Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

Environmental Modelling and Software 147 (2022) 105233

Contents lists available at ScienceDirect

Environmental Modelling and Software


journal homepage: www.elsevier.com/locate/envsoft

Comparative study of fuzzy-AHP and BBN for spatially-explicit prediction


of bark beetle predisposition
Meryem Tahri a, *, Jan Kašpar a, Anders L. Madsen b, c, Roman Modlinger a, Khodabakhsh Zabihi a,
Róbert Marušák a, Harald Vacik d
a
Faculty of Forestry and Wood Sciences, Czech University of Life Sciences Prague (CZU), Kamýcká 129, 16500 Praha 6, Suchdol, Czech Republic
b
HUGIN EXPERT A/S, Gasværksvej 5, 9000, Aalborg, Denmark
c
Department of Computer Science, Aalborg University (AAU), Selma Lagerlof Vej 300, 9220, Aalborg, Denmark
d
Institute of Silviculture, Department of Forest and Soil Sciences, University of Natural Resources and Life Sciences (BOKU), Peter Jordan Str. 82, 1190, Vienna, Austria

A R T I C L E I N F O A B S T R A C T

Keywords: The European spruce bark beetle ‘Ips typographus L.’ is the most serious disturbance agent for European forests.
Forest pest management The complex interactions of many influencing factors need to be integrated into a model-based decision-support
Bark beetle system to reduce the potential loss of forests. This paper compares two methodological approaches for spatially-
Outbreak disturbance
explicit prediction of the predisposition for bark beetle infestations. The fuzzy analytic hierarchy process and the
Fuzzy-AHP
BBN
Bayesian belief networks were used in combination with a geographical information system to manage un­
GIS certainties. Using available data resources, the two approaches were evaluated to produce robust results for
forest practitioners and to support measures to minimize the spread of bark beetles. The findings revealed that
nearly 32% of the sites investigated in a case study were moderately-high or high risk categories. It is concluded
that BBN is more efficient. Both methods can easily be used to analyze environmental problems involving
complex interactions among various criteria.

1. Introduction et al., 2017). Understanding this complex and dynamic environment and
identifying relevant factors can help minimize the risks and uncertainty;
Bark beetle (I. typographus L.) infestations have been steadily the use of modern planning methods and low-cost access to advanced
increasing over recent decades, and insect diseases and climate change geospatial technologies in particular can support short and long-term
are expected to damage about one million hectares of European forest in strategic forest planning (Tahri et al., 2021).
the near future (Seidl et al., 2014; Augustynczik et al., 2021). Droughts In this context, there has been a rapid rise in demand from forest
and the general rise in global temperature have triggered insect out­ practitioners for support in decision-making based on the most recent
breaks and mortality of the Norway spruce species (Pureswaran et al., findings in forest science (Vacik and Lexer, 2014). Forest managers have
2018; Jactel et al., 2019), as well as impacting the ecology, threatening to take the essential measures to avoid damage to trees in the future,
recreational activities and impacting the timber industry (Kunegel-Lion which requires a good understanding of the disturbance regimes. The
and Lewis, 2020b; Hlásny et al., 2019). In the Czech Republic, the available human and financial resources can then be directed towards
annual loss of Norway spruce growing stock through bark beetle attacks protecting the forests with a better chance of protecting the landscape in
is around 3.1%–5.4% for the period 2017–2019 (Hlásny et al., 2021). the future. One of the challenges facing forest managers is the early
The environmental and ecological pressures (i.e. climatic changes, detection of potential disturbance agents and adopting the relevant
acceleration of bark beetle spread, dramatic decline of the Norway strategies to overcome them (Kunegel-Lion and Lewis, 2020a).
spruce species), and related economic losses are important drivers for Various statistical methods, predictive models and expert systems
adaptive forest management (Biedermann et al., 2019; Hlásny et al., have been applied to detect and prevent outbreaks of bark beetle
2021). Forest management requires a rational planning process to (Netherer and Nopp-Mayr, 2005; Pasztor et al., 2014), including
reduce the uncertainty involved in choosing between alternatives (Seidl multi-criteria analysis methods, machine learning algorithms and

* Corresponding author.
E-mail address: tahri@fld.czu.cz (M. Tahri).

https://doi.org/10.1016/j.envsoft.2021.105233
Received 1 April 2021; Received in revised form 13 October 2021; Accepted 22 October 2021
Available online 28 October 2021
1364-8152/© 2021 Elsevier Ltd. All rights reserved.
M. Tahri et al. Environmental Modelling and Software 147 (2022) 105233

remote sensing (Bright et al., 2020; Valdez Vasquez et al., 2020; Hollaus 2. Material and methods
and Vreugdenhil, 2019; Radl et al., 2018; Seidl et al., 2016). Under­
standing the inter-dependencies of forest disturbance agents and 2.1. Criteria for assessing predisposition to bark beetle infestation
involving different stakeholders in decision making considering spatial
and temporal aspects of bark beetle outbreaks is challenging. Many The selection of criteria for evaluating the predisposition of forest
studies have highlighted the need for an integrated model combining the stands to bark beetle infestation was based on a literature review and
methodological advantages of different techniques (Tahri et al., 2020). inputs provided through an online panel discussion. For the panel dis­
In this context, the use of Bayesian Belief Networks (BBN) and/or fuzzy cussion, online questionnaires were developed with input from forest
Analytic Hierarchy Process (fuzzy-AHP) techniques has been recently entomology and forest ecology experts. In total, eight criteria were
proposed in the fields of environmental and industrial problems (Ar considered as important for predicting the predisposition: proportion of
et al., 2020; Shao et al., 2020). Fuzzy-AHP and BBN are well known for Norway spruce (% of basal area); stand age (average years); drought
their strengths in dealing with uncertainties and have proven to be (categories); precipitation (mm/year); mean annual temperature ( C/
robust in solving complex decision-planning problems (Malczewski and year); solar radiation (watt hours/m2); altitude (meters above sea level)
Rinner, 2015; Shao et al., 2020). and soil type (different edaphic factor categories). The list of criteria is in
Radl et al. (2018) studied storm and bark beetle agent disturbances line with many studies related to the assessment of forest data and site
in mountain forests at stand scale applying a BBN, including bark beetle information as predisposing factors for I. typographus L. outbreaks (e.g.
damages as a sub-model, to predict the number of damaged trees. It has Netherer and Nopp-Mayr (2005); Pasztor et al. (2014); Hlásny et al.
been evidenced that an understanding of Bayesian statistics is required (2021). A web-based survey was designed to derive estimates for the
for probabilistic analysis (Kokolakis, 2010), and deterministic analysis is importance of the evaluation criteria in a pairwise comparison matrix.
useful in many applications. Several fuzzy-AHP technique studies have Feedback provided via the online portal by the researchers involved in
proved to be useful in identifying risk areas (e.g. disasters and hazards) the survey was also analyzed. The relation of the relevant criteria is
for management planning (Yariyan et al., 2020; Lyu et al., 2020; Roy shown in the visualization of the Bayesian network diagram in Fig. 1.
and Saha, 2019; Haidara et al., 2019; Eskandari, 2017); these studies
indicate that they are reliably accurate when compared to ground truth, 2.2. Bayesian Belief Networks and GIS
and could provide a useful framework for other case studies. The tech­
niques involve different mathematical theories and require different BBN have been used in many different fields from medical science
input data. BBN is based on the probabilistic interpretation of certain (McLachlan et al., 2020; Hepler et al., 2019; Wang et al., 1999) to
factors as a directed acyclic graph (network process), which can provide environmental sciences (Roostaei et al., 2021; Baldock et al., 2019). The
insight into conditional dependence and independence relations be­ latest literature review illustrates the emerging number of tools and
tween variables by hypothetical observation (Pearl, 1988, 2009). Expert practices developed (De Iuliis et al., 2021; Zhang and Mahadevan, 2021;
knowledge is used in conjunction with empirical data in the graphical Landuyt et al., 2015). In our study the causal influence of the identified
model. Fuzzy-AHP is based on pairwise comparisons as a deterministic predisposition variables on the potential I. typographus L. outbreaks for
matrix from expert perception only (Saaty, 1987; Emrouznejad and Ho, each stand is defined on a regional scale. The aim is to identify whether
2017). However, understanding of the effectiveness of such methods or not a specific stand can be infested by the various Xc (child) consid­
remains limited and no formalized comparison between approaches ered as variable. A survey was carried out among researchers and senior
using fuzzy-AHP and BBN combined with GIS (Geographical informa­ consultants in the field of forest entomology and forest management
tion system) has been carried out. planning with the aim of constructing a framework of influence dia­
Our aim is therefore to compare the fuzzy-AHP and BBN approaches grams for the BBN. To reduce the burden of a large number of estimates
to identify the probability of a forest stand being infested by Ips typog­ for conditional probability distributions by the experts, the elicitations
raphus L. To address the issues of intensified tree mortality with bark were done in accordance with Wisse et al. (2008) applying the HUGIN
beetle infestation, two model-based techniques are proposed for forest EXPERT tool. The experts were asked to build the network structure and
practitioners, to reduce the uncertainty related to the future manage­ identify the names of variables and their relationships within the overall
ment of the forests under changing environmental conditions (Kjærulff ‘bark beetle disturbance’. Once the network was constructed, the elici­
and Madsen, 2008). We will demonstrate that the fuzzy-AHP and BBN tation was accomplished using expert interviews, where experts
approach can be designed to help decision-makers in managing ambi­ expressed their “beliefs” using the verbal statements: 1 = certain; 0.9 =
guity on different temporal and spatial scales regarding bark beetle very probable; 0.5 = fifty-fifty; 0.2 = not probable; 0.1 = unlikely; 0 =
outbreaks. impossible. This allowed a simple assessment in the conditional proba­
With this contribution we (i) introduce a combination of fuzzy-AHP bility tables (CPT) with the experts in face-to-face interviews. The
and BBN techniques with GIS to predict spatially explicit vulnerability to calculation of the conditional probability for each value xk of variable Xc
bark beetle infestations and (ii) compare the characteristics and results was estimated using an elicitation method for BBN, as below:
of the two methodological approaches for a case study in the Czech pa(Xc) is the parent node of the variable Xc and a corresponds to the
Republic to identify the advantages and drawbacks for each method. The variable assignment, which is in our case {Low = 1, Moderately low = 2,
fuzzy-AHP model is used to estimate the importance of the predisposi­ Moderately high = 3 and High = 4}
tion criteria for Ips typographus L. following the procedure described in ∫ imax,k
Tahri et al. (2017). The structure of the probabilistic network of the BBN ∑ i
f (i) × di
P(Xc |pa(Xc ) = a) = wk × min,k
is set up comparatively to the fuzzy-AHP model following the recom­ k:X |pa(X )
k c
i max,k − imin,k
mendations of Radl et al. (2018). A geographical information system is
used to map the estimated probability of bark beetle damages for both Where:
models based on national-scale input data for the Czech Republic. The
approach proposed may help stakeholders gain a better understanding ● f(i) : [0, 1]→[0, 1] were constructed by piecewise linear interpolation
of the relevance of a large set of spatially-explicit prediction parameters function through the points (ijoint (axc ), P(Xc = xc |axc ))
and support the design of prevention measures within a short time ● imin,k = min(ik(xk), ijoint(a))
frame. ● imax,k = max(ik(xk), ijoint(a))
● ijoint(a) is the individual influence factor for each conditioning vari­
able Xk ∈ pa(Xc) (Wisse et al., 2008).

