Mapping the potential distribution of Pinus merkusii Jungh et de Vries. a
vulnerable gymnosperm in eastern Arunachal Pradesh using Maximum Entropy
model
Jyotishman Deka*, Sanjeeb Bharali, Anup Kr Das, Om Prakash Tripathi and
Mohamed Latif Khan
Department of Forestry, North Eastern Regional Institute of Science and Technology,
Itanagar, 791109, Arunachal Pradesh, India
*Corresponding author: jyotishmandeka@gmail.com
1
Mapping the potential distribution of Pinus merkusii Jungh et de Vries. a
vulnerable gymnosperm in eastern Arunachal Pradesh using Maximum Entropy
model
Abstract
Pinus merkusii is a vulnerable gymnosperm and occurs mainly in the eastern part of
Anjaw district of Arunachal Pradesh, northeastern India. However, rapid extraction
from its natural habitat had led to its population decline and is as considered
vulnerable. Documentation and distribution of the species in this part of the country
is lacking till date. As such in order to overcome the constrictions of the region’s
remoteness and rugged terrain for field surveys, the present study utilized the species
occurrence information and remotely sensed environmental variables to generate the
distribution map of P. merkusii and its validation. The AUC of 0.979 and 0.972
observed for the training and test data, respectively, indicated higher success rates.
This study revealed that the Enhanced Vegetation Index (EVI) proposed by the
MODIS play an important role in providing dependable spatial and temporal
information of vegetation cover and can be suitably applied to species distribution
modeling. However multicollinearity among the predictor variables which is
commonly associated with remotely sensed data sets should be emphasized. The
study also focuses on the potentiality of Maximum entropy (Maxent) modeling for
identifying species distributions. It was further hoped that the results drawn in the
current study will acts an effective tool for monitoring as well in preparation of
effective conservation planning of the species in this particular region.
Keywords: Enhanced vegetation index, MODIS, Pinus merkusii, Species distribution,
Vulnerable
2
1 Introduction
Pinus merkusii Jungh et de Vries is the only surviving member of the genus that
naturally extends its distribution across the south equator. In India, it occurs mainly in
the eastern part of Anjaw district of Arunachal Pradesh which is neighboring to the
India–Myanmar Pine Forests Eco-region. It is the dominant tree species in this region.
However, in recent year’s unsustainable extraction for timber, deforestation, clear
felling, logging etc. have built up the pressure on the habitat of the species and as a
result population is decreasing significantly (Bharali et al., 2012). With the increasing
awareness towards the importance of Himalayan biodiversity and alarming rate at
which they are being exploited from natural habitats, there is a need to initiate various
conservation and management strategies to mitigate such uncontrolled resource
exploitation (Ray et al., 2011). As such the various key features such as species
geographic distribution, its habitat and ecological characteristic, must be studied
properly for the species protection / restoration activities.
Locating populations of rare, endangered, and threatened species is an
important consideration in biodiversity conservation and for developing effective
strategies for both in situ and ex situ conservation (Menon et al., 2010). With the
advancement of remote sensing and GIS technologies, use of ecological niche models
(ENM) to generate potential geographic distributions of species has rapidly increased
in ecology, conservation and evolutionary biology (Terribile et al., 2010). The most
common strategy for estimating the actual or potential geographic distribution of a
species is to characterize the environmental conditions that are suitable for the
species, and identify where suitable environments are distributed in space for any
3
particular species (Pearson, 2007). ENM establishes relationships between
occurrences of species with biophysical and environmental conditions (Soberon and
Peterson, 2005; Kumar and Stohlgren, 2009). A large number of species distribution
modeling methods are now available to predict potential distribution for a species but
the basic constrain with most of the model is that they are sensitive to sample size. In
such case with limited sample size, Maxent species distribution models were proven
to be advantageous over the methods (Philips et al., 2006). Maxent is a bioclimatic
model which deals with presence only data and has a number of other aspects that
make it well-suited for species distribution modeling (Chitale and Behera, 2012). As
such a large number of researchers worldwide have implemented Maxent for
predicting the distribution of endangered and threatened species (Elith et al., 2007;
Philips et al., 2006; Yost et al., 2008; Kumar and Stohlgren, 2009). Further free
availability of high to low resolution satellite imageries, environmental variables
databases and interpolated spatial datasets on climate and vegetation from public
domain has enhanced the accuracy of prediction of these models manifold (Adhakari
et al., 2012).
