Segoni 2020
Segoni 2020
Abstract The literature about landslide susceptibility mapping is Many researchers recently presented new improvements in the
rich of works focusing on improving or comparing the algorithms techniques (Huang et al. 2017; Shirzadi et al. 2017; Pham et al. 2019)
used for the modeling, but to our knowledge, a sensitivity analysis or focused on the comparison between different models (Akgun
on the use of geological information has never been performed, 2012; Pham et al. 2016; Youssef et al. 2016; Bueechi et al. 2019), thus
and a standard method to input geological maps into susceptibility improving the state of the art with analysis tools capable of an
assessments has never been established. This point is crucial, increased effectiveness. Some works also performed a sensitivity
especially when working on wide and complex areas, in which a analysis to different model settings (resolution, scale, parameters,
detailed geological map needs to be reclassified according to more or methods to use parameters) that can be a reference for future
general criteria. In a study area in Italy, we tested different con- works to design a correct and robust model configuration (Catani
figurations of a random forest–based landslide susceptibility mod- et al. 2013; Greco and Sorriso-Valvo 2013).
el, accounting for geological information with the use of lithologic, Surprisingly, it seems that these undoubtedly useful and inter-
chronologic, structural, paleogeographic, and genetic units. Differ- esting aspects overshadowed the importance of geology in LSM: to
ent susceptibility maps were obtained, and a validation procedure the best of our knowledge, a sensitivity analysis on the use of
based on AUC (area under receiver-operator characteristic curve) geological information has never been performed, nor a standard
and OOBE (out of bag error) allowed us to get to some conclusions method to input geological maps into susceptibility assessments
that could be of help for in future landslide susceptibility assess- has ever been established.
ments. Different parameters can be derived from a detailed geo- Geological maps depict the spatial distribution on the topo-
logical map by aggregating the mapped elements into broader graphic surface of rocks with different ages, natures, and charac-
units, and the results of the susceptibility assessment are very teristics. The aim is not limited to represent the spatial
sensitive to these geology-derived parameters; thus, it is of para- relationships of different mapped units, but it involves also the
mount importance to understand properly the nature and the conveying of information about the Earth’s crust evolution (Butler
meaning of the information provided by geology-related maps and Bell 1988). Thus, geological maps are not directly conceived to
before using them in susceptibility assessment. Regarding the assist landslide modeling, and sometimes, they can be subdivided
model configurations making use of only one parameter, the best into a very large number of mapped elements, some of which are
results were obtained using the genetic approach, while lithology, not directly related to slope stability (e.g., different units may be
which is commonly used in the current literature, was ranked only defined based on the appearance/disappearance of a fossil species,
second. However, in our case study, the best prediction was ob- even if they have the same lithology and geomechanical
tained when all the geological parameters were used together. characteristics).
Geological maps provide a very complex and multifaceted infor- In the recent literature, it is possible to identify two main
mation; in wide and complex area, this information cannot be methodologies to use geological information in LSM: (i) pre-
represented by a single parameter: more geology-based parame- existing thematic maps about the nature of the bedrock are
ters can perform better than one, because each of them can retrieved and used without further elaborations (Pourghasemi
account for specific features connected to landslide predisposition. and Kerle 2016; Pradhan et al. 2019; Xiao et al. 2019); (ii)
existing detailed digital geological maps undergo a reclassifica-
Keywords Susceptibility . Random tion aimed at reducing the number of classes and at defining a
forest . Comparison . Sensitivity . Geology . Lithology stronger correlation with landsliding. This approach is used
especially when working over large or complex areas, where a
Introduction too detailed geological information needs to be generalized
Landslide susceptibility mapping (LSM henceforth) is a very im- (Bălteanu et al. 2010; Van Den Eeckhaut et al. 2012; Trigila
portant activity in landslide hazard assessment, consisting in et al. 2013). Usually, the reclassification is performed on a
representing over appropriate spatial units the relative spatial lithological or lithotechnical basis, in search for a stronger
probability of landslide occurrence (Brabb 1984). The overwhelm- correlation with landslide predisposition (Catani et al. 2005;
ing literature on landslide susceptibility is rich of works presenting Segoni et al. 2018), and may entail some subjectivity in the
applications to case studies from the local to the global scale choice of classes.
(Youssef et al. 2016; Pradhan et al. 2019; Yang et al. 2019; Trigila In any case, the lithological information is the most recurrent in
et al. 2013; Günther et al. 2013; Hong et al. 2007) obtained mainly LSM literature (Akgun 2012; Youssef et al. 2016): different units are
by means of statistical techniques (Reichenbach et al. 2018, and grouped according to the main lithologies encountered. In turn,
references therein), artificial neural networks (Catani et al. 2005), lithologies are defined based on physical characteristics such as
and machine learning algorithms (Brenning 2005, and references grain size, texture, and constituting minerals, which are clearly
therein; Catani et al. 2013). connected with slope stability.