2
M. Tahri et al. Environmental Modelling and Software 147 (2022) 105233

Fig. 1. Bayesian network diagram for bark beetle disturbance.

When pa(Xc) has positive influence on Xc, the joint influence factor state nodes for the Xc = {Low}. aneg,k+ is pa(Xc) assignment, where (Xk) is
ijoint is as follow: fleqntruemathmargin0pt in the most favorable state and all Xm ∈ pa(Xc) ~ Xk are in the least
∑ favorable. P(Xc = xmaxc |aneg,k+ ) and P(Xc = xc |aneg,k+ ) are the most and
min
k:X ∈pa(Xc ) ik (xk ) × (rank(xk ) − 1)
ijoint (a) = ∑k least favorable for high and low values of (Xc), respectively.
k:Xk ∈pa(Xc ) (rank(xk,max ) − 1)
All the criteria variables represented positive influences, except for
When pa(Xc) has negative influence on Xc, the joint influence factor precipitation and altitude, where high parent values pa(Xc) contributed
ijoint is written: fleqntruemathmargin0pt to the low probability child (Xc). The input data for each compartment
∑ were converted into raster files and subdivided into four states from 1 to
ijoint (a) =
k:Xk ∈pa(Xc ) ik (xk ) × (rank(xk,max ) − rank(xk ))
∑ 4, applying the equal interval classification method in ArcGIS (Appendix
k:Xk ∈pa(Xc ) (rank(xk,max ) − 1) A). To support the calculations for this study, the CPT algorithm was
implemented built and a new CPT program was designed in MATLAB
wk representing the weight for each parent Xk ∈ pa(Xc) is defined as:
software to accelerate the computation modeling. The results were then
1 δ+ 1 δ−k incorporated into HUGIN Researcher software (see network model in
wk = ∑ k
+ ∑ (1)
2 n:Xpn ∈pa(Xc ) δ+
n 2 −
n:Xpn ∈pa(Xc ) δn
Section Software and/or data availability).
Once the rasters were stacked, each raster was linked to a node in the
where δ+
k and δk are the highest and the lowest state of Xc.

model to propagate the evidence by applying the HUGIN plugin to the
δ+ P(Xc = xmax max open-source QGIS (Fig. 2). For each XY geographic location a Bayesian
= c |aneg,k+ ) − P(Xc = xc |aneg )
network case consists of the input data as evidence for the input node (i.
k

δ−k P(Xc = xmin min e. linked to a raster layer). The Bayesian network inserts and propagates
= c |aneg ) − P(Xc = xc |aneg,k+ )
the evidence in the model, resulting in two bands: the probability of a
and aneg corresponds to the best combination assignment of parent state with maximum probability and the other band is an index of state

Fig. 2. General framework for fuzzy-AHP and BBN methodologies.

3
M. Tahri et al. Environmental Modelling and Software 147 (2022) 105233

with maximum probability (YouTube tutorial link: Section Software judgment matrix A ̃ of experts was built, independent of k, defined by:
and/or data availability). ⎡ ⎤
A scenario-based sensitivity analysis of the BBN method was carried 1, 1, 1 l12 , m12 , u12 ⋯ l1n , m1n , u1n
⎢ l21 , m21 , u21 1, 1, 1 ⋯ l2n , m2n , u2n ⎥
out for three different scenarios. The aim of this type of sensitivity ̃=̃
A aij = ⎢


⎦ (2)
⋮ ⋮ ⋮ ⋮
analysis is to assess the results of the model under specific types of
ln1 , mn1 , un1 ln2 , mn2 , un2 ⋯ 1, 1, 1
scenario. In the first scenario, a Bark Beetle Disturbance (BBD) was
considered assuming higher temperatures and old-aged forest stands. Defuzzification of the fuzzy pairwise comparison matrix is given as
For the two other scenarios, mountain forests above 490 m altitude and Liou and Wang (1992) as follows: For i < j
lowland forests beyond 400 m altitude were considered (Table 1). The
aij ) = [μ.fα (lij ) + (1 − μ).fα (uij )],
gα,μ (̃ 0 ≤ α, μ ≤ 1 (3)
second scenario combined all high state nodes in the mountain forest
stand, and the third scenario corresponded to the worst-case scenario of
1
forests in lowland areas with low precipitation (all high state nodes, gα,μ (̃
aij ) = 0 ≤ α, μ ≤ 1 : i>j (4)
gα,μ (̃
aij )
except for low state in terms of altitude and precipitation). In addition,
the entropy was measured to check how far the probability mass was Where α corresponds to an index to define stable or unstable con­
distributed across the states of a random variable (Cover and Thomas; ditions, μ is an index of the degree of pessimism of a decision-maker for
Kjærulff and Madsen, 2008). the judgment matrix A. ̃ While fα(lij) and fα(uij) are derived from α, μ and
from equation (3) and (4) as well (Tahri et al., 2017).
2.3. Fuzzy analytical hierarchy process and GIS Finally, an I. typographus L. risk map was generated by applying
overlay layers in ArcGIS software, applying the geometrical mean values
The fuzzy-AHP approach is a well-known multi-attribute decision- for the evaluation criteria derived by the experts.
making approach (Zadeh, 1965; Saaty, 1987). It enables the expres­
sion of weights for each criterion to be calculated on the basis of expert 2.4. Validation of methodological approaches
knowledge, and it also manages uncertain and imprecise data. Several
steps are involved in applying the fuzzy-AHP model. In this study, the Two different approaches were taken to validate the robustness of
vulnerability of forest areas to Ips typographus L. was assessed by col­ the models: (i) comparison with empirical observations from a field
lecting expert judgments on the importance of the predisposition criteria survey; (ii) comparison with a reference map based on documented bark
in a pairwise comparison matrix (1 = equal importance; 3 = moderate beetle infestations at national level.
importance; 5 = strongly important; 9 = very important) (Saaty, 1977). First, to assess the accuracy of the fuzzy-AHP and BBN maps, a survey
The model algorithm constructed has been used previously in other was complemented by observations both in the field and in situ using
studies (Haidara et al., 2019; Tahri et al., 2017). An online survey (http Google Earth images. Ground truth data were collected between May
s://bpmsg.com/ahp-online-calculator/) was used to retrieve expert and August 2019 at several spots in the study area, to identify trees
knowledge on the importance of the evaluation criteria regarding the infected-uninfected by bark beetles. The spatial locations of 319 trees at
predisposing effect for bark beetle infestations, and speed up the data the green-attack stage were recorded using an ArcGIS collector app
collection process. The values of the pairwise comparison matrix was (ESRI Inc., Redlands, CA) installed on a mobile phone. At the green-
extracted afterwards and incorporated in the fuzzy-AHP computation attack stage, trees do not show any sign of discoloration (Niemann
modeling. The algorithm was implemented in a MATLAB Graphical User and Visintini, 2005), so a visual survey of red-brown boring dust was
Interface (GUI). The application can be found in Section Software carried out to detect traces of beetles entering the bark. Uninfected trees
and/or data availability. were observed in situ, linking the Google Earth image to ArcGIS, where
uninfected trees were clearly visible in situ by traces of salvage logging
As in Buckley (1985), we noted by akij the relative importance of
or afforestation, and validated by an expert in the field. Points were
criteria i and criteria j given by the expert k.
selected to cross-validate the predicted and field results using a confu­
lij = min akij sion matrix.
√̅̅̅̅̅̅̅̅̅̅̅
∏n Second, we used a map based on the mean percentage of trees
akij affected by bark beetles between 2017 and 2019 at national level
n
mij =
(Hlásny et al., 2021). The majority of the input data for this map were
k=1

uij = max akij


based on statistical reports from the state agencies Forest of the Czech
for all i, j = 1, …, n. Republic, Military Forests and Properties, and the Czech Statistical Of­
Where lij, mij and uij are respectively minimum, geometrical mean, fice. The surface area for each class was calculated using ArcGIS for each
and maximum values of expert opinions (pairwise comparison matrix region. The reference map and the map of infected/uninfected trees
among k experts derived from AHP online calculator). The fuzzy were used to compare the outcomes for each model (fuzzy-AHP and BBN
risk maps).
Table 1
A script was built in RStudio software to compute the spatial dif­
Description of scenarios selected for sensitivity analysis. ferences between the two output raster datasets cell by cell (Bishop
et al., 1998; Wealands et al., 2005), and the fuzzy-AHP and BBN maps
Scenario 1 Scenario 2 Scenario 3 (lowland
(mountain areas) areas - worst case)
for the same field area. The two continuous raster variables were
cropped and the data aggregated then resampled for the exact spatial
Temperature 7.66 ◦ C - 8.5 ◦ C > 8.5 ◦ C
resolution, to ensure coherent correlation. Stack and linear model
8.08 ◦ C
Stand age 83 years > 83 years > 83 years functions were also computed to generate a linear regression analysis in
Altitude n/a > 490 m < 382 m RStudio, to define the heterogeneity of the fuzzy-AHP and BBN models.
Drought n/a high (4th category) high In this case, the fuzzy-AHP variable was considered as a linear predictor.
Share n/a high (4th category) high Finally, a Pearson’s correlation coefficient was calculated to compare
Precipitation n/a > 706 mm < 621 mm the related differences and similarities between the fuzzy-AHP and BBN
Soil type n/a high (4th category) high methods.
Solar n/a > 1000 kWh/m2 > 1000 kWh/m2
radiation

n/a = not applied.