In the current study, we used species occurrence records, GIS, environmental
layer (Enhanced vegetation index) and the maximum entropy distribution modeling
approach Phillips et al. (2008) to predict the potential distribution for this vulnerable
gymnosperm in order to develop the strategic planning for its conservation.
4
2 Methodology
2.1 Study site
The study was carried out in Anjaw district of Arunachal Pradesh. The district
is located between 26°55' and 28°40' N latitude and between 92°40' and 94°21' E
longitude and with an altitudinal range of 1000-2000 m asl (Figure 1). It is surrounded
by China on north, Myanmar is on the eastern part of the district, and Lower Dirang
Valley district is in the west and Lohit district in the South. The major rivers are
Lohit, Delei and Tellu. The vegetation of the district falls under four broad climatic
categories and classified in five broad forest types with a sixth type of secondary
forests. These are tropical forests, sub-tropical forests, pine forests, temperate forests
and alpine forests (Kaul and Haridasan, 1987). The climate is sub-tropical, wet and
highly humid in the lower elevations and the valleys; and cold in the higher
elevations. Annual rainfall in the district varies from 3500 to 5500 mm. Soil type was
generally sandy and acidic in nature.
2.2 Selected species and species records
P. merkusii, is the only surviving member of the genus that naturally extends
its distribution across the south equator (Figure 2). It is a medium to large size tree,
grows up to 70 m height making it one of the world’s tallest pines, with a trunk
diameter up to 1 m and pyramidal to conical crown on young age, and flatter and
spreading on old age. The needles are very slender, rigid, straight 15-25 cm long and
less than 1 mm thick, green to yellow color and are found in a pair of two. It flowers
during March-June and the seeds ripen during October-November of the following
5
year. P. merkusii has a large altitudinal range from a few meters to over 1800 m
elevation (Cooling, 1968).
We obtained 30 occurrence records of P. merkusii tree species from field visit.
These records were considered as the total known distribution for the species. All the
coordinated were collected in degree decimal and converted to point vector file for
modeling the distribution of the species.
2.3 Environmental variables and distribution modeling
In the present study, twenty three layers of 16 day composite MODIS EVI
images (MOD13Q1) with a spatial resolution of 250 m were downloaded from Oak
Ridge National Laboratory Distributed Active Archive Centre (http://daac.ornl.gov
/MODIS/modis.html) and were used to characterize environments across the region.
The images were downloaded in Geo-tiff format and converted to ASCII raster grids.
These layers correspond to the year 2011 during which the field survey was
conducted. EVI (Enhanced vegetation index) has been considered as the modified
NDVI with improved sensitivity to high biomass regions and improved vegetation
monitoring capability (Huete et al., 1999). In order avoid cloud cover and other
spurious effects the MODIS EVI products are generated by compositing daily data
every 16 days, resulting in 23 composites per year (Huete et al., 2002). We did not use
elevation layers as their influence on species distribution as P. merkusii has a large
altitudinal range from a few meters to over 1800 m elevation (Cooling, 1968). As
MODIS EVI dataset has been proved to be quite sensitive to the variations in
topographic conditions topographic variables such as slope and aspect was also not
6
included in the model (Matsushita et al., 2007; Adhakari et al., 2012). Further
multicollinearity among the predictor variables which is commonly associated with
remotely sensed data sets can be a source of inaccurate results and hence should be
avoided (Dormann et al., 2007). In order to identify the multicollinearity among of
layers (r > 0.9), all the 23 EVI layers were then subjected to correlation test. From
each of the highly correlated pair of layers one of the EVI layers was then excluded
from the ENM analysis. Finally, 10 environmental data layers were selected for
habitat distribution modelling.
2.4 Model validation
ENMs were built using the maximum entropy niche modeling approach
implemented
in
the
software
Maxent
v3.3.3k
(http://www.cs.princeton.
edu/~schapire/maxent) as it estimates the probability distribution for a species
occurrence based on environmental constraints and is applicable with presence-only
data (Philips et al., 2006). Of the total records, seventy five percent were used for
model training and twenty five percent for testing. To provide a test of model
predictability, jackknife-based approach presented by Pearson et al. (2007) was
applied, which is designed explicitly for situations of small sample size (Menon et al.,
2010). The area under the receiver operating characteristic curve (AUC) value was
evaluated and the model was graded as: poor (AUC < 0.8), fair (0.8 < AUC < 0.9),
good (0.9 < AUC < 0.95) and very good (0.95 < AUC < 1.0) following Thuiller et al.
