Modeling the risk of illegal forest activity and its distribution in the southeastern region of the Sierra Madre Mountain Range, Philippines ...

Page created by Gregory Maxwell
 
CONTINUE READING
Modeling the risk of illegal forest activity and its distribution in the southeastern region of the Sierra Madre Mountain Range, Philippines ...
iForest                                                                                                      Research Article
                                                                                                           doi: 10.3832/ifor3937-014
                                                                                                                    vol. 15, pp. 63-70
          Biogeosciences and Forestry

Modeling the risk of illegal forest activity and its distribution in the
southeastern region of the Sierra Madre Mountain Range, Philippines

Jhun B Barit (1-2),                                Illegal activity within protected forests, such as illegal logging, slash-and-burn
                                                   farming, and agricultural expansion, has brought many plant and animal spe-
Kwanghun Choi (2),                                 cies to the brink of extinction and threatens various conservation efforts. The
Dongwook W Ko (2)                                  Philippine government has introduced a number of actions to combat environ-
                                                   mental degradation, including the use of mobile platforms such as the SMART-
                                                   Lawin system to collect patrol data from the field, which represents a remark-
                                                   able step towards data-driven conservation management. However, it remains
                                                   difficult to control illegal forest activity within protected landscapes due to
                                                   limited patrol and law enforcement resources. A better understanding of the
                                                   spatial distribution of illegal activity is crucial to strengthening and efficiently
                                                   implementing forest protection practices. In the present study, we predicted
                                                   the spatial distribution of illegal activity and identified the associated environ-
                                                   mental factors using maximum entropy modeling (MaxEnt). Patrol data collect-
                                                   ed using the SMART-Lawin system from the Baliuag Conservation Area for the
                                                   period 2017-2019 were used to train and validate the MaxEnt models. We
                                                   tuned the MaxEnt parameter setting using the ENMeval package in R to over-
                                                   come sampling bias, avoid overfitting, and balance model complexity. The re-
                                                   sulting MaxEnt models provided a clear understanding of the overall risk of il-
                                                   legal activity and its spatial distribution within the conservation area. This
                                                   study demonstrated the potential utility of data-driven models developed from
                                                   patrol observation records. The output of this research is beneficial for con-
                                                   servation managers who are required to allocate limited resources and make
                                                   informed management decisions.

                                                   Keywords: Philippines, SMART, Ranger Patrol Data, Illegal Forest Activity, Pro-
                                                   tected Area Management

Introduction                                       diversity in protected areas in the Philip-      east Asia and the seventh-fastest globally
  Due to their diverse landscapes, the for-        pines (Carandang et al. 2013), while the         during this period (Bisson et al. 2003). In
ests in the Philippines are considered one         growing demand for forest resources by lo-       the Philippines, slash-and-burn farming has
of the most significant global biodiversity        cal communities makes it more challenging        converted primary or secondary forest to
hotspots and an important conservation             to manage these areas and their natural re-      agricultural land (Kummer 1992), while in-
target (Lasco et al. 2008). However, Philip-       sources (Jones et al. 2019).                     discriminate illegal mining operations have
pine forests and their biodiversity have             Earlier studies have emphasized that the       also seriously damaged several forest areas
been degraded at an alarming rate due to a         threats to biodiversity are directly associ-     (Finger et al. 1999). Similarly, charcoal pro-
combination of illegal anthropogenic for-          ated with the excessive utilization of re-       duction is a significant threat in the Philip-
est activity, such as illegal logging, slash-      sources, developmental activity, and the in-     pines because fuelwood-dependent indus-
and-burn farming, mining, and charcoal             creasing population in the Philippines (Van      tries rely on illegally gathered charcoal
production, and biophysical phenomena,             Der Ploeg et al. 2003). From 2000 to 2005,       from natural or plantation forests to pro-
such as typhoons, floods, and landslides           the Philippines lost about 0.1 million ha of     duce goods for commercial and domestic
(Rubas-Leal et al. 2017). Illegal activity asso-   its forest cover (2.1% of the overall forested   use (Remedio 2009). A number of strate-
ciated with the extraction of natural re-          area) annually due to illegal logging, the       gies are thus required to prevent illegal
sources has reduced forest cover and bio-          second-fastest deforestation rate in South-      activity in protected areas, including
                                                                                                    strengthening law enforcement monitor-
                                                                                                    ing and patrolling, support for alternative
  (1) Department of Environment and Natural Resources (DENR), Philippines; (2) Depart-              industries in the local community, the re-
ment of Forest Environment and Systems, Kookmin University, South Korea                             duction of forest communities living in
                                                                                                    close proximity to the forests, and the
@ Dongwook W Ko (dwko@kookmin.ac.kr)                                                                strict implementation of laws and reg-
                                                                                                    ulations related to forest conservation
Received: Jul 28, 2021 - Accepted: Dec 16, 2021                                                     (Plumptre et al. 2014, Critchlow et al. 2015).
                                                                                                      Various policies and programs have been
Citation: Barit JB, Choi K, Ko DW (2022). Modeling the risk of illegal forest activity and its      introduced to reduce illegal activity in Phil-
distribution in the southeastern region of the Sierra Madre Mountain Range, Philippines.            ippine forests, but these have been under-
iForest 15: 63-70. – doi: 10.3832/ifor3937-014 [online 2022-02-21]                                  mined by inadequate law enforcement and
                                                                                                    institutionalization (e.g., law enforcement
Communicated by: Maurizio Marchi                                                                    monitoring or ground patrolling – Domingo
                                                                                                    & Manejar 2018, Carandang et al. 2013). Re-

© SISEF https://iforest.sisef.org/                                      63                                                    iForest 15: 63-70
Modeling the risk of illegal forest activity and its distribution in the southeastern region of the Sierra Madre Mountain Range, Philippines ...
Barit JB et al. - iForest 15: 63-70

                                        cently, the Philippine government has            vided a comprehensive spatial analysis of a        data generated using the SMART-Lawin
iForest – Biogeosciences and Forestry