4
M. Tahri et al. Environmental Modelling and Software 147 (2022) 105233

2.5. Study area description and input data The input data for the evaluation criteria of the fuzzy-AHP and BBN
applications are listed in Appendix C. All raster were resampled to cell
An area of around 20 to 30 × 106 m3 trees has been destroyed by size 200 × 200 m, and the output of the spatial raster resolution of the
I. typographus L. calamity in the Czech Republic (out of 480 × 106 m3 of both models were 1 : 0.54. The share of Norway spruce, stand age,
spruce timber in total). Between 1990-1997, the Czech Republic was altitude and edaphic factors in the forest stand were collected from
ravaged by bark beetle outbreaks, following a storm in 1990 brought on Czech national forest inventory. The survey was carried out between
by several warm, dry summers (Schelhaas et al., 2003). Over the period 2011 and 2014 by the Forest Management Institute in the Czech Re­
1964 to 1991, over five times the amount of salvage logging (70 × 106 public. The point data were interpolated using geostatistical tools in the
m3) was carried out than sanitation felling (13 × 106 m3), when intensity ArcGIS software, and the ordinary kriging was applied to provide a
gradually increased (Modlinger and Novotný, 2015). logical framework for interpolation accuracy, which resulted in a
Within this context, a representative study area was selected to reduced mean square error.
include different landscape characteristics to test the approach. The The drought map was extracted as an image from INTERSUCHO
selected central Bohemian region represents an excellent example of portal and was digitalized and geo-referenced in ArcGIS; the droughts
typical forest conditions in the Czech Republic. The study area is situ­ comprised standard classifications: abnormally dry, moderate drought,
ated 35 km south-east of the capital Prague, between 14 41′ 56.4′′ severe drought and exceptional drought (Svoboda et al., 2002). The
longitude and 49 59′ 20.4′′ latitude, and covers an area of 80 000 ha. solar radiation power per unit area was derived from the Digital
The area is characterized by four vegetation types, with coniferous forest Elevation Model (DEM) by applying the spatial analyst tool of ArcGIS.
covering 61% followed by 32% mixed forest; few areas are covered by The DEM data were imported from NASA Earth Science Data (https
broad-leaved forest and transitional woodland-shrub (Fig. 3). ://search.earthdata.nasa.gov/).
The forests are dominated by Norway spruce (Picea abies L.), and the
stand age varies between 34 and 100 years, while the oldest stands are 3. Results
mainly found in the West and the South. The elevation in the study area
ranges between 409 and 493 m and the average altitude is 430 m. 3.1. Fuzzy-AHP model
Climate conditions vary significantly across the seasons and over the
year; the mean annual temperature is 7.9 ◦ C and the average annual 3.1.1. Weight calculation and map
precipitation is 686 mm, although the southeast of the area often has no Following the steps in subsection 2.3 and incorporating the Excel
rainfall during the growing season. A considerable proportion of the data for each expert opinion, the weights were computed using the
area receives high-quantity solar irradiation. The study area in­ defuzzification matrix method in Eq. (3) and Eq. (4). According to the
corporates diverse edaphic conditions; the forest site classification in­ fuzzy-AHP assessment methodology applied, the eigenvalue was 8.60.
dicates acidic, stagnic acidic, deep loamy soils, gleysols, peat and The consistency ratio was 0.06. Therefore, the matrix judgments were
organic soils. All the geographical criteria information appears in Ap­ considered reasonable and acceptable. The derived criteria weights are
pendix B. presented in Table 2.

Fig. 3. Study area location in the Czech Republic (Central Europe); Land use/land cover map in central Bohemia regions.

5
M. Tahri et al. Environmental Modelling and Software 147 (2022) 105233

Table 2 ● All parents had positive influence on ‘Predisposition’ and ‘Bark


Criteria priority results of fuzzy-AHP method based on MATLAB GUI beetle disturbance’ nodes: ijointa(low) = 0; ijointa(m.low) = 0.33; ijointa(m.
application. high) = 0.66 and ijointa(high) = 1
Criteria weights Fuzzy-AHP MATLAB GUI (%) ● and one parent had negative influence on ‘Climate’ and ‘Site specific
Stand age 22
stand’ nodes: ijointa(low) = 0; ijointa(m.low) = 0.11; ijointa(m.high) = 0.44
Share of Norway spruce 20 and ijointa(high) = 1
Temperature 17
Drought 17 Four polygons were derived from the piecewise linear functions to
Precipitation 10
compute the estimation for each CPT node network following the al­
Altitude 7
Soil type 4 gorithm by Wisse et al. (2008). The resulting graphical curves are shown
Solar radiation 3 in Appendix E. Following the network structure design and based on the
mathematical formula provided in subsection 2.3, one master BBN
model for bark beetle infestation was built; the MATLAB CPT automatic
Stand age and share of Norway spruce were more relevant than the computation and network model outcomes are given in section Software
other criteria for detecting bark beetle outbreak predispositions, with and/or data availability.
weights of 22% and 20%, respectively. The average temperature of over Following HUGIN tool, Fig. 4 shows the Bayesian network model for
8.5 ◦ C was considered a very high risk for spruce tree vulnerability. Also estimated bark beetle predisposition which was linked to QGIS to derive
the role of droughts 17% was considered a strong evidence that Norway a thematic risk map for the various factor categories. An example was
spruce trees can be affected by bark beetles. Although the elevation proposed in Table 4 for summer (with extreme climatic variables) and
above sea level (m.a.s.l.) is often considered to be a key parameter winter (with a high precipitation variable), which revealed the maximal
affecting bark beetle infestations, in this case study, the altitude was high disturbance risk as occurring in the summer season, with a prob­
considered less important by the experts. However, there is a strong ability of 0.66. This result would occur only in extreme cases with high
relation to the average temperature, as it follows an altitudinal gradient. temperature, drought and solar radiation variables. The risk of bark
The solar radiation criterion had the lowest weight with only 3%. beetle dropped to only 38% in the winter season.
The different raster layer of the evaluation criteria were combined As shown in the BBN map (Fig. 5b), 10% of the area has a probability
considering the derived weightings for the fuzzy-AHP. The resulting risk of between 31% and 41%, and around 22% of the area has a
map is characterized by four levels of risk: low risk, moderately low risk, moderately low risk. Around 30% of the study area are classified within
moderately high risk and high risk. Areas prone to bark beetle infesta­ the two categories moderately high risk and high risk. The most
tion are indicated in Fig. 5a. With the fuzzy-AHP approach 38% of the vulnerable sites are found in the central zone of the study area and in a
study area is in the moderately high or high risk category, while 25% is few parts in the north; the south and west are considered to be affected
low or moderately low. by a high probability level as well.

3.1.2. Sensitivity analysis of the fuzzy-AHP model 3.2.2. Sensitivity analysis of BBN model
The sensitivity analysis of the fuzzy-AHP criteria weights were By simulating variations of the input values for share of Norway
assessed following the steps outlined by Triantaphyllou and Sánchez spruce and stand age, we computed the minimum, current and
(1997), to demonstrate how changes in the weighting could change the maximum probability values for each BBD risk state. The results are
ranking of the criteria. The most sensitive criteria were temperature, shown in Table 5. The entropy was H(BBD) = 1.09 of BBD risk. This
share of Norway spruce and drought (Appendix: Table D) as any changes shows that observation of share of Norway spruce variable would pro­
to these criteria would influence the criteria ranking. Stand age, solar duce non-significant variations in the posterior belief in states being
radiation and soil type showed a lower sensitivity coefficient. moderately high, moderately low, or low. A slight variation in the
posterior belief was noted for the state of high. However, stand age
3.2. BBN model showed significant variations in posterior belief for all states except
moderately high. The findings show that posterior distribution is very
3.2.1. Network and map robust for variations in input.
Based on expert assessments and according to Eq.(1) in subsection From an investigation of the impact of different subsets of the evi­
2.2. Table 3 shows each parent node weight (wk). The results show that dence on each state of the hypothesis variable, it is clear that the finding
predisposition and climate node are major drivers for bark beetle pre­ ϵSa on Sa (Stand age) acts in favor of the hypothesis H(BBD) = high. On
disposition. The weights of the child nodes specify, that drought, solar the other hand, the evidence ϵT on T (Temperature) acts slightly against
radiation and soil type contribute mostly to the parent nodes. The results the hypothesis, while the normalized likelihood of both evidence ϵT and
(ijoint) were derived using MATLAB software. ϵSa is 1.06, which acts slightly in favor of the hypothesis H(BBD) = high.
The evidence therefore supports the hypothesis that an outbreak
disturbance is driven by temperature and stand age. In the worst-case
scenario, the evidence ϵPr (Precipitation) acts against the hypothesis,
while ϵSr (Solar radiation) acts in favor of the hypothesis H(BBD).
The model behavior representing a variation of the state for each of
Table 3
the variables observed is reported for scenarios 2 and 3 (Appendix
BBN model wk weights for each parent and child node.
Table F). The findings show that posterior distribution is relatively non-
Node Weight (%) Criteria node Weight (%) sensitive to one-way variations in individual nodes. The distribution of
Predisposition node 51 Share of Norway spruce 11 bark beetle probability risk is non-sensitive to variations in the precip­
Stand age 26 itation node. The most sensitive observation is altitude and soil type in
Drought 63 scenario 2, and altitude followed by drought and solar radiation in
Climate node 46 Temperature 22
Solar radiation 61
scenario 3.
Precipitation 17 The parameter sensitivity analysis of the BBD hypothesis shows the
Site specific stand 2 Altitude 6 belief that BBD is high was 0.47 and the belief that it is low was 0.05.
Soil type 94 The altitude, drought, precipitation and solar radiation variables
revealed the most sensitivities and were influential of all connected

6
M. Tahri et al. Environmental Modelling and Software 147 (2022) 105233

Fig. 4. A Bayesian network distribution for probability of bark beetle disturbance.

