegetation cover and health, the normalized difference vegetation index (NDVI) was utilized, derived from Copernicus Sentinel-2 satellite data. This remote sensing technique provides valuable information about vegetation growth and vigor. The NDVI data underwent standard preprocessing steps to ensure data quality and accuracy. These steps included atmospheric correction to remove the influence of atmospheric conditions and cloud masking to identify and exclude cloudy pixels. Subsequently, the NDVI was calculated using a well-established formula that considers the difference between near-infrared and red reflectance values. This calculation allows for the quantification of vegetation density and health. The specific source and resolution of the NDVI and land cover data are provided in Table 1. In particular, the land cover classification in Figure 5 was obtained from the Corine Land Cover (CLC) dataset (see Table 1), which distinguishes different land cover types based on spectral reflectance properties and satellite-derived indices. Forests and semi-natural areas are identified by high NDVI values, reflecting dense vegetation with strong chlorophyll absorption and high near-infrared reflectance. Agricultural surfaces show moderate NDVI values, indicating seasonal or less dense vegetation, often varying with crop cycles. Water bodies are highlighted using NDWI, as water strongly absorbs in the near-infrared spectrum but reflects in the visible range, resulting in low NDVI and high NDWI values. Artificial surfaces are characterized by low NDVI and NDWI values combined with specific spectral reflectance patterns typical of urban and built-up areas. This robust classification ensures accurate and reproducible differentiation of land cover types.
The static and dynamic factors considered in this study for the landslide susceptibility assessment are listed in Table 1 along with the data sources.

3.2. Dynamic Conditioning Factors

Rainfall and soil saturation conditions at the onset of a storm event are two of the main triggering factors for landslides activation [75,76]. An increase in the saturation creates an increase in the stresses within the soil layer, reducing slope stability [77]. Despite the study area being densely instrumented, with more than 100 rain gauges operationally working, localized events as well as local features may impact the rainfall and soil saturation observed fields. Moreover, over the study area, soil moisture monitoring networks were not present and modeled estimates could be impacted by the quality of the input rainfall. The use of hydrometeorological variables obtained through remotely sensed information could mitigate such limitations. To this end, in this study, soil saturation and rainfall data obtained through remotely sensed information were used to train the proposed model. Rainfall data were obtained through the integration of multiple datasets in order to achieve a more reliable and more resolute field. Specifically, the state-of-the-art Integrated Multi-satellitE Retrievals for GPM Late Run product [78], the Climate Prediction Center Global Unified Gauge-Based Analysis of Daily Precipitation [79], and the soil moisture-derived rainfall through SM2RAIN algorithm [80] were integrated and then downscaled to a 1 km of spatial resolution. Table 2 summarizes the main features of the parent products.
The integration of the parent products was carried out by estimating the signal-to-noise ratio (SNR) of each input dataset through triple collocation analysis [81]. The analysis allowed for obtaining a set of integration weights that are used to create a weighted average of the three parent products. Then, through the use of CHELSA climatology [82], the coarse resolution product was downscaled to 1 km of spatial resolution. For more details about the integration and the downscaling procedures, readers are referred to Filippucci et al. (2022) [83]. The soil moisture data were obtained through the GLEAM model [79].
GLEAM is a modeling framework dedicated to the estimation of evaporation over land through satellite data. The model estimates the different components of evapotranspiration, i.e., transpiration, bare-soil evaporation, interception loss, open-water evaporation, and sublimation along with surface and root-zone soil saturation conditions. Specifically, soil moisture data were obtained by taking advantage of a multi-layer balance model that can assimilate satellite observations. The estimates used in the current study were those obtained by using the model with the same product used for retrieving the rainfall conditions, guaranteeing the same temporal and spatial resolutions.

4. Method

To assess landslide susceptibility in the area, the random forest (RF) model was implemented. The first part of the Methodology section (Section 4.1) provides a detailed explanation of the RF algorithm, including its core principles and parameter optimization techniques to enhance model accuracy and efficiency. The dataset was partitioned into training and testing subsets, with the training data being used to construct the decision tree ensemble, and the model’s predictive accuracy being evaluated with the testing data.
The second part (Section 4.2), Feature Impact Analysis, focuses on evaluating the effect of each input feature on the RF model’s performance. The statistical relationships between independent variables were analyzed, and the relative importance of conditioning factors in influencing landslide susceptibility was determined.

4.1. Random Forest Algorithm

The RF model was implemented using the randomForest package in the R 4.6.0 statistical software environment [30]. Input data included approximately 400 landslide and 400 non-landslide locations, with features like slope angle, lithology type, elevation, and rainfall. During training, the RF algorithm built multiple decision trees, each with varied splits based on these features. For example, one tree might start with slope angle as the root, dividing locations with slopes over 30° as more likely to experience landslides. The next node could consider lithology type, with sedimentary rock areas deemed at higher risk. In another tree, elevation could be the root, splitting locations over 1000 m as more susceptible, followed by nodes assessing recent rainfall amounts. This variety in tree construction enables the model to capture complex patterns associated with landslide susceptibility based on observed data from our study area.
Once the trees were built, new data points could be evaluated to assess the probability of a landslide occurring at those locations. The features of the new points—such as slope angle, lithology, elevation, and rainfall—were processed by each tree in the forest to provide a binary output, either yes (indicating a landslide is likely) or no (indicating it is not). The final output was determined by calculating the percentage of trees that predicted yes. For instance, if 70 out of 100 trees predicted yes, a probability of 0.7 (or 70%) for a landslide was assigned by the model. Selecting the optimal train/test split ratio is crucial for improving model performance. A well-defined split ratio reduces the risk of overfitting and enhances the model’s ability to perform effectively on unseen data. To ensure balanced data, 410 non-landslide samples, equal to the number of landslide samples, were utilized in the study. The proportion of training data varied from 0.5 to 0.9, which resulted in an increase in the number of landslide samples considered for model training from 200 to 360.
The classification accuracy of the RF model on the testing dataset, evaluated at different train/test split ratios, is shown in Figure 6A. It was observed that as the training subset ratio increased, the model’s classification accuracy on the testing dataset also improved, peaking at a 0.7 (70/30) split. This finding aligns with the recommendations of Gholamy et al. (2018) [84], who advocate for a 70/30 split as optimal for model training and testing in similar contexts.
To minimize the out-of-bag error rate of bootstrap samples, the hyperparameters ‘ntree’ (number of trees) and ‘mtry’ (number of predictor variables considered at each split) were fine-tuned and optimized for each dataset using the model’s tuning function, as illustrated in Figure 6B,C.
Four key performance indicators were used to evaluate the effectiveness of the random forest (RF) model (Figure 6D): accuracy, recall, precision, and AUC (area under the receiver operating characteristic (ROC) curve). To understand these metrics, we first defined the four prediction types in binary classification for landslide prediction:
  • True positive (TP): a landslide is correctly predicted where one occurred;
  • False positive (FP): a landslide is incorrectly predicted when none occurred;
  • True negative (TN): no landslide is correctly predicted when none occurred;
  • False negative (FN): no landslide is incorrectly predicted when one occurred.
Using these definitions, the following metrics were derived:
  • Accuracy measures the proportion of all correct predictions (both TP and TN) over the total predictions, reflecting the model’s overall performance. A high accuracy score, as shown in Figure 6D, indicates that the model is effective in identifying both landslide and non-landslide instances.
  • Recall (also known as sensitivity or true positive rate) is the proportion of actual landslides that are correctly predicted, defined as TP/(TP + FN). High recall suggests that landslide events are effectively captured by the model, reducing missed occurrences.
  • Precision represents the proportion of correctly predicted landslides among all positive predictions (TP/(TP + FP)). High precision, as shown in the model’s results, indicates that most landslides predicted by the model are accurate, reducing false positives.
  • AUC (area under the receiver operating characteristic (ROC) curve) reflects the model’s ability to distinguish between landslide and non-landslide instances. An AUC close to 1 suggests high discriminative power, meaning the model effectively separates landslide-prone areas from stable regions.
Together, these metrics indicate that the RF model performed with high accuracy, recall, and precision, and that the AUC confirmed its strong classification performance. Figure 6D presents the values for these metrics, demonstrating the model’s capability to reliably identify landslide-prone areas.