                                        sought to protect the remaining forest           wide range of illegal forest activity, instead     system can be used to identify the risk and
                                        cover through ground patrolling and law          focusing only on the accurate assessment           potential locations of illegal activity within
                                        enforcement activity using the Spatial           of the dynamics of a specific illegal activity     a landscape and analyze the spatial rela-
                                        Monitoring and Reporting Tool (SMART).           (Critchlow et al. 2015).                           tionship between this distribution and vari-
                                        The Department of Environment and Natu-            Conservation area managers require in-           ous environmental and infrastructure vari-
                                        ral Resources (DENR), in partnership with        formation on the potential risk and spatial        ables, such as the elevation, slope, wet-
                                        the United States Agency for International       distribution of illegal forest activity within a   ness, land cover, and proximity to rivers,
                                        Development (USAID), has developed the           protected area in order to devise optimal          roads, and villages (Keane et al. 2011, Plum-
                                        SMART-Lawin Forest and Biodiversity Pro-         management strategies with limited re-             ptre et al. 2014, Critchlow et al. 2015, 2017,
                                        tection System (hereafter, SMART-Lawin           sources. Several studies have attempted to         Denninger Snyder et al. 2019).
                                        system). Through ground patrolling, this         predict the spatial distribution of illegal ac-       In this study, we used MaxEnt models to
                                        system efficiently gathers information on        tivity using Bayesian spatially explicit occu-     predict the spatial patterns of illegal forest
                                        forest conditions, threats to the environ-       pancy models (Critchlow et al. 2015) and           activity and identify environmental and hu-
                                        ment (i.e., illegal activity), and indicator     maximum entropy (MaxEnt) models (Plum-             man drivers that are likely to influence this
                                        species of forest health using cutting-edge      ptre et al. 2014, Denninger Snyder et al.          activity in the SMMR. An advantage of the
                                        technology. The SMART-Lawin system can           2019). The MaxEnt approach has been em-            MaxEnt approach is that it only requires
                                        be employed to track illegal forest activity     ployed in many studies to examine spatial          presence data for an item of interest, while
                                        and locate their hotspots of illegal activity    patterns of the risk of illegal forest activity    other approaches require both presence
                                        within protected areas, thus allowing time-      within various national parks (Plumptre et         and absence data to identify spatial pat-
                                        ly, informed decisions to be made based on       al. 2014, Denninger Snyder et al. 2019).           terns. MaxEnt models substitute absence
                                        the rapid transmission of data. The primary      These studies have found that the lack of          data with randomly generated background
                                        purpose of this system is to establish ap-       manpower, logistic and financial resources,        data, which is particularly useful for pre-
                                        propriate strategies that maximize patrol        and extensive patrol coverage hinder the           dicting the spatial distribution of a species
                                        effectiveness in deterring illegal forest ac-    effective patrolling/monitoring of forest re-      due to the difficulties associated with
                                        tivity.                                          sources. The accurate assessment and               obtaining absence data (Elith & Leathwick
                                          Previous research on illegal forest activity   modeling of the spatial distribution of the        2009, Torres et al. 2016, Muscarella et al.
                                        has mainly focused on a single specific ac-      potential risk of illegal activity is thus nec-    2017, Zhang et al. 2018, Tumaneng et al.
                                        tivity, for instance, illegal logging (Van Der   essary to strengthen forest protection and         2019). We then used the results to recom-
                                        Ploeg et al. 2011), the illegal hunting and      effectively and efficiently manage individ-        mend improvements in the patrol strategy
                                        gathering of wildlife resources (Minter et       ual conservation areas.                            for the region.
                                        al. 2014), and charcoal production (Geron-         In the Sierra Madre Mountain Range                  In summary, the main objectives of the
                                        imo et al. 2016). These studies have been        (SMMR), ranger patrol activity has gener-          present study were: (i) to develop MaxEnt
                                        useful in that they provide relevant infor-      ated large volumes of data in the SMART-           models for illegal forest activity within the
                                        mation on the extent and intensity of an il-     Lawin system. However, few attempts                SMMR utilizing ranger patrol data collect-
                                        legal activity within the conservation area      have been made to utilize this crucial data        ed via the SMART-Lawin system to under-
                                        and how it has changed over time. How-           to better allocate resources for efficient         stand the spatial patterns of this activity;
                                        ever, most of these studies have not pro-        and strategic patrolling. The ranger patrol        (ii) to identify significant environmental
                                                                                                                                            variables that determine the distribution of
                                                                                                                                            illegal forest activity; and (iii) to assess the
                                         Fig. 1 - Location                                                                                  risk of illegal forest activity in this region
                                              of and land                                                                                   and thus determine the patrol coverage
                                            cover within                                                                                    that is required and improve the general
                                              the Baliuag                                                                                   patrol strategy for the conservation area.
                                           Conservation
                                                     Area.                                                                                  Materials and methods
                                                                                                                                            Study area
                                                                                                                                              The SMMR is the longest mountain range
                                                                                                                                            in the Philippines and is surrounded by nine
                                                                                                                                            provinces: Isabela, Nueva Vizcaya, Cagay-
                                                                                                                                            an, Quirino, Nueva Ecija, Aurora, Bulacan,
                                                                                                                                            Rizal, and Quezon (Van Der Ploeg et al.
                                                                                                                                            2003). It is home to 240 bird species includ-
                                                                                                                                            ing the Philippine eagle (Pithecophaga jef-
                                                                                                                                            feryi) and the endemic Isabela oriole (Ori-
                                                                                                                                            olus isabellae – Mallari & Jensen 1993). A
                                                                                                                                            considerable number of commercially sig-
                                                                                                                                            nificant but severely threatened tree spe-
                                                                                                                                            cies are also found in the conservation
                                                                                                                                            area, including members of the dipterocarp
                                                                                                                                            family such as Shorea spp. and Hopea spp.
                                                                                                                                            (Torres et al. 2016). This region is domi-
                                                                                                                                            nated by the Dumagat ethnic group (62%),
                                                                                                                                            and all of the residents in the area are de-
                                                                                                                                            pendent on natural resources for their live-
                                                                                                                                            lihoods (DENR 2015).
                                                                                                                                              We selected the Baliuag Conservation
                                                                                                                                            Area (BCA) in Bulacan, which is in the
                                                                                                                                            southeastern region of the SMMR, as our
                                                                                                                                            study area (Fig. 1). The total coverage of

