论文信息
论文解读
AI 辅助整理- 方法
- 地质地貌调查、遥感与MaxEnt机器学习模型,约100平方公里。
- 发现
- 岩性和断层密度是主要控制因素,天坑由高水力梯度、裂隙带与渗透性差异共同驱动。
- 机会点
- 可将MaxEnt易发性建模思路迁移至农业区地下水侵蚀或土壤塌陷风险预警研究。
论文摘要
西里诺山(意大利南部巴西利卡塔大区)西南坡具有复杂的地质条件,易发非均质天坑(sinkhole)现象。最突出的地貌特征为西里诺湖,该湖形成于大型滑坡堆积体之上,频繁受到管涌相关天坑现象的影响。本研究采用综合方法,结合地质与地貌调查、遥感及机器学习建模,探讨西里诺湖及其周边地区(约100 km²)天坑形成的成因机制。野外工作识别出10处天然天坑,主要与上覆于低渗透性基底之上的碎屑沉积物中的深部管涌过程有关。研究采用最大熵(MaxEnt)算法生成研究区天然天坑敏感性图。分析表明,岩性和断裂密度是主要的易发因素。结果表明,西里诺湖地区天坑的发育受高水力梯度、破碎带及渗透性差异层之间相互作用的控制。
全文
Rendiconti online della Società Geologica Italiana1
Special Section - XIX Convegno Nazionale GIT - Milazzo 2025
Sinkhole susceptibility and genesis at the southwestern slope
of the Mount Sirino (Basilicata, Southern Italy): insights from field
survey and machine learning modeling
Patrizio Petricca1*, Stefania Nisio1, Pantaleone De Vita2, Giancarlo Ciotoli3 & Vincenzo Allocca2
1 ISPRA–Servizio Geologico d’Italia, Via Vitaliano Brancati 60, Rome, Italy.
2 Department of Earth, Environmental and Resources Sciences, University of Naples Federico II, Naples 80126, Italy.
3 C.N.R. – Istituto di Geologia Ambientale e Geoingegneria, 00015 Monterotondo, Rome, Italy.
PP, 0000-0003-1186-4296; SN, 0000-0002-5422-3694; PDV, 0000-0002-0692-8630; GC, 0000-0002-9496-3021; VA, 0000-0001-7146-2438.
Rend. Online Soc. Geol. It., Vol. 72 (2027), pp. …-…, 3 figs. https://doi.org10.3301/ROL.2026.39
Short Note
Corresponding author e-mail: patrizio.petricca@isprambiente.it
Citation: Petricca P., Nisio S., De Vita P., Ciotoli G. & Allocca V. (2027) - Sinkhole susceptibility and genesis at the southwestern slope of the Mount Sirino (Basilicata,
Southern Italy): insights from field survey and machine learning modeling. Rend. Online Soc. Geol. It., 72, XX-XX, https://doi.org10.3301/ROL.2026.39.
Associate Editor: Alice Busetti
Submitted: 16 January 2026
Accepted: 10 July 2026
Published online: XX September 2026
Copyright: © Società Geologica Italiana ETS, 2027.
This article is distributed under the terms of the Creative Commons Attribution License, which
permits use, distribution and reproduction in any medium, provided the original work is properly cited.
ABSTRACT
The southwestern slope of the Mount Sirino (Basilicata, Southern Italy) is
characterized by complex geological conditions favouring heterogeneous
sinkhole phenomena. The most prominent geomorphological feature is Lake
Sirino, formed on a large landslide deposit and frequently affected by piping-
related sinkhole phenomena. This study uses an integrated approach combining
geological and geomorphological surveys, remote sensing and machine learning
modelling to investigate the genetic mechanisms of sinkhole formation at Lake
Sirino and its surrounding areas (c. 100 km²). Fieldwork identified ten natural
sinkholes, mainly associated with deep-seated piping processes in detrital
deposits overlying low-permeability substrate. The Maximum Entropy (MaxEnt)
algorithm was used to create a susceptibility map of natural sinkholes in the
study area. This analysis identified lithology and fault density as the primary
predisposing factors. The results highlight that sinkholes development in the
Lake Sirino area is controlled by the interaction between high hydraulic gradients,
fractured zones, and contrasting permeability layers.
KEYWORDS: natural sinkhole, deep piping, susceptibility mapping,
MaxEnt, Mount Sirino.
INTRODUCTION
Sinkholes represent one of the most hazardous geological
phenomena in areas characterized by karst processes or
subsurface erosion mechanisms (Nisio, 2003, 2008; Nisio et al.,
2007; Caramanna et al., 2008; Ciotoli et al., 2026; Gutierrez et al.,
2014; Gutiérrez, 2016; De Waele & Gutiérrez, 2022). While karst-
related sinkholes are well-documented (Sauro, 2003; Gutierrez
et al., 2014; Gutiérrez, 2016; De Waele & Gutiérrez, 2022), those
generated by piping processes in non-karstic settings remain less
studied, particularly in tectonically active mountain regions (Nisio
et al., 2007; Caramanna et al., 2008).
Deep piping sinkholes represent a specific subtype of dropout
sinkholes, not explicitly defined in the international classification
(Ford & Williams,1989; Waltham et al., 2005; Gutierrez et al. 2014),
but widely recognized in several Italian alluvial and coastal plains.