Landslides
Technical Note
Some studies (Catani et al. 2005; Manzo et al. 2013) use a each other because of compressive tectonic forces, and since
lithotechnical approach that moves from a lithological basis but the Upper Tortonian, an extensional tectonic regime dissected
defines classes according to the typical values (measured, estimat- the aforementioned units producing a horst and graben struc-
ed, or supposed) of their shear strength parameters. The advantage ture. Later, in the Pliocene and Quaternary, marine and fluvio-
of this approach is that a stronger correlation between classifica- lacustrine basins fostered the sedimentation of terrains, filling
tion and landsliding should be provided by the focus on the the tectonic depressions.
geotechnical parameters. The drawback is that spatially distributed This tectonic evolution has now left the presence of NW-SE
maps of geotechnical properties of surface deposits are usually trending ridges and a pervasive pattern of faults, thrusts, and folds.
unavailable, with the exception of some recent studies (e.g., Roughly speaking, the area can be divided in two sectors. In the
Bicocchi et al. 2019). Other studies use also geological units, which western one, the bedrock is mainly constituted by carbonaceous
are mainly based on the age of deposition of the units (Van Den rocks and metamorphic rocks (mainly metamorphic sandstones
Eeckhaut et al. 2012; Xiao et al. 2019) even if in local-scale studies and phyllitic schists), giving rise to a sharper relief with higher
this could also partially or totally reflect lithological aspects slope gradients than in the eastern sector, which is characterized
(Youssef et al. 2016; Cárdenas and Mera 2016; Pourgashemi and by sedimentary rocks (mainly flysch-alternating layers of different
Kerle 2016). Besides, a number of alternate approaches can be lithologies and textures). The mountainsides are mantled by a
found in the recent literature about landslide susceptibility map- residual colluvium with a thickness ranging from a few centime-
ping: as instance, Meinhardt et al. (2015) use a broad classification ters to a few meters (Mercogliano et al. 2013), which exhibit a
into 5 rock types subdivided by the different genesis (e.g., loose marked contrast in geotechnical properties with respect to the
soils and metamorphic, intrusive, extrusive, and sedimentary bedrock (Tofani et al. 2017).
rocks); Myronidis et al. (2016) make use of units classified based The area is a significant hotspot of landslides (Battistini et al.
on the geological structure; Camarinha et al. (2014), to study 2013) and it is affected by landslides of different typologies (Segoni
shallow landslide susceptibility in a test site in Brazil, used a et al. 2014): rotational and translational slides in the colluvium and
rock-type classification based on a complex criterion complex movements (slides evolving into slow flows) are the most
encompassing geology, lithology, and pedology. representative, but also, debris flows and rockfalls are present. The
The examples above show that a standard method to consider typical areal extension of the landslides ranges from 102 to 106 m2
geology in LSM has not been defined, nor a sensitivity analysis to and rainfall is the main triggering factor. The area receives an
the parameterization of geology has ever been performed (to the average annual precipitation around 2000 mm/year in the moun-
best of our knowledge). tains and 1100 mm/year in the plains, with short and intense
In a large and complex area, a detailed geological map could be rainstorms increasing their frequency and severity as a conse-
classified using different approaches. All of them could make sense quence of the recent trends in climate change (Segoni et al. 2015).
from a geological point of view and all of them could be put in
relation with landslide predisposition. The main objective of this Susceptibility assessment
work is to test several of these approaches and to try to answer the The susceptibility assessment was carried out using an updated
following research questions: (i) do the derived susceptibility out- version of the software program named Claret (Lagomarsino
puts show noticeable differences? (ii) Which approach performs et al. 2017), which is based on the Treebagger random forest
better? (iii) Can we devise some best practices to handle geological machine learning technique and which automates several pas-
data for future landslide susceptibility assessments? sages of the analysis, including random selection of the
To pursue this objective, we selected a 3000-km2 test site in training/test dataset, evaluation and ranking of the input var-
Italy characterized by a very complex geological setting, we classi- iables according to their explanatory power, and identification
fied the mapped geological units according to six different ap- and application of the optimal regression model, model
proaches, and we used them as input variables in several tests of a validation.
landslide susceptibility model based on the Treebagger random The random forest is a technique that has been applied to
forest algorithm (Breiman 2001; Catani et al. 2013). Afterwards, we landslide susceptibility only recently (Brenning 2005; Catani
compared the results of the tests in terms of AUC (area under et al. 2013); however, it can be considered well-established because
receiver-operator characteristic curve) and OOBE (out of bag its use has been consolidated through many applications in differ-
error). Results were then critically analyzed and discussed, pro- ent case studies and because it has often shown better perfor-
viding new insights on the effectiveness of different parameteriza- mances when compared with other state-of-the-art
tion approaches and on the possible use of geological information methodologies (Trigila et al. 2013; Xiao et al. 2019). In addition, it
in landslide susceptibility assessment. is very flexible and straightforward to apply, because it can handle
at the same time both numerical and categorical variables, it
Material and methods implicitly accounts for mutual dependency between variables, it
reduces overfitting, and it does not require particular assumptions
Study area on the statistical distribution of the values of the data.
The study area is located in Northern Tuscany, Italy, and it is a Based on the experience learnt from past sensitivity studies
3100-km 2 territory composed by hills, mountains (up to (Catani et al. 2013), the following model configuration was used
2000 m of elevation), and limited alluvial plains (Fig. 1). The in this work:
geological setting is quite complex (Vai and Martini 2001): the
area belongs to the Northern Apennine fold and thrust belt, – Training data were sampled randomly in the landslide and in
and since the Tertiary different tectonic units were stacked on the non-landslide areas;
Landslides
Fig. 1 Study area and landslides
– In the sampled data, landslide conditions and no-landslide movements together. Another argument supporting this decision
conditions are balanced 50–50% to guarantee a balanced is that in the inventory, the difference among these two categories
prediction; in small, thus the category in which a landslide is classified may
– The sampled dataset was randomly split into a training (70%) vary according to the subjective interpretation of the surveyors,
and a test (30%) subset, in which landslide and no-landslide which were different from province to province. About landslides
conditions are still equally balanced. with unspecified movements, this is a shortcoming of the classifi-
– We performed a regression using a 200-tree forest structure, cation in the IFFI inventory and it can be easily resolved observing
and for each analysis, the model was run 20 times to get more the geometrical features of the polygons and the analogy with the
robust conclusions. dominant landslide typology of the area: these landslides can be
also grouped together with slides and complex movements (Catani
et al. 2016; Segoni et al. 2016). Our analyses do not account for
The dependent variable used in susceptibility assessments is rockfalls and debris flows, because their number in the study area
represented by landslide and no-landslide conditions. To train and is limited and their spatial distribution is confined only to some
validate the susceptibility model, we used an excerpt of the Italian spots. These shortcomings prevent form having a good sample for
national inventory of landslides (IFFI) at 1:10,000 scale, which is a rigorous susceptibility analysis regarding falls and debris flows.