                                        64                                                                                                                              iForest 15: 63-70
Modeling the risk of illegal forest activity and its distribution in the southeastern region of the Sierra Madre Mountain Range, Philippines ...
Modeling the risk of illegal forest activity in SE Philippines

the BCA is approximately 90,448 ha, which

                                                                                                                                                     iForest – Biogeosciences and Forestry
                                                 Tab. 1 - Summary of the environmental variables used in this study. (1): ASTER-Global
includes closed forest (34,849 ha), open
                                                 Digital Elevation Model (GDEM) 30 m of NASA; (2): derived from the DEM; (3): http://
forest (18,531 ha), and non-forest (37,069
                                                 www.geofabrik.de; (4): National Mapping Resources Information Authority / Forest
ha) areas (DENR-FMB 2018). The BCA com-
                                                 Management Bureau.
prises three protected areas, the Angat
Watershed Forest Reserve, Biak-na-Bato
National Park, and Doña Remedios Water-            Variable     Variable Unit              Purpose                                       Source
shed, which are designated for protection         Elevation     Meters above sea level    Accessibility: high elevations can be           (1)
and conservation (DENR-PAWB 2009, DEN-                                                    difficult to access and detect illegal
R-FMB 2018). The entire BCA is a vital                                                    activity
source of fresh water for irrigation and in-      Slope         Slope of the terrain,     Accessibility: terrain with steep slopes        (2)
dustrial purposes (e.g., hydroelectric pow-                     percent                   can be difficult to access
er generation) and serves as a critical eco-
                                                  Road          Distance in meters to     Accessibility: roads facilitate access to a     (3)
logical corridor within the SMMR.                               the nearest road          conservation area, leading to an increase
                                                                (Euclidean distance)      in detection risk
Data collection
                                                  River         Distance in meters to     Accessibility: riverbanks or streambeds         (3)
   Currently, the BCA is managed by 23 for-
                                                                the nearest river         areas provide access for illegal activity
est rangers registered in the SMART-Lawin                       (Euclidean distance)      and denser vegetation suitable for illegal
system, though only 18 members regularly                                                  settlements
join the patrols. These forest rangers are
grouped into four teams, each in charge of        Settlement    Distance in meters to     Accessibility: most accessible areas if the     (3)
                                                                the nearest settlement    (permanent/temporary)
patrolling one of the four patrol sectors                       (Euclidean distance)      settlements are in close proximity to the
over an average distance of 6 km. Each                                                    conservation area
team conducts three patrols a month on
average (8 h per patrol). With the aid of         Land cover    Generalized categories Availability of resources: areas of intact         (4)
                                                                for Philippine land cover forest cover provide more available
the SMART-Lawin system, the forest
                                                                classification (i.e closed resources and suitable habitat for wildlife
rangers recorded 3554 observations (ap-                         forest, open forest, etc.)
proximately 1184 per year) within the BCA
over the three years from 2017 to 2019.           Topographic Index of soil moisture      Availability of resources: high wetness         (2)
                                                  wetness                                 and close proximity to water are
   We obtained data for the occurrence of il-
                                                                                          associated with a high animal density
legal activity over the entire BCA from the
SMART-Lawin system for the period of Jan-
uary 2017 to December 2019, which were
collected by the forest rangers during their    variance inflation factor (VIF) was used to      the corrected delta-Akaike information cri-
patrols. The spatial data were recorded in      test the multicollinearity of the predictors;    terion (ΔAICc), with priority given to those
the SMART-Lawin database using Cyber-           any predictor with a VIF greater than 5 was      with an ΔAICc < 2 as the best model for
Tracker, an open-source mobile application      removed from further analysis (Hair et al.       each illegal activity category. We then used
(USAID/B+WISER 2018).                           2014).                                           the area under the receiver operating
   We converted the occurrence data for il-                                                      curve (AUCTEST) to select the most appropri-
legal forest activity into spatial point data   Model tuning and processing                      ate model from among the competitive al-
and then categorized these points into            We optimized the models used to identify       ternatives (Denninger Snyder et al. 2019).
three categories: agricultural expansion, in-   the spatial distribution of illegal forest ac-     Applying the optimized FC and RM, we
frastructure expansion, and forest product      tivity using ENMeval. Though MaxEnt mod-         built MaxEnt models to predict the spatial
extraction. We removed unwanted pres-           els are known to be a useful and robust ap-      distribution patterns of the illegal activity
ence records that could influence the prob-     proach to spatial prediction (Elith et al.       categories. Based on the results, we gener-
ability distribution of the models based on     2006, Phillips et al. 2006), overfitting, com-   ated predictive maps and analyzed the in-
three criteria: (i) the point location was      plexity, and bias are common issues, and         fluence of environmental covariates, ob-
outside the study area; (ii) the records        these can be effectively addressed in ENM-       tained evaluation metrics, and evaluated
were duplicated (i.e., same date and coor-      eval (Radosavljevic & Anderson 2014,             the contribution of each predictor variable
dinates); and (iii) thinning of the data was    Merow et al. 2013). We investigated the op-      to the final model (Phillips et al. 2004).
possible (i.e., only retaining one occurrence   timal parameter settings, particularly the
point per grid pixel). The final dataset con-   regularization multiplier (RM) and feature       Model evaluation
sisted of 259 occurrence records, with 65,      classes (FCs), to overcome these issues.           We used k-fold cross-validation to evalu-
147, and 47 occurrences of agricultural ex-     For MaxEnt modeling and validation, we           ate the model by partitioning the occur-
pansion, infrastructure expansion, and for-     used 80% of the presence data as training        rence data into training and testing sets.
est product extraction, respectively. These     data and the remaining 20% for testing to        MaxEnt randomly partitions training and
were used to model the distribution of the      determine the distribution of illegal activity   testing sets based on user-defined k values,
illegal activity categories.                    and the accuracy of each model.                  which can inflate model performance met-
                                                  Initially, we identified the optimal combi-    rics due to spatial autocorrelation (Rado-
Environmental predictors                        nation of FCs and the RM for the three ille-     savljevic & Anderson 2014). To avoid over-
  Seven environmental variables at a 30 ×       gal forest activity categories using ENM-        fitting, we partitioned illegal activity occur-
30 m resolution (elevation, slope, wetness,     eval, which generates a series of MaxEnt         rence and background points into four sets
land cover, and distance from roads, rivers,    model candidates for particular parameter        (coarse-grained and fine-grained) using the
and settlements – Tab. 1) were used as po-      settings. There were six FCs (L, LQ, H, LQH,     “checkerboard2” partitioning scheme
tential predictors of illegal forest activity   LQHP, and LQHPT), where L, Q, H, P, and T        (Denninger Snyder et al. 2019, Muscarella
(Plumptre et al. 2014, Critchlow et al. 2015,   represent linear, quadratic, hinge, product,     et al. 2014). The AUCTEST (Elith et al. 2011)
2017, Denninger Snyder et al. 2019). All spa-   and threshold features, respectively, and        and true skill statistic (TSS – Allouche et al.
tial data were processed for input into the     we tested RM values of 0.5, 1, 1.5, 2, 2.5, 3,   2006) were used to assess the perfor-
Ecological Niche Model Evaluation (ENM-         3.5, and 4. The parameters used for the          mance of the final models for each illegal
eval) v. 0.3.0 package (Muscarella et al.       MaxEnt models are summarized in Tab. 2.          activity category.
2014) in R v. 3.5.1 (R Core Team 2020). The     The optimal models were selected using             The AUC is a strong independent thresh-

iForest 15: 63-70                                                                                                                               65
Barit JB et al. - iForest 15: 63-70