4.2. Feature Impact Analysis

In the second part of evaluating the performance of the RF, the statistical relationships between independent variables were assessed to determine potential correlation issues, prior to the conditioning factors importance analysis being investigated. Spearman’s correlation matrix [85] was employed for this purpose (Figure 3), due to its suitability for both continuous and ordinal data [86] being recognized. Multicollinearity, indicated by a high degree of relationship between variables, was identified by Spearman correlation values exceeding 0.7. Figure 7 reveals strong correlations among daily and antecedent rainfall factors (e.g., 7 day and 15 day antecedent rain). Such correlations have been shown to have minimal impact on the predictive performance of machine learning models [87], but they were still taken into consideration.
The RF model assessed the relative importance of conditioning factors for landslide susceptibility prediction using mean decrease accuracy (MDA) and mean decrease GINI (MDG), a criterion designed to improve class purity, used to determine the best slit at each node. MDA quantifies the reduction in model accuracy when a factor is excluded, while MDG measures a factor’s contribution to node purity. Higher values of MDA and MDG indicate greater importance for a given factor (Figure 8).
All selected conditioning factors were observed to have positive values for landslide prediction model learning, although their degree of contribution differed. Specifically, among static topographic factors, elevation was noted to be of the highest importance, followed by slope angle, profile, and plan curvature. Surface cover materials, indicated by land cover and NDVI, were found to play a more significant role than subsurface materials when compared to geological factors.
Regarding dynamic rainfall conditions, daily rainfall was identified as more important than antecedent rainfall factors (i.e., 7 day and 15 day). This is consistent with the understanding that most landslides in Italy are classified as shallow, primarily triggered by short, intense rainstorms. Within these factors, soil moisture was determined to have a relatively greater influence than antecedent conditions.
Distance to road and geology were found to be the least significant among the 14 selected conditioning factors. However, it should be noted that the determined importance of road distribution in the model predictions may have been affected by the exclusion of human-made landslide triggering events from the input landslide dataset. The geological features were not found to be as helpful in predicting landslides as other factors. This could be because landslides were observed in areas with very different types of rocks and soil, making it difficult to find a single geological cause. It was suggested that separating the geology data into smaller groups and considering the effects of weathering might make the predictions better.
Although the removal of conditioning factors with null predictive value has been suggested in some studies, all 14 landslide conditioning factors in this study were found to have positive predictive capability. Furthermore, it has been indicated in literature (see e.g., [33]) that the importance of a single conditioning factor like slope angle can be site-specific and dependent on the scale of analysis and selection method. Therefore, all 14 conditioning factors were utilized in the analysis for the development of the model.

5. Prediction of Landslide Susceptibility in the Case Study and Discussion of the Results

This section presents the results of the landslide susceptibility assessment conducted using the random forest model, focusing on two major rainfall events that occurred in northern Italy in 2019. These events, marked by intense rainfall and resulting landslides, had significant impacts on infrastructure and communities.
The spatial distribution of rainfall, soil moisture, and landslide occurrences was analyzed to evaluate the model’s ability to identify susceptible areas. Predicted susceptibility maps were compared with actual landslide locations to investigate the influence of various hydrological factors, such as rainfall and soil moisture, on landslide susceptibility.

5.1. Rainfall Events

To evaluate the susceptibility map, two major rainstorms with the highest number of induced landslides were selected for analysis. The rainfall hydrogram, obtained from the procedures developed by CNR-IRPI in Perugia, is shown in Figure 9.
The first event (R1) occurred on 20 October 2019. It brought intense rainfall and subsequent landslides to northwestern Italy, causing widespread disruption and displacement. Locations such as San Bartolomeo del Bosco, Osigli, and areas surrounding Genoa were particularly affected. The consequences included road closures, isolated communities, evacuations, and disruptions to public transportation. Infrastructure damage was also reported, with a bridge collapsing in Capriata d’Orba. In total, this rainstorm resulted in 39 natural terrain landslides and 2 fatalities, with a maximum reported rainfall of 73 mm. Figure 10, besides showing the values of maximum daily rainfall, illustrates the average daily rainfall in the studied area and the antecedent 15 day cumulative average rainfall over the study area between October and late January.
The second event (R2), on 23 November 2019, Impacted a broader range of areas from the northeast to the south of the studied region. This event led to significant financial and human losses, particularly in the Liguria area, resulting in 45 natural terrain landslides and directly affecting more than 120 people. As shown in Figure 9, the maximum rainfall amounts in both events were similar, reaching close to 70 mm, but the cumulative average rainfall on 23 November reached higher values compared to the event in October.