the most complete catalogue of landslides available in Italy (Trigila Moreover, since the triggering mechanism and predisposing fac-
et al. 2010), as recently updated by means of radar satellite inter- tors of rockfalls and debris flows may be very different from those
ferometry technique (Rosi et al. 2018). However, before preparing characterizing slides, these typologies could not be grouped with
the landside dataset for the susceptibility model, some preliminary slides in the same susceptibility analysis. As a consequence, only
considerations on the mapped landslide typologies are needed. In the mapped landsides that can be associated with the “slides”
the study area, the inventory includes 3452 slides, 3294 complex typology of movement were extracted from the IFFI database
movements, 16 falls, 540 rapid flows, and 522 landslides of unspec- and used in the susceptibility assessment.
ified typology. According to the general knowledge of the charac- Concerning the independent variables, there is no consensus on
teristics of the IFFI inventory in the Tuscany Apennines, and the number of parameters to use in a susceptibility assessment:
according to some previous studies (Trigila et al. 2013; Segoni many studies have been published that make use of a high number
et al. 2016), unspecified movements and complex movements were of parameters (Lee and Pradhan 2007; Nefeslioglu et al. 2011;
aggregated with the “slides” typology for the following reasons. In Segoni et al. 2015; Xiao et al. 2019) and many that use only a few
the study area, complex movements are typically represented by (Hong et al. 2007; Akgun 2012; Manzo et al. 2013). In this study, we
compound slides evolving into slow flows: since the triggering decided to use a very basic and reduced set of parameters, because
mechanism and predisposing factors are the same, the suscepti- the objective is to explore the sensitivity to geology and to focus
bility assessment can consider both slides and complex the discussion on the impact of this parameter. Using too many
Landslides
Technical Note
parameters is not useful to this purpose; rather, it could cast a However, it is worth being tested, as it could relate to landslide
shadow on the effect of geology. Consequently, 7 parameters have susceptibility since each structural unit through geological times
been selected that are broadly used in literature and that resulted was subject to a particular tectonic stress-strain history, including
in a high predictive power in other susceptibility assessment per- uplifting, folding, faulting, displacement, and thrusting. All these
formed in this study area (Segoni et al. 2016): elevation, slope processes may be responsible of weakening the bedrock and pre-
gradient, land cover, flow accumulation, aspect, planar curvature, disposing it to instabilities.
geology. •Broader structural approach (5 classes)—The same as before,
All morphologic and hydrologic parameters (elevation, slope but a broader classification was performed, grouping together
gradient, flow accumulation, aspect, and planar curvature) similar units to obtain a number of classes (five) comparable with
were derived by a digital elevation model with 10-m spatial the one used for all the other criteria (Fig. 2d).
resolution. Land cover was derived by the 1:50,000 Corine •Paleogeographic approach (Fig. 2e)—The Apennine has been
Land Cover maps, adopting a site-specific reclassification into traditionally subdivided also in paleogeographic units, according
9 classes (urban areas, crops, grasslands, heterogenic rural mainly to the paleogeographic environment where rocks were
areas, broad-leaved forests; conifer forests; shrubs; bare rocks; originally formed. We therefore used the main paleogeographic
humid areas) already used in previous landslide susceptibility units outcropping in the test area (all recent deposits were
works carried out in the same test site (Segoni et al. 2016; grouped together). To our knowledge, a similar approach has
Segoni et al. 2018). Concerning geology, its use as an explana- never been used in LSM. We decided to test it because rocks with
tory variable is more complex and it is explained in detail in similar lithologies could have different mineralogical or textural
the following section. characteristics according to the environment of deposition, thus
potentially providing a slightly different predisposition to
Geological information landsliding.
The basic information on geology is a 1:10,000 scale digital •Chronological approach (Fig. 2f)—In this case, the geological
geological map produced and made available by the Tuscany units were grouped according to the age of deposition. This crite-
Region. In the study area, it encompasses 194 lithostratigraphic rion has been already used in the LSM literature, even if quite
units. Such a detailed information cannot be used in a land- rarely; however it could be regarded as a potential predisposing
slide susceptibility assessment based on statistical machine factor because broadly speaking, the older a geological unit, the
learning algorithms: the number of classes is too high and higher the degree of weathering and the exposition to tectonic
some of the classes have a limited spatial extension, thus stress, and this can be linked to the reduction in its strength
posing problems for an adequate model calibration. This is a parameters.
