Ecological niche modeling to find potential habitats of Vanda thwaitesii , a notified endangered orchid of Western Ghats, India

: Conservation planning of a threatened species essentially requires information on its present population and extent of distribution. Ecological niche modeling (ENM) is a suitable machine learning technique to predict potential distribution of a species enabling to identify suitable habitat for conservation action. The present study undertook distribution modeling of Vanda thwaitesii , a notified endangered orchid species of the Indian Western Ghats and Sri Lanka using maxent software. Geographical coordinates of 54 occurrence points at 1 km resolutions gathered during the study were utilized for modeling. A total of 37 variables including bioclimatic, topographical, and seasonal climate subjected to principal component analysis extracted into three components based on temperature and precipitation. Four representative variables from each component in all possible combinations resulted consistent output showing distribution of the species extending from Gavi in Periyar Tiger Reserve of Kerala to Chikkamagalur of Karnataka. Habitat suitability was confined to the cooler regions receiving an average 3,400 mm annual mean precipitation, 22.7 0 C annual mean temperature, and 290 mm summer precipitation. A total of 2,557 km 2 in Kerala and Karnataka mostly outside protected forests demonstrated as the highly suitable habitats. Silent Valley National Park, Idukki Wildlife Sanctuary, Periyar Tiger Reserve, and Brahmagiri Wildlife Sanctuary in addition to a few reserve forests hold sufficient area for reinforcement of diversity of V. thwaitesii from vulnerable locations. The present study revealed niche modeling as a useful tool to find suitable habitats for V. thwaitesii in the Western Ghats.


INTRODUCTION
Orchids are a group of plants belonging to Orchidaceae with 29,335 accepted species worldwide (POWO 2021). They have wide climatic preferences ranging from tropical to alpine habitats. However, due to habitat loss, habitat fragmentation, over-exploitation, and unrestrained illegal activities, many orchids have limited distribution range and population strength as other threatened plant species (Agustini et al. 2016;Warghat et al. 2016;Bachman et al. 2019;Lughadha et al. 2020). The existence of biodiversity have prime importance for the stability of an ecosystem and thus developing effective strategies for their conservation is the serious concern of conservation biologists . Besides, conservation planning of a threatened species essentially requires information on its present population and extent of distribution (Radosavljevic & Anderson 2014;Štípková et al. 2020). Mapping the identified geographical locations and predicting potential distribution of a species out of that is known to be useful in identifying critical regions that may need conservation action (Warren & Seifert 2011).
Vanda thwaitesii Hook.f. is one among the 81 species of Vanda reported worldwide (POWO 2021). It is endemic to the southern Western Ghats and Sri Lanka and is endangered mainly due to habitat loss and fragmentation. It was first collected by Thwaites from Sri Lanka in 1898 and remained elusive for over a century which forced to declare the species as extinct in 1981 (Sathishkumar & Sureshkumar 1998). During the period from 1982 to 1997, the species was collected from Silent Valley and Wayanad, Kerala thus confirming its presence in India (Sathishkumar & Sureshkumar 1998). The latter authors could also locate reference sample of V. thwaitesii collected in 1885 from Mananthavady in Wayanad District of Kerala. As per reports, the species have distribution in seven localities in Kerala. The species is distributed in narrow pockets with restricted numbers and later under section 38 of the Biological Diversity Act 2002, the Central Government notified that V. thwaitesii is on the verge of extinction and prohibited/regulated collection (MOEF 2009). The ministry also called for studies on all aspects of the notified species for holistic understanding and propagation of the species for the purpose of in situ and ex situ conservation and rehabilitation.
Plant distribution modeling/ ecological niche modeling (ENM) is recognized as an efficient tool to understand potential distribution of a plant/animal species and maximum entropy method (MEM; Phillips et al. 2006) is widely used for the purpose (Elith et al. 2011;Peterson et al. 2011). These models establish relationships between occurrences of species and biophysical and environmental conditions in the study area thus predicting suitable habitat for survival and existence of a species. This technique was successfully applied to find potential distribution and identify environmental niches for several plant species including orchids Vanda wightii (Decruse 2014), V. bicolor (Deb et al. 2017), Paphiopedilum javanicum (Romadlon et al. 2021), and Habenaria suaveolens (Jalal & Singh 2017). Based on the current understanding and the requirement of devising conservation action, the present study is framed to understand the habitat suitability of the species in localities other than that described earlier and the reported localities in Western Ghats region of Kerala, Tamil Nadu, and Karnataka states in India.

Study site and field survey
As per the cited literature, Wayanad, Silent Valley, and Periyar Tiger Reserve in Kerala are the reported localities of V. thwaitesii in India (Agustine 1995; Sathishkumar & Sureshkumar 1998). The species have distribution at altitudes 500-1,060 m in moist deciduous to evergreen forests. Therefore, similar habitats in Idukki District of Kerala to Coorg District of Karnataka were surveyed to record geographical coordinates of occurrence points and score population status. Periyar Tiger Reserve, Idukki WS, Thirunelli, and Silent Valley are the protected forests surveyed. The inhabited land mainly covered is along roadways in North and South Forest Division of Wayanadu and adjoining places in Coorg District of Karnataka and Nilgiri District of Tamil Nadu. The surveys were conducted during 2011-2014 periods. The presence of V. thwaitesii was confirmed through close observation of the morphological feature (Image 1) of the species including fruit and flowers with the help of binoculars. Geographical co-ordinates of the presence location were recorded using Garmin GPSMap 60CSx.