Fig. 5. Mapping of fuzzy-AHP and BBN approaches bark beetle risk.

7
M. Tahri et al. Environmental Modelling and Software 147 (2022) 105233

Table 4
Seasonal factor probability monitors.

Table 5 Table 6
Posterior evidence sensitivity of ‘bark beetle disturbance risk probability hy­ Summary of the linear regression model of fuzzy-AHP and BBN raster outputs
pothesis’ variable to findings on the ‘share of Norway spruce’ variable. computed by RStudio.
Minimum Current Maximum Residuals Min Median Max

High 0.6 0.62 0.62 Values − 0.19 0.002 0.15


Moderately high 0.21 0.21 0.21 Coefficients Estimate Standard Error t-value
Moderately low 0.11 0.11 0.11 (Intercept) 0.39 0.03 116.86
Low 0.08 0.08 0.09 Values 0.03 0.00 45.71

nodes (Tahri et al., 2020). A tornado graph of all the root variables compared with reference data and a field survey. The overall confusion
(except T and Sa) is shown in the (Appendix Fig. G) showing the possible matrix with a confidence interval of 95% gave a result of nearly 87%
variation of each state of BBD. Altitude is the variable with the greatest (kappa coefficient 71%) for the BBN model, while the overall confusion
potential impact on the probability distribution of BBD, and shows that matrix for the fuzzy-AHP model was 81% (kappa coefficient 56%). Both
solar radiation, drought and precipitation also have the greatest po­ models showed appropriate accuracy with the field survey. This was also
tential impact on BBD probability distribution. confirmed with the reference map (Hlásny et al., 2021), where the Kolin
and Praha regions were characterized by serious yearly losses of tree
3.3. Spatial comparative analysis stock, representing more than 73% of moderately-high and highly
vulnerable areas for both models (Appendix Table H). Although the
The two methodological approaches were compared in two steps: (i) Benesov region was more uncertain with both models, the surrounding
summary of the linear regression model and spatial correlation between border was shown to be characterized by a high density of Norway
the two resulting raster layer, (ii) interpretation of the results of the two spruce species, ancient forest stands with high temperature, and low
approaches in comparison with reference data sets (Appendix Table H). precipitation after checking the criteria maps, which confirms the ac­
curacy of the predicted results. The mismatch can probably be explained
3.3.1. Comparison of risk maps by the different research scales used in the research: the reference map
A summary of the linear regression models resulting from the raster was on the national scale, while our own study was based on a regional
comparison of the fuzzy-AHP and BBN outputs is given in Table 6. A scale. The Havlickuv Brod location gave a contradictory outcome with
comparison of the two maps resulted in a reduced standard error coef­ the fuzzy-AHP technique, with climatic parameters assuming a high
ficient (0.03). The result also reveals that the median was close to zero, weighting and dominating the southern area. In summary, the BBN
at 0.002, while the t coefficient values (116.86; 45.71) were relatively model gave more accurate prediction than the Fuzzy-AHP technique.
high and are greatly relative to the standard error, which could signify
the existence of a relationship. The residual standard error was (0.049). 4. Discussion
Fig. 6 shows the result of RStudio analysis; note the moderate positive
relationship between the two raster maps, fuzzy-AHP and BBN, with a To our knowledge, this comparative research is among the first ex­
correlation coefficient of 0.66 with a confidence interval of 95%. amples, where a regional bark beetle disturbance mapping was done
Spatial correlation analysis of the difference maps is shown in the with a fuzzy-AHP and BBN modeling approach in a GIS.
Fig. 6. The BBN model was more highly correlated with fuzzy-AHP Comparing the criteria ranks of both applications indicate, that the
model, which also can seem similar by visual observation. Most dark most relevant criteria in the fuzzy-AHP approach were quite similar to
red color (High risk) areas are located in the north, central and south those with the BBN approach. The parameters contributing to bark
where old forest cover mainly dominates. The small differences beetle disturbance in both models were share of Norway spruce followed
observed were seen in the South and South-West (Fig. 5). by temperature, precipitation and drought. In this study a lower risk was
observed for young stands with a small share of Norway Spruce. This
3.3.2. Validation of data with ground truth interpretation concurs with works published previously (Stereńczak
The accuracy assessment between the fuzzy-AHP and BBN maps was et al., 2020; Radl et al., 2018), where high-risk disturbance sites were

8
M. Tahri et al. Environmental Modelling and Software 147 (2022) 105233

Fig. 6. Spatial correlation between two rasters, BBN and fuzzy-AHP (location: central Bohemian region, Czech Republic).

Christiansen et al., 1987).


Table 7
Matthews et al. (2018) also perceived a dependence on whether or
Comparison with previous similar bark beetle stands by Radl et al. (2018).
not a drought had occurred in the region. Netherer et al. (2019) inves­
Research Worst scenario Best scenario Stand A* tigated environmental drivers for bark beetle spread in the case of
studies
Austrian Norway spruce stands. The authors concluded that the warm
This study 0.67 high state 0.49 high state 0.47 high temperature and acute disposition of the trees to attack caused by
findings probability probability state
drought stress was likely to increase the risk of bark beetle infestation.
probability
Radl et al. High probability Low probability found Result The results of the work by Radl et al. (2018) were used to validate the
(2018) corresponds to high in high bark beetle identical to results of this study. For comparison purposes the probability level based
bark beetle predisposition level best scenario on four criteria was extracted: stand age, share of Norway spruce,
predisposition level drought and soil type. The BBN model of this study allows to reproduce
Note *: corresponds to the selected forest stand, characterized by 60–79 forest similar results (Table 7). Despite the different input models and selected
age stand, 25%–49% of share of Norway spruce proportion and cambisol soils criteria in the Austrian and Czech cases, the overall outcomes of the
type. studies have some similarities: older stands and a higher proportion of
Norway spruce in combination with an extreme drought event seem to
characterized by ancient stand age and proportion of Norway spruce as a contribute to higher probabilities for damages. The worst and best sce­
dominant species. Mezei et al. (2014) also demonstrated in their study narios revealed the estimated probability distributions, according to the
that forest stand age and proportion of Norway spruce are among the most up-to-date scientific literature findings on bark beetle disturbance.
parameters that cause spruce mortality. This research indicates that the The applied methodological approaches are different in terms of the
denser the spruce tree cover and the more advanced the age of the forest, underlying mathematical theories. While fuzzy-AHP requires a hierar­
the more vulnerable the trees are to bark beetle. This has also been chical process mode, BBN is based on a network process, taking into
demonstrated by other studies (Holeksa et al., 2017; Hlásny and account inter-dependencies between variables. Nevertheless, the major
Turčáni, 2013), where these parameters were considered as major advantages and drawbacks of both methods are identified along with the
driving factors for intense damage in the disturbed areas. Also precipi­ potential trade-offs. A summary of the differences between the two
tation, drought and temperature were often identified as relevant factors methods, based on literature reviews and the present research experi­
(Černý et al., 2020; Marini et al., 2017), and many reviews consider ment, is given in Table 8.
climate variables such as temperature and precipitation as the two main Concerning calculations, both approaches deal with uncertainty and
drivers leading to mass outbreaks of bark beetle (Brandl et al., 2020; imprecise data and require more complex systems. For this reason, the

9
M. Tahri et al. Environmental Modelling and Software 147 (2022) 105233

Table 8
Summary of differences between fuzzy-AHP and BBN approaches based on literature reviews.
Fuzzy-AHP method BBN method

Expert elicitation Data input source Expert judgment (qualitative) Expert judgment and/or empirical data (qualitative and/or
process quantitative)
Data input type Deterministic Probabilistic
Input data size Unlimited number of experts (supports a large pairwise Limited number of experts
survey matrix)
Data input Saaty’s numerical ratio scale 1-9 Categorical; Boolean; numbered (continuous/discrete) and Interval
subtype
Participatory Broad range of stakeholders/decision-makers can participate Broad range of stakeholders/decision-makers can participate
process
System Complex Transparency of the BBN graphical model enhances understanding
of system
Difficulty Pairwise comparison process can be time-consuming with a Confusion around expressing input probability data with a large
large number of criteria number of sub-models
Computation results Software Proposed user-friendly fuzzy-AHP MATLAB GUI (See Section HUGIN and available software; limited and less user-friendly open-
Software and/or data availability) source BBN software
Output treatment Allows to consider uncertainty implicitly Deals with uncertainty in environmental change and estimates the
uncertainty of unobserved variables
Output result Fixed output weights Dynamic output probability modeling in real-time depending on
evidence updating