common issue in landslide susceptibility assessments over
large areas, and it is commonly solved by grouping the avail- Description of the tests
able units into broader classes (Bălteanu et al. 2010; Van Den All data were imported in a GIS, where 15,500 random points were
Eeckhaut et al. 2012; Manzo et al. 2013). In this study, the generated (50% inside landslide polygons, 50% outside landslide
lithostratigraphic units were grouped into classes following polygons) and used to sample the values of dependent and inde-
six different approaches. Each approach uses a specific criteri- pendent variables; 70% of them were used to train the model and
on related to a single geological characteristic; therefore, two 30% were used for test. The random forest model was run several
geological units may be grouped in the same class according to times, varying the configuration used to account for the geologic
a criterion and may be separated into different classes accord- information: a base test in which no information on geology was
ing to another criterion. All partitioning approaches make used, one test for each of the classification criteria described in the
sense from the standpoint of geological reasoning, and some previous section (six tests), an additional test using all of them
functional links can be found to relate them to landslide together.
predisposition (even if their actual connection to landslide The used model has an internal test procedure that allows for
susceptibility is to be objectively and quantitatively assessed an objective evaluation of the predictive capacity of each variable
in the following sections): used and of the model itself. Two parameters have been
•Lithologic approach—Geological units were classified accord- considered:
ing to their prevailing lithology. This is probably the criterion that
has been more widely used in LSM and has years of literature – AUC (area under ROC curve)—this parameter estimates the
relating different lithologies to different degrees of landslide sus- predictive effectiveness of the regression performed. The near
ceptibility (Fig. 2a). the AUC value is to 1, the better the regression approximates
•Genetic approach—Geological units were grouped into five the experimental data as an expression of reality. It can be used
broad classes according to the genetic process that gave them to quantitatively compare different model configurations in
birth: magmatic rocks, metamorphic rocks, clastic rocks, terms of their prediction effectiveness.
organogenic rocks, soils (Fig. 2b). – OOBE (out of bag error)—this parameter is an estimate of
•Detailed structural approach (Fig. 2c)—The Apennine chain the relative error that would be committed if the variable to
has a very complex structure, and geologists have subdivided it whom it is referred were not used in the regression. During
into several structural units according to their evolution and each test, it is automatically calculated for all independent
response to tectonic forcing. In the study area, 10 different struc- variables and allows quantitative assessment of their explan-
tural units outcrop, and they were used in this criterion. To our atory power. OOBE can be used to rank the variables ac-
knowledge, a similar approach has never been used in LSM. cording to their importance.
Landslides
Fig. 2 Reclassification of the geological map according to six different criteria: lithological (a), genetic (b), detailed structural (c), structural (d), paleaogeographic (e),
chronological (f)
Landslides
Technical Note
While AUC allows for a comparison among different model Table 1 Ranking of the model configurations according to their effectiveness (a-
configurations, OOBE allows for a comparison among the vari- ssessed by the AUC values)
ables used in a specific configuration. These parameters are cal- Test Mean AUC Max AUC
culated for each model configuration after running and averaging
20 times, to get stable results and to account for the inherent BASE configuration (no geology) 0.606 0.630
chaotic nature of the random forest algorithm. BASE + chronological units 0.635 0.656
In addition, a preliminary run of the model was performed
BASE + structural (detailed) 0.640 0.652
adding to the full configuration a random control variable. This
variable was obtained by applying a random value (from 1 to 5) in BASE + paleogeographic units 0.648 0.665
each pixel of the study area, thus simulating a completely random BASE + structural units (broad) 0.657 0.676
reclassification of the spatial units used in the susceptibility as- BASE + lithologic units 0.661 0.682
sessment. The objective of this test is evaluating if the random
reclassification outperforms some of the other approaches used. In BASE + genetic units 0.700 0.724
case it happens, an approach with a predictive effectiveness lower Full configuration (base + all criteria) 0.752 0.774
than random values would be deemed inappropriate and
discarded. As shown in Fig. 3, all variables had an OOB higher
than random variable; therefore, all of them were used in the tests maximum AUC 0.724). However, the configuration that includes
as potentially suitable predictors. in the modeling of all six reclassification criteria is by far the most
effective, with mean and maximum AUC of 0.752 and 0.774,
Results respectively. Further insights on these outcomes are provided in
Figure 3 shows the susceptibility maps obtained with the eight the “Discussion” section. The ranking of the configurations is
configurations that were tested. Susceptibility values range from identical in case mean AUC or max AUC is taken into account,
0 to about 0.70, and different spatial distributions of the values can thus demonstrating the stability of the results obtained in the tests.
be observed with a visual inspection. During the tests, the accuracy It is worth remembering that the objective of the present work is
of these maps was assessed to understand which one is the most not only to produce a susceptibility map of the area but to also
reliable. The results of the validation are summarized in Table 1, provide a sensitivity analysis of the susceptibility model with
which shows the mean and the maximum AUC values obtained respect to the geological information. The AUC values are lower
with the 20 runs of each configuration. AUC can be used to than the one obtained in past works on the same test site (Segoni
estimate the forecasting effectiveness of each configuration and et al. 2016), because a model configuration using a limited number
thus to rank them accordingly. of parameters is used here, to better focus the sensitivity analysis
From Table 1, it can be seen that the worst performance is on the impact of geology.
obtained with the configuration that neglects geological informa- In Fig. 4, the importance of each variable inside each configu-
tion (base configuration) and that all the reclassification criteria ration is assessed by means of the OOBE. It can be seen that for the
used in the tests, when used singularly, provide an amelioration geological parameters, the importance is always relatively high
with respect to the base configuration. Among these, the best (OOBE around 1.40 and geological variables ranked among the
forecasting effectiveness is obtained when the geological units first ones), except when chronological units (and, to a lesser
are aggregated following a genetic approach (mean AUC 0.700, extent, lithologic units) are taken into account.