                                                                                                                                       Spatial analysis
iForest – Biogeosciences and Forestry

                                         Tab. 2 - List of significant MaxEnt parameters used to fine-tune the models for each             We analyzed the predictive model for the
                                         illegal activity category. (*): FC and RM parameters are the output of the ENMeval            spatial distribution of each illegal activity
                                         tuning process.                                                                               by: (i) assessing the spatial extent of each
                                                                                                                                       illegal activity by its coverage; and (ii) esti-
                                             MaxEnt                                       MaxEnt    Tuned Model                        mating the overall risk of illegal activity
                                                                     Selection
                                             Parameter                                    Default   Parameters
                                                                                                                                       across the landscape. First, we assessed
                                          Feature class (FC)        Linear (L)            Auto:     *Agricultural Expansion = LQHPT    the potential spatial extent of each illegal
                                                                    Quadratic (Q)         LQH       *Infrastructure Expansion = LQHP   activity by reclassifying the MaxEnt output
                                                                    Product (P)                     *Forest Product Extraction = LQ    of each pixel into likely and unlikely areas
                                                                    Threshold (T)
                                                                                                                                       for the illegal activity using the 10 th per-
                                                                    Hinge (H)
                                                                    Auto                                                               centile logistic threshold. Second, we as-
                                                                                                                                       sessed the overall risk of illegal activity
                                          Regularization            Number >=0.5 in       1.0       *Agricultural Expansion = 2.5      across the landscape by summing the num-
                                          multiplier (RM)           steps of 0.5                    *Infrastructure expansion = 1
                                                                                                                                       ber of illegal activity occurrences exceed-
                                                                                                    *Forest Product Extraction = 1.5
                                                                                                                                       ing the threshold for each pixel, as deter-
                                          Maximum number of         Number                10,000    20,000                             mined above; we represented the likeli-
                                          background points                                                                            hood of any illegal activity taking place at a
                                          Maximum iterations        Number                500       1,000                              particular location with four risk levels
                                          Output format             Loglog, Logistic,     Cloglog   Raw and Logistic                   from 0 (none risk) to 3 (high risk).
                                                                    Cumulative, Raw
                                                                                                                                       Results

                                         Tab. 3 - MaxEnt parameter settings used in the study and the evaluation metric re-            Model selection and performance
                                         sults. (§): selected optimal model with ΔAICc < 2 for each illegal forest activity. (- -):    evaluation
                                         values were not computed because it was not considered an optimal model; (Th*):              We developed 144 MaxEnt models (48 for
                                         the logistic threshold value for the 10th percentile minimum training presence.            each illegal activity category) for various
                                                                                                                                    combinations of the RM and FC (Tab. S1 in
                                          Illegal Activity                                          Avg
                                                                                                                                    Supplmentary material). We obtained opti-
                                                                n       FC          RM     ΔAICc             AUCTRAIN TSS     Th*   mal parameter sets for each illegal activity
                                          Category                                                 AUCTEST
                                                                                            §
                                                                                                                                    according to the model selection criteria
                                          Agricultural          65      LQHPT       2.5   0        0.8880 0.9158 0.6130 0.083
                                                                                                                                    (Tab. 3). We employed RM values of 2.5, 1,
                                          expansion
                                                                                                                                    and 1.5, and the FC classes LQHPT, LQHP,
                                          Infrastructure       147      LQHP        1     0§       0.8055 0.8594 0.4420 0.191       and LQ for agricultural expansion, infra-
                                          expansion                                                                                 structure expansion, and forest product
                                          Forest product        47      LQH         2.5   0        0.6790       --      --     --   extraction, respectively. The models for all
                                          extraction                    LQH         2     0.3589 0.6788         --      --     --   categories of illegal activity with optimal
                                                                                                 §
                                                                                                                                    parameter settings exhibited acceptable
                                                                        LQ          1.5   0.5948 0.6973 0.7259 0.2551 0.214
                                                                                                                                    model performance (Denninger Snyder et
                                                                        LQ          2     1.1455 0.6916         --      --     --   al. 2019). The agricultural and infrastruc-
                                                                        LQ          2.5   1.8373 0.6849         --      --     --   ture expansion models exhibited good per-
                                                                                                                                    formance (mean AUCTEST = 0.88 and 0.80,
                                                                                                                                    and TSS = 0.613 and 0.442, respectively),
                                        old value that evaluates the ability of a         forms better than random chance, and a while that of the product extraction model
                                        model to discriminate presence from ab-           high TSS represents the degree of agree- was relatively poor (mean AUC TEST = 0.69
                                        sence points (i.e., background points). The       ment between the samples generated by and TSS = 0.26).
                                        resulting average AUC for each model was          the MaxEnt model and the background
                                        compared and ranked. The TSS was used             prediction model. Therefore, models with Analysis of environmental predictors
                                        as an additional metric because it considers      both a high AUC and a high TSS would have   The environmental predictors differed in
                                        both omission and commission errors. A            greater reliability and robustness.       their impact on each illegal forest activity
                                        high AUC indicates that the model per-                                                      model, with land cover and proximity to
                                                                                                                                    roads and rivers having the strongest influ-
                                                                                                                                    ence (Fig. 2). Agricultural expansion was
                                                                                                                                    strongly affected by elevation (36%), fol-
                                                                                                                                    lowed by land cover (32%) and roads (19%).
                                                                                                                                    Infrastructure expansion was primarily af-
                                                                                                                                    fected by land cover (34%), roads (27%), and
                                                                                                                                    rivers (12%). Similarly, product extraction
                                                                                                                                    was substantially influenced by land cover
                                                                                                                                    (58%), rivers (18%), and roads (11%).