5.2. Modeling Results

Using the collected data, which included conditioning factors in the study area as well as values extracted from satellite data (including daily rainfall, soil moisture, 7 and 15 day cumulative rainfall at each point), two landslide susceptibility maps were created for the two events under review using the optimized RF model. Figure 10D and Figure 11D, respectively, show the predicted landslide susceptibility results for these two events. The locations of landslides triggered by each rainfall event are indicated in both figures. From the predicted landslide susceptibility and the spatial distribution of rainfall and soil moisture parameters for the two events (Figure 9 and Figure 10), some interesting results can be identified.
In the October event, the concentration of rainfall in the northern and central sectors of the area can be clearly observed in Figure 10A, and to better understand this, the points where landslides occurred on that day are marked. By comparing daily rainfall and examining the 15 day cumulative rainfall, a spatial shift in rainfall concentration from the central areas to the western and southern parts of the study area (Figure 10B) can be observed. The concentration of maximum rainfall was in the northern part of the study area, where no significant landslides were reported. Conversely, by analyzing the cumulative rainfall distribution for this event, as well as the distribution of soil moisture, a clear alignment with the locations of landslide occurrences can be observed. This 15 day rainfall concentration in the central areas could have a very interesting correlation with the spatial distribution of soil moisture values on the day of the event (Figure 10C). By examining the predicted landslide susceptibility results for October 2019, a meaningful correlation can be observed where both predicted high-risk landslide zones and the concentration of the 15 day rainfall and soil moisture values are located. Soil moisture and rainfall conditions play an important role in spatially identifying landslides and can vary significantly with different rainfall events. This difference can be observed in the spatial distribution of maximum rainfall on this day and the location of landslides. On 20 of October, the maximum rainfall was 70 mm, which occurred in the northern part of the study area, but the landslides occurred in the central areas. These results clearly emphasize the importance of considering rainfall and soil moisture factors in accurately understanding and identifying high-risk areas.
As shown in Figure 11, on 23 November the intensity of rainfall was almost the same as on 20 October, with both experiencing approximately 70 mm of rainfall. However, on 23 November, the spatial distribution of rainfall was concentrated in the western and southwestern regions and also affected a larger area from north to south. This difference can also be observed in the average daily rainfall presented in Figure 9 for these two events. On this day, the average rainfall across the entire area was nearly twice the reported average rainfall for 20 October. By examining the distribution of landslides that occurred on 23 November and the spatial distribution of soil moisture and rainfall on that day, the accuracy of the model in predicting high-susceptibility areas can be recognized. Similarly, the maximum predicted susceptibility index on 23 November is higher than on 20 October (0.93). Although these calculated susceptibility values may not be quantitatively accurate, it has been proven that the proposed model can correctly consider various intensities of soil moisture and rainfall-related parameters. For 23 November, the high-susceptibility area predicted was not concentrated in a specific region but was able to identify points that were in the eastern part of the study area, where the spatial distribution of soil moisture and intense rainfall accurately predicted the location of landslides. The proposed spatiotemporal landslide prediction model effectively accounted for the impacts of rainfall and soil moisture with distinct spatial distributions, aligning with the principles described by Bogaard et al. [88] and Greco et al. [89]. These studies emphasized that, for shallow landslides, antecedent soil moisture conditions regulate rainwater infiltration into the soil, ultimately triggering landslide events.
Moreover, the findings of this study are consistent with the work of Moreno [90], who reported the significance of a 14 day cumulative precipitation parameter in landslide susceptibility assessments. Similarly, Smith et al. (2023) [91] identified a 10 day pre-event rainfall accumulation as a critical factor influencing landslide occurrences.

5.3. Reliability Results

To better understand the presented model, it is essential to investigate the model’s ability to differentiate and assess the impact of hydrological factors on the predicted values of landslide susceptibility. To this end, the susceptibility difference values for each pixel in two events (Figure 12), along with the difference values of daily rainfall (Figure 13A), 15 day cumulative rainfall (Figure 13B), and soil moisture (Figure 14), were extracted and presented.
In the eastern part of the study area, higher daily and 15 day cumulative rainfall values were observed in November compared to the 20 October event, leading to an increase in areas with a high probability of landslides (red dots) in the eastern regions. Additionally, an examination of the obtained soil moisture values for this event compared to the previous one (Figure 14) indicates higher soil moisture in the second event, thus increasing the probability of landslides in this area. As reported in the landslide database, four landslides occurred in these areas on this day, all of which were correctly classified by the model as very high risk.
However, a noteworthy point in this area is that in high-altitude regions like Monte Carmo, located in the southeast, the measured moisture values at the peak were higher in the first event than in the second event, yet the susceptibility values measured in this area were higher in the second event. To investigate the causes of this phenomenon, Figure 13, which respectively shows the daily (Figure 13A) and 15 day cumulative rainfall (Figure 13B) in this area, can be examined. It can be observed that the cumulative rainfall in this area was approximately 240 mm higher in the second event than in the corresponding 15 day period of the first event. This demonstrates the model’s ability to consider multiple hydrological factors simultaneously in identifying areas with a high probability of landslides. However, upon examining the susceptibility values of these two events in these areas, it is noted that although the differences in susceptibility values seem significant, Figure 12 reveals that these areas, due to their specific landscape characteristics, have a low probability of landslides.
Furthermore, on 20 October, an analysis of the central and southern parts of the study area revealed higher soil moisture values (Figure 14) compared to the second event (R2).
Additionally, by examining the reported landslides on this day and analyzing the cumulative rainfall and spatial distribution pattern of daily rainfall in this area, it can be concluded that the model was able to identify the high-probability areas in this region (yellow dots in Figure 12). By considering only daily rainfall values for the 20 October event (maximum rainfall in the northern and central regions, see Figure 13A) without considering the other dynamic parameters like soil moisture and 15 day cumulative rainfall, this prediction would have been less accurate.
On 23 November, a large number of landslides occurred in the western part of the study area. Examination of the daily rainfall distribution on this day reveals a concentration of rainfall in the western and southern parts of the study area (Figure 11A). The extracted soil moisture values for this region also show higher values in the landslide-prone areas compared to the first event (Figure 14). In this region, there is a wider distribution of areas with a high probability of landslides compared to the 20 October event, indicating the successful performance of the model in identifying high-risk areas considering rainfall and moisture scenarios.
In the northwestern part of the study area, a higher probability of landslides was predicted for the 23 November event compared to 20 October, despite lower soil moisture values on this day compared to 20 October. This can be well explained by examining the daily rainfall distribution in this area, which shows higher rainfall values on 23 November. Considering this, the model was able to predict areas with a higher risk class for this day. It is also worth mentioning that the 15 day cumulative rainfall values were higher in the second event compared to the first event (Figure 13B).

6. Conclusions

Since the 1980s, hundreds of studies have contributed to the understanding of landslide susceptibility across various geological, climatic, and geographical settings. A variety of methods, both direct and indirect, have been employed, incorporating both qualitative and quantitative approaches.
In this study, a method for predicting landslide probability using the RF algorithm was presented. Significant predictive accuracy was achieved by integrating static landslide conditioning factors with high-resolution dynamic variables, such as soil moisture and both daily and cumulative rainfall derived from satellite data. An area along the Po River region in Italy was analyzed, and the model, validated using two rainstorm events from 2019, successfully identified areas with a high probability of landslides by considering multiple hydrogeological factors and effectively captured the distinct characteristics of rainfall and soil moisture distribution and intensity. The predicted high probability zones closely aligned with the observed spatial distribution of landslides, particularly during severe rainstorms.
While a general methodology for spatial and temporal landslide prediction was introduced, the results are site-specific. It is recommended that the method be recalibrated for other applications to identify the optimal configuration of parameters, especially when considering different types of landslides or dynamic variables. Future work could focus on examining the relationship between the time series of dynamic variables—rainfall and soil moisture—and landslide occurrences within specified time windows.

