![Multicollinearity test of preselected variables [38].](https://www.researchgate.net/publication/362698193/figure/tbl3/AS:11431281079435439@1660704614144/Multicollinearity-test-of-preselected-variables-38_Q320.jpg)
Available via license: CC BY 4.0
Content may be subject to copyright.
Citation: Rezende, M.; Madˇera, P.;
Vahalík, P.; Van Damme, K.;
Habrová, H.; Riccardi, T.; Attorre, F.;
De Sanctis, M.; Sallemi, G.;
Malatesta, L. Identifying Suitable
Restoration and Conservation Areas
for Dracaena cinnabari Balf.f. in
Socotra, Yemen. Forests 2022,13, 1276.
https://doi.org/10.3390/f13081276
Academic Editor: Bartolomeo
Schirone
Received: 8 June 2022
Accepted: 7 August 2022
Published: 12 August 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Article
Identifying Suitable Restoration and Conservation Areas for
Dracaena cinnabari Balf.f. in Socotra, Yemen
Marcelo Rezende 1, * , Petr Madˇera 2, Petr Vahalík3, Kay Van Damme 2,4 , Hana Habrová2,
Tullia Riccardi 1, Fabio Attorre 1, Michele De Sanctis 1, Grazia Sallemi 1and Luca Malatesta 1
1
Department of Environmental Biology, Sapienza University of Rome, Piazzale Aldo Moro 5, 00185 Rome, Italy
2Department of Forest Botany, Dendrology and Geobiocoenology, Faculty of Forestry and Wood
Technology (FFWT), Mendel University in Brno (MENDELU), Zemˇedˇelská3, 613 00 Brno, Czech Republic
3Department of Forest Management and Applied Geoinformatics, Faculty of Forestry and Wood
Technology (FFWT), Mendel University in Brno (MENDELU), Zemˇedˇelská3, 613 00 Brno, Czech Republic
4Centre for Academic Heritage and Archives & Ghent University Botanical Garden, Ghent University,
K.L. Ledeganckstraat 35, 9000 Ghent, Belgium
*Correspondence: marceloarvore@gmail.com
Abstract:
We examine the distribution of Dracaena cinnabari, the Socotran Dragon’s Blood Tree, an
endangered species endemic to the island of Socotra (Yemen)—and we propose an accessibility
approach to its conservation, taking the proximity of local communities and land users into account.
Using the present occurrence of D. cinnabari, we applied a machine learning algorithm (random forest
classifier) to estimate the potential distribution of the species across the island (overall validation
accuracy of 0.91) based on available climatic and physiographic parameters. In parallel, we used an
accessibility methodology to generate a map of the energy cost of accessing potential areas from the
villages. This community-focused accessibility map, combined with the potential distribution map
of Dracaena cinnabari, could contribute to decision-making processes related to long-term ecological
restoration and reforestation activities. With our case study, we wish to emphasize that user-focused
efforts and the implementation of sustainable land practices should play key roles in conserving
endangered tree species.
Keywords: Dracaena cinnabari; Socotra; species distribution model; accessibility model
1. Introduction
Dragon trees are a small group of arborescent Dracaena taxa [
1
] among the more than
190 species in the genus [
2
]. Most dragon tree species are endangered, and often endemic
with isolated, insular populations with low abundances [
3
]. Dracaena cinnabari Balf.f. is
an endemic Cenozoic relict of the main island of Socotra (Yemen), well known for its red
resin called dragon’s blood used since ancient time [
4
,
5
]. The population of this species is
currently estimated at ca. 80,000 individuals [
6
], but the age structure is unbalanced, with a
prevalence of overmature trees [
7
–
9
]. Juvenile trees are missing across the island due to
a long-term population decline caused by overgrazing [
10
–
12
], resin harvesting [
4
], and
manifestation of global climate change [
6
,
13
]. Miller [
14
] assessed D. cinnabari as vulnerable
according to IUCN criteria (IUCN Red List). This species is not threatened with extinction
in the coming years [
15
]; however general models of future population development predict
a decline by 40% in 100 years and extinction over maximally ca. 500 years [6,16].
The iconic Socotran Dragon’s Blood trees provides a wide range of ecosystem ser-
vices [
3
]. Rejžek, et al. [
17
] considered D. cinnabari as a nurse tree with high ecological
importance for biodiversity maintenance; every fallen tree can lead to decreases in the pop-
ulations of other endemic plant species. Due to its ability to capture horizontal precipitation,
the decline of the D. cinnabari population may increase aridification of the land which, in
addition to overgrazing, makes natural regeneration even more difficult even though this
Forests 2022,13, 1276. https://doi.org/10.3390/f13081276 https://www.mdpi.com/journal/forests
Forests 2022,13, 1276 2 of 12
species is well adapted to droughts [
12
,
18
,
19
]. Furthermore, D. cinnabari plays a crucial
cultural role for the indigenous communities of Socotra, providing valuable resin and other
non-timber plant products that have a wide range of medicinal and other ethnobotanical
uses [5].
Other authors referred to the decline of different dragon tree species populations
in the 20th century [
13
,
15
,
20
–
25
], for different reasons. Hence, there is a clear need for
conservation activities to support natural regeneration or artificial reforestation [
3
]. A recent
interesting recovery of Dracaena ombet in Baryakay (Sudan) was documented after mass
death events in the 20th century with a shift of saplings to the higher altitudes and coastal
areas compared to the distribution of adult trees [
21
]. Moreover, in an exclosure experiment
on the island of Socotra, natural regeneration of Dracaena cinnabari was reported [
26
].
Regarding artificial afforestation, there are only a few examples of rescue programs [
3
]. For
D. tamaranae, an effective rescue conservation program has been running in Gran Canaria
(Spain) for over 15 years [
27
]. On the Canary Islands [
28
] and on the Cape Verde Islands [
29
],
many dragon trees (Dracaena draco s.l.) have been cultivated near houses, hotels, in garden
and in parks. On Socotra, efforts to reforest D. cinnabari in situ have been documented since
2006, when more than 700 three-year old seedlings were planted in the Ras Ayre area [
12
].
Taking into account the ongoing decline of the terrestrial ecosystems on Socotra
for various reasons, ecological restoration and reforestation activities of this endemic
tree species are urgently needed [
12
,
30
,
31
]. To identify the most suitable areas for such
interventions, a model, based on the Dragon tree as a case study and applicable to other
species, integrating suitability and accessibility analyses is proposed in this study.