Deep piping sinkholes represent a specific subtype driven by
upward-directed groundwater flow from pressurized deep aquifers
through thick heterogeneous cover sequences (Nisio et al., 2007;
Caramanna et al., 2008; Forti, 2002), and are distinct from the
shallow piping processes documented in the present study.
In this study, the term “piping” is used in a broader sense to
indicate subsurface erosion processes driven by groundwater flow,
leading to the progressive removal of fine particles and eventual
collapse. However, the sinkholes observed in the Mount Sirino
area are interpreted as predominantly shallow piping phenomena
occurring within detrital and pelitic deposits, rather than deep
piping processes associated with pressurized flow from confined
aquifers.
Accepted manuscript
2
P. PETRICCA ET AL.
The southwester slope of the Mount Sirino, with the highest
peak at the Mount Papa (2005 m a.s.l.) is characterized by a
complex geological setting characterized by Lagonegro and Liguridi
tectonic units (Fig. 1a) favoring geomorphological processes such
as well-known deep-seated landslides and sinkhole phenomena. In
this context, the Lake Sirino (784 m a.s.l.) occupies a depression
developed within a large deposit of a deep-seated landslide and
has been repeatedly affected by sudden drainage events caused by
piping-induced collapse along its southwestern shore (Giampaolo
et al., 2016; Grassi et al., 2001) i.e. subsurface erosion processes
driven by groundwater flow, leading to the progressive removal of
fine particles and eventual collapse of the overlying material. These
events pose significant risks to surrounding infrastructure and
population.
Fig. 1 - a) Geological map of the Lake Sirino area (modified from CARG Sheet 521 – Lauria; ISPRA, 2003). Legend: a1 - landslide deposits; STSa - Siliceous
Schists (Scisti Silicei fm); SLCa - Cherty Limestones (Calcari Selciferi fm); FYGa - Galestrino Flysch (Lagonegro I complex); FAC - Monte Facito Formation;
FOP - Monte Foraporta Dolomites and Limestones; DBS - Upper Dolomite (Dolomia superiore fm); UAS - Nemoli Complex (Liguridi). Red box indicates
Lake Sirino location; purple box represents the study area; black dashed line is the trace of the geological section (A-B) represented under the map. b)
The NNE-SSW section shows the antiform structure, with the core consisting of the Cherty Limestones of the LAGONEGRO I tectonic unit (SLCa); moving
towards the flanks of the structure, progressively higher tectonic units are encountered.
Accepted manuscript
3
SINKHOLE GENESIS AND SUSCEPTIBILITY AT MT. SIRINO
Previous studies have documented the hydrogeological
framework (D’Ecclesiis et al., 1990), slope dynamics (Guida et
al., 2006; Coppola et al., 2013), and geophysical characterization
(Giampaolo et al., 2016) of the area. Although several authors
have proposed either karstic or piping mechanisms (Nisio
2008, 2014, 2023; Nisio et al., 2007, Caramanna et al., 2008;
Giampaolo et al., 2016), the genetic process responsible for
the sinkhole phenomena in the surroundings of the Lake Sirino
remains debated. However, a comprehensive analysis of sinkhole
distribution and susceptibility in the surroundings of the lake and
along the southwestern slope of the Mount Sirino has not yet been
carried out.
This study aims to (i) identify and characterize the genetic
mechanisms of sinkholes occurring in a sector of the southwester
slope of the Mount Sirino, extended over about 100 km², through
detailed field validation, and (ii) produce a sinkhole susceptibility
map based on Machine Learning modelling, providing a tool to
support risk assessment and land-use planning.
In recent years, susceptibility mapping has become a widely
used approach for assessing the spatial likelihood of hazardous
geomorphological processes. In the context of sinkholes, both
statistical and machine learning methods, such as Maximum
Entropy (MaxEnt; Philips et al., 2025), Random Forest and
Support Vector Machines, have been successfully applied to
integrate multiple environmental predictors (lithology, structure,
topography, hydrogeology) with a georeferenced inventory of
known occurrences (e.g., Al-Kouri et al., 2013; Ciotoli et al., 2026;
Taheri et al., 2019). These approaches are particularly useful in
complex geological settings, such as tectonically active and non-
karstic environments, where subsurface processes are difficult to
directly be observed and quantitative tools for risk assessment
are still limited.
GEOLOGICAL AND GEOMORPHOLOGICAL SETTING
The study area is located within the CARG Sheet 521 – Lauria
(Servizio Geologico d’Italia, 2003), in one of the axial culminations
of the southern Apennines where the Lagonegro Units outcrop in
a tectonic window (Fig. 1b). Mount Sirino forms the core of an
antiform structure exposing the Lagonegro Units, flanked by
progressively higher tectonic units including Apennine Platform
carbonates and Liguridi Units (Monaco et al., 1998).
The Lagonegro Units consist of a basinal succession that
begins with the Monte Facito Formation (FAC), composed of
Lower-Middle Triassic red and gray claystones with fine-grained
sandstones and limestones. It is overlain by Cherty Limestones
(SLC – “Calcari Selciferi”) of Upper Triassic-Liassic age, followed
by Siliceous Schists (STS – “Scisti Silicei) of Jurassic age, and
the Galestrino Flysch (FYG), consisting of Upper Jurassic-Lower
Cretaceous siliceous marly-clayey sediments.