                                                                                                                                       Potential distribution of illegal forest
                                                                                                                                       activity
                                                                                                                                          The predicted spatial distribution for each
                                                                                                                                       illegal activity category varied across the
                                                                                                                                       landscape (Fig. 3). The threshold values for
                                                                                                                                       the presence and absence of agricultural
                                                                                                                                       expansion, infrastructure expansion, and
                                                                                                                                       forest product extraction were 0.083,
                                                                                                                                       0.191, and 0.214, respectively. Forest prod-
                                         Fig. 2 - Importance of each predictor variable to the illegal forest activity models.         uct extraction was the most common ille-

                                        66                                                                                                                         iForest 15: 63-70
Modeling the risk of illegal forest activity in SE Philippines

gal activity across the landscape (66%), fol-

                                                                                                                                                     iForest – Biogeosciences and Forestry
                                                                                                                            Fig. 3 - Spatial cov-
lowed by infrastructure expansion (44%)
                                                                                                                            erage of the
and agricultural expansion (30% – Tab. 4).
                                                                                                                            potential distribu-
Of the land cover types, brush/shrub was
                                                                                                                            tion of each illegal
most likely to be damaged by illegal activity
                                                                                                                            forest activity cat-
(59,007 ha), followed by open forest
                                                                                                                            egory (a-c) and the
(28,073 ha) and grasslands (16,405 ha –
                                                                                                                            level of risk pre-
Tab. 5). The predicted model revealed that
                                                                                                                            dicted across the
product extraction was most detrimental
                                                                                                                            study area (d).
to open and closed forest (25,463 ha), fol-
lowed by infrastructure expansion (13,901
ha) and agricultural expansion (173 ha).
  The overall risk assessment, which repre-
sents the total frequency of all illegal activ-
ity occurrences, revealed that 25% of the
conservation area was at high risk, 20% at
moderate risk, 25% at low risk, and 30% at
no risk (Fig. 3d).

Discussion
   This study demonstrated a method for
predicting the spatial patterns of illegal ac-
tivity in Philippine forests utilizing MaxEnt
models. We classified illegal activity into
three categories: agricultural expansion, in-
frastructure expansion, and forest product
extraction. Predictions of the distribution
of these illegal activity categories were car-
ried out using MaxEnt models based on
seven environmental predictors (i.e., eleva-
tion, slope, roads, rivers, settlements, land
cover, and topographic wetness) and 259
occurrences of illegal activity obtained
from the SMART-Lawin patrolling system.
The optimal MaxEnt models for agricul-
tural expansion and infrastructure expan-
sion produced a high prediction perfor-
mance, with an average AUCTEST of over
0.80, higher than the model for forest              According to our results, each illegal ac-      reported a very low relationship between
product extraction (AUCTEST = 0.697). We          tivity was affected by different environ-         land cover and forest product extraction,
found that the distribution and potential         mental variables. Agricultural and infra-         our results revealed that land cover had a
occurrence of illegal activity were also in-      structure expansion demonstrated similar          very large influence on the product extrac-
fluenced by the sample size and their spa-        patterns in terms of the main environmen-         tion model. This contrast in the results
tial extent of distribution. A higher sample      tal variables affecting the models. They          could be due to a difference in the land
size tends to increase model performance,         were evenly affected by land cover and            cover types in the respective study areas.
as represented by the AUC, but the pre-           roads and slightly affected by the proximity      Denninger Snyder et al. (2019) conducted
dicted coverage of each illegal activity had      of settlement areas, indicating the gradual       their study in the African savanna, where
the opposite effect on the AUC. Because           expansion of both types of illegal activity.      grassland is the dominant land cover type.
product extraction had fewer observed oc-         On the other hand, forest product extrac-         On the other hand, our study was carried
currences and its predicted coverage area         tion was mainly affected by land cover and        out in a region mainly covered by forests.
was almost twice as large as the two other        the proximity of roads and rivers, which          Because trees are evenly and sparsely dis-
illegal activities, its model had a lower AUC.    can be used to transport forest products.         tributed in savanna regions, the land cover
Further monitoring of illegal activity in for-      Although Denninger Snyder et al. (2019)         is not greatly impacted by tree logging; in-
ested areas using the SMART-Lawin system
could help to improve the accuracy of the
model.                                             Tab. 5 - Coverage of illegal activity categories predicted for each land cover type
                                                   within the conservation area. (1): Annual crop; (2): Brush/shrubs; (3): Built-up; (4):
                                                   Closed forest; (5): Grassland; (6): Inland water; (7): Open forest; (8): Open barren; (9):
 Tab. 4 - Coverage of each illegal activity        Perennial crop.
 category predicted across the conserva-
 tion area.
                                                    Illegal Activity                        Land Classification                        Coverage
                                                    Class                1         2    3       4        5        6    7     8     9     (ha)
  Illegal Activity        Area    Coverage
  Category                (ha)      (%)             Agricultural        976    19999   99      35    3734    2062     138    79   34    27256
  Forest product         59.913       66            expansion
  extraction                                        Infrastructure     1166    18247   03    3405    3978    2005 10496     159   63    39722
                                                    expansion
  Infrastructure         39.721       44
  expansion                                         Forest product     1790    20761   15    8024    8693    2214 17439     471   07    59914
  Agricultural           27.256       30            extraction
  expansion                                         Total (ha)         3932    59007 317 11464 16405         6281 28073     709 704    126892

iForest 15: 63-70                                                                                                                               67
Barit JB et al. - iForest 15: 63-70

                                        stead, proximity to villages and manage-          right predictors is critical when seeking to      activity. The present research demon-
iForest – Biogeosciences and Forestry