Distribution modeling
A total of 37 environmental variables including 19 bioclimatic, six topographic, 11 seasonal climatic, and one vegetation variable (Table 1) influencing survival of a plant species were analyzed for their importance in modeling studies. As peninsular India receive monsoon in four distinct season, i.e., January-February (Dry period),

J TT
March-May (Summer precipitation), June-September (South-west monsoon), and October-December (Northeast monsoon) the monthly precipitation data obtained from world climate (https://www.worldclim.org) was reconstructed to obtain seasonal precipitation variables of the region. All extraction, clipping, and recalculation were undertaken using QGIS software version 3.10 The climatic variables are derived from temperature and precipitation data and thus multi co-linearity always exists among different variables. Therefore, principal component analysis (PCA) is often recommended (Junior & Nóbrega 2018) to control the negative effects of co-linearity and as a more objective solution for the problem of variable selection. The data points of all the 37 variables corresponding to the geo-reference points of V. thwaitesii occurrence in 1-km spatial resolution were extracted using point sampling tool in QGIS and PCA analysis undertaken in SPSS 16 software to sort out the variables having significant contribution to the model. Accordingly, three major components were extracted (

Field Survey
Extensive field surveys revealed the distribution of V. thwaitesii altitudes from 489 m to 1,168 m mostly on evergreen trees exposed to sunlight. A total of 93 occurrence points were recorded in Kerala, Tamil Nadu, and Karnataka. The surveys in Periyar Tiger Reserve (PTR), Idukki Wildlife Sanctuary, and Wayanad revealed inhabited land in Wayanad Plateau as the dominant distribution area and their population near forest segments is meager. Therefore, it is clear that the dominant population in the most ideal habitat was lost due to habitat loss and extension of their population was withheld due to habitat fragmentation.

Modeling studies
Out of the 93 occurrence points recorded during our field surveys, 54 geo-reference points were at 1-km spatial resolution (Table 3). Hunnasgiria in Sri Lanka the reported type locality retrieved from Google Earth was also included for modeling. Thus 55 occurrence points were available for modeling. The 37 variables (19 bioclimatic, six topography, 11 seasonal, and one vegetation) subjected to principal component analysis, extracted into three components explaining 96.5% variance ( Table 2). The final model based on the four selected variables revealed summer precipitation in the months of March to May (PMM), annual precipitation (Bio_12), and annual mean temperature (Bio_1), have significant contribution to the model (Table 4). Jack Knife test (Figure 2) revealed PMM as the environmental variable with highest gain and significant drop when it is omitted. The species flowers during April-May when receives summer showers, after a dry period and thus precipitation during March-May may be critical for the survival and spread of the species. In the distribution model generated (Figure 3), one occurrence point at 489 m altitude in Kannur District was in the least probable region. This is the only one occurrence record below 700 m still establishing a solitary colony with 15 individuals with the incidence of fruit set and new recruit. However, 41% of the points fall in high probable (0.8-1) region and 32.6% in the 0.6-0.8 region (Figure 4). The whole peninsular India was modeled where only 0.59% area (11,561 km 2 ; Figure 4) confined to the Western Ghats region of Kerala, Karnataka, and Tamil Nadu emerged as suitable habitats for V. thwaitesii. Out of the total area (2,557 km 2 ) in the high probable region (0.6-1.0), Wayanad District of Kerala and Coorg District of Karnataka together constitute 1,461 km 2 (57%). Idukki District (353), Nelliyampathy (195), and Sholayar (137)  In addition, a small area in Central Province of Sri Lanka also has habitat suitability. Different variables extracted in the two principal components have equal variance of temperature and precipitation in the occurrence localities and thus the distribution models appeared very similar; V. thawaitesii inhabits cooler regions receiving annual mean temperature in the range 19.75-24.3 o C. Similarly, annual precipitation in the occurrence points are in the range 2020-4794 (Table 5). The occurrence points also    (Table 4) and jackknife test (Figure 2).

DISCUSSION
Conservation assessment essentially requires sufficient field surveys and gathering of primary data on the distribution of species and their population attributes. This is an exhaustive process requiring substantial effort and investment of human and financial   resources which has not been taken seriously in India and therefore the availability of primary data is limited. Conservation planning, action monitoring and evaluation of a species initially require prioritization through threat assessment (Mace & Lande 1991;Master 1991;Moran & Kanemoto 2017). Vanda thwaitesii is an epiphytic orchid endemic to Indian peninsula and Sri Lanka hitherto unknown for over 100 years until 1998 when a few populations were discovered from Kerala (Sathishkumar & Sureshkumar 1998). Based on the information on herbarium data, the Government of India notified the species as endangered and invited studies on all aspects of the notified species for propagation and conservation. Based on the extensive field surveys, we found significant populations of the species outside protected forests and the present habitats are highly fragmented. Thus conservation through rehabilitation or translocation into safer localities appeared very essential, preventing further loss before their economic value is deciphered. In spite of its unique aromatic and exquisite flowers it is underutilized in breeding as it is lesser known to orchid breeders. The present study delivers a habitat suitability model for conservation of the species prepared as part of a sponsored project supported by DBT, Government of India during 2010-16.