Two main tectonic subunits are distinguished based on their
structural position: Lagonegro I, representing deposits from the
axial sector of the basin and cropping out in the Lake Sirino area,
and Lagonegro II, which includes marginal facies deposited at the
base of the basin scarps.
Large debris accumulations and Würmian glacial-phase
landslide deposits are widespread throughout the region. The
southwestern slope of Mount Sirino exhibits a typical dip-slope
morphology, with strata dipping approximately 30° toward the
south and southwest. Quaternary brittle tectonics is evidenced
by normal and trastensive faults with N-S and N130° orientations
(Schiattarella, 1998; Papanikolaou & Roberts, 2007). These faults
play a significant role in controlling groundwater circulation and
slope stability.
Lake Sirino occupies a depression bounded by Siliceous
Schists that detached from the upper slope. This was likely caused
by a major rotational-translational deep-seated rockslide occurring
between 11,000-9,000 years BP (Guida et al., 2006; Coppola et
al., 2013). The lake covers an area of approximately 3 hectares,
reaches a maximum depth of 6 m, and is fed by a perennial spring
(Spring “del Lago”) discharging approximately 130 L/s (Fig. 2).
Water exits the system through two downstream springs (Springs
“sotto il Lago I” and “sotto il Lago II”) which emerge within the
Galestrino Flysch, releasing approximately 2-3 L/s and 100 L/s,
respectively, under normal conditions (Giampaolo et al., 2016).
The hydrogeological setting is characterized by high hydraulic
conductivity within the fractured cherty limestones and siliceous
schists, which constitute the main aquifer of the Mount Sirino
system (see Fig. 2 cross sections). In contrast, the Galestrino
Flysch, with its prevalent pelitic and marly composition, acts as
the main aquiclude. Electrical resistivity tomography (ERT) surveys
(Giampaolo et al., 2016) revealed zones of high fracturing and
elevated hydraulic conductivity near the southwestern shore of the
lake (Fig. 2).
MATERIALS AND METHODS
Field surveys and sinkhole inventory
This study integrates field surveys, remote sensing analysis,
and machine learning modeling to investigate sinkhole distribution
and susceptibility in the study area. The methodological workflow
includes: (i) identification and validation of sinkholes through
field surveys and remote sensing; (ii) compilation of a sinkhole
inventory; (iii) susceptibility modeling using the MaxEnt algorithm.
In this study, the term “sinkhole” is used as a general term, while “
collapse sinkhole” refers specifically to collapse features developed
in carbonate lithologies. Two field campaigns were carried out to
validate and characterize the preliminary recognition of sinkholes.
The first campaign, conducted in August 2024, focused on the
geological validation and documentation of a fresh sinkhole event
that occurred in July 2024 along the southwestern shore of Lake
Sirino (red square in Fig. 2). The second campaign, carried out
in February 2025, involved a detailed investigation of potential
sinkhole sites previously identified through remote sensing analysis
over an area extending about 100 km 2 in the surroundings of the
Sirino lake.
The remote sensing analysis integrates high-resolution
orthophotos at 1:5.000 scale (updated to 2013), recent satellite
imageries, and ground motion data from the European Ground
Motion Service (EGMS) covering the period 2019-2023. Drone
Accepted manuscript
4
P. PETRICCA ET AL.
surveys (model DJI Mavic mini 4k) were locally used to support field
observations and to document selected sinkhole sites.
Susceptibility modeling
The field and drone surveys provided a validated dataset of
ten natural sinkholes, which formed the basis for the susceptibility
analysis. Maximum Entropy modeling (MaxEnt; Phillips et al., 2025)
was applied to assess the susceptibility to piping-related collapses
across the study area (Fig. 1). MaxEnt is a presence-only machine
learning method based on the principle of maximum entropy,
widely used in both ecological and geological applications for
spatial prediction. The model was trained using six confirmed piping
sinkholes, and 10,005 background points randomly distributed across
the study area. Background points represent randomly distributed
locations across the study area used to characterize the full range of
environmental conditions available within it. The MaxEnt algorithm
estimates susceptibility by comparing the environmental signature
of confirmed sinkhole locations against that of the background,
identifying predictor combinations statistically associated with
piping sinkhole occurrence. The total of 10,005 reflects the default
behavior of MaxEnt v.3.4.1, which automatically includes the
presence localities within the background sample. The high density
of background points ensured adequate sampling of environmental
gradients across the study area, which was essential for the study.
The environmental predictors considered by the modeling
were:
• Lithology, as a categorical variable (derived from geological
map of Italy 1:50k scale; ISPRA, 2003);
• Fault density, calculated using a 200 m buffer (derived from
faults mapped geological map of Italy 1:50k scale; ISPRA,
2003);
• Modified Normalized Difference Water Index (MNDWI), derived
from satellite imagery using ad-hoc deep-learning algorithms
developed for this study (Petricca, 2025) to estimate soil
moisture conditions;
• Topographic Position Index (TPI), describing local topographic
position, derived from the TINITALY Digital Elevation Model
(Tarquini et al., 2023) at 10 m spatial resolution;
• Elevation, extracted from the 5×5 m Digital Terrain Model
(DTM) obtained from the Geoportal of the Basilicata Region
(https://rsdi.regione.basilicata.it/geonetwork/);
• Kernel density map (buffer 500 m) of water springs (derived
from geological map of Italy 1:50k scale; ISPRA, 2003)
computed with a 500 m search radius to account for local
hydrogeological influence.