                                        ment strategies have a greater effect on il-      produce accurate and reliable models. In          strates the use of a data-driven response
                                        legal activity. However, in the tropical for-     this study, we limited our focus to seven         to facilitate a better understanding of an
                                        ests in our study area, protected areas usu-      important variables that are likely to affect     entire conservation area and to provide ac-
                                        ally conserve forests. Therefore, illegal ac-     the occurrence of illegal activity. We found      curate data that can be used to make in-
                                        tivity tends to occur at locations where it is    that land cover was a crucial environmen-         formed conservation management deci-
                                        difficult to detect but where it is easy to       tal factor that affected the spatial distribu-    sions. Applying MaxEnt models that were
                                        transport the products quickly. Therefore,        tion of all three illegal activity categories.    based on illegal activity occurrence data
                                        forest product extraction tends to be af-         Elevation and proximity to roads and rivers       collected from the field using the SMART-
                                        fected by the land cover and proximity to         were also very important to the accuracy          Lawin patrolling system, we successfully
                                        roads and rivers.                                 of the model predictions. However, conser-        identified the critical environmental drivers
                                                                                          vation managers are increasingly recogniz-        determining illegal activity within the study
                                        Applicability of the model                        ing the dynamic nature of illegal activity,       site. However, this does not mean that the
                                           Although the MaxEnt model is generally         especially in response to patrolling or law       seven environmental variables that we in-
                                        used to estimate the spatial distribution of      enforcement, which is extremely difficult         vestigated are best suited for all categories
                                        a species, it has been utilized in many stud-     to include in the MaxEnt framework. On            of illegal activity. Future research should in-
                                        ies to estimate the spatial patterns of non-      the other hand, there have been calls to fo-      clude significant environmental predictors
                                        biological items, such as illegal forest activ-   cus patrolling efforts on areas with greater      that were not used in this study but that
                                        ity, because non-biological items are also        conservation value (Plumptre et al. 2014),        are likely to affect the occurrence of illegal
                                        determined by environmental factors (Liao         so that such illegal activity would eventu-       activity. An improved version of the ap-
                                        et al. 2017, Denninger Snyder et al. 2019).       ally shift to areas with less conservation        proach showcased in this study could be
                                        Therefore, this method is widely used for         value.                                            implemented in other priority conservation
                                        developing predictive models to under-                                                              areas with wider coverage using a large
                                        stand the potential distribution of illegal       Improving the patrol strategy                     sample size of illegal forest activity gener-
                                        forest activity to assist in the efforts to          Our study area in the BCA is 90,448 ha in      ated over a longer time period, allowing
                                        manage protected areas (Denninger Sny-            size, which is covered by four patrol teams.      for more effective patrol strategies. How-
                                        der et al. 2019). However, it is difficult to     This means that each team is responsible          ever, it is important to note that the behav-
                                        minimize the effects of spatial autocorrela-      for patrolling over 20,000 ha. Given the lim-     ior of poachers or violators is likely to
                                        tion, particularly when the occurrences of        ited human and logistic resources, it is al-      change in response to changes in patrol
                                        illegal activity are clustered. Therefore, this   most impossible to cover the entire area in       strategies (Keane et al. 2011, Critchlow et
                                        approach may work well in large conserva-         a systematic manner. Unfortunately, this          al. 2017). Our study has also made it possi-
                                        tion areas, especially when patrol efforts        lack of conservation resources is not un-         ble to predict locations with a high poten-
                                        are systematically distributed over the en-       common for most protected areas in the            tial for illegal activity, which is helpful for
                                        tire region. In our study, identifying areas      tropics. The large patrol coverage area and       improving the patrol strategy within pro-
                                        that lack patrol coverage and variables that      the limited budget and human resources,           tected areas.
                                        are poorly represented and including these        hinder the effective implementation of law
                                        in future patrols will be critical to improv-     enforcement strategies (Plumptre et al.           Acknowledgements
                                        ing the modeling approach.                        2014, Denninger Snyder et al. 2019).                This research was supported by the Na-
                                           The modeling approach presented here              Under these circumstances, we believe          tional Research Foundation of Korea (NRF)
                                        also needs to include other relevant predic-      our results can be used to improve pa-            grant funded by the Korea government
                                        tor variables such as (i) human and live-         trolling by shifting focus to three specific      (MSIT, no. NRF-2017R1A2B4010460), and
                                        stock density, (ii) employment and income         goals. The first should be to focus efforts       R&D Program for Forest Science Technol-
                                        levels, which have been linked to illegal re-     on a specific illegal forest activity. In this    ogy (Project No. 2019150B10-21230301)
                                        source use (Critchlow et al. 2015), and (iii)     case, the output map for the extent of the        funded by the Korea Forest Service (Korea
                                        distance to ranger camps and patrol posts         illegal activity can be used to identify the      Forestry Promotion Institute).
                                        (Denninger Snyder et al. 2019). However,          target areas for a particular illegal activity.
                                        any rapid changes, such as the shift in the       The second goal should be detecting as            List of Abbreviations
                                        location of illegal activity hotspots occur-      many types of illegal activity as possible at       The following abbeviations have been
                                        ring in response to patrolling, land-use          once. In this case, the moderate- to high-        used along the paper:
                                        changes, or other socio-economic reasons,         risk areas identified in the risk assessment      • AICc: Corrected Akaike Information Crite-
                                        may be difficult to include in the model, es-     map can provide the best information. The           rion;
                                        pecially if the patrols are ground-based. A       third goal should be to focus on conserving       • ΔAICc: Corrected Delta Akaike Informa-
                                        combination of low-cost and easily deploy-        and protecting the most valuable areas,             tion Criterion;
                                        ed unmanned aerial vehicles (UAVs) and            such as protected areas and vulnerable ar-        • ASCII: American Standard Code II;
                                        high-resolution, freely available remote          eas covered with intact forest (closed/           • ASTER: Advanced Spaceborne Thermal
                                        sensing images with SMART-Lawin data              open forest). In this case, overlaying the          Emission and Reflection Radiometer;
                                        would thus be beneficial.                         risk map for each illegal activity with land      • AUC: Area Under the Curve;
                                                                                          cover or forest maps will be useful.              • AUCTEST: Area Under the Curve for the
                                        Optimal strategy for mitigating illegal                                                               Test Dataset;
                                        activity                                          Conclusion                                        • BCA: Baliuag Conservation Area;
                                          The results of this study can be used to          It is important for conservation area man-      • DEM: Digital Elevation Model;
                                        deploy patrol teams that prioritize high-risk     agers to identify the drivers determining         • ENMeval: Ecological Niche Model Evalua-
                                        areas. Managers can either target the de-         the occurrence of illegal activity and the lo-      tion;
                                        terrence of a specific illegal activity or a      cations where it is most likely to occur          • FC: Feature Classes;
                                        combination of multiple illegal activities.       within their areas of jurisdiction in order to    • FCA: Forest Conservation Area;
                                        However, our results are limited by the           effectively implement forest protection           • FMB: Forest Management Bureau;
                                        range of variables used in developing the         and law enforcement. Because there are            • MaxEnt: Maximum Entropy;
                                        model. According to Critchlow et al. (2015),      vast areas of forest with high biodiversity       • RM: Regularization Multiplier;
                                        the selection of variables plays a decisive       in the Philippines, the optimal spatial allo-     • SMART: Spatial Monitoring and Reporting
                                        role in the resulting distribution of predict-    cation of limited patrolling resources must         Tool:
                                        ed illegal activity. Therefore, identifying the   be carried out to effectively prevent illegal     • SMART-Lawin: Lawin Forest and Biodiver-