2. Materials and Methods
2.1. Datasets
Although lacking detailed resolution, remote sensing products containing climatic
data are currently widely available [
32
]. For this study we used bioclimatic variables [
33
],
due to their prompt availability as a collection in Google Earth Engine (Table 1).
Table 1.
Bioclimatic variables [
33
] considered in this study (if scale values equal zero, it indicates that
no conversion is applied to the pixel value).
Bands Description Unit Scale
bio01 Annual mean temperature ◦C 0.1
bio02 Mean diurnal range (mean of monthly (max–min temperature)) ◦C 0.1
bio03 Isothermality (bio02/bio07) % 0
bio04 Temperature seasonality (standard deviation ×100) ◦C 0.01
bio05 Max temperature of warmest month ◦C 0.1
bio06 Min temperature of coldest month ◦C 0.1
bio07 Annual temperature range (bio05–bio06) ◦C 0.1
bio08 Mean temperature of wettest quarter ◦C 0.1
bio09 Mean temperature of driest quarter ◦C 0.1
bio10 Mean temperature of warmest quarter ◦C 0.1
bio11 Mean temperature of coldest quarter ◦C 0.1
bio12 Annual precipitation mm 0
bio13 Precipitation of wettest month mm 0
bio14 Precipitation of driest month mm 0
bio15 Precipitation seasonality (coefficient of variation) CoV 0
bio16 Precipitation of wettest quarter mm 0
bio17 Precipitation of driest quarter mm 0
bio18 Precipitation of warmest quarter mm 0
bio19 Precipitation of coldest quarter mm 0
Topographic parameters are key in describing species distribution across the landscape.
This is relevant for Socotra, where physiography plays a key role in creating different
microclimates across the island [
5
]. We used the Shuttle Radar Topography Mission (SRTM)
Forests 2022,13, 1276 3 of 12
digital elevation data, at a resolution of 1 arc-second (approximately 30 m), to derive slope,
elevation, and aspect to be considered in the distribution model [
34
]. To complement the
selected variables, three soil-related variables were included to evaluate their potential
contribution to increase the model’s accuracy. Soil characteristics can also facilitate the
stratification of potential restoration sites and allow for more tailored restoration practices
and recommendations (Table 2).
Table 2. Soil-related variables considered in this study.
Bands Description Unit Scale
CEC Cation-exchange capacity (at pH 7) (0–5 cm depth) mmol(c)/kg 0.1
pH Potential of hydrogen (0–5 cm depth) pH*10 0.1
Sand Sand content (0–5 cm depth) g/kg 0
In order to harmonize all environmental layers, we used bicubic interpolation neighbor
embedding for interpolating data points to create a collection of “super high resolution”
images for analysis (Supplementary Figure S1) [
35
]. This operation was performed using
Google Earth Engine, and is available through the code in the Results section [36].
2.2. Data Analysis—Species Distribution
To model the potential distribution of Dracaena cinnabari on Socotra, we used Google
Earth Engine to process the layers, create testing and training sample plots, train a Random
Forest classifier, and display the accuracy and final results. Multicollinearity tests were
performed using STATISTICA Version 12 [37].
Individual trees were identified using visual interpretation of satellite images, cor-
rected with ground truthing, and then used to map the current distribution of Dracaena
cinnabari on Socotra (Supplementary Figure S2) [
6
]. This database, containing 80,247 tree
occurrences, was transformed into a categorical image of the presence and absence of
Dracaena cinnabari trees [
6
]. The operation was performed applying a 100 m buffer from the
tree location (presence = 1), and mapping the remaining pixels as zero (absence = 0).
First, a systematic grid of plots was created, spaced at 100 ×100 m from one another.
These plots (368,707 in total) were exported from Google Earth Engine to be tested for
multicollinearity (Table 3). This test of predicting variables was performed using tolerance
and the variance inflation factor (VIF) [38–42].
Table 3. Multicollinearity test results, including tolerance, variance, and R-squared.
Variables Description Tolerance Variance R-Squared
Aspect Aspect
0.9825247
1.0178 0.0174753
“bio01” Annual mean temperature
0.0004400
2272.7789 0.9995600
“bio02” Mean diurnal range
0.0073897
135.3231 0.9926103
“bio03” Isothermality (bio02/bio07)
0.0742739
13.4637 0.9257261
“bio04” Max temperature of warmest month
0.1134717
8.8128 0.8865283
“bio05” Min temperature of coldest month
0.0006309
1585.0028 0.9993691
“bio06” Annual temperature range
(bio05-bio06)
0.0003536
2827.8492 0.9996464
“bio07” Mean temperature of wettest quarter
0.0000000
“bio08” Mean temperature of driest quarter
0.0004968
2012.9537 0.9995032
“bio09” Mean temperature of warmest quarter
0.0004151
2408.8149 0.9995849
“bio10” Mean temperature of coldest quarter
0.0000000
“bio11” Annual precipitation
0.0003522
2839.0579 0.9996478
“bio12” Precipitation of wettest month
0.0016195
617.4925 0.9983805
“bio13” Precipitation of driest month
0.0054004
185.1713 0.9945996
“bio14” Precipitation seasonality
0.0649898
15.3870 0.9350102
Forests 2022,13, 1276 4 of 12
Table 3. Cont.
Variables Description Tolerance Variance R-Squared
“bio15” Precipitation of wettest quarter
0.0897891
11.1372 0.9102109
“bio16” Precipitation of driest quarter
0.0016349
611.6759 0.9983651
“bio17” Precipitation of warmest quarter
0.0129514
77.2116 0.9870486
“bio18” Precipitation of coldest quarter
0.0000000
“bio19” Precipitation of coldest quarter
0.0026119
382.8649 0.9973881
CEC Cation-exchange capacity
0.2232478
4.4793 0.7767522
pH
0.1014427
9.8578 0.8985573
Sand Sand content
0.1351782
7.3976 0.8648218
Slope
0.8471080
1.1805 0.1528920
Considering the above results and the ecological coherence of the variables, highly
correlated variables and others with high VIF values (i.e., above 10) were discarded, priori-
tizing variables that are commonly recorded in weather stations and ensuring that both
temperature and precipitation were considered in the final selection. Simulations with
different variables and combinations were also performed to assist with variable selection.