One of the environmental predictors used in the MaxEnt model
— the Modified Normalized Difference Water Index (MNDWI) —
was derived from Sentinel-2 multispectral imagery using a custom
deep-learning algorithm developed specifically for this study
(HydraNet; Petricca, 2025). HydraNet is a convolutional neural
network (CNN) architecture trained to automatically detect and
map wet areas (saturated soils, shallow water bodies, and areas
of persistent moisture) from multispectral satellite imagery. The
network was trained on a labelled dataset of Sentinel-2 Level-2A
images (10 m spatial resolution, bands B3 and B11) acquired over
a range of geological and land-cover conditions in southern Italy.
Fig. 2 - Conceptual cross-section of Lake Sirino (modified from Giampaolo et al., 2016). White arrows on the map indicate the groundwater flow direction
through the fractured zone beneath the SW shore (black arrow on the right sections). Blue drops indicate the locations of the springs (“del Lago” north of
the lake, and “sotto il lago I” and “sotto il lago II” to the south). The red square indicates the location of the piping event documented in July 2024. Red
dashed lines represent the traces of the geological sections.
Accepted manuscript
5
SINKHOLE GENESIS AND SUSCEPTIBILITY AT MT. SIRINO
The model produces a continuous soil moisture probability raster
(0–1) that was used as a proxy for local hydrogeological influence
on sinkhole susceptibility. The trained model and inference code
are openly available at https://github.com/BaterHub/HydraNet-
Wetland-Detection (Petricca, 2025). This approach provides a
scalable, data-driven method for estimating soil moisture conditions
as a predictor variable in geohazard susceptibility models where in-
situ hydrological monitoring data are unavailable.
All raster predictors were resampled and harmonized to a
common spatial resolution, extent, and coordinate reference
system before model execution.
Model performance was evaluated using the Area Under the
Curve (AUC) of the Receiver Operating Characteristic (ROC), which
ranges from 0.5 (random prediction) to 1.0 (perfect discrimination).
The contribution of the predictors and response curves were
analyzed to identify the main predisposing factors controlling
the occurrence of piping sinkholes and to understand how
susceptibility varies as a function of each environmental variable.
These variables represent the lithological, structural, topographic
and hydrogeological controls that govern subsurface erosion and
piping development in the study area.
RESULTS
Sinkhole inventory and characterization
Field surveys and drone-based photographic documentation
were carried out at the main sinkhole sites, including Lake Sirino,
Vignale, Molino, and Vaieto, to validate and document morphometric
features, surface evidence of instability, and possible signs of
reactivation. These investigations provided new geomorphological
data integrated with satellite imagery and EGMS ground motion
analysis to characterize each site. Among the 14 depressions
initially identified, 10 were validated as natural sinkholes following
field surveys, whereas 4 were excluded after being recognized as
anthropogenic features (e.g., agricultural terraces or waste disposal
sites). The validated sinkholes were subsequently classified
according to their genetic origin, distinguishing between piping
sinkholes in pelitic and detrital formations, and cover collapse
sinkholes (sensu Gutierrez et al., 2008) developed in carbonate
or dolomitic lithologies. Morphometric features of the investigated
sites are summarized in table 1.
The main morphometric and geological characteristics of
the identified sinkholes are summarized within the following
descriptions. Ten natural sinkholes were validated within the
approximately 100 km² study area (Fig. 1a), with six attributed
to piping mechanisms and four representing rock collapse
sinkholes (Nisio, 2023) in carbonate-dolomitic lithologies. Lake
Sirino itself represents the most significant feature, with well-
documented recurrent piping events along its southwestern
shore. The most thoroughly documented event occurred in July
1994, when a sinkhole approximately 1 m in diameter opened
along the southwestern shore about 3 meters from the water’s
edge (Giampaolo et al., 2016). The resulting water flow through
the cavity caused the lake level to drop by nearly 2 m while
simultaneously increasing the discharge of the Sorgente sotto il
Lago II to approximately 300 L/s. The deep piping phenomenon
ceased naturally after the partial collapse of the shore material that
Table 1 - Inventory of sinkhole forms in the Monte Sirino area, summarizing their main morphometric parameters
and geomorphological characteristics, including elevation, diameter, depth, cover properties, shape classification, activity status.
Toponym Elevation (m a.s.l.) Diameter (m) Depth (m) Cover Thickness Cover Type Shape Classification Activity Status
Lake Sirino 780 100 3-6 30-50
Clayey–detrital
cover over carbonate
bedrock
Piping sinkhole Active
Monte Papa 1500 50 n.d. n.d. Glacial till over carbonate bedrock Cover collapse sinkhole Inactive
Vignale 1 670 60 1-2 n.d.
Clayey–detrital
cover over fractured
carbonate bedrock
Piping sinkhole n.d.
Vignale 2 650 25 1-2 n.d. Fractured carbonate bedrock Piping sinkhole n.d.
Vignale 3 540 20 1-2 n.d. Dolomitic bedrock Bedrock collapse sinkhole Active
Arenazza 460 15 n.d. n.d. Dolomitic bedrock Bedrock collapse sinkhole n.d.
Molino 530 70 2-3 n.d. Arenitic-pelitic detrital Piping sinkhole Active
Vaieto 1 700 20 n.d. n.d. Arenitic-pelitic detrital Piping sinkhole Active
Vaieto 2 690 10 n.d. n.d. Arenitic-pelitic detrital Piping sinkhole Active
Fortino 1150 55 n.d. n.d. Carbonate bedrock Bedrock collapse sinkhole n.d.