Author Contributions

Conceptualization, Y.P., E.V., L.C., E.C.; methodology, Y.P., E.V., L.C., E.C; formal analysis, Y.P., investigation, Y.P., E.V.; data curation, Y.P., E.V., L.C., E.C.; writing—original draft preparation, Y.P., E.V.; writing—review and editing, L.C., E.C. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Data Availability Statement

Data are contained within the article.

Acknowledgments

This work was supported by the European Space Agency “4DMed-Hydrology” project (contract n. 4000136272/21/I-EF). The authors would also like to thank the municipality of Castel San Vincenzo, Isernia, (Italy) for expressing interest in the findings of this research.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Ardizzone, F.; Gariano, S.L.; Volpe, E.; Antronico, L.; Coscarelli, R.; Manunta, M.; Mondini, A.C. A Procedure for the Quantitative Comparison of Rainfall and DInSAR-Based Surface Displacement Time Series in Slow-Moving Landslides: A Case Study in Southern Italy. Remote Sens. 2023, 15, 320. [Google Scholar] [CrossRef]
  2. Volpe, E.; Gariano, S.L.; Ciabatta, L.; Peiro, Y.; Cattoni, E. Expected Changes in Rainfall-Induced Landslide Activity in an Italian Archaeological Area. Geosciences 2023, 13, 270. [Google Scholar] [CrossRef]
  3. Salciarini, D.; Brocca, L.; Camici, S.; Ciabatta, L.; Volpe, E.; Massini, R.; Tamagnini, C. Physically based approach for rainfall-induced landslide projections in a changing climate. Proc. Inst. Civ. Eng.-Geotech. Eng. 2019, 172, 481–495. [Google Scholar] [CrossRef]
  4. Volpe, E.; Ciabatta, L.; Salciarini, D.; Camici, S.; Cattoni, E.; Brocca, L. The Impact of Probability Density Functions Assessment on Model Performance for Slope Stability Analysis. Geosciences 2021, 11, 322. [Google Scholar] [CrossRef]
  5. Mwakapesa, D.S.; Lan, X.; Mao, Y. Landslide susceptibility assessment using deep learning considering unbalanced samples distribution. Heliyon 2024, 10, e30107. [Google Scholar] [CrossRef]
  6. Duan, G.; Zhang, J.; Zhang, S. Assessment of Landslide Susceptibility Based on Multiresolution Image Segmentation and Geological Factor Ratings. Int. J. Environ. Res. Public Health 2020, 17, 7863. [Google Scholar] [CrossRef]
  7. Gao, J.; Shi, X.; Li, L.; Zhou, Z.; Wang, J. Assessment of Landslide Susceptibility Using Different Machine Learning Methods in Longnan City, China. Sustainability 2022, 14, 16716. [Google Scholar] [CrossRef]
  8. Trigila, A.; Iadanza, C.; Esposito, C.; Scarascia-Mugnozza, G. Comparison of Logistic Regression and Random Forests techniques for shallow landslide susceptibility assessment in Giampilieri (NE Sicily, Italy). Geomorphology 2015, 249, 119–136. [Google Scholar] [CrossRef]
  9. Chen, W.; Li, W.; Chai, H.; Hou, E.; Li, X.; Ding, X. GIS-based landslide susceptibility mapping using analytical hierarchy process (AHP) and certainty factor (CF) models for the Baozhong region of Baoji City, China. Environ. Earth Sci. 2016, 75, 63. [Google Scholar] [CrossRef]
  10. Dou, J.; Yunus, A.P.; Bui, D.T.; Merghadi, A.; Sahana, M.; Zhu, Z.; Chen, C.-W.; Khosravi, K.; Yang, Y.; Pham, B.T. Assessment of advanced random forest and decision tree algorithms for modeling rainfall-induced landslide susceptibility in the Izu-Oshima Volcanic Island, Japan. Sci. Total Environ. 2019, 662, 332–346. [Google Scholar] [CrossRef]
  11. Reichenbach, P.; Rossi, M.; Malamud, B.D.; Mihir, M.; Guzzetti, F. A review of statistically-based landslide susceptibility models. Earth-Sci. Rev. 2018, 180, 60–91. [Google Scholar] [CrossRef]
  12. Aleotti, P.; Chowdhury, R. Landslide hazard assessment: Summary review and new perspectives. Bull. Eng. Geol. Env. 1999, 58, 21–44. [Google Scholar] [CrossRef]
  13. Guzzetti, F.; Carrara, A.; Cardinali, M.; Reichenbach, P. Landslide hazard evaluation: A review of current techniques and their application in a multi-scale study, Central Italy. Geomorphology 1999, 31, 181–216. [Google Scholar] [CrossRef]
  14. Carrara, A. Multivariate models for landslide hazard evaluation. Math. Geol. 1983, 15, 403–426. [Google Scholar] [CrossRef]
  15. Salciarini, D.; Morbidelli, R.; Cattoni, E.; Volpe, E. Physical and numerical modelling of the response of slopes under different rainfalls, inclinations and vegetation conditions. Riv. Ital. Di Geotec. 2022, 1229, 47–61. [Google Scholar] [CrossRef]
  16. Salciarini, D.; Volpe, E.; Cattoni, E. Probabilistic vs. Deterministic Approach in Landslide Triggering Prediction at Large–Scale. In Geotechnical Research for Land Protection and Development; Calvetti, F., Cotecchia, F., Galli, A., Jommi, C., Eds.; Springer International Publishing: Cham, Switzerland, 2020; pp. 62–70. [Google Scholar] [CrossRef]
  17. Salciarini, D.; Lupattelli, A.; Cecinato, F.; Cattoni, E.; Volpe, E. Static and seismic numerical analysis of a shallow landslide located in a vulnerable area. RIG-2022-2-2 2022, 1236, 5–23. [Google Scholar] [CrossRef]
  18. Johari, A.; Peiro, Y. Determination of stochastic shear strength parameters of a real landslide by back analysis. Int. J. Reliab. Risk Safety Theory Appl. 2021, 4, 7–16. [Google Scholar] [CrossRef]
  19. Charrière, M.; Humair, F.; Froese, C.; Jaboyedoff, M.; Pedrazzini, A.; Longchamp, C. From the source area to the deposit: Collapse, fragmentation, and propagation of the Frank Slide. Geol. Soc. Am. Bull. 2015, 128, B31243.1. [Google Scholar] [CrossRef]
  20. 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]
  21. Goetz, J.N.; 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]
  22. Youssef, A.M.; El-Haddad, B.A.; Skilodimou, H.D.; Bathrellos, G.D.; Golkar, F.; Pourghasemi, H.R. Landslide susceptibility, ensemble machine learning, and accuracy methods in the southern Sinai Peninsula, Egypt: Assessment and Mapping. Nat. Hazards 2024, 1–32. [Google Scholar] [CrossRef]