(2005).
7
3 Results and discussion
Maxent model was run with the following configurations: Hinge features,
perform jackknife tests, logistic output format with a threshold rule of 10 percentile
training presence. As the program is already calibrated on a wide range of species
datasets, other parameters were set to default8. Hinge features can effectively replace
quadratic, product and threshold features and was found suitable with any presenceonly dataset and are used by default in the software whenever there are at least 15
presence sites (Phillips and Dudik, 2008).
The model calibration test for P. merkusii generated acceptable results, AUC
train = 0.979 and AUC test = 0.972 with standard deviation of 0.007 (Figure 3). The
environmental variable with highest gain when used in isolation is evi_081 and
contributed 23.4% to the MaxEnt mode which therefore appears to have the most
useful information. The environmental variable that decreases the gain the most when
it is omitted is evi_273, which therefore appears to have the most information that
isn't present in the other variables. It was found from the literature and also from the
field survey that the layers (evi_081 and evi_273) correspond to the period of
flowering and fruiting phage of the species. Sensitivity of EVI layers towards
flowering and fruiting phage of the species Ilex khasiana Purk., was also reported by
Adhikari et al. (2012). Menon et al. (2010) also used 11 EVI layers to characterize
environments variables for predicting the distribution of Gymnocladus assamicus in
Tawang and West Kameng district of Arunachal pradesh. Waring et al. (2006)
reported that EVI has more advantages over other vegetation index in that it is
sensitive to site-specific variations and may be more sensitive to local variation in
8
canopy leaf area and chlorophyll concentrations. This shows that the EVI proposed by
the MODIS play an important role in providing dependable spatial and temporal
information of vegetation cover and can be suitably applied to species distribution
modeling. Other layers of EVI together contributed 76.6% to the distribution model of
P. merkusii of which evi_273, evi_193 and evi_017 contributed 23.8%, 21.4% and
15.9% respectively. Approximations of percent contributions and permutation
importance of different environmental variables are given in Table 1.
The Maxent model predicted potential distribution for P. merkusii with high
success rates and are well agreed with the known distribution of the species (Figure
4). Most suitable habitat for P. merkusii was predicted in the north eastern parts of the
district i.e., Kibithoo and Walong, forming almost a continuous belt and few other
scattered areas towards Hawaii, Goiliang and Chaglagam. Using arbitrarily defined
scale we grouped the probability values into four classes as high (0.6-1.0), medium
(0.3-0.6), small (0.1-0.3) and no distribution (0.0-0.1). The highly distributed P.
merkusii area class was found to be 130 km2; medium 354 km2 and low 847 km2.
Thus study showed that ENMs can be used to predict the potential distribution of
species in areas with relatively few available data (Stockwell, D.R.B. and Peterson,
(2002); Pearson, 2007; Babar et al., 2012).
This study makes available the first predicted potential distribution map for a
plant species (P. merkusii) in India. The study showed that with the use of small
number of occurrence records and environmental variables derived from remotely
sensed data; potential habitat as well as distribution patterns of plant species such as
P. merkusii can be suitably modeled using Maxent. Since the species is confined to a
9
very small geographical area, results of this study will play a major role in building
strategic planning for the conservation of species. The potential habitat distribution
map for P. merkusii can be helpful in land use management planning around its
existing populations, identify conservation sites and also in selecting new sites for
reintroduction of the species in its natural habitat. More research in the near future is
necessary in order to safeguard the species and its habitat.
4 Conclusion
The main objective of this research is to model the geographic distribution of P.
merkusii in Anjaw district of Arunachal Pradesh and also to identify the most
important environmental variables and predict the species richness distribution using
Maxent. The present study utilized the species occurrence information and remotely
sensed environmental variables to generate the distribution map of the target species
and its validation. The AUC of 0.979 and 0.972observed for the training and test data
respectively, indicated higher success rates. Of the 10 predictor variables used to build
our model, evi_081 appeared as the strongest predictors. Hence it has also been
observed that environmental variables like MODIS enhanced vegetation index images
derived from remotely sensed data
can plays a key role in determining the
distribution of potential habitats of P. merkusii. The result further indicates that
species distribution models offer new understandings on the geographical bionetwork
of these species. Mapping and modeling the geographic range of a species grips its
potentiality in conservation biology. It was further hoped that the predicted potential
10
distribution habitat of the species in this study would help in developing conservation
strategies for monitoring and managing the species in its native range.