Accepted manuscript
6
P. PETRICCA ET AL.
sealed the opening (Nisio et al. 2007). During the fieldwork carried
out for this study, a similar fresh collapse was documented in July
2024 at nearly the same location (Fig. 2), confirming the ongoing
activity of these processes.
In the Vignale area, two closely spaced sinkholes were
documented during the field survey and drone-based
photogrammetric campaign. Both features occur in dolomitic
lithologies of the Monte Foraporta Formation (Fig. 3A). The first is
a circular to slightly elongated depression measuring 40 by 60 m,
while the second exhibits a horseshoe shape with dimensions of 17
by 26 m. Field and drone observations confirmed that both features
are located adjacent to a fault trending N140° (dashed line in Figure
3). Although the depressions have been partially modified for waste
disposal, their geometric characteristics and structural setting
suggest an original natural genesis related to piping processes
enhanced by structural discontinuities.
The Molino site (Fig. 3A) consists of an elliptical depression
measuring 56 by 69 m at 533 m elevation, developed in the pelitic
and arenitic lithologies of the Nemoli Complex (UAS-Liguridi).
The feature is situated on relatively flat terrain with evidence of
subsurface water circulation indicated by a small stream emerging
downslope. Comparison between satellite imagery and recent drone
photography reveals recent reactivation evidenced by subsidence
affecting the asphalt road adjacent to the depression margin.
In the Vaieto area, two circular ponds with diameters of 20 m
and 10 m respectively occur in Nemoli Complex deposits (Fig. 3A).
Ground motion data from EGMS reveals approximately 20 mm of
subsidence in this area during the 2019-2022 period, supporting
active ground deformation. While the landowner reported that
these features were artificially enlarged and filled with diverted
stream water to create reservoirs, the initial depressions appear to
have formed naturally through piping mechanisms.
The four rock collapse sinkholes occur in carbonate and
dolomitic lithologies where dissolution and mechanical collapse
processes dominate (Gray triangles in Fig. 3A). These include: a
50 m wide depression at Mount Sirino (~1500 m a.s.l.), developed
within glacial deposits overlying fractured limestones; another 50
m diameter depression at Fortino site; and two smaller depressions
(15-20 m in diameter) along the Roccazzo slope within the Upper
Dolomite formation.
The spatial distribution of all validated sinkholes shows a clear
structural control, aligned with the NW-SE trending Mercure Fault
System and secondary transtensive structures, oriented NE-SW,
i.e. perpendicular to the main trend (red dashed line in Fig. 3A).
This is precisely where sinkholes develop preferentially,
suggesting the presence of subsurface flow pathways.
This structural and lithological framework establishes the
hydrogeological conditions that favour concentrated groundwater
flow and localized subsurface erosion. These are key predisposing
factors for sinkhole related to deep piping.
Susceptibility mapping
The MaxEnt model achieved high predictive performance,
with a training AUC value of 0.893, indicating good discriminative
capacity. The variable contribution analysis revealed that fault
Fig. 3 - A) Sinkhole susceptibility (MaxEnt model). Warm colors indicate higher susceptibility (scale 0-1). Collapse sinkholes (gray triangles) and piping
sinkholes (gray circles) position overlays the susceptibility map. The red dashed line represents the trace of a fault recognized during the field survey
phases. The black dashed line represents the average direction of capable Nemoli fault (NF), Vignale fault (VF) and Rivello fault (RF) belonging to the
Mercure fault system and reported in the "ITHACA" database (https://diss.ingv.it/ithdiss/#). A semi-transparent shaded-relief representation has been
overlaied. B) Numerical outcomes of the preferred model for variable with contribution > 0 in the preferred model.
Accepted manuscript
7
SINKHOLE GENESIS AND SUSCEPTIBILITY AT MT. SIRINO
density (Fig. 3B - panel a) and lithology (panel b) dominate the
model, accounting for 41.3% and 30.2% contribution, whereas
other variables account for progressively lower contributions.
The response curve of fault density indicates that while lithology
controls the baseline susceptibility, the presence of structural
discontinuities strongly modulates the actual likelihood of sinkhole
occurrence. Areas at higher elevations are more prone to sinkhole
occurrence (Fig. 3B – panel c). The MNDWI index contributed 6.3%
(Fig. 3B - panel d), with susceptibility increasing in areas showing
positive MNDWI values, corresponding to higher soil moisture.
Conversely, TPI (Fig. 3B - panel e), showed a minimal effect (0.7%),
suggesting that local topographic position plays a relatively minor
role compared with lithological and structural factors.
The response curves (Fig. 3B) show that pelitic and marly
lithologies of the Liguridi Units (values 11 and 12 in the partial
dependance plot of Fig. 3B panel b), particularly the Nemoli
Complex (UAS), along with detrital landslide deposits, highlight the
highest susceptibility values. These materials are characterized
by cohesion values sufficient to maintain stable the ceiling of the
caves, even if lacking significant cementation to prevent internal
erosion due to groundwater flow. In contrast, massive carbonate
and dolomitic units display low susceptibility to piping, confirming
that the dominant processes in the study area are primarily
mechanical and hydraulic rather than karstic. In particular, the
genetic mechanisms of sinkholes developed in the area—formed
within cohesive clay-rich sediments with depths of 50 m —are
consistent with deep piping processes, i.e., upward-propagating
erosion driven by pressurized groundwater. In this context, the
karstified carbonate bedrock occurs at considerable depth and
does not directly control surface collapse, although localized
dissolution processes may still contribute to collapse sinkholes
formation in carbonate lithologies.