J TT
Niche modelling is an economical and effective tool to prepare guide maps for intended plant survey and delineation of conservation areas for selected species (Adhikari et al. 2012) thus improving the availability of primary data on the distribution of species and their population attributes for improved threat assessment and more accurate categorization of endangered species (Adhikari et al. 2018). Georeferenced occurrence points and environmental data pertaining to the distribution area are the two prerequisites for habitat modeling. The 55 occurrence points used are more than sufficient to undertake such study. Environmental variable as 19 bioclimatic variables and digital elevation are observed as biologically meaningful to define ecophysiological tolerances of a species (Graham & Hijimans 2006;Murienne et al. 2009) and generally utilized for modeling studies. Topographic variables are derived from digital elevation model from satellite data and are also known to influence the distribution of plant species (Wang et al. 2014). However, those variables and normalized digital vegetation index, another remote sensing data variable are least influencing the distribution of V. thwaitesii as they did not extract as a significant factor in PCA. Nevertheless, the bioclimatic variables and seasonal climatic variables are derived from temperature and precipitation data and thus often expresses multi colinearity and thus often difficult to select the decisive environmental variables and their contributions. Principal component analysis and regression analysis executed on them reduced the number of factors into three still explaining 96.5% variance. Environmental variables showing VIF less than 10 are also included as the existing co-linearity is less significant (Naimi et al. 2014). In this study, we could extract three components with six, four, and one variable and excluding variables with VIF greater than 10, retained four variables (Table  2) to determine habitat suitability for V. thwaitesii in peninsular India.
The distribution map thus obtained showed high resolution with AUC 0.997 and therefore is having high prediction efficiency. The AUC ranges from 0.5-1.0 for models that are no better than random to perfect predictive ability. It is also clear that 73.3% of the occurrence records falls in the high suitability region (0.6-1.0) and only 1% in the least suitable region (0-0.2).
Besides, a few occurrence records in Chikkamagallur, Hassan and Nelliyampathy later gathered from online resources and not used for modeling fall in the suitable area predicted. Thus, the presence of V. thwaitesii could be confirmed in most of the prediction area proving the robustness of the model. Elevation and temperature are often the most determining variable in habitat modeling as revealed in the terrestrial orchids as Dactylorhiza hatagirea (Wani et al. 2021) and Oeceoclades maculata (Kolanowska 2013). Precipitation is also an important variable that influences habitat modeling in some species, such as Habenaria suaveolens (Jalal & Singh 2017) and Zanthoxylum armatum (Xu et al. 2019). However, it seems that the determining factor in habitat modeling is species-specific and not exclusively confined to any of the variables. Work undertaken in epiphytic orchids as Vanda wightii from Western Ghats, India (Decruse 2014) and Vanda bicolor from northeastern region of India (Deb et al. 2017) indicate that precipitation warmest quarter (Bio_18) is the most influential factor in the model. While Polystachya concreta, a pantropic epiphytic orchid (Kolanowska et al. 2020), is reported to prefer different temperature and precipitation factors as far as Asian, African, and American regions are concerned. As reported, temperature seasonality (Bio_4), isotheramality (Bio_3), and precipitation seasonality (Bio_15) have significant contribution in Asian region while temperature factors alone (Bio_2, 4, 1) in American region and precipitation factors alone (Bio_12,18,14) in African region. Therefore, a single factor can't be considered important for global distribution of a particular species. The model output

J TT
for peninsular India and Sri Lanka in the present study revealed that temperature and precipitation contribute significantly to the distribution of V. thwaitesii providing a robust model with high prediction efficiency. Therefore, the predicted areas in protected forests are highly suitable for conservation of V. thwaitesii. The generated model is also a guide map to find new populations from locations other than those reported earlier.

CONCLUSION
Vanda thwaitesii can sustain in the regions receiving 3,400 mm average annual precipitation and 290 mm in summer. In addition, they prefer cool climate with an average annual mean temperature 22.27 o C. The most ideal climatic conditions (0.6-1.0 class) prevail mostly in Wayanad, Idukki, and Palakkad districts of Kerala in addition to Coorg District of Karnataka. However, most of the modeled area in Western Ghats is outside protected forests. Still, there are sufficient locations in reserve forests in Kerala and Karnataka in addition to the sanctuaries as Silent Valley National Park, Idukki Wildlife Sanctuary, Periyar Tiger Reserve, and Brahmagiri Wildlife Sanctuary for reinforcement of diversity from vulnerable locations as inhabited land, plantations, and wayside trees.