Discussion
The examination of the AUC values obtained in our tests (Table 1)
reveals that geology is very important in landslide susceptibility
assessment: the model configuration that does not encompass any
geological information is by far the one providing the worst
prediction. This result was expected; what we consider more im-
portant is to discuss the sensitivity of a landslide susceptibility
model to the different approaches that can be used to parameterize
the geological information. A geological classification based only
on the age of formation of the units is the worst performing one in
our case of study, but, still, it does provide an improvement over
the basic version of the model. In a few words, none of the six
configurations tested is completely unrelated to landslides. This
highlights an important operational and technical consideration:
when working on areas with limited geological information, a
susceptibility assessment may still gain from the inclusion of
geological mapping among the preparatory factors, whatever its
accuracy, scale, or mapped units. Of course, this does not mean
that geological information could be entered in the modeling
Fig. 3 Estimation of the relative importance of the predictors, including a control carelessly; on the contrary, our tests show that the approach used
random variable. Predictors with geological significance are highlighted in orange to define the geologic units may have a deep impact in the spatial
Landslides
Fig. 4 Susceptibility maps obtained with the different configurations tested: base configuration (a), full configuration using all geological information (b), using
paleogeographic units (c), using detailed structural units (d), using genetic units (e), using broad structural units (f), using lithological units (g), using chronological units
(h)
Landslides
Technical Note
distribution of susceptibility values and in the effectiveness of the formed in lacustrine or sea environment, providing slightly differ-
modeling. Therefore, during the data collection phase, the inves- ent responses to the geomorphological processes acting on the
tigators should be aware of the geological meaning of the units hillslopes. The same applies for structural units that could differ-
defined in their maps. In our case study, the best outcomes are entiate similar rocks, even if formed in similar geologic times,
obtained when using genetic units, which reclassified the original according to their different tectonic histories. Following this rea-
lithostratigraphic units into broad classes according to the pro- soning, different assumptions could be considered not mutually
cesses leading to the formation of rocks and soils. The most widely exclusive and could be used jointly, to better encompass the
used lithologic approach is ranked only as the second-best ap- complex geological characteristics of a given test site. The separa-
proach, with performance indicators close to those obtained by the tion of different characteristics of geological units into different
broad structural approach. The distinction between the detailed independent variables may also have the additional advantage of
and the broad structural approaches deserves attention in our providing the statistical engine with an orderly set of data to
opinion. All the criteria used in this work have a very similar classify from, avoiding complex, multi-component variables. The
number of classes (5 or 6); this characteristic ensures that the drawback is that different parameters related to geology would be
sensitivity analysis is not influenced by different degrees of detail used in the modeling at the same time, thus bringing problems of
in the classification schemes pursued. The only exception is the collinearity and interdependence among the variables. However, it
structural criterion, in which originally, 10 classes were present, should be noted that many advanced machine learning algorithms
and which was reclassified into 5 classes (broad structural criteri- (such as the random forest) are widely credited for not being
on) to be consistent with the other criteria. However, we kept in a negatively influenced in similar cases. In addition, we observe that
separate configuration also the detailed structural criterion (based in the international literature, the use of more than one parameter
on 10 classes), to test how the degree of detail in the classification describing morphology is widely accepted (and the same can be
affects the results. The use of broader structural units provides said for hydrology). Moreover, also, the use of morphological or
better results than using the detailed ones (higher AUC and OOBE hydrological variables that are strongly correlated is a common
values) (Fig. 5); this result can have two interpretations: either the practice, e.g., topographic wetness index and drainage area, or
regression model performs a better regression with a “simple” and stream power index and slope gradient, or different definitions
more generic level of classification, or in the detailed classification, of curvature, may be used simultaneously.
some units are defined that are scarcely significant in describing This work explores only a few approaches (namely, six) to
the predisposition to landsliding. parameterize the geological information and proposes a procedure
One of the most significant outcomes of this work is that the to select the optimal ones. Of course, in other study areas, other
validation results show that the susceptibility map with the highest approaches could be defined according to the test site character-
forecasting effectiveness is obtained when all the geological clas- istics and to the information available. In addition, it should be
sification schemes are used together: the AUC is markedly higher stressed that even when using the same approaches used here,
than the AUCs obtained with any other model configurations different results could be obtained in other locations. Thus, we
relying on a single geological parameter. We explain this outcome recommend implementing the susceptibility assessment with a
observing that geological information may be faceted and very forward selection of parameters: using this technique, the pejora-
difficult to encompass into susceptibility models, because there are tive or redundant parameters would be automatically identified
many features related to geology that may influence the spatial and filtered off.
distribution of landslides; therefore, more features related to ge- The examination of the results in terms of OOBE provided
ology should be taken into account with different classification interpretations consistent with the findings in the earlier
schemes; conversely, if only one approach is used, only a partial discussion (Fig. 5). Considering OOBE to rank the importance of
information is entered into the modeling. For instance, most of the the variables used inside each model configuration, it can be
works found in the literature use a lithological classification, with a observed that geology is one of the most important parameters,
strong supporting reason: landslide activity is clearly related to the generally the 2nd or 3rd among 7 parameters. It is important to
shear strength parameters of the hillslope material, and broadly highlight some exceptions: when geology is the 1st parameter, the
speaking, different lithologies have different ranges of strength configuration provides the highest AUC (configuration based on
parameters values. However, one may argue that two different genetic units), thus confirming that when geology is parameterized
units with the same prevailing lithology may have very different for the best, its explanatory power is fully exploited by the regres-
weathering degree and thus different mechanical characteristics: sion algorithm to get to optimal results. Conversely, when the
from this point of view, the age of the formation could also be importance of geology is lower (4th rank), a reduction in the
related to landslide susceptibility. However, our tests showed that model performance can be observed (as in the case of the use of
a chronological classification alone is of little use (it is the worst chronological units). This highlights the importance of a correct
aggregation criterion among all the tested ones), probably because parameterization of the geological information to get a reliable
in similar ages very different rocks or terrains could have been susceptibility assessment.