  23. Sharma, N.; Saharia, M.; Ramana, G.V. High resolution landslide susceptibility mapping using ensemble machine learning and geospatial big data. CATENA 2024, 235, 107653. [Google Scholar] [CrossRef]
  24. Huang, F.; Zhang, J.; Zhou, C.; Wang, Y.; Huang, J.; Zhu, L. A deep learning algorithm using a fully connected sparse autoencoder neural network for landslide susceptibility prediction. Landslides 2020, 17, 217–229. [Google Scholar] [CrossRef]
  25. Yao, J.; Qin, S.; Qiao, S.; Che, W.; Chen, Y.; Su, G.; Miao, Q. Assessment of Landslide Susceptibility Combining Deep Learning with Semi-Supervised Learning in Jiaohe County, Jilin Province, China. Appl. Sci. 2020, 10, 5640. [Google Scholar] [CrossRef]
  26. Tiwari, A.; Shirzaei, M. A novel machine learning and deep learning semi-supervised approach for automatic detection of InSAR-based deformation hotspots. Int. J. Appl. Earth Obs. Geoinf. 2024, 126, 103611. [Google Scholar] [CrossRef]
  27. Bui, D.T.; Shirzadi, A.; Shahabi, H.; Geertsema, M.; Omidvar, E.; Clague, J.; Pham, B.T.; Dou, J.; Asl, D.T.; Ahmad, B.B.; et al. New Ensemble Models for Shallow Landslide Susceptibility Modeling in a Semi-Arid Watershed. Forests 2019, 10, 743. [Google Scholar] [CrossRef]
  28. Pham, B.T.; Prakash, I.; Dou, J.; Singh, S.K.; Trinh, P.T.; Tran, H.T.; Le, T.M.; Van Phong, T.; Khoi, D.K.; Shirzadi, A.; et al. A novel hybrid approach of landslide susceptibility modelling using rotation forest ensemble and different base classifiers. Geocarto Int. 2020, 35, 1267–1292. [Google Scholar] [CrossRef]
  29. Wang, C.; Jia, H.; Zhang, S.; Ma, Z.; Wang, X. A dynamic evaluation method for slope safety with monitoring information based on a hybrid intelligence algorithm. Comput. Geotech. 2023, 164, 105772. [Google Scholar] [CrossRef]
  30. Breiman, L. Random forests. Mach. Learn. 2001, 45, 5–32. [Google Scholar] [CrossRef]
  31. Xu, K.; Zhao, Z.; Chen, W.; Ma, J.; Liu, F.; Zhang, Y.; Ren, Z. Comparative study on landslide susceptibility mapping based on different ratios of training samples and testing samples by using RF and FR-RF models. Nat. Hazards Res. 2024, 4, 62–74. [Google Scholar] [CrossRef]
  32. Cutler, D.R.; Edwards, T.C.; Beard, K.H.; Cutler, A.; Hess, K.T.; Gibson, J.; Lawler, J.J. Random forests for classification in ecology. Ecology 2007, 88, 2783–2792. [Google Scholar] [CrossRef] [PubMed]
  33. Chen, W.; Xie, X.; Wang, J.; Pradhan, B.; Hong, H.; Bui, D.T.; Duan, Z.; Ma, J. A comparative study of logistic model tree, random forest, and classification and regression tree models for spatial prediction of landslide susceptibility. CATENA 2017, 151, 147–160. [Google Scholar] [CrossRef]
  34. Vorpahl, P.; Elsenbeer, H.; Märker, M.; Schröder, B. How can statistical models help to determine driving factors of landslides? Ecol. Model. 2012, 239, 27–39. [Google Scholar] [CrossRef]
  35. Brenning, A. Spatial prediction models for landslide hazards: Review, comparison and evaluation. Nat. Hazards Earth Syst. Sci. 2005, 5, 853–862. [Google Scholar] [CrossRef]
  36. Quevedo, R.P.; Velastegui-Montoya, A.; Montalván-Burbano, N.; Morante-Carballo, F.; Korup, O.; Rennó, C.D. Land use and land cover as a conditioning factor in landslide susceptibility: A literature review. Landslides 2023, 20, 967–982. [Google Scholar] [CrossRef]
  37. Lima, P.; Steger, S.; Glade, T.; Murillo-García, F.G. Literature review and bibliometric analysis on data-driven assessment of landslide susceptibility. J. Mt. Sci. 2022, 19, 1670–1698. [Google Scholar] [CrossRef]
  38. Zhang, J.; Ma, X.; Zhang, J.; Sun, D.; Zhou, X.; Mi, C.; Wen, H. Insights into geospatial heterogeneity of landslide susceptibility based on the SHAP-XGBoost model. J. Environ. Manag. 2023, 332, 117357. [Google Scholar] [CrossRef]
  39. Guzzetti, F.; Reichenbach, P.; Cardinali, M.; Galli, M.; Ardizzone, F. Probabilistic landslide hazard assessment at the basin scale. Geomorphology 2005, 72, 272–299. [Google Scholar] [CrossRef]
  40. Di Napoli, M.; Tanyas, H.; Castro-Camilo, D.; Calcaterra, D.; Cevasco, A.; Di Martire, D.; Pepe, G.; Brandolini, P.; Lombardo, L. On the estimation of landslide intensity, hazard and density via data-driven models. Nat. Hazards 2023, 119, 1513–1530. [Google Scholar] [CrossRef]
  41. Wang, N.; Zhang, H.; Dahal, A.; Cheng, W.; Zhao, M.; Lombardo, L. On the use of explainable AI for susceptibility modeling: Examining the spatial pattern of SHAP values. Geosci. Front. 2024, 15, 101800. [Google Scholar] [CrossRef]
  42. Steger, S.; Moreno, M.; Crespi, A.; Gariano, S.L.; Brunetti, M.T.; Melillo, M.; Peruccacci, S.; Marra, F.; De Vugt, L.; Zieher, T.; et al. Adopting the margin of stability for space–time landslide prediction—A data-driven approach for generating spatial dynamic thresholds. Geosci. Front. 2024, 15, 101822. [Google Scholar] [CrossRef]
  43. Magliulo, P.; Di Lisio, A.; Russo, F. Comparison of GIS-based methodologies for the landslide susceptibility assessment. Geoinformatica 2009, 13, 253–265. [Google Scholar] [CrossRef]
  44. Shano, L.; Raghuvanshi, T.K.; Meten, M. Landslide susceptibility evaluation and hazard zonation techniques—A review. Geoenviron Disasters 2020, 7, 18. [Google Scholar] [CrossRef]
  45. Nocentini, N.; Rosi, A.; Segoni, S.; Fanti, R. Towards landslide space-time forecasting through machine learning: The influence of rainfall parameters and model setting. Front. Earth Sci. 2023, 11, 1152130. [Google Scholar] [CrossRef]
  46. Dahal, A.; Tanyas, H.; Van Westen, C.; Van Der Meijde, M.; Mai, P.M.; Huser, R.; Lombardo, L. Space–time landslide hazard modeling via Ensemble Neural Networks. Nat. Hazards Earth Syst. Sci. 2024, 24, 823–845. [Google Scholar] [CrossRef]
  47. Lee, J.-J.; Song, M.-S.; Yun, H.-S.; Yum, S.-G. Dynamic landslide susceptibility analysis that combines rainfall period, accumulated rainfall, and geospatial information. Sci. Rep. 2022, 12, 18429. [Google Scholar] [CrossRef]