Upon selecting the variables (“bio01”, “bio04”, “bio15”, “cec”, “ph”, “dist”, “sand”, “slope”),
another multicollinearity test was performed [
42
]. When considering only the selected
variables, the multicollinearity was drastically reduced to values below VIF = 10 (Table 4).
Table 4. Multicollinearity test of preselected variables [38].
Bands Description Tolerance Variance R-Squared
“bio01” Annual mean temperature
0.4317074
2.3163837 0.5682926
“bio04” Max temperature of warmest month
0.4734176
2.1123002 0.5265824
“bio15” Precipitation of wettest quarter
0.2625159
3.8092930 0.7374841
CEC Cation-exchange capacity
0.3358107
2.9778686 0.6641893
Sand Sand content
0.1435930
6.9641276 0.8564070
slope Slope
0.8719841
1.1468100 0.1280159
pH pH
0.1060253
9.4317117 0.8939747
Google Earth Engine was used to run a Random Forest (RF) classifier and test its
accuracy in estimating the potential distribution of the species. 24,646 training plots were
used for training and 10,524 plots for testing; 500 decision tree iterations were used in the
training of the model.
The number of plots predicted by the classifier (RF) with presence from the train-
ing sample was 17,585, while the number of plots predicted with absence was 351,122.
Previous studies found that RF improved its performance with balanced classes, and
there was a relevant disproportion between the two classes, with the absence class being
over-represented [
43
,
44
]. An under-sampling method was used to balance the datasets by
limiting the number of absence plots as a function of the number of presence plots [
45
,
46
]
and cross-validation was applied dividing the training and testing sample data into a
70:30 proportion, of which 70% were used to “train” the model (training data set) and the
remaining 30% (validation data set) were used to evaluate its predictive capacity. According
to the confusion matrix, reported in Table 5, the overall validation accuracy was 0.91.
Table 5. Classification matrix (testing dataset).
Classified 0 Classified 1
Observed 0 4570 740
Observed 1 192 5022
The traditional metric of overall accuracy is no longer adequate for describing classifier
performance [
45
,
47
]; therefore, the confusion matrix and its values were used to calculate
Forests 2022,13, 1276 5 of 12
some performance metrics. In the confusion matrix, the precision (TP/TP + FN) is the
number of correctly-identified presence divided by the total number of times the model
predicted presence, while the recall (TN/TN + FP) is the ratio of actual absence plots that
were predicted incorrectly as presence. The results were precision = 0.87 and recall = 0.96.
Therefore, in 87% of the cases, RF correctly predicts presence and in 96% of cases, it correctly
predicted absence. Precision and recall cannot be high at the same time; if an optimization
is performed on one, then the other will decrease. Therefore, when applying these metrics,
we must choose which of them we want to be more precise.
For the objectives of this research a better prediction of true presence is a priority
compared to the prediction of true absence, especially considering the difficulty of reaching
some areas of the island. A high probability of success is required in order to efficiently
allocate efforts and resources. A greater significance of precision is supported by the
analysis of Peterson et al. [
48
], who states that “in a niche-modeling framework, a model that
errors by omitting known points of presence is more seriously flawed than one that predicts areas
not known to be inhabited”.
This is also the reason we decided to use precision and recall instead of specificity
and sensitivity—two other performance metrics commonly used in studies on species
distribution models [
47
]. Precision focuses on true presence, while it overlooks true absence.
Therefore, an 87% degree of correct predictions of true presence allows for a theoretical
framework identifying the suitable areas for Dracaena reforestation projects; however,
further experiments could be implemented with different sampling strategies [
49
], or
different algorithms as also recommended by Qiao et al. [
50
], to obtain a higher precision
value and consequently a better prediction capacity.
2.3. Data Analysis—Accessibility
In addition to the current Dracaena cinnabari distribution map, this study proposes
an accessibility model focused on the locations of villages/settlements on Socotra which
are relevant to help in choosing the most suitable restoration and conservation areas for
these trees. With these two maps in hand, decision makers can start to plan interventions
and investments for reforestation/conservation activities taking into consideration the
metabolic energy cost for the land users.
For an accessibility map to be focused on the local human population, we first need
to take into consideration how far the settlements are from other points of the island.
The distance from settlements to other parts of the island in meters was calculated using
recent available geolocation data of villages across the islands of Socotra (Vahalík et al.,
in preparation).
Similarly to Riccardi et al. [
51
], we developed an accessibility model focused on the
location of villages/settlements on Socotra to identify the areas most suitable for reforesta-
tion/conservation actions. Even though we are aware that not all villages/settlements are
occupied throughout the year and that local people often move from and to different areas
for part of the year, we considered the maximal potential occupancy of villages at this point
in time. First, the distance from settlements to other parts of the island, in meters, was
calculated (Supplementary Figure S3).
Considering the trade-offs in place regarding the time available to allocate for restora-
tion efforts, we can assume that land users weigh the costs of accessing a site based on
the distance from their household. Any restoration activity must rely on the contribu-
tion of its surrounding communities to achieve success. This can be in the form of active
contribution—such as tree planting and maintenance, improved management practices,
degradation controls, etc.—or indirectly—for example, the exclusion of livestock and peo-
ple through fencing. Aside from distance, terrain ruggedness is also a discouraging factor
for moving through a landscape. In order to account for this factor, a metabolic energy cost
image was created based on slope.
Forests 2022,13, 1276 6 of 12
Slope and its links to the energy required to move have been extensively
researched [52–54]
.
In this study, we used an equation proposed by Minetti et al. [
54
] to relate the metabolic
energy cost of walking to slope, as follows:
Cw = 280.5 m5−58.7 m4−76.8 m3+ 51.9 m2+ 19.6 m + 2.5 (1)
where m is the slope and Cw (J
·
kg
−1·
m
−1
) is the energy cost of moving one unit of mass a
horizontal distance equivalent to one unit of vertical displacement. Notice that due to the
polynomial formulation, this equation is valid for a range of slopes approximately between
−
0.5 and 0.5, outside of which the behavior of the function becomes counterintuitive [
55
].
The above approach was implemented using Google Earth Engine for image processing and
SRTM as the base elevation model for the calculation of slope (resampled) (Supplementary
Figure S1).