formed. This corroborates the hypothesis that the joint use of When coming to observe the OOBE values in the configuration
lithological and chronological information in the parameterization using all geological classification schemes together, the values and
of geology should not be regarded as a redundant information. the ranking of the geological variables may seem exceptionally low.
Rather, the two information are complimentary. The same exam- This is not in contrast with the conclusion that this is the best
ple could be extended to other classification approaches: geolog- model configuration: the geological information provided to the
ical units with the same lithology could have been originated in model is more complete than in other configurations, thus pro-
very different depositional environment, e.g. clays could have been viding the highest AUC value, the low OOBE values of geological
Landslides
Fig. 5 Out of bag error (OOBE) of the variables used in every test. Predictors with geological significance are highlighted in orange
Landslides
Technical Note
factors become because the importance of geology is shared thus providing an advanced and complete information to the
among six different factors. If the OOBE values of the geological susceptibility model that, in turn, provided a better prediction.
factors are summed up, a weight of 2.9 is obtained that is similar to
the weight obtained summing up the contribution of the factors
with a morphological (slope, elevation) and a hydrological (aspect,
planar curvature and flow accumulation) meaning (2.7 and 3.0, Acknowledgments
respectively), thus confirming that the geological information is This work was supported by Florence University in the framework
among the most important ones in a susceptibility assessment and of the project SEGONISAMUELERICATEN19.
that the task of providing such information can be shared among
different parameters.
Open Access This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing,
Conclusion
adaptation, distribution and reproduction in any medium or for-
Since there is no agreement on how geological information should
mat, as long as you give appropriate credit to the original au-
be included in LSM activities, we performed a sensitivity analysis
thor(s) and the source, provide a link to the Creative Commons
and a series of tests in a study area with a complex geological
licence, and indicate if changes were made. The images or other
setting to (i) ascertain the sensitivity of the results to different
third party material in this article are included in the article's
approaches used to reclassify geological units and (ii) to get some
Creative Commons licence, unless indicated otherwise in a credit
insights on the best approaches to pursue in future susceptibility
line to the material. If material is not included in the article's
assessments.
Creative Commons licence and your intended use is not permitted
The study area is located in northern Tuscany (Italy) and a
by statutory regulation or exceeds the permitted use, you will need
regional 1:10000 map, when clipped on the area of interests,
to obtain permission directly from the copyright holder. To view a
encompassing 194 different lithostratigraphic units. Such a de-
copy of this licence, visit http://creativecommons.org/licenses/by/
tailed information was reclassified into broader units according
4.0/.
to six different criteria; all of them somehow are potentially con-
nected with landslide susceptibility: lithology, age, paleogeograph-
ic environment of formation, genetic process; tectonic history
References
(according to two different degree of detail).
Using the well-established random forest technique, we per- Akgun A (2012) A comparison of landslide susceptibility maps produced by logistic
formed several landslide susceptibility assessments, varying the regression, multi-criteria decision, and likelihood ratio methods: a case study at İzmir,
approach to use the geological information. The resulting maps Turkey. Landslides 9(1):93–106. https://doi.org/10.1007/S10346-011-0283-7
Bălteanu D, Chendeş V, Sima M, Enciu P (2010) A country-wide spatial assessment of
were validated, and the different configurations were evaluated
landslide susceptibility in Romania. Geomorphology 124:102–112. https://doi.org/
with a series of internal tests. The comparison of the results 10.1016/j.geomorph.2010.03.005
supported the following conclusions: Battistini A, Segoni S, Manzo G, Catani F, Casagli N (2013) Web data mining for
automatic inventory of geohazards at national scale. Appl Geogr 43:147–158.