  48. Ahmed, M.; Tanyas, H.; Huser, R.; Dahal, A.; Titti, G.; Borgatti, L.; Francioni, M.; Lombardo, L. Dynamic rainfall-induced landslide susceptibility: A step towards a unified forecasting system. Int. J. Appl. Earth Obs. Geoinf. 2023, 125. [Google Scholar] [CrossRef]
  49. Halter, T.; Lehmann, P.; Wicki, A.; Aaron, J.; Stähli, M. Optimising landslide initiation modelling with high-resolution saturation prediction based on soil moisture monitoring data. Landslides 2024. [Google Scholar] [CrossRef]
  50. Marino, P.; Peres, D.J.; Cancelliere, A.; Greco, R.; Bogaard, T.A. Soil moisture information can improve shallow landslide forecasting using the hydrometeorological threshold approach. Landslides 2020, 17, 2041–2054. [Google Scholar] [CrossRef]
  51. Schilirò, L.; Marmoni, G.M.; Fiorucci, M.; Pecci, M.; Mugnozza, G.S. Preliminary insights from hydrological field monitoring for the evaluation of landslide triggering conditions over large areas. Nat. Hazards 2023, 118, 1401–1426. [Google Scholar] [CrossRef]
  52. Peiro, Y.; Ciabatta, L.; Volpe, E.; Cattoni, E. Spatiotemporal Modelling of Landslide Susceptibility Using Satellite Rainfall and Soil Moisture Products through Machine Learning Techniques. In Proceedings of the EGU General Assembly 2024, Vienna, Austria, 14–19 April 2024. [Google Scholar] [CrossRef]
  53. Sun, Y.; Wendi, D.; Kim, D.E.; Liong, S.-Y. Deriving intensity–duration–frequency (IDF) curves using downscaled in situ rainfall assimilated with remote sensing data. Geosci. Lett. 2019, 6, 17. [Google Scholar] [CrossRef]
  54. Basumatary, V.; Sil, B. Generation of Rainfall Intensity-Duration-Frequency curves for the Barak River Basin. Meteorol. Hydrol. Water Manag. 2017, 6, 47–57. [Google Scholar] [CrossRef]
  55. Mondini, A.C.; Guzzetti, F.; Melillo, M. Deep learning forecast of rainfall-induced shallow landslides. Nat. Commun. 2023, 14, 2466. [Google Scholar] [CrossRef] [PubMed]
  56. Levy, M.C.; Cohn, A.; Lopes, A.V.; Thompson, S.E. Addressing rainfall data selection uncertainty using connections between rainfall and streamflow. Sci. Rep. 2017, 7, 219. [Google Scholar] [CrossRef]
  57. Brunetti, M.T.; Melillo, M.; Gariano, S.L.; Ciabatta, L.; Brocca, L.; Amarnath, G.; Peruccacci, S. Satellite rainfall products outperform ground observations for landslide prediction in India. Hydrol. Earth Syst. Sci. 2021, 25, 3267–3279. [Google Scholar] [CrossRef]
  58. Pellarin, T.; Román-Cascón, C.; Baron, C.; Bindlish, R.; Brocca, L.; Camberlin, P.; Fernández-Prieto, D.; Kerr, Y.H.; Massari, C.; Panthou, G.; et al. The Precipitation Inferred from Soil Moisture (PrISM) Near Real-Time Rainfall Product: Evaluation and Comparison. Remote Sens. 2020, 12, 481. [Google Scholar] [CrossRef]
  59. Yang, H.; Hu, K.; Zhang, S.; Liu, S. Feasibility of satellite-based rainfall and soil moisture data in determining the triggering conditions of debris flow: The Jiangjia Gully (China) case study. Eng. Geol. 2023, 315, 107041. [Google Scholar] [CrossRef]
  60. Hong, Y.; Adler, R.; Huffman, G. Evaluation of the potential of NASA multi-satellite precipitation analysis in global landslide hazard assessment. Geophys. Res. Lett. 2006, 33, 2006GL028010. [Google Scholar] [CrossRef]
  61. Bucci, F.; Santangelo, M.; Fongo, L.; Alvioli, M.; Cardinali, M.; Melelli, L.; Marchesini, I. A new digital lithological map of Italy at the 1:100 000 scale for geomechanical modelling. Earth Syst. Sci. Data 2022, 14, 4129–4151. [Google Scholar] [CrossRef]
  62. Calvello, M.; Pecoraro, G. FraneItalia: A catalog of recent Italian landslides. Geoenviron. Disasters 2018, 5, 13. [Google Scholar] [CrossRef]
  63. Peruccacci, S.; Gariano, S.L.; Melillo, M.; Solimano, M.; Guzzetti, F.; Brunetti, M.T. The ITAlian rainfall-induced LandslIdes CAtalogue, an extensive and accurate spatio-temporal catalogue of rainfall-induced landslides in Italy. Earth Syst. Sci. Data 2023, 15, 2863–2877. [Google Scholar] [CrossRef]
  64. Po River Basin Authority. Caratteristiche del Bacino del Fiume Po e Primo Esame dell’ Impatto Ambientale Delle Attivitá Umane Sulle Risorse Idriche (Characteristics of Po River Catchment and First Investigation of the Impact of Human Activities on Water Resources), (n.d.); 2016. Available online: https://www.adbpo.it/PBI/PBI_progetto_piano/01_PBI_Po_Relazione_Generale_V06_10_2016.pdf (accessed on 1 November 2024).
  65. Vezzoli, R.; Mercogliano, P.; Pecora, S.; Zollo, A.L.; Cacciamani, C. Hydrological simulation of Po River (North Italy) discharge under climate change scenarios using the RCM COSMO-CLM. Sci. Total Environ. 2015, 521–522, 346–358. [Google Scholar] [CrossRef]
  66. Achour, Y.; Pourghasemi, H.R. How do machine learning techniques help in increasing accuracy of landslide susceptibility maps? Geosci. Front. 2020, 11, 871–883. [Google Scholar] [CrossRef]
  67. Setargie, T.A.; Tsunekawa, A.; Haregeweyn, N.; Tsubo, M.; Fenta, A.A.; Berihun, M.L.; Sultan, D.; Yibeltal, M.; Ebabu, K.; Nzioki, B.; et al. Random Forest–based gully erosion susceptibility assessment across different agro-ecologies of the Upper Blue Nile basin, Ethiopia. Geomorphology 2023, 431, 108671. [Google Scholar] [CrossRef]
  68. Bourenane, H.; Bouhadad, Y.; Guettouche, M.S.; Braham, M. GIS-based landslide susceptibility zonation using bivariate statistical and expert approaches in the city of Constantine (Northeast Algeria). Bull. Eng. Geol. Environ. 2015, 74, 337–355. [Google Scholar] [CrossRef]
  69. Cantarino, I.; Carrion, M.A.; Goerlich, F.; Ibañez, V.M. A ROC analysis-based classification method for landslide susceptibility maps. Landslides 2019, 16, 265–282. [Google Scholar] [CrossRef]
  70. QGis, (n.d.). Available online: https://www.qgis.org (accessed on 10 April 2024).
  71. Lehner, B.; Verdin, K.; Jarvis, A. HydroSHEDS Technical Documentation, Version 1.2; World Wildlife Fund US: Washington, DC, USA, 2008. [Google Scholar]