                                        68                                                                                                                             iForest 15: 63-70
Modeling the risk of illegal forest activity in SE Philippines

  sity Protection System;                               prove prediction of species’ distribution from           tion 5: 1198-1205. - doi: 10.1111/2041-210X.12261

                                                                                                                                                                     iForest – Biogeosciences and Forestry
• SMMR: Sierra Madre Mountain Range;                    occurrence data. Ecography 29 (2): 129-151. -           Muscarella R, Galante PJ, Soley-Guardia M, Boria
• TSS: True Skill Statistics;                           doi: 10.1111/j.2006.0906-7590.04596.x                    RA, Kass JM, Uriarte M, Anderson RP (2017).
• VIF: Variance Inflation Factor.                      Elith J, Leathwick JR (2009). Species distribution        Package “ENMeval”, version 0.3.0. Web site.
                                                        models: ecological explanation and prediction            [online] URL: http://cran.r-project.org/web/pa
References                                              across space and time. Annual Review of Ecol-            ckages/ENMeval/index.html
Allouche O, Tsoar A, Kadmon R (2006). Assess-           ogy Evolution and Systematics 40: 677-697. -            Phillips SJ, Dudík M, Schapire RE (2004). A maxi-
 ing the accuracy of species distribution models:       doi: 10.1146/annurev.ecolsys.110308.120159               mum entropy approach to species distribution
 prevalence, kappa and the true skill statistic        Elith J, Phillips SJ, Hastie T, Dudík M, Chee YE,         modelling. In: Proceedings of the “21st Interna-
 (TSS). Journal of Applied Ecology 43: 1223-1232.       Yates CJ (2011). A statistical explanation of            tional Conference on Machine Learning” (Grei-
 - doi: 10.1111/j.1365-2664.2006.01214.x                MaxEnt for ecologists. Diversity and Distribu-           ner R, Shuurmans D eds). ACM Press, Banff,
Bisson J, Guiang ES, Walpole P, Tolentino D             tions 17 (1): 43-57. - doi: 10.1111/j.1472-4642.2010.    Canada, New York, USA, pp. 655-662.
 (2003). Better governance critical to reversing        00725.x                                                 Phillips SJ, Anderson RP, Schapire RE (2006).
 forest degradation in Southeast Asia. In: Pro-        Finger A, Stolton S, Dudley N (1999). Metals              Maximum entropy modelling of species geo-
 ceedings of the “12th World Forestry Congress”.        from the forests. Mining and forest degrada-             graphic distributions. Ecological Modelling 190:
 UN-FAO (Québec City, Canada), no. 0837-A4.             tion. IUCN and WWF, Gland, Switzerland. pp.              231-259. - doi: 10.1016/j.ecolmodel.2005.03.026
 [online] URL: http://www.fao.org/3/XII/0837-A4         36.                                                     Plumptre AJ, Fuller RA, Rwetsiba A, Wanyama F,
 .htm                                                  Geronimo MC, Marie Grace Cabansag S, Agnes                Kujirakwinja D, Driciru M, Nangendo G, Watson
Carandang AP, Bugayong LA, Dolom PC, Garcia             Reyes S (2016). Indigenous utilization of re-            JEM, Possingham HP (2014). Efficiently target-
 LN, Villanueva MMB, Espiritu NO (2013). Analy-         sources and conservation practices of the Agta           ing resources to deter illegal activities in pro-
 sis of key drivers of deforestation and forest         of Lupigue, Ilagan City, Isabela, Philippines. In:       tected areas. Journal of Applied Ecology 51 (3):
 degradation in the Philippines. Deutsche Gesell-       “EDUCARE” International Journal for Educa-               714-725. - doi: 10.1111/1365-2664.12227
 schaft für Internationale Zusammenarbeit (GIZ)         tional Studies, Minda Masagi Press and UMP              R Core Team (2020). R: a language and environ-
 GmbH, Manila, Philippines, pp. 110.                    PurwokertoBandung, Indonesia, Vol. 8 (2), pp.            ment for statistical computing. R Foundation
Critchlow R, Plumptre AJ, Driciru M, Rwetsiba A,        115-128. [online] URL: http://www.journals.mind          for Statistical Computing, Vienna, Austria. [on-
 Stokes EJ, Tumwesigye C, Wanyama F, Beale              amas.com/index.php/educare/article/view/749              line] URL: http://www.r-project.org/
 CM (2015). Spatiotemporal trends of illegal ac-       Hair JF, Black WC, Babin BJ, Anderson RE (2014).         Radosavljevic A, Anderson RP (2014). Making
 tivities from ranger-collected data in a Ugan-         Multivariate data analysis. Pearson Education            better Maxent models of species distributions:
 dan national park. Conservation Biology 29 (5):        Limited, Harlow, UK, pp. 734.                            complexity, overfitting and evaluation. Journal
 1458-1470. - doi: 10.1111/cobi.12538                  Jones KR, Venter O, Fuller RA, Allan JR, Maxwell          of Biogeography 41 (4): 629-643. - doi: 10.1111/
Critchlow R, Plumptre AJ, Alidria B, Nsubuga M,         SL, Negret PJ, Watson JEM (2019). One-third of           jbi.12227
 Driciru M, Rwetsiba A, Wanyama F, Beale CM             global protected land is under intense human            Remedio EM (2009). An analysis of sustainable
 (2017). Improving law-enforcement effective-           pressure. Science 360: 788-791. - doi: 10.1126/sci       fuelwood and charcoal production systems in
 ness and efficiency in protected areas using           ence.aap9565                                             the Philippines: a case study. In: “Criteria and
 ranger-collected monitoring data. Conservation        Keane A, Jones JPG, Edwards-Jones G, Milner-              indicators for sustainable woodfuels: case stud-
 Letters 10: 572-580. - doi: 10.1111/conl.12288         Gulland EJ (2011). Encounter data in resource            ies from Brazil, Guyana, Nepal, Philippines and
Denninger Snyder K, Mneney PB, Wittemyer G              management and ecology: pitfalls and possibili-          Tanzania” (Rose S, Remedio EM, Trossero MA
 (2019). Predicting the risk of illegal activity and    ties. Journal of Applied Ecology 48: 1164-1173. -        eds), FAO, Rome, Italy, pp. 133-194. [online]
 evaluating law enforcement interventions in            doi: 10.1111/j.1365-2664.2011.02034.x                    URL: http://www.fao.org/3/i1321e/i1321e08.pdf
 the western Serengeti. Conservation Science           Kummer DM (1992). Upland agriculture, the land           Rubas-Leal LC, Washington-Allen RA, Popescu
 and Practice 1 (9): 534. - doi: 10.1111/csp2.v1.9      frontier and forest decline in the Philippines.          SC, Conner JR (2017). Estimation of above-
DENR-PAWB (2009). New conservation areas in             Philippine Agroforestry Systems Publications             ground biomass in Mt. Apo natural park in the
 the Philippines project (NewCAPP). Threats to          18: 31-46. - doi: 10.1007/BF00114815                     Southern Philippines using terrestrial LiDAR
 Philippine biodiversity, Department of Environ-       Lasco RD, Pulhin FB, Sanchez PJ, Villamor GB, Vil-        system. BANWA Series B 12: res005. [online]
 ment and Natural Resources-Protected Areas             legas KAL (2008). Climate change and forest              URL: http://ojs.upmin.edu.ph/index.php/banwa-
 and Wildlife Bureau, Manila, Philippines, pp. 3-       ecosystems in the Philippines: vulnerability,            b/article/view/325
 5.                                                     adaptation and mitigation. Journal of Environ-          Torres AM, Pulhin FB, Lasco RD, Tiburan CL, Es-
DENR (2015). Integrated watershed manage-               mental Science and Management 11 (1): 1-14.              lava DF (2016). Modeling future spatial distribu-
 ment plan (IWMP) for Angat - Watershed for-           Liao Y, Lei Y, Ren Z, Chen H, Li D (2017). Predict-       tion of Shorea palosapis (Blanco) Merr and
 est reserve. Department of Environment and             ing the potential risk area of illegal vaccine           Shorea polysperma (Blanco) Merr in Northern
 Natural Resources- Region 3, San Fernando,             trade in China. Scientific Reports 7 (1): 3883.          Sierra Madre natural park using Maxent. Jour-
 Pampanga, Philippines, pp. 10-18.                     Mallari NMD, Jensen A (1993). Biological diver-           nal of Environmental Science and Management
DENR-FMB (2018). Philippine forestry statistics         sity in norther Sierra Madre, Philippines: its im-       2016 (1): 15-30. [online] URL: http://ovcre.uplb.
 2018. Department of Environment Natural Re-            plication for conservation and management.               edu.ph/journals-uplb/index.php/JESAM/article/v
 sources - Forest Management Bureau R, Mani-            Asia Life Sciences 2 (2): 101-112.                       iew/485
 la, Philippines, pp. 92. [online] URL: http://        Merow C, Smith MJ, Silander JA (2013). A practi-         Tumaneng MR, Tumaneng R, Tiburan CL (2019).
 forestry.denr.gov.ph/index.php/statistics/philip       cal guide to MaxEnt for modeling species’ dis-           Modeling species distribution of Shorea guiso
 pines-forestry-statistics                              tributions: what it does, and why inputs and             (blanco) Blume and Parashorea malaanonan
Domingo SN, Manejar AJA (2018). Forest protec-          settings matter. Ecography 36: 1058-1069. - doi:         (blanco) Merr in mount Makiling forest reserve
 tion in the Philippines. Discussion Paper Series       10.1111/j.1600-0587.2013.07872.x                         using Maxent. ISPRS - International Archives of
 no. 2018-53, Philippine Institute of Develop-         Minter T, Van Der Ploeg J, Pedrablanca M (2014).          the Photogrammetry, Remote Sensing and
 ment Studies, Quezon City, Philippines, pp. 48-        Limits to indigenous participation: the Agta and         Spatial Information Sciences, pp. 441-448. [on-
 52.                                                    the Northern Sierra Madre natural park, the              line] URL: http://noa.gwlb.de/servlets/MCRFile
Elith J, Graham CH, Anderson RP, Dudik M, Fer-          Philippines. Human Ecology 42 (5): 769-778. -            NodeServlet/cop_derivate_00049506/isprs-arc
 rier S, Guisan A, Hijmans RJ, Huettmann F,             doi: 10.1007/s10745-014-9673-5                           hives-XLII-4-W19-441-2019.pdf
 Leathwick JR, Lehmann A, Li J, Lohmann LG,            Muscarella R, Galante PJ, Soley-Guardia M, Boria         USAID/B+WISER (2018). The Lawin forest and
 Loiselle BA, Manion G, Moritz C, Nakamura M,           RA, Kass JM, Uriarte M, Anderson RP (2014).              biodiversity protection system manual. Module
 Nakazawa Y, Overton Mc JC, Peterson AT, Phil-          ENMeval: An R package for conducting spatial-            2: purpose-driven patrolling. Chemonics Inter-
 lips SJ, Richardson KS, Scachetti-Pereira R,           ly independent evaluations and estimating opti-          national Inc, Washington, DC, USA. [online]
 Schapire RE, Soberón J, Williams S, Wisz MS,           mal model complexity for MAXENT ecological               URL: http://forestry.denr.gov.ph/index.php/pub
 Zimmermann NE (2006). Novel methods im-                niche models. Methods in Ecology and Evolu-              lications/2020