fuzzy-AHP MATLAB GUI application is proposed for the non-specialist as it meticulous transparency and produces dynamic results (Li et al., 2020;
provides straightforward guidance for a broad range of decision-makers Laurila-Pant et al., 2019), and also proves to be more precise at stand
(academics, ecologists, consultants, etc.) to facilitate the decision- and local levels. Building a sample regional BBN model generates mul­
making process for better understanding of forest issues. While using tiple scenarios based on the various knowledge sources (Höfer et al.,
BBN in our study research, the CPT MATLAB program was found to be 2020), which provides an insight into the relationship between the
useful but limited to two or three directed node graphs. A new CPT al­ forest components and the output nodes for each scenario. The func­
gorithm is needed, requiring an additional long process in cases with tional model can be applied to other local and/or regional forest en­
four (or more) interconnected nodes. On the other hand, BBN can sup­ terprises by simply changing the input data. The spatial BBN represents a
port a large number of criteria in the model. Unlike fuzzy-AHP, the trade-off between assessing the impact of bark beetle on different
method has a threshold on the number of criteria that can be ecosystem services and stakeholders’ knowledge of the effects on forest
incorporated. management practices (Morris et al., 2017).
A pairwise comparison questionnaire was carried out in fuzzy-AHP, After overlaying and resampling different criteria, spatial patterns at
applying a free web survey integrated with mobile devices available on regional and local scales were easy to access for both methods and gave
the market, the online questionnaire offering a user-friendly remote tool closely-matching results in compiling thematic maps. This research
to involve various experts independently. The fuzzy-AHP was conse­ corroborates the work of Hlásny et al. (2021), where the Kolin, Praha
quently quicker at collecting expert judgments and avoided conflicting and Kutna Hora regions were well-matched in three models. According
compromises between parties more readily. The fundamental limitation to the confusion matrix results, the BBN model gives better results than
of fuzzy-AHP is that the required data input is restricted (expert judg­ the fuzzy-AHP model.
ment only, according to the Saaty 9-point scale). In addition, the pair­ It is argued that the maps produced provide both support for forest
wise comparison questionnaire offers non-familiarity among management planning to prevent further disturbance, and monitoring
stakeholders (Tahri et al., 2017). Although there are no pairwise com­ for efficient, strategic decision-making. An improved understanding of
parison matrices with the BBN technique, its main strength is the ability regulated timber harvesting is therefore needed, which can be achieved
to acquire probabilistic interdependency between uncertain variables by the introduction of salvage-logging management (Vanická et al.,
(Dai et al., 2021), based on non-homogeneous variables and many 2020; Dobor et al., 2020), and the reforestation and plantation of trees in
different types of input data in one functional model. Expressing prob­ the most suitable areas (Thorn et al., 2017; Macek et al., 2017; Nováková
ability beliefs demanded enormous concentration but required a limited and Edwards-Jonášová, 2015).
number of experts (max. 2 per case study) to retrieve expert perceptions. The results of ground truth and reference data validations reveal that
The major weaknesses associated with the BBN method compared to BBN is more efficient than the fuzzy-AHP, but it cannot be concluded
the fuzzy-AHP technique are its highly complicated processes; it was that BBN is more effective than the other model, as either of them may
time-consuming to implement and retrieving the input data to convert predominate depending on processing time, knowledge source, and goal
into conditional probability data was inherently challenging. This was priority. The BBN technique requires computing capabilities and is time-
also confirmed by the majority of previous studies (Barton et al., 2020; consuming to implement, but is more accurate and flexible in forest
Nascimento et al., 2019), although the overwhelming relationship be­ stands since it takes into account many different types of dataset. Unlike
tween dependence-independence probability nodes was problematic. fuzzy-AHP, MATLAB GUI is a user-friendly tool offering rapid processing
One of the key advantages of the BBN method is that it offers and is practical at large spatial and temporal scales. Most importantly,

10
M. Tahri et al. Environmental Modelling and Software 147 (2022) 105233

both methods are reproducible for use in other types of case study warning tool, while the BBN approach proved to be effective for dy­
involving stakeholder insights and future priority research. namic and small-scale landscapes.

5. Conclusion Software and/or data availability

This research will be of great help for forest managers but also for The fuzzy-AHP MATLAB GUI is freely available via the GitHub link
conducting further field research in this area. The results obtained from https://github.com/Meyem4/foman-FA. It was developed by Haytham
overlaying raster maps of the spatial multi-criteria decision-making Tahri (haytham.tahri@gmail.com) and Meryem Tahri (see co-author
method and Bayesian belief network analyses showed that most loca­ contact information) and launched in 2020.
tions had a similar level of relevance in the region. The current situation Conditional probability tables and I. typographus L. network model
is challenging in terms of rethinking spatial planning, time available for outcomes are provided via the GitHub link https://github.com/M
data collection and the high costs of field research. Two decision-support eyem4/foman-ITL-bbn. HUGIN EXPERT software is required to run
system methods were selected to detect sites potentially disturbed by the I. typographus L. network model. This was developed by Meryem
bark beetles. The work carried out required input of a geo-referenced Tahri under the supervision of Prof. Anders L. Madsen (see co-author
data set. Several meteorological, topographical and forest-stand pa­ contact information). A YouTube tutorial on the HUGIN plugin for
rameters were obtained, including the extraction of forest vegetation QGIS (mapping) produced by Prof. Anders L. Madsen is provided at
and use of the national forest inventory database. However, once the https://www.youtube.com/watch?v=ZCDfTMj5tOU.
methods and data are provided, it is possible to update the input data
layer quickly and perform scenario analysis by changing the importance
of the evaluation criteria. This will allow a more robust and less sensitive Declaration of competing interest
prediction of bark beetle infestations in the future.
The new MATLAB code was effective for both techniques, facilitating The authors declare that they have no known competing financial
accurate computational algorithms and programs. The conditional interests or personal relationships that could have appeared to influence
probability table code and the fuzzy-AHP MATLAB user interface tools are the work reported in this paper.
both useful for application in other study fields. The CPT code greatly
speeds up processing time and provides a dynamic output model for Acknowledgments
each stand, depending on seasonal distribution. Since the fuzzy-AHP
code can be converted to a simple graphical user interface which no This research was funded by grant ”EVA4.0”, No. CZ 02.1.01/0.0/
longer requires a fuzzy specialist to compute the model, the fixed output 0.0/16_019/0000803 financed by OP RDE. The authors gratefully
weight could be used in forest areas worldwide. The results of the acknowledge Prof. Dr. Daniel Ames (Editor-in-Chief) and anonymous
comparison show that both applications can easily be used to provide reviewers for their scientific expertise. We would like to thank consul­
solutions for complex environmental decision-making problems. The tants who participated in our survey, in particular thank to Dr. Andrew
efficiency of fuzzy-AHP MATLAB GUI was tested for use as an early Liebhold from US Forest Service Northern Research Station.

Appendix A. Criteria and/or node description, states

Criteria/variable Description State


Parameter node
1: Low risk 2: Moderate - 3: Moderate - 4: High risk
low risk high risk

Predisposition Share of Norway Proportion of Norway spruce in the stand: dense Low density Moderate High density Very high
spruce (%) spruce areas result in stronger I. typographus L. density density
attacks on trees
Drought Lack of rainfall: drought has significant influence Abnormally dry Moderate Severe drought Exceptional
on bark beetle drought drought
Stand age (years) Age of stand: the older the forest the higher the <50 50–64 64–78 >78
risk
Climate Annual Average yearly air temperature: high 6.3–7.6 7.6–8.0 8.0–8.50 8.50–9.2
Temperature (◦ C) temperature accelerates tree vulnerability stress
Solar radiation Solar radiation power per unit area: greater 322k - 864k 864k - 956k 956k - 1,021k 1,021k -
(Watt hours/m2) propagation attracts I. typographus L. 1,191k
Annual Annual precipitation: high precipitation reduces 706–820 660–706 622–660 550–622
precipitation (mm) risk of I. typographus L.
Site specific Altitude (m) Elevation above sea level: lower areas have >526 464–526 400–464 338–400
stand negative impact
Edaphic category Edaphic or ecological Gley soil, waterlogged Rich in water Fertile soils Acid soils
(soil type) soil, peats, organic soil soils

11
M. Tahri et al. Environmental Modelling and Software 147 (2022) 105233

Appendix B. Raster criteria considered in the application of fuzzy-AHP and BBN models in the Central Bohemian Region

Appendix C. Source table for data used

Parameter Institution Data origin Access format Cell size resolution

Share of Norway spruce (%) Forest Management Institute Czech National Forest Inventory 2011–2014 Database (*.accdb) 200 m
Stand age (years) Forest Management Institute Czech National Forest Inventory 2011–2014 Database (*.accdb) 200 m
Annual temperature (◦ C) Czech Hydrometeorological Institute Meteorological station http://portal.chmi.cz/ ASCII Raster (*.asc) 500 m
Annual precipitation (mm) Czech Hydrometeorological Institute Meteorological station http://portal.chmi.cz/ ASCII Raster (*.asc) 150 m
Solar radiation (Wh/m2) NASA ASTER sensor https://search.earthdata.nasa. Raster: Digital Elevation model 30 m
gov/
Drought INTERSUCHO https://www.intersucho.cz/ PDF map: Digitalization 500 m
Altitude (m a s l) Forest Management Institute Czech National Forest Inventory 2011–2014 Database (*.accdb) 30 m
Edaphic categories Forest Management Institute Czech National Forest Inventory 2011–2014 Database (*.accdb) 200 m

Appendix D. Sensitivity analysis results of fuzzy-AHP decision-making

Solar radiation Share of Norway spruce Stand age Altitude Temperature Soil type Precipitation Drought

Criticality degree of criterion (Ck) 369 1.8 9.95 5.9 1.7 9.74 2.77 3.55
Sensitivity coefficient of criterion (1/Ck) 0.00 0.56 0.10 0.17 0.59 0.10 0.36 0.28
Ranking 8 2 7 5 1 6 3 4

12
M. Tahri et al. Environmental Modelling and Software 147 (2022) 105233

Appendix E. Graphical curve outcome; examples to derive estimation of CPT distributions

Appendix F. Sensitivity analysis; min. – max. analysis of each observed node given the remaining observed nodes