  72. CIESIN: Center for International Earth Science Information Network (CIESIN), Columbia University. High-Resolution Settlement Layer (HRSL), 10-meter Resolution. NASA Socioeconomic Data and Applications Center (SEDAC). 2021. Available online: https://sedac.ciesin.columbia.edu/data/collection/hrsl[EC1] (accessed on 8 July 2024).
  73. Volpe, E.; Gariano, S.L.; Ardizzone, F.; Fiorucci, F.; Salciarini, D. A Heuristic Method to Evaluate the Effect of Soil Tillage on Slope Stability: A Pilot Case in Central Italy. Land 2022, 11, 912. [Google Scholar] [CrossRef]
  74. Salciarini, D.; Volpe, E.; Di Pietro, L.; Cattoni, E. A Case-Study of Sustainable Countermeasures against Shallow Landslides in Central Italy. Geosciences 2020, 10, 130. [Google Scholar] [CrossRef]
  75. Ponziani, F.; Pandolfo, C.; Stelluti, M.; Berni, N.; Brocca, L.; Moramarco, T. Assessment of rainfall thresholds and soil moisture modeling for operational hydrogeological risk prevention in the Umbria region (central Italy). Landslides 2012, 9, 229–237. [Google Scholar] [CrossRef]
  76. Wicki, A.; Jansson, P.-E.; Lehmann, P.; Hauck, C.; Stähli, M. Simulated or measured soil moisture: Which one is adding more value to regional landslide early warning? Hydrol. Earth Syst. Sci. 2021, 25, 4585–4610. [Google Scholar] [CrossRef]
  77. Felsberg, A.; De Lannoy, G.J.M.; Girotto, M.; Poesen, J.; Reichle, R.H.; Stanley, T. Global Soil Water Estimates as Landslide Predictor: The Effectiveness of SMOS, SMAP, and GRACE Observations, Land Surface Simulations, and Data Assimilation. J. Hydrometeorol. 2021, 22, 1065–1084. [Google Scholar] [CrossRef]
  78. Huffman, G.J.; Bolvin, D.T.; Braithwaite, D.; Hsu, K.-L.; Joyce, R.J.; Kidd, C.; Nelkin, E.J.; Sorooshian, S.; Stocker, E.F.; Tan, J.; et al. Integrated Multi-satellite Retrievals for the Global Precipitation Measurement (GPM) Mission (IMERG). In Satellite Precipitation Measurement; Levizzani, V., Kidd, C., Kirschbaum, D.B., Kummerow, C.D., Nakamura, K., Turk, F.J., Eds.; Springer International Publishing: Cham, Switzerland, 2020; pp. 343–353. [Google Scholar] [CrossRef]
  79. Xie, P.; Chen, M.; Yang, S.; Yatagai, A.; Hayasaka, T.; Fukushima, Y.; Liu, C. A Gauge-Based Analysis of Daily Precipitation over East Asia. J. Hydrometeorol. 2007, 8, 607–626. [Google Scholar] [CrossRef]
  80. Brocca, L.; Ciabatta, L.; Massari, C.; Moramarco, T.; Hahn, S.; Hasenauer, S.; Kidd, R.; Dorigo, W.; Wagner, W.; Levizzani, V. Soil as a natural rain gauge: Estimating global rainfall from satellite soil moisture data. JGR Atmos. 2014, 119, 5128–5141. [Google Scholar] [CrossRef]
  81. Stoffelen, A. Toward the true near-surface wind speed: Error modeling and calibration using triple collocation. J. Geophys. Res. 1998, 103, 7755–7766. [Google Scholar] [CrossRef]
  82. Karger, D.N.; Lange, S.; Hari, C.; Reyer, C.P.O.; Conrad, O.; Zimmermann, N.E.; Frieler, K. CHELSA-W5E5: Daily 1 km meteorological forcing data for climate impact studies. Earth Syst. Sci. Data 2023, 15, 2445–2464. [Google Scholar] [CrossRef]
  83. Filippucci, P. High-Resolution Remote Sensing for Rainfall and River Discharge Estimation. Ph.D. Thesis, Florence University—UNIFI, Firenze, Italy, Technischen Universität Wien—TUWIEN, Vienna, Austria, 2022. [Google Scholar]
  84. Gholami, H.; Mohammadifar, A. Novel deep learning hybrid models (CNN-GRU and DLDL-RF) for the susceptibility classification of dust sources in the Middle East: A global source. Sci. Rep. 2022, 12, 19342. [Google Scholar] [CrossRef]
  85. Corder, G.W.; Foreman, D.I. Nonparametric Statistics: A Step-by-Step Approach; John Wiley & Sons: Hoboken, NJ, USA, 2014. [Google Scholar]
  86. Hauke, J.; Kossowski, T. Comparison of Values of Pearson’s and Spearman’s Correlation Coefficients on the Same Sets of Data. Quaest. Geogr. 2011, 30, 87–93. [Google Scholar] [CrossRef]
  87. Garg, A.; Tai, K. Comparison of statistical and machine learning methods in modelling of data with multicollinearity. Int. J. Model. Identif. Control 2013, 18, 295. [Google Scholar] [CrossRef]
  88. Bogaard, T.; Greco, R. Invited perspectives: Hydrological perspectives on precipitation intensity-duration thresholds for landslide initiation: Proposing hydro-meteorological thresholds. Nat. Hazards Earth Syst. Sci. 2018, 18, 31–39. [Google Scholar] [CrossRef]
  89. Greco, R.; Marino, P.; Bogaard, T.A. Recent advancements of landslide hydrology. WIREs Water 2023, 10, e1675. [Google Scholar] [CrossRef]
  90. Moreno, M.; Lombardo, L.; Crespi, A.; Zellner, P.J.; Mair, V.; Pittore, M.; Van Westen, C.; Steger, S. Space-time data-driven modeling of precipitation-induced shallow landslides in South Tyrol, Italy. Sci. Total Environ. 2024, 912, 169166. [Google Scholar] [CrossRef] [PubMed]
  91. Smith, H.G.; Neverman, A.J.; Betts, H.; Spiekermann, R. The influence of spatial patterns in rainfall on shallow landslides. Geomorphology 2023, 437, 108795. [Google Scholar] [CrossRef]
Geosciences 14 00330 g003
Figure 3. Elevation distribution in the study area (A) and slope distribution (B). Red dots show the occurred landslides in the study area.
Figure 3. Elevation distribution in the study area (A) and slope distribution (B). Red dots show the occurred landslides in the study area.
Geosciences 14 00330 g003
Geosciences 14 00330 g004
Figure 4. Spatial distribution of the distance to river parameter in the study area.
Figure 4. Spatial distribution of the distance to river parameter in the study area.
Geosciences 14 00330 g004
Geosciences 14 00330 g005
Figure 5. Distribution of land cover (A) and the spatial distribution of NDVI (B).
Figure 5. Distribution of land cover (A) and the spatial distribution of NDVI (B).
Geosciences 14 00330 g005
Geosciences 14 00330 g006
Figure 6. RF model accuracy for different train/test split ratios (A); model’s tuning function (B,C); model’s performance (D).
Figure 6. RF model accuracy for different train/test split ratios (A); model’s tuning function (B,C); model’s performance (D).