– Geology is one of the most important parameters in LSM. https://doi.org/10.1016/j.apgeog.2013.06.012
Bicocchi G, Tofani V, D’Ambrosio M, Tacconi-Stefanelli C, Vannocci P, Casagli N, Lavorini
– Geological maps provide a very complex and multifaceted
G, Trevisani M, Catani F (2019) Geotechnical and hydrological characterization of
information, in wide and complex areas this information can- hillslope deposits for regional landslide prediction modeling. Bull Eng Geol Environ
not be effectively represented by a single parameter. 81:122–4891. https://doi.org/10.1007/s10064-018-01449-z
– Different parameters can be derived from a detailed geological Brabb EE (1984) Innovative approaches to landslide hazard mapping, 1st edn. Proceed-
map by aggregating the mapped elements into broader units ings 4th International Symposium on Landslides, Toronto, pp 307–324
Breiman L (2001) Random forests. Mach Learn 45:5–32
(in our study: lithologic, chronological, paleogeographic, struc-
Brenning A (2005) Spatial prediction models for landslide hazards: review, comparison
tural, and genetic units). and evaluation. Nat Hazards Earth Syst Sci 5:853–862. https://doi.org/10.5194/nhess-
– The results of the susceptibility assessment are very sensitive to 5-853-2005
these geology-derived parameters; thus, it is of paramount Bueechi E, Klimeš J, Frey H, Huggel C, Strozzi T, Cochachin A (2019) Regional-scale
importance to understand properly the nature and the mean- landslide susceptibility modelling in the Cordillera Blanca, Peru—a comparison of
different approaches. Landslides 16:395–407. https://doi.org/10.1007/s10346-018-
ing of the information provided by geology-related maps be-
1090-1
fore using them in susceptibility assessment. Butler BCM, Bell JD (1988) Interpretation of geological maps. Longman earth science
– Regarding the configurations making use of only one parame- series. Longman Scientific & Technical; Wiley, Harlow, New York
ter, the best results were obtained using the genetic approach, Camarinha PIM, Canavesi V, Alvalá RCS (2014) Shallow landslide prediction and analysis
while lithology, which is commonly used in the current litera- with risk assessment using a spatial model in a coastal region in the state of São
ture, was ranked only second. Paulo. Brazil Nat Hazards Earth Syst 14(9):2449–2468
Cárdenas NY, Mera EE (2016) Landslide susceptibility analysis using remote sensing and
– However, in our case of study, the best prediction was obtained GIS in the western Ecuadorian Andes. Nat Hazards 81(3):1829–1859
when all the geological parameters were used together. Catani F, Casagli N, Ermini L, Righini G, Menduni G (2005) Landslide hazard and risk
– Different geology-based parameters can perform better than mapping at catchment scale in the Arno River basin. Landslides 2:329–342. https://
only a geological parameter, because each of them can account doi.org/10.1007/s10346-005-0021-0
for specific settings connected to landslide predisposition. In Catani F, Lagomarsino D, Segoni S, Tofani V (2013) Landslide susceptibility estimation by
random forests technique: sensitivity and scaling issues. Nat Hazards Earth Syst Sci
our case of study, the use of six different geological parameters 13:2815–2831. https://doi.org/10.5194/nhess-13-2815-2013
allowed to account for lithology, tectonic stress, age and envi- Catani F, Tofani V, Lagomarsino D (2016) Spatial patterns of landslide dimension: a tool
ronment of formation, and subsequent degree of weathering, for magnitude mapping. Geomorphology 273:361–373
Landslides
Greco R, Sorriso-Valvo M (2013) Influence of management of variables, sampling zones Segoni S, Rossi G, Rosi A, Catani F (2014) Landslides triggered by rainfall: a semi-
and land units on LR analysis for landslide spatial prevision. Nat Hazards Earth Syst Sci automated procedure to define consistent intensity–duration thresholds. Comput
13:2209–2221. https://doi.org/10.5194/nhess-13-2209-2013 Geosci 63:123–131. https://doi.org/10.1016/j.cageo.2013.10.009
Günther A, Reichenbach P, Malet J-P, van den Eeckhaut M, Hervás J, Dashwood C, Segoni S, Tofani V, Lagomarsino D, Moretti S (2016) Landslide susceptibility of the
Guzzetti F (2013) Tier-based approaches for landslide susceptibility assessment in Prato–Pistoia–Lucca provinces, Tuscany, Italy. Journal of Maps 12:401–406. https://
Europe. Landslides 10:529–546. https://doi.org/10.1007/s10346-012-0349-1 doi.org/10.1080/17445647.2016.1233463
Hong Y, Adler R, Huffman G (2007) Use of satellite remote sensing data in the mapping Segoni S, Tofani V, Rosi A, Catani F, Casagli N (2018) Combination of rainfall thresholds
of global landslide susceptibility. Nat Hazards 43:245–256. https://doi.org/10.1007/ and susceptibility maps for dynamic landslide hazard assessment at regional scale.
s11069-006-9104-z Front Earth Sci 6:21. https://doi.org/10.3389/feart.2018.00085
Huang F, Yin K, Huang J, Gui L, Wang P (2017) Landslide susceptibility mapping based Shirzadi A, Bui DT, Pham BT, Solaimani K, Chapi K, Kavian A, Shahabi H, Revhaug I (2017)
on self-organizing-map network and extreme learning machine. Eng Geol 223:11–22. Shallow landslide susceptibility assessment using a novel hybrid intelligence ap-
https://doi.org/10.1016/j.enggeo.2017.04.013 proach. Environ Earth Sci 76:1515. https://doi.org/10.1007/s12665-016-6374-y
Lagomarsino D, Tofani V, Segoni S, Catani F, Casagli N (2017) A tool for classification and Tofani V, Bicocchi G, Rossi G, Segoni S, D’Ambrosio M, Casagli N, Catani F (2017) Soil