iForest 15: 63-70                                                                                                                                              69
Barit JB et al. - iForest 15: 63-70

                                        Van Der Ploeg J, Van Weerd M, Masipiqueña AB,          of the “4th Regional Conference on Environ-          totenv.2018.04.112
iForest – Biogeosciences and Forestry

                                         Persoon GA (2011). Illegal logging in the North-      ment and Development” (van der Ploeg J,
                                         ern Sierra Madre Natural Park, the Philippines.       Masipiquena AB, Bernando EC eds). CVPED, Is-        Supplementary Material
                                         Conservation and Society 9 (3): 202-215. [on-         abela, The Philippines, pp. 443.
                                         line] URL: http://www.jstor.org/stable/263930        Zhang K, Yao L, Meng J, Tao J (2018). Maxent         Tab. S1 - ENMeval results for the three ille-
                                         43                                                    modeling for predicting the potential geo-          gal activity classifications.
                                        Van Der Ploeg J, Masipiquena AB, Bernando EC           graphical distribution of two peony species un-
                                         (2003). The Sierra Madre mountain range:              der climate change. Science of The Total Envi-      Link: Barit_3937@suppl001.xlsx
                                         global relevance, local realities. In: Proceedings    ronment 634 (10): 1326-1334. - doi: 10.1016/j.sci

                                        70                                                                                                                                   iForest 15: 63-70
You can also read