Geosciences 14 00330 g006
Geosciences 14 00330 g007
Figure 7. Spearman’s correlation between pairs of 14 condition factors.
Figure 7. Spearman’s correlation between pairs of 14 condition factors.
Geosciences 14 00330 g007
Geosciences 14 00330 g008
Figure 8. Ranking of conditioning factors importance.
Figure 8. Ranking of conditioning factors importance.
Geosciences 14 00330 g008
Geosciences 14 00330 g009aGeosciences 14 00330 g009b
Figure 9. Bar chart showing for each day the maximum rainfall (red lines) and the average rainfall (blue line). Pink bars show the cumulative 15 d rainfall and daily rainfall, in the studied period (A). Distribution of landslide occurrences in the study area from 2016 to late 2021. The stars in the figure mark the two chosen events (B).
Figure 9. Bar chart showing for each day the maximum rainfall (red lines) and the average rainfall (blue line). Pink bars show the cumulative 15 d rainfall and daily rainfall, in the studied period (A). Distribution of landslide occurrences in the study area from 2016 to late 2021. The stars in the figure mark the two chosen events (B).
Geosciences 14 00330 g009aGeosciences 14 00330 g009b
Geosciences 14 00330 g010
Figure 10. In relation to the R1 rainfall event, the images show the spatial distribution of average daily rainfall (A), the cumulative 15 d rainfall (B), soil moisture (C), and landslide susceptibility (D). The red circles symbolize the landslides that occurred.
Figure 10. In relation to the R1 rainfall event, the images show the spatial distribution of average daily rainfall (A), the cumulative 15 d rainfall (B), soil moisture (C), and landslide susceptibility (D). The red circles symbolize the landslides that occurred.
Geosciences 14 00330 g010
Geosciences 14 00330 g011
Figure 11. In relation to the R2 rainfall event, the images show the spatial distribution of average daily rainfall (A), the cumulative 15 d rainfall (B), soil moisture (C), and landslide susceptibility (D). The red circles symbolize the landslides that occurred.
Figure 11. In relation to the R2 rainfall event, the images show the spatial distribution of average daily rainfall (A), the cumulative 15 d rainfall (B), soil moisture (C), and landslide susceptibility (D). The red circles symbolize the landslides that occurred.
Geosciences 14 00330 g011
Geosciences 14 00330 g012
Figure 12. Spatial distribution of susceptibility difference evaluated in relation to the two rainfall events, R1 and R2, for each pixel of study area.
Figure 12. Spatial distribution of susceptibility difference evaluated in relation to the two rainfall events, R1 and R2, for each pixel of study area.
Geosciences 14 00330 g012
Geosciences 14 00330 g013
Figure 13. For each pixel in the study area, the figures show the difference of daily rain (A) and the difference of cumulative 15 d rain (B) related to R1 and R2.
Figure 13. For each pixel in the study area, the figures show the difference of daily rain (A) and the difference of cumulative 15 d rain (B) related to R1 and R2.
Geosciences 14 00330 g013
Geosciences 14 00330 g014
Figure 14. For each pixel in the study area and in relation to the R1 and R2 events, the figure shows the soil moisture difference.
Figure 14. For each pixel in the study area and in relation to the R1 and R2 events, the figure shows the soil moisture difference.
Geosciences 14 00330 g014
Table 1. Static and dynamic triggering factors considered in the analyses.
Table 1. Static and dynamic triggering factors considered in the analyses.
FactorDescriptionSource,
Scale/Resolution
Elevation Digital elevation of the terrain surface DTM, 10 m
Slope angleAngle of the slope inclination DTM, 10 m
Aspect Compass direction of the slope exposure DTM, 10 m
Plan curvatureCurvature perpendicular to the slope, indicating concave or convex surface DTM, 10 m
Profile curvatureCurvature parallel to the slope, indicating concave or convex surfacesDTM, 10 m
GeologyLithology of the surface materialGeo-Map 1:100,000
Land coverphysical material on the surface of the EarthCORINE Land Cover (CLC), 100 m
NVDIAn index to quantify the growth of green vegetation on land coverSentinel-2, 10 m
Distance to riverDistance to riverHyrdoSHED(SRTM),10 m
Distance to roadDistance to roadCIESIN,10 m
Soil moistureAmount of soil water content GLEAM 4DMED, 1 km
1 day rainAmount of cumulative 1 d antecedent rainfall4DMED, 1 km
7 day RainAmount of cumulative 7 d antecedent rainfall4DMED, 1 km
15 day Rain Amount of cumulative 15 d antecedent rainfall4DMED, 1 km
Table 2. Main characteristics of the products considered.
Table 2. Main characteristics of the products considered.
Product Spatial Resolution Temporal Resolution Temporal Coverage Source
IMERG-LR 0.1° 0.5 h 2002–to date NASA
CPC 0.5° Daily 1981–to date NOAA
SM2RAIN 1 km Daily 2017–2022 TUWIEN
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

Peiro, Y.; Volpe, E.; Ciabatta, L.; Cattoni, E. High Resolution Precipitation and Soil Moisture Data Integration for Landslide Susceptibility Mapping. Geosciences 2024, 14, 330. https://doi.org/10.3390/geosciences14120330

AMA Style

Peiro Y, Volpe E, Ciabatta L, Cattoni E. High Resolution Precipitation and Soil Moisture Data Integration for Landslide Susceptibility Mapping. Geosciences. 2024; 14(12):330. https://doi.org/10.3390/geosciences14120330

Chicago/Turabian Style

Peiro, Yaser, Evelina Volpe, Luca Ciabatta, and Elisabetta Cattoni. 2024. "High Resolution Precipitation and Soil Moisture Data Integration for Landslide Susceptibility Mapping" Geosciences 14, no. 12: 330. https://doi.org/10.3390/geosciences14120330

APA Style

Peiro, Y., Volpe, E., Ciabatta, L., & Cattoni, E. (2024). High Resolution Precipitation and Soil Moisture Data Integration for Landslide Susceptibility Mapping. Geosciences, 14(12), 330. https://doi.org/10.3390/geosciences14120330

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