Acknowledgement
Firstly the authors are grateful to Oak Ridge National Laboratory Distributed Active
Archive Centre for free distribution the MODIS data. Thanks are due to Biranjoy
Basumatary and Bijit Basumatary for their help during the field study.
References
1.
Bharali, S., Deka, J., Saikia, P., Khan, M.L., Paul, A., Tripathi, O.P., Singha,
L.B. and Uma Shankar., 2012, Pinus merkusii Jungh et de Vries - a vulnerable
gymnosperm needs conservation. Nebio, 3, 94–95.
2.
Ray, R., Gururaja, K.V. and Ramchandra, T.V., 2011, Predictive distribution
modeling for rare Himalayan medicinal plant Berberis aristata DC. Journal of
Environmental Biology, 32, 725–730.
3.
Menon, S., Choudhury, B.I., Khan, M.L. and Peterson, A.T., 2010, Ecological
niche modeling and local knowledge predict new populations of Gymnocladus
assamicus a critically endangered tree species. Endangered Species Research,
11, 175–181.
4.
Terribile, L.C., Diniz‑Filho, J.A.F. and De Marco Jr., P., 2010, How many
studies are necessary to compare niche-based models for geographic
distributions? Inductive reasoning may fail at the end. Brazilian Journal of
Biology, 70, 263–269.
11
5.
Pearson, R.G., 2007, Species Distribution Modeling for Conservation
Educators and Practitioners. Report, American Museum of Natural History,
Available at http://ncep.amnh.org.
6.
Soberon, J. and Peterson, A.T., 2005, Interpretation of models of fundamental
ecological niches and species distributional areas. Biodiversity Informatics, 2,
1–10.
7.
Kumar, S. and Stohlgren, T.J., 2009, Maxent modeling for predicting suitable
habitat for threatened and endangered tree Canacomyrica monticola in New
Caledonia. Journal of Ecology and The Natural Environment, 1, 094–098.
8.
Phillips, S.J., Anderson, R.P. and Schapire, R.E., 2006, Maximum entropy
modeling of species geographic distributions. Ecological Modelling, 190,
231–259.
9.
Chitale, V.S. and Behera, M.D., 2012, Can the distribution of sal (Shorea
robusta Gaertn. f.) shift in the northeastern direction in India due to changing
climate?. Current Science, 2, 1126–1135.
10.
Elith, J., Graham, C.H., Anderson, R.P., Dudik, M., Ferrier, S., Guisan, A.,
Hijmans, R.J., Huettmann, F., Leathwick, J.R., Lehmann, A., Li, J., Lohmann,
L.G., Loiselle, B.A., Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y.,
Overton, J.M., Peterson, A.T., Phillips, S.J., Richardson, K., Scachetti Pereira,
R., Schapire, R.E., Soberon, J., Williams, S., Wisz, M.S. and Zimmermann,
N.E., 2006, Novel methods improve prediction of species' distributions from
occurrence data. Ecography, 29, 129–151.
12
11.
Yost, A.C., Petersen, S.L., Gregg, M. and Miller, R., 2008, Predictive
modeling and mapping sage grouse (Centrocercus urophasianus) nesting
habitat using Maximum Entropy and a long-term dataset from Southern
Oregon. Ecological Informatics, 3, 375–386.
12.
Adhikari, D., Barik, S.K. and Upadhaya, K., 2012, Habitat distribution
modelling for reintroduction of Ilex khasiana Purk., a critically endangered
tree species of northeastern India. Ecological Engineering, 40, 37– 43.
13.
Kaul, R.N. and Haridasan, K., 1987, Forest types of Arunachal Pradesh, a
preliminary study. Journal of Economic and Taxonomic Botany, 9, 379–389.
14.
Cooling, E.N., 1968, Pinus merkusii. Fast Growing Timber Trees of the
Lowland Tropics. Commonwealth Forestry Institute, Oxford, 4, 169–176.