The final susceptibility map displays continuous values from
0 to 1, with warm colors indicating areas of higher probability of
piping sinkhole occurrence (Fig. 3A). The spatial pattern highlights
elevated susceptibility along the southwestern slope of Mount
Sirino, particularly where detrital and pelitic deposits, with depths
ranging between 30 and 50 m coincide with zones of high fault
density. The model successfully predicts all validated piping
sinkholes (gray circles in Figure 3), which fall within moderate to
high susceptibility zones. Moreover, the model successfully predicts
previously unrecognized high-susceptibility areas to be studied in
the future with detailed investigations, especially along the main
structural lineaments where similar geological and hydrological
conditions are present.
DISCUSSION
Interpretation of the Lake Sirino sinkhole mechanism
Lake Sirino originated as a collapse feature and represents a
key site for understanding sinkhole processes in the study area.
Lake Sirino is here considered as a natural laboratory to investigate
the interplay between lithological, structural, and hydrogeological
factors. The geological setting consists of permeable landslide
debris, approximately 30 m thick, overlying the low-permeability
Galestrino Flysch (Coppola et al., 2013), creating a perched aquifer
system. The hydraulic gradient, exceeding 200 m between the
lake and downstream springs, provides a strong driving force for
groundwater flow.
Geophysical investigations (Giampaolo et al., 2016) identified
a high-conductivity corridor beneath the southwestern shore,
associated with intense fracturing and representing a preferential
pathway for subsurface flow. The recurrence of drainage events
along a restricted sector of the southwestern shore indicates
a persistent hydrogeological control on sinkhole development.
Recent and historical events (e.g., 1994 and 2024) confirm the
ongoing activity of these processes in the same sector. Additional
evidence of instability, including recent subsidence affecting
infrastructure downslope, suggests that piping processes extend
beyond the immediate lakeshore.
Overall, the available evidence supports a deep piping
mechanism acting within the landslide deposits, rather than a deep
karstic or tectonically controlled collapse process.
Regional-scale controls on sinkhole development
At the regional scale, the results indicate that sinkholes
primarily originate from piping processes, governed by the
combined influence of lithological, structural, and hydrogeological
factors. Among these, lithological control emerges as the dominant
one. Pelitic and marly units, along with detrital deposits, provide
the critical combination of sufficient cohesion to maintain cavity
roofs, but insufficient cementation to prevent particle mobilization
by groundwater flow. In contrast, massive carbonate and dolomitic
formations are less affected by piping, and the few depressions
observed occur exclusively within these lithologies, where
dissolution and mechanical collapse prevail.
Structural control is supported by the alignment of sinkholes
along the NW-SE and NE-SW fault systems, particularly the Mercure
Fault System and the Nemoli-Vignale Fault Complex. Although the
statistical model assigned relatively low formal contribution to fault
density as an independent predictor, the response curves indicate
that faults exert substantial influence by creating preferential
pathways for groundwater circulation and zones of mechanical
weakness, where local stress concentrations facilitate cavity
development and propagation. The spatial coincidence between
high-susceptibility zones and fault corridors in the final map
underscores this structural control.
Slope dynamics may play an additional contributing factor
that deserves further investigation. The active landslide complex
described by Coppola et al. (2013) immediately downslope from
Lake Sirino could enhance piping processes by destabilizing the
debris system and maintaining or increasing the permeability
through deformation. The temporal coincidence between the July
2024 Lake shore collapse and recent road failures downslope
suggests a possible link between slope movements and piping
activation, though confirming such causation would require
continuous hydrogeological and geotechnical monitoring.
The limited number of validated events (n=10) represents an
inherent limitation for statistical modeling and raises the risk of
overfitting, where the model fits the training data well but may not
Accepted manuscript
8
P. PETRICCA ET AL.
generalize optimally to new locations. However, the relatively high
AUC value of 0.893 together with the accurate spatial prediction
of known sinkhole locations, supports the reliability of the
susceptibility model. The dominance of lithology and fault density
as key predictors is also consistent with the mechanics and physics
of the understood mechanisms of piping initiation and evolution.
Future studies should aim to prioritize the increase of the training
dataset through systematic regional surveys and the extension of
the study area to comparable geological settings.
The coexistence of both piping sinkholes and collapse within
the study area reflects the marked lithological diversity of the
region. Piping mechanism dominates in pelitic, marly, and detrital
formations, where particle mobilization by groundwater flow
governs the process, whereas rock collapse sinkholes occur in
carbonate and dolomitic units, where dissolution and mechanical
rock failure of cavern roofs prevail. This dual genesis highlights the
importance of careful field validation and genetic classification,
rather than relying solely on morphometric criteria for sinkhole
identification.
CONCLUSION
This study provides new insights into sinkhole development
in a tectonically active mountain environment, integrating field
surveys, remote sensing analyses and machine learning modeling
to investigate the mechanisms responsible for sinkholes formation
along the southwestern slope of Mount Sirino. In this highly
complex geological and geomorphological framework Lake Sirino
is comprised. It is formed by a landslide-dammed depression
and affected by recurrent piping-related collapses along its
southwestern shore. The 2024 collapse event, documented in this
work, confirms that the area remains highly susceptible to future
piping events.