Multiplying the distance to settlements by the metabolic energy cost (Supplementary
Figure S4), we estimate the energy required for moving one unit of mass (1 kg) outwards
from the settlements’ center points. We assume that the higher the energy required to
access an area, the lower the probability of community individuals commuting to it for
reforestation and ecological restoration purposes (Figure 1).
Forests 2022, 13, x FOR PEER REVIEW 7 of 13
Figure 1. Accessibility map using energy requirement (J⋅kg
−1
) to access different areas of the island
of Socotra.
The areas of Socotra with the highest energy requirement for the local populations to
access are the Hageher Mountains, the escarpments along the southern part of the island,
and the rugged mountain ranges in the west of the island. Flat areas both on the coast and
inland, where many population centers are found, have the estimated lowest energy
requirements to access.
3. Results
Random Forest proved to be an efficient classifier, with an overall validation
accuracy of 0.91. Among the predictors, pH was identified as the most important,
followed by the precipitation of the wettest quarter, maximum temperature of the
warmest month, annual mean temperature, cation exchange capacity, sand content, and
slope (Table 6).
Table 6. Classification predictor importance.
Variable Importance
pH 1.000000
bio15 0.875614
bio04 0.677423
bio01 0.675045
CEC 0.548218
Sand 0.498888
Slope 0.325849
The area where Dracaena cinnabari can potentially occur is over 25% of the total area
of the island (Figure 2). The potential distribution of Dragon’s blood trees overlaps with
more than 80 settlement locations.
Figure 1.
Accessibility map using energy requirement (J
·
kg
−1
) to access different areas of the island
of Socotra.
The areas of Socotra with the highest energy requirement for the local populations to
access are the Hageher Mountains, the escarpments along the southern part of the island,
and the rugged mountain ranges in the west of the island. Flat areas both on the coast
and inland, where many population centers are found, have the estimated lowest energy
requirements to access.
3. Results
Random Forest proved to be an efficient classifier, with an overall validation accuracy
of 0.91. Among the predictors, pH was identified as the most important, followed by the
precipitation of the wettest quarter, maximum temperature of the warmest month, annual
mean temperature, cation exchange capacity, sand content, and slope (Table 6).
Forests 2022,13, 1276 7 of 12
Table 6. Classification predictor importance.
Variable Importance
pH 1.000000
bio15 0.875614
bio04 0.677423
bio01 0.675045
CEC 0.548218
Sand 0.498888
Slope 0.325849
The area where Dracaena cinnabari can potentially occur is over 25% of the total area of
the island (Figure 2). The potential distribution of Dragon’s blood trees overlaps with more
than 80 settlement locations.
Forests 2022, 13, x FOR PEER REVIEW 8 of 13
Figure 2. Potential distribution of Dracaena cinnabari (green: presence) on Socotra (Yemen), using
the parameters considered in this study.
By overlapping the potential distribution map (Figure 2) with that of accessibility
from villages calculated as the energy requirement (Figure 1), a map of
suitability/accessibility for Dracaena cinnabari was produced (Figure 3).
Figure 3. Energy requirement (J⋅kg
−1
) within the potential distribution of Dracaena cinnabari, with
warmer colors indicating relatively higher energy needed to reach the areas.
The most suitable areas for theoretical community-driven ecological restoration and
reforestation interventions for Dracaena cinnabari are shown in Figure 3. Notice that
although the island has vast areas of easy accessibility (Figure 1, white to yellow colors),
the average energy required to access areas where our modeling indicates the potential
for Dracaena cinnabari distribution is 20% higher than that of the total island area. This is
due to the overall physiography of the landscape (mainly steep valleys, gorges, cliffs and
the Hageher Mountains), coupled with fewer and more disperse settlements.
Figure 2.
Potential distribution of Dracaena cinnabari (green: presence) on Socotra (Yemen), using the
parameters considered in this study.
By overlapping the potential distribution map (Figure 2) with that of accessibility from
villages calculated as the energy requirement (Figure 1), a map of suitability/accessibility
for Dracaena cinnabari was produced (Figure 3).
The most suitable areas for theoretical community-driven ecological restoration and
reforestation interventions for Dracaena cinnabari are shown in Figure 3. Notice that al-
though the island has vast areas of easy accessibility (Figure 1, white to yellow colors), the
average energy required to access areas where our modeling indicates the potential for
Dracaena cinnabari distribution is 20% higher than that of the total island area. This is due
to the overall physiography of the landscape (mainly steep valleys, gorges, cliffs and the
Hageher Mountains), coupled with fewer and more disperse settlements.
Forests 2022,13, 1276 8 of 12
Forests 2022, 13, x FOR PEER REVIEW 8 of 13
Figure 2. Potential distribution of Dracaena cinnabari (green: presence) on Socotra (Yemen), using
the parameters considered in this study.
By overlapping the potential distribution map (Figure 2) with that of accessibility
from villages calculated as the energy requirement (Figure 1), a map of
suitability/accessibility for Dracaena cinnabari was produced (Figure 3).
Figure 3. Energy requirement (J⋅kg
−1
) within the potential distribution of Dracaena cinnabari, with
warmer colors indicating relatively higher energy needed to reach the areas.
The most suitable areas for theoretical community-driven ecological restoration and
reforestation interventions for Dracaena cinnabari are shown in Figure 3. Notice that
although the island has vast areas of easy accessibility (Figure 1, white to yellow colors),
the average energy required to access areas where our modeling indicates the potential
for Dracaena cinnabari distribution is 20% higher than that of the total island area. This is
due to the overall physiography of the landscape (mainly steep valleys, gorges, cliffs and
the Hageher Mountains), coupled with fewer and more disperse settlements.
Figure 3.
Energy requirement (J
·
kg
−1
) within the potential distribution of Dracaena cinnabari, with
warmer colors indicating relatively higher energy needed to reach the areas.
4. Discussion
4.1. Why Attempt Reforestation of Socotran Dragon’s Blood Trees?
Based on circumstantial evidence, the forests of Dragon Blood trees in the island were
most likely distributed over a substantially larger area in the past [
13
,
56
]. The decline of D.
cinnabari forests can be attributed to the combination of long-term harvesting of resin [
4
],
rapid growth of the human population on Socotra after the Second World War (resulting in
land-use changes), and the effects of current climate change [
3
,
13
]. Using niche modelling,
Attorre et al. [
13
] developed a first model of the potential distribution of this species, which
extends far beyond its current distribution. Some areas currently without these trees are
areas of potential previous occurrence according to a survey using phytotoponyms [
56
].