Scenario 3 Scenario 2

Min Max Min Max

Precipitation Bark beetle disturbance risk = High 0.74 0.75 0.63 0.65
Bark beetle disturbance risk = Moderately high 0.17 0.18 0.2 0.21
Bark beetle disturbance risk = Moderately low 0.05 0.06 0.09 0.09
Bark beetle disturbance risk = Low 0.03 0.03 0.06 0.07
Soil type Bark beetle disturbance risk = High 0.73 0.75 0.53 0.63
Bark beetle disturbance risk = Moderately high 0.17 0.18 0.21 0.23
Bark beetle disturbance risk = Moderately low 0.05 0.06 0.09 0.13
Bark beetle disturbance risk = Low 0.03 0.03 0.07 0.12
Altitude Bark beetle disturbance risk = High 0.65 0.75 0.63 0.73
Bark beetle disturbance risk = Moderately high 0.17 0.2 0.18 0.2
Bark beetle disturbance risk = Moderately low 0.05 0.08 0.06 0.09
Bark beetle disturbance risk = Low 0.03 0.06 0.03 0.07
Drought Bark beetle disturbance risk = High 0.7 0.75 0.57 0.63
Bark beetle disturbance risk = Moderately high 0.17 0.19 0.21 0.22
Bark beetle disturbance risk = Moderately low 0.05 0.06 0.09 0.12
Bark beetle disturbance risk = Low 0.03 0.04 0.07 0.1
Stand age Bark beetle disturbance risk = High 0.72 0.75 0.6 0.63
(continued on next page)

13
M. Tahri et al. Environmental Modelling and Software 147 (2022) 105233

(continued )
Scenario 3 Scenario 2

Min Max Min Max

Bark beetle disturbance risk = Moderately high 0.18 0.17 0.21 0.21
Bark beetle disturbance risk = Moderately low 0.05 0.06 0.09 0.1
Bark beetle disturbance risk = Low 0.02 0.03 0.07 0.08
Temperature Bark beetle disturbance risk = High 0.73 0.75 0.52 0.63
Bark beetle disturbance risk = Moderately high 0.17 0.17 0.21 0.22
Bark beetle disturbance risk = Moderately low 0.05 0.05 0.09 0.13
Bark beetle disturbance risk = Low 0.02 0.03 0.07 0.12
Solar radiation Bark beetle disturbance risk = High 0.7 0.75 0.56 0.64
Bark beetle disturbance risk = Moderately high 0.17 0.18 0.2 0.22
Bark beetle disturbance risk = Moderately low 0.05 0.06 0.09 0.12
Bark beetle disturbance risk = Low 0.02 0.04 0.07 0.1
Share of Norway spruce Bark beetle disturbance risk = High 0.73 0.75 0.61 0.63
Bark beetle disturbance risk = Moderately high 0.17 0.17 0.21 0.21
Bark beetle disturbance risk = Moderately low 0.05 0.05 0.09 0.1
Bark beetle disturbance risk = Low 0.02 0.03 0.07 0.08

Appendix G. Tornado graph of all root variables except temperature and stand age nodes

14
M. Tahri et al. Environmental Modelling and Software 147 (2022) 105233

Appendix H. Comparison with reference map

Reference data (Hlásny et al., 2021) This study

Region % of trees stock loss (yearly) Categorie % of surface area using fuzzy-AHP % of surface area using BBN

Kolin 5%–7% Low risk 0 0


Moderately low risk 5 2
Moderately high risk 65 24
High risk 30 73
Praha-Vychod 3%–5% Low risk 0 10
Moderately low risk 15 17
Moderately high risk 73 32
High risk 12 41
Kutna Hora 3%–5% Low risk 48 25
Moderately low risk 32 32
Moderately high risk 21 24
High risk 0 19
Benesov 2%–3% Low risk 6 8
Moderately low risk 35 27
Moderately high risk 54 34
High risk 5 32
Havlickuv Brod 5%–7% Low risk 48 16
Moderately low risk 32 41
Moderately high risk 21 34
High risk 0 9

References jclepro.2021.127480. URL: https://linkinghub.elsevier.com/retrieve/pii/S0959652


621016991.
De Iuliis, M., Kammouh, O., Cimellaro, G., Tesfamariam, S., 2021. Quantifying
Ar, I.M., Erol, I., Peker, I., Ozdemir, A.I., Medeni, T.D., Medeni, I.T., 2020. Evaluating the
restoration time of power and telecommunication lifelines after earthquakes using
feasibility of blockchain in logistics operations: a decision framework. Expert Syst.
Bayesian belief network model. Reliab. Eng. Syst. Saf. 208 https://doi.org/10.1016/
Appl. 158, 113543. https://doi.org/10.1016/j.eswa.2020.113543. URL: http://
j.ress.2020.107320.
www.sciencedirect.com/science/article/pii/S0957417420303675.
Dobor, L., Hlásny, T., Rammer, W., Zimová, S., Barka, I., Seidl, R., 2020. Is salvage
Augustynczik, A.L.D., Dobor, L., Hlásny, T., 2021. Controlling landscape-scale bark
logging effectively dampening bark beetle outbreaks and preserving forest carbon
beetle dynamics: can we hit the right spot? Landsc. Urban Plann. 209, 104035.
stocks? J. Appl. Ecol. 57, 67–76. https://doi.org/10.1111/1365-2664.13518. URL:
https://doi.org/10.1016/j.landurbplan.2020.104035. URL: https://www.sciencedi
https://besjournals.onlinelibrary.wiley.com/doi/abs/10.1111/1365-2664.13518.
rect.com/science/article/pii/S016920462031519X.
Emrouznejad, A., Ho, W., 2017. Fuzzy Analytic Hierarchy Process. CRC Press, Boca
Baldock, T.E., Shabani, B., Callaghan, D.P., 2019. Open access Bayesian Belief Networks
Raton.
for estimating the hydrodynamics and shoreline response behind fringing reefs
Eskandari, S., 2017. A new approach for forest fire risk modeling using fuzzy AHP and
subject to climate changes and reef degradation. Environ. Model. Software 119,
GIS in Hyrcanian forests of Iran. Arabian Journal of Geosciences 10, 190. https://
327–340. https://doi.org/10.1016/j.envsoft.2019.07.001. URL: https://www.scienc
doi.org/10.1007/s12517-017-2976-2. URL:
edirect.com/science/article/pii/S1364815219305547.
Haidara, I., Tahri, M., Maanan, M., Hakdaoui, M., 2019. Efficiency of fuzzy analytic
Barton, D., Sundt, H., Bustos, A., Fjeldstad, H.P., Hedger, R., Forseth, T., Berit, K., Aas, .,
hierarchy process to detect soil erosion vulnerability. Geoderma 354, 113853.
Alfredsen, K., Madsen, A., 2020. Multi-criteria decision analysis in Bayesian
https://doi.org/10.1016/j.geoderma.2019.07.011. URL: http://www.sciencedirect.
networks - diagnosing ecosystem service trade-offs in a hydropower regulated river.
com/science/article/pii/S0016706118317361.
Environ. Model. Software 124. https://doi.org/10.1016/j.envsoft.2019.104604.
Hepler, S., McKnight, E., Bonny, A., Kline, D., 2019. A latent spatial factor Approach for
Biedermann, P.H., Müller, J., Grégoire, J.C., Gruppe, A., Hagge, J., Hammerbacher, A.,
Synthesizing opioid-associated deaths and treatment Admissions in Ohio counties.
Hofstetter, R.W., Kandasamy, D., Kolarik, M., Kostovcik, M., Krokene, P., Sallé, A.,
Epidemiology 30, 365–370. https://doi.org/10.1097/EDE.0000000000000978.
Six, D.L., Turrini, T., Vanderpool, D., Wingfield, M.J., Bässler, C., 2019. Bark beetle
Hlásny, T., Turčáni, M., 2013. Persisting bark beetle outbreak indicates the
population dynamics in the Anthropocene: challenges and solutions. Trends Ecol.
unsustainability of secondary Norway spruce forests: case study from Central
Evol. 34, 914–924. https://doi.org/10.1016/j.tree.2019.06.002. URL: https://linki
Europe. Ann. For. Sci. 70, 481–491. https://doi.org/10.1007/s13595-013-0279-7.
nghub.elsevier.com/retrieve/pii/S0169534719301673.
URL: https://hal.archives-ouvertes.fr/hal-01201492.
Bishop, G.D., Church, M.R., Aber, J.D., Neilson, R.P., Ollinger, S.V., Daly, C., 1998.
Hlásny, T., Krokene, P., Liebhold, A., Montagné-Huck, C., Müller, J., Qin, H., Raffa, K.,
A comparison of mapped estimates of long-term runoff in the northeast United
Schelhaas, M.J., Seidl, R., Svoboda, M., Viiri, H., European Forest Institute, 2019.
States. J. Hydrol. 206, 176–190. https://doi.org/10.1016/S0022-1694(98)00113-9.
Living with Bark Beetles: Impacts, Outlook and Management Options. From Science
URL: https://www.sciencedirect.com/science/article/pii/S0022169498001139.
to Policy. European Forest Institute. https://doi.org/10.36333/fs08. URL: https
Brandl, S., Paul, C., Knoke, T., Falk, W., 2020. The influence of climate and management
://www.efi.int/publications-bank/living-bark-beetles-impacts-outlook-and-mana
on survival probability for Germany’s most important tree species. For. Ecol. Manag.
gement-options.
458, 117652. https://doi.org/10.1016/j.foreco.2019.117652. URL: https://linkingh
Hlásny, T., Zimová, S., Merganičová, K., Štěpánek, P., Modlinger, R., Turčáni, M., 2021.
ub.elsevier.com/retrieve/pii/S0378112719316068.
Devastating outbreak of bark beetles in the Czech Republic: drivers, impacts, and
Bright, B., Hudak, A., Meddens, A., Egan, J., Jorgensen, C., 2020. Mapping multiple
management implications. For. Ecol. Manag. 490, 119075. https://doi.org/10.1016/
insect outbreaks across large regions annually using landsat time series data. Rem.
j.foreco.2021.119075. URL: https://www.sciencedirect.com/science/article/pii/
Sens. 12 https://doi.org/10.3390/rs12101655.
S037811272100164X.
Buckley, J.J., 1985. Fuzzy hierarchical analysis. Fuzzy Set Syst. 17, 233–247. https://doi.
Höfer, S., Ziemba, A., El Serafy, G., 2020. A Bayesian approach to ecosystem service
org/10.1016/0165-0114(85)90090-9. URL: http://www.sciencedirect.com/science/
trade-off analysis utilizing expert knowledge. Environ. Syst.Decis. 40, 67–83.
article/pii/0165011485900909.
https://doi.org/10.1007/s10669-019-09742-2. URL: http://link.springer.com/10.1
Černý, J., Pokorný, R., Vejpustková, M., Šrámek, V., Bednář, P., 2020. Air temperature is
007/s10669-019-09742-2.
the main driving factor of radiation use efficiency and carbon storage of mature
Holeksa, J., Jaloviar, P., Kucbel, S., Saniga, M., Svoboda, M., Szewczyk, J., Szwagrzyk, J.,
Norway spruce stands under global climate change. Int. J. Biometeorol. https://doi.
Zielonka, T., Żywiec, M., 2017. Models of disturbance driven dynamics in the West
org/10.1007/s00484-020-01941-w. http://link.springer.com/10.1007/s00484-02
Carpathian spruce forests. For. Ecol. Manag. 388, 79–89. https://doi.org/10.1016/j.
0-01941-w.
foreco.2016.08.026. URL: http://www.sciencedirect.com/science/article/pii/
Christiansen, E., Waring, R.H., Berryman, A.A., 1987. Resistance of conifers to bark
S0378112716304406.
beetle attack: Searching for general relationships. For. Ecol. Manag. 22, 89–106.
Hollaus, M., Vreugdenhil, M., 2019. Radar Satellite imagery for detecting bark beetle
https://doi.org/10.1016/0378-1127(87)90098-3. URL: http://www.sciencedirect.
outbreaks in forests. Current Forestry Reports 5, 240–250. https://doi.org/10.1007/
com/science/article/pii/0378112787900983.
s40725-019-00098-z. URL: http://link.springer.com/10.1007/s40725-019-00098-z.
Cover, T.M., Thomas, J.A., 1991. Elements of Information Theory 774.
Jactel, H., Koricheva, J., Castagneyrol, B., 2019. Responses of forest insect pests to
Dai, L., Han, Q., de Vries, B., Wang, Y., 2021. Applying Bayesian Belief Network to
climate change: not so simple. Current Opinion in Insect Science 35, 103–108.
explore key determinants for nature-based solutions’ acceptance of local
https://doi.org/10.1016/j.cois.2019.07.010. URL: https://linkinghub.elsevier.
stakeholders. J. Clean. Prod. 310, 127480. https://doi.org/10.1016/j.
com/retrieve/pii/S2214574519300227.