15.
Huete, A., Justice, C. and van Leeuwen, W., 1999, MODIS vegetation index
(MOD13) algorithm theoretical basis document. Ver. 3. Available at
http://modis.gsfc.nasa.gov/data/ atbd/atbd_mod13.pdf
16.
Huete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao X. and Ferreira, L.G.,
2002, Overview of the radiometric and biophysical performance of the
MODIS vegetation indices. Remote Sensing of Environment, 83, 195–213.
17.
Matsushita, B., Yang, W., Chen, J., Onda, Y. and Qiu, G., 2007, Sensitivity of
the Enhanced Vegetation Index (EVI) and Normalized Difference Vegetation
Index (NDVI) to Topographic Effects: A Case Study in High-Density Cypress
Forest. Sensors, 7, 2636–2651.
18.
Dormann, C.F., Mcpherson, J.M., Araujo, M.B., Bivand, R., Bolliger, J., Carl,
G., Davies, R.G., Hirzel, A., Jetz, W., Kissling,D.W.,Kuhn, I.,Ohlemuller, R.,
13
Peres-Neto, P.R., Reineking, B., Schroder, B., Schurr, F.M. and Wilson, R.,
2007, Methods to account for spatial autocorrelation in the analysis of species
distributional data: a review. Ecography, 30, 609–628.
19.
Thuiller, W., Richardson, D.M., Pysek, P., Midgley, G.F., Hughes, G.O. and
Rouget, M., 2005, Niche-based modelling as a tool for predicting the risk of
alien plant invasions at a global scale. Global Change Biology, 11, 2234–
2250.
20.
Phillips, S.J. and Dudik, M., 2008, Modeling of species distributions with
MaxEnt: new extensions and a comprehensive evaluation. Ecography, 31,
161–175.
21.
Waring, R.H., Coops, N.C., Fan, W. and Nightingale, J.M., 2006, MODIS
enhanced vegetation index predicts tree species richness across forested
ecoregions in the contiguous U.S.A. Remote Sensing of Environment, 103,
218–226.
22.
Stockwell, D.R.B. and Peterson, A.T., 2002, Effects of sample size on
accuracy of species distribution models. Ecological Modelling, 148, 1–13.
23.
Babar, S., Amarnath, G., Reddy, C.S., Jentsch, A. and Sudhakar, S., 2012,
Species distribution models: ecological explanation and prediction of an
endemic and endangered plant species (Pterocarpus santalinus L.f.). Current
Science, 102, 1157–1165.
14
List of Table:
1. Table1: Percent contributions and permutation importance of different
environmental variables.
Table1: Percent contributions and permutation importance of different environmental
variables.
Percent
Permutation
Environmental Variable
contribution
importance
evi_081 (22 March - 6 April)
23.4
30.6
evi_273 (30 Sept - 15 Oct)
23.8
24.1
evi_017 (17 Jan - 1 Feb)
15.9
16.6
evi_193 (12 July - 27 July)
21.4
16.1
evi_241 (29 Aug - 13 Sept)
5.5
9.9
evi_257 (14 Sept - 29 Sept)
0.6
1.1
evi_161 (10 June - 25 June)
3.4
0.8
evi_145 (25 June - 9 June)
0.1
0.5
evi_225 (14 Aug - 28 Aug)
5.6
0.3
evi_097 (7 April - 22 April)
0.3
0
15
List of Figures:
1. Figure 1: Map of study area showing Anjaw district in Arunachal Pradesh
2. Figure 2: Natural stand of P. merkusii in Anjaw district of Arunachal Pradesh
3. Figure 3: (a) Jackknife of regularized training gain for P. merkusii (b)
Sensitivity vs 1- Specificity for P. merkusii
4. Figure 4: Predicted potential distribution for P. merkusii in Anjaw district,
Arunachal Pradesh
Figure 1: Map of study area showing Anjaw district in Arunachal Pradesh
16
Figure 2: Natural stand of P. merkusii in Anjaw district of Arunachal Pradesh
17
(a)
(b)
Figure 3: (a) Jackknife of regularized training gain for P. merkusii (b) Sensitivity vs
1- Specificity for P. merkusii
18
Figure 4: Predicted potential distribution for P. merkusii in Anjaw district, Arunachal
Pradesh
19