Our model of the potential distribution of Dracaena on Socotra shows a relatively smaller
area than the first one published by Attorre et al. [
13
], possibly due to the input of different
and updated parameters. In addition, it is also possible that global climate change resulted
in the input of different climate characteristics as the models were updated. Reforestation
is a major strategy to stop the decline of the tree population to avoid the risk of extinction
of this endemic species.
Moreover, these trees play an important ecological role for the entire island. This
tree species helps to maintain biodiversity, has broad ethnobotanical uses and, as a part of
cloud forests brings huge input of water for the hydrogeological cycle of Socotra [5,18,57].
The unique umbrella-shape of the crown constitutes a fundamental natural system that
traps humidity brought by the ocean breeze and helps recharge the shallow aquifers
of the island [
18
]. Therefore, the disappearance of these dragon’s blood trees would
have a significant impact on the freshwater resources [
18
], biodiversity [
58
], and touristic
attractiveness of Socotra [
30
], in addition to the potential loss of key human–nature linkages
that the island is known for.
The major cause of the decline of these trees is the absence of natural regeneration due
to overgrazing, resulting in an unbalanced age structure of the entire population [
6
,
8
,
9
].
Without regeneration the forest decline will continue towards global extinction through the
gradual disappearance of local subpopulations [6,16].
4.2. Suitability Models
Suitability models are becoming widely used to guide and support conservation and
reforestation programs targeting endangered tree species. These models are particularly rel-
evant when considering the current and future impacts of climate change [
59
–
62
]. However,
Forests 2022,13, 1276 9 of 12
when such models include the involvement of local communities for their implementation,
an estimation of the accessibility of suitable areas must be considered. In this study, we
applied an approach similar to that of Riccardi et al. [
51
], based on an assessment of the
energy required to access an area from the surrounding villages (Figure 1). Based on the
model, about 25% of Socotra is potentially suitable for Dracaena cinnabari (Figure 2), but the
accessibility varies greatly depending on the slope and distance from settlements (Figure 3).
4.3. Reforestation and Conservation Programs
Intense reforestation programs, using a community forestry approach in the most
accessible areas, should be urgently implemented based on the findings of exclosure
experiments, which have already been established in several areas on the island [
6
,
12
,
26
].
Considering time that it takes for these trees to escape browsing risk from goats, care
by local communities is a long-term commitment (of several decades, or perhaps even a
century) [
12
]. Different methods must be applied to protect natural regeneration, as well as
seedlings produced in local nurseries, until these plants can escape the browsing zone and
the effects of drought [
12
,
26
,
31
]. These methods include fences, agroforestry, individual
protection, walling, and watering, but potentially also other approaches assisting natural
regeneration and, above all, long-term community involvement [63,64].
Conversely, in less accessible areas, such as the Hageher Mountains, which were
already considered a potential refugium of Dracaena cinnabari [
5
,
13
], strict conservation
measures would be best put in place together with the revision of the Socotra Conservation
Zoning Plan [
31
] to enlarge the boundaries of the nature sanctuaries and include all of the
remaining dragon tree forests on the island. Based on the Socotra Conservation Zoning
Plan, it is important to enable local communities to participate in the development and
implementation of such management plans to foster ownership and secure sustainability
of the interventions [65].
5. Conclusions
Dracaena cinnabari is experiencing a fast change related to ecosystem dynamics, and
recent studies have predicted its extinction if the current trends are not reversed. This
methodology seeks to enhance science-based approaches to improve the selection of areas
for restoration and conservation of Dracaena cinnabari, with a strong focus on community
engagement and participatory processes.
Supplementary Materials:
The following supporting information can be downloaded at: https:
//www.mdpi.com/article/10.3390/f13081276/s1, Figure S1. Slope imagery showing (a) the original
layer (slope with a resolution of 30 m [
66
]); (b) high-resolution image of the area [
67
]; (c) resampled
slope image applying a bicubic interpolation function in Google Earth Engine. See the Supplementary
Materials section for an interactive geoportal. Figure S2. Current distribution of Dracaena cinnabari in
Socotra, based on [
6
]. Figure S3. Relative distance from human settlements (blue = close to settlement,
orange = far from settlements), modified based on [
51
]. Figure S4. Metabolic energy cost (J
·
kg
−1·
m
−1
)
for land users on the island of Socotra, based on slope.
Author Contributions:
Conceptualization, M.R., F.A., P.M. and K.V.D.; methodology, M.R., T.R., P.V.
and L.M.; software, M.R. and G.S.; validation, M.R., P.M. and K.V.D.; formal analysis, M.R., F.A. and
L.M.; investigation, M.R., K.V.D., H.H. and P.M.; resources, M.R., H.H., P.V. and P.M.; data curation,
M.R. and P.V.; writing—original draft preparation, M.R., F.A. and P.M.; writing—review and editing,
M.R., K.V.D., M.D.S., F.A. and L.M.; visualization, M.R.; supervision, F.A., P.M. and K.V.D.; project
administration, F.A., P.M. and K.V.D.; funding acquisition, F.A. and P.M. All authors have read and
agreed to the published version of the manuscript.
Funding:
This research was partly funded by the Franklinia Foundation (2020–2023) “Conservation
of the endangered endemic Boswellia trees on Socotra Island (Yemen)”, grant number 2020-03.
Informed Consent Statement:
Informed consent was obtained from all subjects involved in
the study.
Forests 2022,13, 1276 10 of 12
Data Availability Statement:
The datasets used in this study can be accessed directly in Google
Earth Engine, available at https://developers.google.com/earth-engine/datasets/ (accessed on
10 August 2022). Settlement locations used in the study are not yet publicly available.
Conflicts of Interest: The authors declare no conflict of interest.
References
1.
Marrero, A.; Almeida, S.R.; Martín-González, M. A new species of the wild dragon tree, Dracaena (Dracaenaceae) from Gran
Canaria and its taxonomic and biogeographic implications. Bot. J. Linn. Soc. 1998,128, 291–314. [CrossRef]
2.