regression using random forest methodology: applications to landslide susceptibility characterization for shallow landslides modeling: a case study in the Northern
mapping and soil thickness modeling. Environ Model Assess 22:201–214. https:// Apennines (Central Italy). Landslides 14:755–770. https://doi.org/10.1007/s10346-
doi.org/10.1007/s10666-016-9538-y 017-0809-8
Lee S, Pradhan B (2007) Landslide hazard mapping at Selangor, Malaysia using Trigila A, Frattini P, Casagli N, Catani F, Crosta G, Esposito C, Iadanza C, Lagomarsino D,
frequency ratio and logistic regression models. Landslides 4:33–41. https://doi.org/ Mugnozza GS, Segoni S, Spizzichino D, Tofani V, Lari S (2013) Landslide susceptibility
10.1007/s10346-006-0047-y mapping at National Scale: the Italian case study. In: Margottini C, Canuti P, Sassa K
Manzo G, Tofani V, Segoni S, Battistini A, Catani F (2013) GIS techniques for regional- (eds) Landslide science and practice, Landslide inventory and susceptibility and
scale landslide susceptibility assessment: the Sicily (Italy) case study. Int J Geogr Inf hazard zoning, vol 1. Springer, Berlin, pp 287–295
Sci 27(7):1433–1452 Trigila A, Iadanza C, Spizzichino D (2010) Quality assessment of the Italian landslide
Meinhardt M, Fink M, Tünschel H (2015) Landslide susceptibility analysis in central inventory using GIS processing. Landslides 7:455–470. https://doi.org/10.1007/
Vietnam based on an incomplete landslide inventory: comparison of a new method to s10346-010-0213-0
calculate weighting factors by means of bivariate statistics. Geomorphology 234:80– Vai GB, Martini IP (eds) (2001) Anatomy of an Orogen: the Apennines and adjacent
97. https://doi.org/10.1016/j.geomorph.2014.12.042 Mediterranean basins. Kluwer Academic, Dordrecht, The Netherlands, p 632
Mercogliano P, Segoni S, Rossi G, Sikorsky B, Tofani V, Schiano P, Catani F, Casagli N Van den Eeckhaut M, Hervás J, Jaedicke C, Malet J-P, Montanarella L, Nadim F (2012)
(2013) Brief communication “a prototype forecasting chain for rainfall induced Statistical modelling of Europe-wide landslide susceptibility using limited landslide
shallow landslides”. Nat Hazards Earth Syst Sci 13:771–777. https://doi.org/10.5194/ inventory data. Landslides 9:357–369. https://doi.org/10.1007/s10346-011-0299-z
nhess-13-771-2013 Xiao T, Yin K, Yao T, Liu S (2019) Spatial prediction of landslide susceptibility using GIS-
Myronidis D, Papageorgiou C, Theophanous S (2016) Landslide susceptibility mapping based statistical and machine learning models in Wanzhou County, Three Gorges
based on landslide history and analytic hierarchy process (AHP). Nat Hazards 81:245– Reservoir, China. Acta Geochim 38:654–669. https://doi.org/10.1007/s11631-019-
263. https://doi.org/10.1007/s11069-015-2075-1 00341-1
Nefeslioglu HA, Gokceoglu C, Sonmez H, Gorum T (2011) Medium-scale hazard mapping Yang Y, Yang J, Xu C, Xu C, Song C (2019) Local-scale landslide susceptibility mapping
for shallow landslide initiation: the Buyukkoy catchment area (Cayeli, Rize, Turkey). using the B-GeoSVC model. Landslides 16:1301–1312. https://doi.org/10.1007/
Landslides 8:459–483. https://doi.org/10.1007/s10346-011-0267-7 s10346-019-01174-y
Pham BT, Jaafari A, Prakash I, Bui DT (2019) A novel hybrid intelligent model of support Youssef AM, Pourghasemi HR, Pourtaghi ZS, Al-Katheeri MM (2016) Landslide suscepti-
vector machines and the MultiBoost ensemble for landslide susceptibility modeling. bility mapping using random forest, boosted regression tree, classification and
Bull Eng Geol Environ 78:2865–2886. https://doi.org/10.1007/s10064-018-1281-y regression tree, and general linear models and comparison of their performance at
Pham BT, Pradhan B, Tien Bui D, Prakash I, Dholakia MB (2016) A comparative study of Wadi Tayyah Basin, Asir Region, Saudi Arabia. Landslides 13:839–856. doi: https://
different machine learning methods for landslide susceptibility assessment: a case doi.org/10.1007/s10346-015-0614-1
study of Uttarakhand area (India). Environ Model Softw 84:240–250. https://doi.org/
10.1016/j.envsoft.2016.07.005
S. Segoni ()) : T. Luti : F. Catani
Pourghasemi HR, Kerle N (2016) Random forests and evidential belief function-based
landslide susceptibility assessment in Western Mazandaran Province, Iran. Environ
Earth Sci 75:23. https://doi.org/10.1007/s12665-015-4950-1 Department of Earth Sciences,
Pradhan AMS, Lee S-R, Kim Y-T (2019) A shallow slide prediction model combining University of Firenze,
rainfall threshold warnings and shallow slide susceptibility in Busan, Korea. Landslides Florence, Italy
16:647–659. https://doi.org/10.1007/s10346-018-1112-z Email: [email protected]
Reichenbach P, Rossi M, Malamud BD, Mihir M, Guzzetti F (2018) A review of statistically-
based landslide susceptibility models. Earth Sci Rev 180:60–91. https://doi.org/ G. Pappafico
10.1016/j.earscirev.2018.03.001 Department of Pure and Applied Sciences,
Rosi A, Tofani V, Tanteri L, Stefanelli CT, Agostini A, Catani F, Casagli N (2018) The new University of Urbino,
landslide inventory of Tuscany (Italy) updated with PS-InSAR: geomorphological Urbino, Italy
features and landslide distribution. Landslides 15(1):5–19
Segoni S, Battistini A, Rossi G, Rosi A, Lagomarsino D, Catani F, Casagli N (2015) Technical Department of Earth Sciences,
note: an operational landslide early warning system at regional scale based on space– University of Firenze,
time-variable rainfall thresholds. Nat Hazards Earth Syst Sci 15(4):853–861. https:// Florence, Italy
doi.org/10.5194/nhessd-2-6599-2014 Department of Pure and Applied Sciences,
University of Urbino,
Urbino, Italy
Landslides