At scale of the whole investigated area, extended over 100
km2, a total of 10 natural sinkholes were validated including
six attributed to piping mechanisms and four classified as rock
collapse sinkholes within carbonate lithologies. The MaxEnt model
provided a good performance (AUC=0.893) in predicting piping-
sinkhole susceptibility and identified lithology as the dominant
controlling predictor, followed by fault density. The predominance
of pelitic and detrital units within high-susceptibility zones supports
the interpretation that mechanical and hydraulic processes, rather
than karst-dominated dissolution, control sinkhole genesis in this
setting. The resulting susceptibility map constitutes a valuable
tool for identifying areas potentially prone to sinkhole occurrence,
supporting territorial planning, river basin management, and
seismic microzonation studies, as well as guiding the identification
of areas requiring further investigation and mitigation measures.
Future work should focus on establishing a hydrological monitoring
network comprising continuous monitoring of spring discharge,
lake level, and subsurface flow dynamics along the southwestern
shore and downstream sectors. Further, expanding field surveys
to adjacent areas would increase the training dataset and improve
model reliability. In addition, the integration of geochemical
tracers and continuous hydrogeological monitoring would enhance
understanding of flow pathways and support the early warning
capabilities of piping initiation. A bathymetric and sub-bottom
profiling survey of Lake Sirino, combined with underwater video
documentation, would represent a valuable next step to identify
morphological indicators of active piping at the lake bottom and to
better constrain subsurface flow pathways.
The methodological approach developed in this study,
combining field validation, remote sensing interpretation and
machine learning modeling, offers a replicable approach for sinkhole
risk assessment in comparable geological contexts characterized by
complex interactions among tectonics, lithology, and groundwater
circulation. Overall, the results highlight the value of integrated,
multi-source approaches that merge empirical observations with
quantitative spatial modeling to improve the understanding and
management of geological hazards in vulnerable landscapes.
ACKNOWLEDGEMENTS
The authors acknowledge the use of a locally optimized AI agent based on
Anthropic Claude Sonnet 4.5 to assist in drafting and refining the manuscript
text for improving clarity, grammar, and style. All scientific content,
interpretation, and conclusions were generated and verified by the authors
and remain their sole responsibility. This research has been conducted as part
of a Master’s thesis funded by European Union - NextGenerationEU - Mission
4 “Education and Research” - Component 2 “From Research to Business” -
Investment 3.1 “Fund for the realization of an integrated system of research
and innovation infrastructures” - Project IR0000037 - GeoSciences IR - CUP
I53C22000800006, and organized in collaboration between Università degli
Studi di Napoli Federico II, Università degli Studi della Tuscia, Università degli
Studi di Trieste and ISPRA.
REFERENCES
Al-Kouri O., Al-Fugara A.K., Al-Rawashdeh S., Sadoun B. & Pradhan
B. (2013) - Geospatial modeling for sinkholes hazard map based
on GIS & RS data. J. of Geograp. Inf. Sys., 5(06), 584, https://doi.
org/10.4236/jgis.2013.56055.
Caramanna G., Ciotoli G. & Nisio S. (2008) - A review of natural sinkhole
phenomena in Italian plain areas. Nat. Hazards 45, 145–172 (2008),
https://doi.org/10.1007/s11069-007-9165-7.
Ciotoli G., Ruggiero L., Mori F., Annunziatellis A., Di Salvo C., Finoia
M.G., Varone C., Moscatelli M. & Nisio S. (2026) - Integrating
susceptibility, vulnerability and exposure for screening level sinkhole
risk mapping in Metropolitan Rome (Italy), Eng. Geology, 368,
108740, https://10.1016/j.enggeo.2026.108740.
Coppola L., Filaridi A. & Bromhead E.N. (2013) - Landslide and flood
hazard from the Lago Sirino, Basilicata, Italy. Italian J. Eng. Geol.
Environ., 393-398.
D’Ecclesiis G., Grassi D., Sdao F. & Tadolini T. (1990) - Potenzialità
e vulnerabilità delle risorse idriche sotterranee del Monte Sirino
(Basilicata). Geol. Appl. Idrogeol., 25, 195-219.
De Vita P., Allocca V., Celico F., Fabbrocino S., Mattia C., Monacelli G.,
Musilli I., Piscopo V., Scalise A.R., Summa G., Tranfaglia G. & Celico
P. (2018) - Hydrogeology of continental southern Italy. J. Maps,
14(2), 230-241.
De Waele J. & Gutiérrez F. (2022) - Karst Hydrogeology,
Geomorphology and Caves. Wiley & Sons Ltd. https://doi.
org/10.1002/9781119605379.
Ford D.C. & Williams P.W. (1989) - Karst Geomorphology and Hydrology.
Unwin Hyman, London, 601 pp.
Accepted manuscript
9
SINKHOLE GENESIS AND SUSCEPTIBILITY AT MT. SIRINO
Forti P. (2002) - Evoluzione ipercarsica all’interno di acquiferi termali
e in rapporto a possibili problemi di suffosione. In: Le voragini
catastrofiche, un nuovo problema per la Toscana. Att. Conv. 31
marzo 2000, GR. Regione Toscana,11-26.