Govaerts, R.; Zonneveld, B.J.M.; Zona, S.A. World checklist of Asparagaceae. Facilitated by the Royal Botanic Gardens, Kew.
Available online: http://apps.kew.org/wcsp/ (accessed on 4 November 2021).
3.
Madˇera, P.; Forrest, A.; Hanáˇcek, P.; Vahalík, P.; Gebauer, R.; Plichta, R.; Jupa, R.; Van Rensburg, J.J.; Morris, M.; Nadezhdina, N.;
et al. What We Know and What We Do Not Know about Dragon Trees? Forests 2020,11, 236. [CrossRef]
4.
Al-Okaishi, A. Local Management System of Dragon’s Blood Tree (Dracaena cinnabari Balf. f.) Resin in Firmihin Forest, Socotra
Island, Yemen. Forests 2020,11, 389. [CrossRef]
5.
Miller, A.G.; Morris, M.; Diccon, A.; Atkinson, R. Ethnoflora of the Soqotra Archipelago, 1st ed.; Royal Botanic Garden: Edinburgh,
UK, 2004; p. 759.
6.
Madˇera, P.; Volaˇrík, D.; Patoˇcka, Z.; Kalivodová, H.; Divín, J.; Rejžek, M.; Vybíral, J.; Lvonˇcík, S.; Jeník, D.; Hanáˇcek, P.; et al.
Sustainable Land Use Management Needed to Conserve the Dragon’s Blood Tree of Socotra Island, a Vulnerable Endemic
Umbrella Species. Sustainability 2019,11, 3557. [CrossRef]
7.
Adolt, R.; Habrova, H.; Madera, P. Crown age estimation of a monocotyledonous tree species Dracaena cinnabari using logistic
regression. Trees-Struct. Funct. 2012,26, 1287–1298. [CrossRef]
8.
Adolt, R.; Madˇera, P.; Abraham, J.; ˇ
Cupa, P.; Svátek, M.; Matula, R.; Šebesta, J.; ˇ
Cermák, M.; Volaˇrík, D.; Koutecký, T.; et al. Field
Survey of Dracaena cinnabari Populations in Firmihin, Socotra Island: Methodology and Preliminary Results. J. Landsc. Ecol.
2013
,
6, 7–34. [CrossRef]
9.
Adolt, R.; Pavlis, J. Age structure and growth of Dracaena cinnabari populations on Socotra. Trees-Struct. Funct.
2004
,18, 43–53.
[CrossRef]
10.
Hubálková, I.; Madˇera, P.; Volaˇrík, D. Growth dynamics of Dracaena cinnabari under controlled conditions as the most effective
way to protect endangered species. Saudi J. Biol. Sci. 2017,24, 1445–1452. [CrossRef]
11.
Madˇera, P.; Habrová, H.; ˇ
Cermák, M.; Adolt, R.; Volaˇrík, D. Age estimation of juvenile stages of Dracaena cinnabari Balf. F., the last
stone in the mosaic of knowing its lifespan. Rend. Lincei Sci. Fis. Nat. 2020,31, 677–686. [CrossRef]
12.
Madˇera, P.; Habrová, H.; Šenfeldr, M.; Kholová, I.; Lvonˇcík, S.; Ehrenbergerová, L.; Roth, M.; Nadezhdina, N.; Nˇemec, P.;
Rosenthal, J.; et al. Growth dynamics of endemic Dracaena cinnabari Balf. f. of Socotra Island suggest essential elements for a
conservation strategy. Biologia 2019,74, 339–349. [CrossRef]
13.
Attorre, F.; Francesconi, F.; Taleb, N.; Scholte, P.; Saed, A.; Alfo, M.; Bruno, F. Will dragonblood survive the next period of climate
change? Current and future potential distribution of Dracaena cinnabari (Socotra, Yemen). Biol. Conserv.
2007
,138, 430–439.
[CrossRef]
14.
Miller, A. Dracaena cinnabari. The IUCN Red List of Threatened Species 2004: E.T30428A9548491. Available online: https:
//www.iucnredlist.org/species/30428/9548491 (accessed on 18 November 2021). [CrossRef]
15. Habrova, H.; Cermak, Z.; Pavlis, J. Dragon’s blood tree–Threatened by overmaturity, not by extinction: Dynamics of a Dracaena
cinnabari woodland in the mountains of Soqotra. Biol. Conserv. 2009,142, 772–778. [CrossRef]
16.
Hubálková, I. Prediction of Dragon’s Blood Tree (Dracaena cinnabari Balf.) Stand Sample Density on Soqotra Island. J. Landsc. Ecol.
2011,4, 5–17. [CrossRef]
17.
Rejžek, M.; Svátek, M.; Šebesta, J.; Adolt, R.; Madˇera, P.; Matula, R. Loss of a single tree species will lead to an overall decline in
plant diversity: Effect of Dracaena cinnabari Balf. f. on the vegetation of Socotra Island. Biol. Conserv.
2016
,196, 165–172. [CrossRef]
18.
Kalivodová, H.; Culek, M.; ˇ
Cermák, M.; Madˇera, P.; Habrová, H. Potential importance of Socotra dragon’s blood tree cloud forests
and woodlands for capturing horizontal precipitation. Rend. Lincei Sci. Fis. Nat. 2020,31, 607–621. [CrossRef]
19.
Nadezhdina, N.; Al-Okaishi, A.; Madera, P. Sap Flow Measurements in a Socotra Dragon’s Blood Tree (Dracaena cinnabari) in its
Area of Origin. Trop. Plant Biol. 2018,11, 107–118. [CrossRef]
20.
Almeida Pérez, R.S. Dracaena draco (L.) L. In Atlas y Libro Rojo de la Flora Vascular Amenazada de España, 2nd ed.; Bañares, A., Blanca,
G., Güemes, J., Moreno, J.C., Ortiz, S., Eds.; Publicaciones de O.A.P.N.: Madrid, Spain, 2004; pp. 680–681.
21.
Andersen, G.L.; Krzywinski, K.; Talib, M.; Musa, M.A.M. Endangered Dracaena ombet Population in the Red Sea Hills, Sudan,
Recovers After Abrupt Change. Front. Environ. Sci. 2022,10, 793583. [CrossRef]
22. Bari, E. Sudan Conservation of Vegetation in Africa South of the Sahara. Acta Phytogeogr. Suec. 1968,54, 59–64.
23.