15
M. Tahri et al. Environmental Modelling and Software 147 (2022) 105233

Kjærulff, U.B., Madsen, A.L., 2008. Bayesian Networks and Influence Diagrams: A Guide Niemann, K.O., Visintini, F., 2005. Assessment of potential for remote sensing detection
to Construction and Analysis. Information Science and Statistics. Springer-Verlag, of bark beetle-infested areas during green attack: a literature review 2005-02. URL:
New York. https://doi.org/10.1007/978-0-387-74101-7. URL: https://www.sprin http://cfs.nrcan.gc.ca/publications?id=25269.
ger.com/gp/book/9781441925466. Nováková, M., Edwards-Jonášová, M., 2015. Restoration of central-european mountain
Kokolakis, G., 2010. Bayesian statistical analysis. In: Peterson, P., Baker, E., McGaw, B. Norway spruce forest 15 years after natural and anthropogenic disturbance. For.
(Eds.), International Encyclopedia of Education, third ed. Elsevier, Oxford, Ecol. Manag. 344, 120–130. https://doi.org/10.1016/j.foreco.2015.02.010.
pp. 37–45. https://doi.org/10.1016/B978-0-08-044894-7.01308-7. URL: https Pasztor, F., Matulla, C., Rammer, W., Lexer, M.J., 2014. Drivers of the bark beetle
://www.sciencedirect.com/science/article/pii/B9780080448947013087. disturbance regime in Alpine forests in Austria. For. Ecol. Manag. 318, 349–358.
Kunegel-Lion, M., Lewis, M., 2020a. Mountain pine beetle outbreak duration and pine https://doi.org/10.1016/j.foreco.2014.01.044. URL: https://www.sciencedirect.co
mortality depend on direct control effort. J. Environ. Manag. 260 https://doi.org/ m/science/article/pii/S0378112714000681.
10.1016/j.jenvman.2020.110167. Pearl, J., 1988. Chapter 4 - belief updating BY network propagation. In: Pearl, J. (Ed.),
Kunegel-Lion, M., Lewis, M.A., 2020b. Factors governing outbreak dynamics in a forest Probabilistic Reasoning in Intelligent Systems. Morgan Kaufmann, San Francisco
intensively managed for mountain pine beetle. Sci. Rep. 10, 7601. https://doi.org/ (CA), pp. 143–237. https://doi.org/10.1016/B978-0-08-051489-5.50010-2. URL:
10.1038/s41598-020-63388-8. URL: https://www.nature.com/articles/s41598-02 https://www.sciencedirect.com/science/article/pii/B9780080514895500102.
0-63388-8. Pearl, J., 2009. Probabilistic reasoning in intelligent systems: networks of plausible
Landuyt, D., Van der Biest, K., Broekx, S., Staes, J., Meire, P., Goethals, P.L.M., 2015. inference. In: The Morgan Kaufmann Series in Representation and Reasoning. Rev. 2.
A GIS plug-in for Bayesian belief networks: towards a transparent software Morgan Kaufmann, San Francisco, Calif transferred to digital printing ed.
framework to assess and visualise uncertainties in ecosystem service mapping. Pureswaran, D.S., Roques, A., Battisti, A., 2018. Forest insects and climate change.
Environ. Model. Software 71, 30–38. https://doi.org/10.1016/j. Current Forestry Reports 4, 35–50. https://doi.org/10.1007/s40725-018-0075-6.
envsoft.2015.05.002. URL: https://www.sciencedirect.com/science/article/pii/ URL: http://link.springer.com/10.1007/s40725-018-0075-6.
S1364815215001334. Radl, A., Lexer, M.J., Vacik, H., 2018. A bayesian belief network approach to predict
Laurila-Pant, M., Mäntyniemi, S., Venesjärvi, R., Lehikoinen, A., 2019. Incorporating damages caused by disturbance agents. Forests 9, 15. https://doi.org/10.3390/
stakeholders’ values into environmental decision support: a Bayesian Belief Network f9010015. URL: https://www.mdpi.com/1999-4907/9/1/15.
approach. Sci. Total Environ. 697, 134026. https://doi.org/10.1016/j. Roostaei, J., Colley, S., Mulhern, R., May, A., Gibson, J., 2021. Predicting the risk of
scitotenv.2019.134026. URL: https://linkinghub.elsevier.com/retrieve/pii/S00489 GenX contamination in private well water using a machine-learned Bayesian
69719340033. network model. J. Hazard Mater. 411 https://doi.org/10.1016/j.
Li, M., Wang, H., Wang, D., Shao, Z., He, S., 2020. Risk assessment of gas explosion in jhazmat.2021.125075.
coal mines based on fuzzy AHP and bayesian network. Process Saf. Environ. Protect. Roy, J., Saha, S., 2019. Landslide susceptibility mapping using knowledge driven
135, 207–218. statistical models in Darjeeling District, West Bengal, India. Geoenvironmental
Liou, T.S., Wang, M.J.J., 1992. Ranking fuzzy numbers with integral value. Fuzzy Set Disasters 6. https://doi.org/10.1186/s40677-019-0126-8.
Syst. 50, 247–255. https://doi.org/10.1016/0165-0114(92)90223-Q. URL: http Saaty, T.L., 1977. A scaling method for priorities in hierarchical structures. J. Math.
://www.sciencedirect.com/science/article/pii/016501149290223Q. Psychol. 15, 234–281. https://doi.org/10.1016/0022-2496(77)90033-5. URL: http
Lyu, H.M., Shen, S.L., Yang, J., Yang, J., Zhou, A.N., 2020. Risk assessment of s://www.sciencedirect.com/science/article/pii/0022249677900335.
earthquake-triggered geohazards surrounding Wenchuan, China. Nat. Hazards Rev. Saaty, R.W., 1987. The analytic hierarchy process—what it is and how it is used. Math.
21 https://doi.org/10.1061/(ASCE)NH.1527-6996.0000375. Model. 9, 161–176. https://doi.org/10.1016/0270-0255(87)90473-8. URL: http
Macek, M., Wild, J., Kopecký, M., Červenka, J., Svoboda, M., Zenáhlíková, J., Bru˙na, J., ://www.sciencedirect.com/science/article/pii/0270025587904738.
Mosandl, R., Fischer, A., 2017. Life and death of Picea abies after bark-beetle Schelhaas, M.J., Nabuurs, G.J., Schuck, A., 2003. Natural disturbances in the European
outbreak: ecological processes driving seedling recruitment: Ecological. Ecol. Appl. forests in the 19th and 20th centuries. Global Change Biol. 9, 1620–1633. https://
27, 156–167. https://doi.org/10.1002/eap.1429. doi.org/10.1046/j.1365-2486.2003.00684.x. URL:
Malczewski, J., Rinner, C., 2015. Multicriteria decision analysis in geographic Seidl, R., Schelhaas, M.J., Rammer, W., Verkerk, P.J., 2014. Increasing forest
information science. In: Advances in Geographic Information Science. Springer- disturbances in Europe and their impact on carbon storage. Nat. Clim. Change 4,
Verlag, Berlin Heidelberg. https://doi.org/10.1007/978-3-540-74757-4. URL: http 806–810. https://doi.org/10.1038/nclimate2318. URL: https://www.nature.
s://www.springer.com/gp/book/9783540747567. com/articles/nclimate2318.
Marini, L., Økland, B., Jönsson, A.M., Bentz, B., Carroll, A., Forster, B., Grégoire, J.C., Seidl, R., Müller, J., Hothorn, T., Bässler, C., Heurich, M., Kautz, M., 2016. Small beetle,
Hurling, R., Nageleisen, L.M., Netherer, S., Ravn, H.P., Weed, A., Schroeder, M., large-scale drivers: how regional and landscape factors affect outbreaks of the
2017. Climate drivers of bark beetle outbreak dynamics in Norway spruce forests. European spruce bark beetle. J. Appl. Ecol. 53, 530–540. https://doi.org/10.1111/
Ecography 40, 1426–1435. https://doi.org/10.1111/ecog.02769. URL: https 1365-2664.12540. URL: https://besjournals.onlinelibrary.wiley.com/doi/abs/1
://onlinelibrary.wiley.com/doi/abs/10.1111/ecog.02769. 0.1111/1365-2664.12540.
Matthews, B., Netherer, S., Katzensteiner, K., Pennerstorfer, J., Blackwell, E., Seidl, R., Thom, D., Kautz, M., Martin-Benito, D., Peltoniemi, M., Vacchiano, G., Wild, J.,
Henschke, P., Hietz, P., Rosner, S., Jansson, P.E., Schume, H., Schopf, A., 2018. Ascoli, D., Petr, M., Honkaniemi, J., Lexer, M.J., Trotsiuk, V., Mairota, P.,
Transpiration deficits increase host susceptibility to bark beetle attack: experimental Svoboda, M., Fabrika, M., Nagel, T.A., Reyer, C.P.O., 2017. Forest disturbances
observations and practical outcomes for Ips typographus hazard assessment. Agric. under climate change. Nat. Clim. Change 7, 395–402. https://doi.org/10.1038/
For. Meteorol. 263, 69–89. https://doi.org/10.1016/j.agrformet.2018.08.004. nclimate3303.
McLachlan, S., Dube, K., Hitman, G., Fenton, N., Kyrimi, E., 2020. Bayesian networks in Shao, Z., Huq, M.E., Cai, B., Altan, O., Li, Y., 2020. Integrated Remote Sensing and GIS
healthcare: distribution by medical condition. Artif. Intell. Med. 107 https://doi.org/ Approach Using Fuzzy-AHP to Delineate and Identify Groundwater Potential Zones
10.1016/j.artmed.2020.101912. in Semi-arid Shanxi Province, China, vol. 134. Environmental Modelling & Software,
Mezei, P., Grodzki, W., Blaženec, M., Jakuš, R., 2014. Factors influencing the wind–bark p. 104868. https://doi.org/10.1016/j.envsoft.2020.104868. URL: http://www.sci
beetles’ disturbance system in the course of an Ips typographus outbreak in the Tatra encedirect.com/science/article/pii/S1364815220309257.
Mountains. For. Ecol. Manag. 312, 67–77. https://doi.org/10.1016/j. Stereńczak, K., Mielcarek, M., Kamińska, A., Kraszewski, B., Piasecka Miścicki, S.,
foreco.2013.10.020. URL: http://www.sciencedirect.com/science/article/pii/ Heurich, M., 2020. Influence of selected habitat and stand factors on bark beetle Ips
S0378112713006877. typographus (L.) outbreak in the Białowieża Forest. For. Ecol. Manag. 459 https://
Modlinger, R., Novotný, P., 2015. Quantification of Time Delay between Damages doi.org/10.1016/j.foreco.2019.117826.
Caused by Windstorms and by Ips Typographus, p. 11. Svoboda, M., LeComte, D., Hayes, M., Heim, R., Gleason, K., Angel, J., Rippey, B.,
Morris, J.L., Cottrell, S., Fettig, C.J., Hansen, W.D., Sherriff, R.L., Carter, V.A., Clear, J.L., Tinker, R., Palecki, M., Stooksbury, D., Miskus, D., Stephens, S., 2002. The drought
Clement, J., DeRose, R.J., Hicke, J.A., Higuera, P.E., Mattor, K.M., Seddon, A.W.R., monitor. Bull. Am. Meteorol. Soc. 83, 1181–1190. https://doi.org/10.1175/1520-
Seppä, H.T., Stednick, J.D., Seybold, S.J., 2017. Managing bark beetle impacts on 0477-83.8.1181. URL: https://journals.ametsoc.org/view/journals/bams/83/8/
ecosystems and society: priority questions to motivate future research. J. Appl. Ecol. 1520-0477-83_8_1181.xml.
54, 750–760. https://doi.org/10.1111/1365-2664.12782. URL: https://besjournals. Tahri, M., Maanan, M., Maanan, M., Bouksim, H., Hakdaoui, M., 2017. Using Fuzzy
onlinelibrary.wiley.com/doi/abs/10.1111/1365-2664.12782. Analytic Hierarchy Process multi-criteria and automatic computation to analyse
Nascimento, N., West, T.A.P., Biber-Freudenberger, L., Sousa-Neto, E.R.d., Ometto, J., coastal vulnerability. Prog. Phys. Geogr.: Earth Environ. 41, 268–285. https://doi.
Börner, J., 2019. A Bayesian network approach to modelling land-use decisions org/10.1177/0309133317695158. URL: http://journals.sagepub.com/doi/10.11
under environmental policy incentives in the Brazilian Amazon. J. Land Use Sci. 77/0309133317695158.
1–15. https://doi.org/10.1080/1747423X.2019.1709223. URL: https://www.tandf Tahri, M., Kašpar, J., Vacik, H., Marušák, R., 2020. Predict Potential Risk of Bark Beetle
online.com/doi/full/10.1080/1747423X.2019.1709223. Disturbance Applying Bayesian Belief Networks. Technical Report EGU2020-21348.
Netherer, S., Nopp-Mayr, U., 2005. Predisposition assessment systems (PAS) as Copernicus Meetings. https://doi.org/10.5194/egusphere-egu2020-21348. URL:
supportive tools in forest management—rating of site and stand-related hazards of https://meetingorganizer.copernicus.org/EGU2020/EGU2020-21348.html.
bark beetle infestation in the High Tatra Mountains as an example for system Tahri, M., Kaspar, J., Vacik, H., Marusak, R., 2021. Multi-attribute decision making and
application and verification. For. Ecol. Manag. 207, 99–107. https://doi.org/ geographic information systems: potential tools for evaluating forest ecosystem
10.1016/j.foreco.2004.10.020. URL: https://www.sciencedirect.com/science/artic services. Ann. For. Sci. 78, 41. https://doi.org/10.1007/s13595-021-01049-0. URL:
le/pii/S0378112704007303. https://link.springer.com/10.1007/s13595-021-01049-0.
Netherer, S., Panassiti, B., Pennerstorfer, J., Matthews, B., 2019. Acute drought is an Thorn, S., Bässler, C., Svoboda, M., Müller, J., 2017. Effects of Natural Disturbances and
important driver of bark beetle infestation in Austrian Norway spruce stands. Salvage Logging on Biodiversity – Lessons from the Bohemian Forest Q, vol. 7. Forest
Frontiers in Forests and Global Change 2. https://doi.org/10.3389/ffgc.2019.00039. Ecology and Management.
URL: https://www.frontiersin.org/articles/10.3389/ffgc.2019.00039/full. Triantaphyllou, E., Sánchez, A., 1997. A sensitivity analysis approach for some
deterministic multi-criteria decision-making methods. Decis. Sci. J. 28, 151–194.