Giampaolo V., Capozzoli L., Rizzo E. & Grimaldi S. (2014) - Geophisical
and hydrogeological characterization of Sirino Lake (Basilicata,
Italy). XXXIII convegno GNGTS, Bologna, 25-27 novembre 2014.
Giampaolo V., Capozzoli L., Grimaldi S. & Rizzo E. (2016) - Sinkhole risk
assessment by ERT: The case study of Sirino Lake (Basilicata, Italy).
Geomorphology, 253, 1-9.
Grassi D., Grimaldi S., Sdao G. & Spilotro G. (2001) - Ambiente
idrogeologico e fenomeni di sifonamento relativi al Lago di frana
Sirino (Basilicata). Atti Geoitalia, 320-322.
Guida D. & Siervo V. (2005) - La carta inventario delle frane del Foglio
521 “Lauria” (Italia meridionale). Giorn. Geol. Appl., 2, 58-64.
Guida D., Nocera N. & Siervo V. (2006) - Analisi morfoevolutiva sulla
riattivazione di sistemi franosi a cinematismo intermittente in
Appennino campano-lucano (Italia meridionale). Giornale di
Geologia Applicata, 3, 114-122, https://doi.org/10.1474/GGA.2006-
03.0-15.0108.
Gutiérrez F. (2016) - Sinkhole Hazards. Oxford Research Encyclopedia of
Natural Hazard Science.
Gutiérrez F., Parise M., De Waele J. & Jourde H. (2014) - A review on
natural and human-induced geohazards and impacts in karst.
Earth-Science Reviews, 138, 61-88, https://doi.org/10.1016/j.
earscirev.2014.08.002
Nisio S. (2003) - I fenomeni di sprofondamento: stato delle conoscenze
ed alcuni esempi in Italia Centrale. Il Quaternario, 16(1) 2003,
121-132.Nisio S. (2008) - I fenomeni naturali di sinkhole nelle aree
di pianura italiane. Monografia Memorie descrittive della Carta
Geologica d’It. Vol. LXXXV; 475pp.
Nisio S. (2014) - I sinkholes tra storia, mito e leggenda. Mem. Descr.
Carta Geol. D’IT. 96, 271-296. Monaco C., Tortorici L. & Paltrinieri
W. (1998). Structural evolution of the Lucanian Apennines, southern
Italy, Journal of Structural Geology, 20(5), 617-638, https://doi.
org/10.1016/S0191-8141(97)00105-3.
Nisio S. (2023) - I sinkhole stato delle conoscenze e problematiche
aperte. Geologia dell’Ambiente 3/2023 ISSN 1591-5352, 91-96.
Nisio S., Caramanna G. & Ciotoli G. (2007) - Sinkholes in Italy: first results
on the inventory and analysis. In: Parise & Gunn (eds) Natural and
Anthropogenic Hazards in Karst areas: Recognition, Analysis and
Mitigation. Geological Society, London, Special Publications, 279,
23-45, https://doi.org/10.1144/SP279.40305-8719/07/$15.00.
Papanikolaou I.D. & Roberts G.P. (2007) - Geometry, kinematics and
deformation rates along the active normal fault system in the
southern Apennines: Implications for fault growth. J. Struct. Geol.,
29, 1, 166-188.
Petricca P. (2025) - CNN_wetland_detection, authomatic survey of
wet areas from Sentinel-2 images, https://github.com/BaterHub/
HydraNet-Wetland-Detection.
Phillips S.J., Dudík M. & Schapire R.E. (2025) - Maxent software for
modeling species niches and distributions (Version 3.4.1). Available
at: https://biodiversityinformatics.amnh.org/open_source/maxent/.
Sauro U. (2003) - Dolines and Sinkholes: aspects of evolution and
problems of classification. Acta carsologica, 32/2, 41-52.
Schiattarella M. (1998) - Quaternary tectonics of the Pollino Ridge,
Calabria-Lucania boundary, southern Italy. Geol. Soc. London Spec.
Publ., 135, 341-354.
Servizio Geologico d’Italia (2003) - Carta Geologica d’Italia alla scala
1:50.000 Foglio 521 “Lauria” . Servizio Geologico d’Italia - ISPRA,
Roma.
Taheri K., Shahabi H., Chapi K., Shirzadi A., Gutiérrez F. & Khosravi K.
(2019) - Sinkhole susceptibility mapping: A comparison between
Bayes-based machine learning algorithms. Land Degradation &
Development, 30(7), 730-745, https://doi.org/10.1002/ldr.3255.
Tarquini S., Isola I., Favalli M., Battistini A. & Dotta G. (2023) - TINITALY,
a digital elevation model of Italy with a 10 meters cell size (Version
1.1). Istituto Nazionale di Geofisica e Vulcanologia (INGV), https://
doi.org/10.13127/tinitaly/1.1.
Waltham T., Bell F. & Culshaw M. (2005) - Sinkholes and Subsidence:
Karst and Cavernous Rocks in Engineering and Construction.
Springer, Berlin. Warren, J.K., 2006. Evaporites: Sediments,
Resources and Hydrocarbons. Springer, Berlin, ISBN 3-540-20725-
2 Springer-Verlag Berlin Heidelberg New York, 382 pp., https://doi.
org/10.1007/b138363.
Accepted manuscript
论文原文与出版信息
Rendiconti online della Società Geologica Italiana