Elnoby, S.K.; Moustafa, A.A.; Mansour, S.R. Impact of climate change on the endangered Nubian dragon tree (Dracaena ombet) in
the South Eastern of Egypt. Catrina 2017,16, 23–28. [CrossRef]
24.
Ghazaly, U. Effect of Climatic Change on the endangered Dracaena ombet tree in the Gabel Elba Protected Area in Egypt. Mt.
Forum Bull. 2008,VIII, 11–12.
Forests 2022,13, 1276 11 of 12
25.
Kamel, M.; Ghazaly, U.M.; Callmander, M.W. Conservation status of the Endangered Nubian dragon tree Dracaena ombet in
Gebel Elba National Park, Egypt. Oryx 2015,49, 704–709. [CrossRef]
26.
Habrova, H.; Pavlis, J. Dynamic response of woody vegetation on fencing protection in semi-arid areas; Case study: Pilot
exclosure on the Firmihin Plateau, Socotra Island. Saudi J. Biol. Sci. 2017,24, 338–346. [CrossRef] [PubMed]
27.
Marrero, A.; Almeida, R.S.; Roca, A. Diáspora y rescate: Acciones para la conservación del Drago de Gran Canaria, Dracaena
tamaranae. In Proceedings of the V Congreso de Biología de la Conservación de Plantas, Minorca, Spain, 28 September–1
October 2011; Available online: https://www.researchgate.net/profile/Rafael-S-Perez/publication/270822517_Diaspora_
y_rescate_acciones_para_la_conservacion_del_drago_de_Gran_Canaria_Dracaena_tamaranae/links/54b58bbf0cf28ebe92e7
913f/Diaspora-y-rescate-acciones-para-la-conservacion-del-drago-de-Gran-Canaria-Dracaena-tamaranae.pdf (accessed on
1 January 2022).
28.
Almeida Pérez, R.S. Sobre la presencia de Dracaena draco (L.) L. En Gran Canaria (Islas Canarias): Aportación corológica, estado
actual y significación biogeográfica. Bot. Macarónesica 2003,24, 17–38.
29.
Marrero, A.; Almeida, S.R. A new subspecies, Dracaena draco (L.) L. subsp. caboverdeana Marrero Rodr. & R. Almeida (Dracae-
naceae) from Cape Verde Islands. Int. J. Geobot. Res. 2012,2, 35–40. [CrossRef]
30.
Attorre, F.; Van Damme, K. Twenty years of biodiversity research and nature conservation in the Socotra Archipelago (Yemen).
Rend. Fis. Acc. Lincei 2020,31, 563–569. [CrossRef]
31.
Van Damme, K.; Banfield, L. Past and present human impacts on the biodiversity of Socotra Island (Yemen): Implications for
future conservation. Zoöl. Middle East 2011,54, 31–88. [CrossRef]
32.
Agrillo, E.; Filipponi, F.; Pezzarossa, A.; Casella, L.; Smiraglia, D.; Orasi, A.; Attorre, F.; Taramelli, A. Earth Observation and
Biodiversity Big Data for Forest Habitat Types Classification and Mapping. Remote Sens. 2021,13, 1231. [CrossRef]
33.
Hijmans, R.J.; Cameron, S.E.; Parra, J.L.; Jones, P.G.; Jarvis, A. Very high resolution interpolated climate surfaces for global land
areas. Int. J. Climatol. 2005,25, 1965–1978. [CrossRef]
34.
Farr, T.G.; Rosen, P.A.; Caro, E.; Crippen, R.; Duren, R.; Hensley, S.; Kobrick, M.; Paller, M.; Rodriguez, E.; Roth, L.; et al. The
Shuttle Radar Topography Mission. Rev. Geophys. 2007,45, RG2004. [CrossRef]
35.
Chang, H.; Yeung, D.-Y.; Xiong, Y. Super-resolution through neighbor embedding. In Proceedings of the 2004 IEEE Computer
Society Conference on Computer Vision and Pattern Recognition, Washington, DC, USA, 27 June–2 July 2004; Volume 1. [CrossRef]
36.
Gorelick, N.; Hancher, M.; Dixon, M.; Ilyushchenko, S.; Thau, D.; Moore, R. Google Earth Engine: Planetary-scale geospatial
analysis for everyone. Remote Sens. Environ. 2017,202, 18–27. [CrossRef]
37. Weiß, C.H. StatSoft, Inc., Tulsa, OK.: STATISTICA, Version 8. AStA Adv. Stat. Anal. 2007,91, 339–341. [CrossRef]
38.
Belsley, D.A. Conditioning Diagnostics: Collinearity and Weak Data Regression, 1st ed.; Wiley-Interscience: Baltimore, MD, USA, 1991;
p. 396. [CrossRef]
39. Craney, T.A.; Surles, J.G. Model-Dependent Variance Inflation Factor Cutoff Values. Qual. Eng. 2002,14, 391–403. [CrossRef]
40. Etikan, I.; Kabiru, B. Sampling and Sampling Methods. Biom. Biostat. Int. J. 2017,5, 00149. [CrossRef]
41. Hair, J.F. Multivariate Data Analysis with Readings, 4th ed.; Prentice Hall: Englewood Cliffs, NJ, USA, 1995; pp. 352–389.
42.
Tay, R. Correlation, variance inflation and multicollinearity in regression model. J. East. Asia Soc. Transp. Stud.
2017
,12, 2006–2015.
43.
Barbet-Massin, M.; Jiguet, F.; Albert, C.H.; Thuiller, W. Selecting pseudo-absences for species distribution models: How, where
and how many? Methods Ecol. Evol. 2012,3, 327–338. [CrossRef]
44.
Liu, C.; Newell, G.; White, M. The effect of sample size on the accuracy of species distribution models: Considering both presences
and pseudo-absences or background sites. Ecography 2019,42, 535–548. [CrossRef]
45. Ali, A.; Shamsuddin, S.M.; Ralescu, A.L. Classification with class imbalance problem. Int. J. Adv. Soft Compu. Appl. 2013,5, 3.
46.
Wisz, M.S.; Hijmans, R.J.; Li, J.; Peterson, A.T.; Graham, C.H.; Guisan, A.; NCEAS Predicting Species Distributions Working
Group. Effects of sample size on the performance of species distribution models. Divers. Distrib. 2008,14, 763–773. [CrossRef]
47.