16
M. Tahri et al. Environmental Modelling and Software 147 (2022) 105233

https://doi.org/10.1111/j.1540-5915.1997.tb01306.x. URL: https://onlinelibrary. Wealands, S.R., Grayson, R.B., Walker, J.P., 2005. Quantitative comparison of spatial
wiley.com/doi/abs/10.1111/j.1540-5915.1997.tb01306.x. fields for hydrological model assessment—-some promising approaches. Adv. Water
Vacik, H., Lexer, M.J., 2014. Past, current and future drivers for the development of Resour. 28, 15–32. https://doi.org/10.1016/j.advwatres.2004.10.001. URL: https://
decision support systems in forest management. Scand. J. For. Res. 29, 2–19. https:// www.sciencedirect.com/science/article/pii/S0309170804001575.
doi.org/10.1080/02827581.2013.830768. URL: Wisse, B.W., Gosliga, S.P.v., Elst, N.P.v., Barros, A.I., TNO Defensie en Veiligheid, 2008.
Valdez Vasquez, M., Chen, C.F., Lin, Y.J., Kuo, Y.C., Chen, Y.Y., Medina, D., Diaz, K., Relieving the elicitation burden of bayesian belief networks. URL: http://resolver.
2020. Characterizing spatial patterns of pine bark beetle outbreaks during the dry tudelft.nl/uuid:4ac862cd-4576-4271-b8e4-2d72cb9ac048.
and rainy season’s in Honduras with the aid of geographic information systems and Yariyan, P., Zabihi, H., Wolf, I., Karami, M., Amiriyan, S., 2020. Earthquake risk
remote sensing data. For. Ecol. Manag. 467 https://doi.org/10.1016/j. assessment using an integrated fuzzy analytic hierarchy process with Artificial
foreco.2020.118162. neural networks based on GIS: a case study of Sanandaj in Iran. Int. J. Disaster .Risk.
Vanická, H., Holuša, J., Resnerová, K., Ferenčík, J., Potterf, M., Véle, A., Grodzki, W., Reduct 50. https://doi.org/10.1016/j.ijdrr.2020.101705.
2020. Interventions have limited effects on the population dynamics of Ips Zadeh, L.A., 1965. Fuzzy sets. Inf. Control 8, 338–353. https://doi.org/10.1016/S0019-
typographus and its natural enemies in the Western Carpathians (Central Europe). 9958(65)90241-X. URL: http://www.sciencedirect.com/science/article/pii/
For. Ecol. Manag. 470–471. https://doi.org/10.1016/j.foreco.2020.118209. S001999586590241X.
Wang, X.H., Zheng, B., Good, W.F., King, J.L., Chang, Y.H., 1999. Computer-assisted Zhang, X., Mahadevan, S., 2021. Bayesian network modeling of accident investigation
diagnosis of breast cancer using a data-driven Bayesian belief network. Int. J. Med. reports for aviation safety assessment. Reliab. Eng. Syst. Saf. 209 https://doi.org/
Inf. 54, 115–126. https://doi.org/10.1016/S1386-5056(98)00174-9. URL: https:// 10.1016/j.ress.2020.107371.
www.sciencedirect.com/science/article/pii/S1386505698001749.

17

You might also like