Chawla, N.V.; Japkowicz, N.; Kotcz, A. Special issue on learning from imbalanced data sets. ACM SIGKDD Explor.
2004
,6, 1–6.
[CrossRef]
48.
Peterson, A.T.; Pape¸s, M.; Soberón, J. Rethinking receiver operating characteristic analysis applications in ecological niche
modeling. Ecol. Model. 2008,213, 63–72. [CrossRef]
49.
Grimmett, L.; Whitsed, R.; Horta, A. Presence-only species distribution models are sensitive to sample prevalence: Evaluating
models using spatial prediction stability and accuracy metrics. Ecol. Model. 2020,431, 109194. [CrossRef]
50.
Qiao, H.; Soberon, J.; Peterson, A.T. No silver bullets in correlative ecological niche modelling: Insights from testing among many
potential algorithms for niche estimation. Methods Ecol. Evol. 2015,6, 1126–1136. [CrossRef]
51.
Riccardi, T.; Malatesta, L.; Van Damme, K.; Suleiman, A.S.; Farcomeni, A.; Rezende, M.; Vahalík, P.; Attorre, F. Environmental
factors and human activity as drivers of tree cover and density on the Island of Socotra, Yemen. Rend. Lincei Sci. Fis. Nat.
2020
,
31, 703–718. [CrossRef]
52. Bobbert, A.C. Energy expenditure in level and grade walking. J. Appl. Physiol. 1960,15, 1015–1021. [CrossRef]
53.
Marcaria, R. Sulla fisiologia e specialmente sul consumo energetico della marcia e della corsa a varie velocita ed inclinazioni del
terreno. Atti Accad. Naz. Lincei Mem. Cl. Sci. Fis. Mat. Nat. Sez 1938,3, 299–368.
54.
Minetti, A.E.; Moia, C.; Roi, G.S.; Susta, D.; Ferretti, G. Energy cost of walking and running at extreme uphill and downhill slopes.
J. Appl. Physiol. 2002,93, 1039–1046. [CrossRef]
Forests 2022,13, 1276 12 of 12
55.
Páez, A.; Anjum, Z.; Dickson-Anderson, S.E.; Schuster-Wallace, C.J.; Ramos, B.M.; Higgins, C.D. Comparing distance, time,
and metabolic energy cost functions for walking accessibility in infrastructure-poor regions. J. Transp. Geogr.
2020
,82, 102564.
[CrossRef]
56.
Al-Okaishi, A. Exploring the historical distribution of Dracaena cinnabari using ethnobotanical knowledge on Socotra Island,
Yemen. J. Ethnobiol. Ethnomed. 2021,17, 22. [CrossRef]
57.
Mies, B.A.; Beyhl, F.E. The vegetation ecology of Soqotra. In Conservation and Sustainable Use of Biodiversity of Soqotra Archipelago,
Proceedings of the First International Symposium on Soqotra Island: Present and Future, Aden, Yemen, 24–28 March 1996; Dumont, H.,
Ed.; UNEP/GEF Publications: New York, NY, USA, 1998; Technical Series; Volume 1, pp. 35–81.
58.
Vasconcelos, R.; Pujol-Buxó, E.; Llorente, G.A.; Saeed, A.; Carranza, S. Micro-Hotspots for Conservation: An Umbrella Tree
Species for the Unique Socotran Reptile Fauna. Forests 2020,11, 353. [CrossRef]
59.
Hu, W.; Wang, Y.; Dong, P.; Zhang, D.; Yu, W.; Ma, Z.; Chen, G.; Liu, Z.; Du, J.; Chen, B.; et al. Predicting potential mangrove
distributions at the global northern distribution margin using an ecological niche model: Determining conservation and
reforestation involvement. For. Ecol. Manag. 2020,478, 118517. [CrossRef]
60.
Jensen, D.A.; Rao, M.; Zhang, J.; Grøn, M.; Tian, S.; Ma, K.; Svenning, J.-C. The potential for using rare, native species in
reforestation–A case study of yews (Taxaceae) in China. For. Ecol. Manag. 2021,482, 118816. [CrossRef]
61.
López-Tirado, J.; Vessella, F.; Schirone, B.; Hidalgo, P.J. Trends in evergreen oak suitability from assembled species distribution
models: Assessing climate change in south-western Europe. New For. 2018,49, 471–487. [CrossRef]
62.
Mateo, R.G.; Gastón, A.; Aroca-Fernández, M.J.; Broennimann, O.; Guisan, A.; Saura, S.; García-Viñas, J.I. Hierarchical species
distribution models in support of vegetation conservation at the landscape scale. J. Veg. Sci. 2019,30, 386–396. [CrossRef]
63.
Gornish, E.; Arnold, H.; Fehmi, J. Review of seed pelletizing strategies for arid land restoration. Restor. Ecol.
2019
,27, 1206–1211.
[CrossRef]
64.
Moore, M.-L.; Riddell, D.; Vocisano, D. Scaling Out, Scaling Up, Scaling Deep: Strategies of Non-profits in Advancing Systemic
Social Innovation. J. Corp. Citizsh. 2015,1, 67–84. [CrossRef]
65.
Campos-Silva, J.V.; Hawes, J.E.; Freitas, C.T.; Andrade, P.C.M.; Peres, C.A. Community-Based Management of Amazonian
Biodiversity Assets. In Participatory Biodiversity Conservation: Concepts, Experiences, and Perspectives; Springer: Cham, Switzerland,
2020; pp. 99–111. [CrossRef]
66.
Jarvis, A.; Reuter, H.I.; Nelson, A.; Guevara, E. Hole-Filled SRTM for the Globe Version 4: Data Grid. CGIAR Consortium for
Spatial Information. 2008. Available online: http://srtm.csi.cgiar.org (accessed on 10 May 2022).
67.
Google Maps. Available online: https://www.google.com/maps/place/12%C2%B023\T1\textquoteright31.4%22N+53%C2%B0
59\T1\textquoteright56.1%22E/@12.3917117,53.9986001,1078m/data=!3m1!1e3!4m6!3m5!1s0x0:0x4564d658ba0e9d14!7e2!8m2
!3d12.3920582!4d53.9989207 (accessed on 27 May 2021).
