Predict Location(s) of Apis dorsata Nesting Sites Using Remote Sensing and Geographic Information System in Melaleuca Forest

: Problem statement: The Asiatic Giant honeybees ( Apis dorsata ) are seasonally abounded throughout the upland and lowland rainforests in South East Asia. The colonies of A. dorsata are found to nest in aggregates on tall bee trees (tree emergent) in the open, as well as, nesting singly in concealed locations when nesting low, especially in the submerged forest of Melaleuca cajuputi as in the vast hectare of Melaleuca forest along the coastal areas of Terengganu. The trunk and branches of the Melaleuca trees are almost upright and the bark are papery and loose that it is unsuitable to serve as nesting support for A. dorsata . Therefore, other trees with sloping branches in the vicinity of the Melaleuca forest are chosen as nesting supports for the A. dorsata colonies. Melaleuca cajuputi and Acacia auriculiformis trees are major sources of nectar and pollen for A. dorsata . Approach: A pilot study were conducted to determine the vegetation distribution area of Melaleuca cajuputi and Acacia auriculiformis , colony distribution and density Apis dorsata colonies using the remote-sensing and field survey by GPS hand held in the submerged Melaleuca forest around the village of Gong Beris, Marang district, Terengganu (Upper left 5°2’19.223”N, 103°10’4.092”E and lower down 4°53’42.31”N, 103°22’37.90”E). Result: Based on the map and field survey, it is clear that almost all Apis dorsata nesting sites were located in the Acacia auriculiformis areas Melaleuca cajuputi is mostly using by Apis dorsata as source of nectar and pollen in Marang district, Malaysia Conclusion: The outcome on the distribution area and feral colonies nesting sites will be used to propose the locating of possible A. dorsata nesting sites and understand better nesting behavior of Apis dorsata to improvement searching ability for honey hunters in Marang district.


INTRODUCTION
Throughout the range of Southeast Asia, upland to lowland forests, giant honey bees (Apis dorsata) are found to build their nests in aggregates on tall bee trees, as well as nesting singly in concealed locations when nesting low. From 13000 BC, based on a rock painting, nests of Apis dorsata at Asia and Southeast of Asia have been exploited to produce the bulk of honey and wax and being traded for ages. In Malaysia, professional honey collectors or Honey Hunters whom using traditional techniques, are collecting honey from Apis dorsata colonies in ages. Honey hunters have to climb steep cliffs or ascend tall "bee trees" by hand-made ladders during the darkness of the night then kept away with smoke and cut away the comb completely for collecting honey. This technique is a very strenuous, laborious and demanding job (Mardan, 1989;Othman, 1997).
In low land rainforest along the coastal corridor of Kelantan and North Terengganu, Melaleuca cajuputi (Gelam) and Acacia auriculiformis (Acacia) are the two dominant trees which are major sources of nectar and pollen for giant honey bee (Apis dorsata) (Kiew and Muid, 1991). The honey from these trees has strong flavor and weak density. It granulates quickly, with grain varying from fine creamy to coarse brown and dark if more than moderate heat is used during extraction, straining or packing; on the other hand, the trunk and branches of the Melaleuca trees are almost upright and the bark are papery and loose that it is unsuitable to serve as nesting support for A. dorsata. Therefore, other smooth barked trees with sloping branches in the vicinity of the Melaleuca forest are chosen as nesting supports for the A dorsata colonies.
Honey from Giant honey bees provides significant supplementary income source to the local people in Marang district. Furthermore, it is important to predict the locations of nesting by Apis dorsata to improve searching ability honey hunters in forest. One of the suitable techniques to understand and predict Apis dorsata nesting behavior is mapping distribution of vegetation in Melaleuca forest and surveys the Apis dorsata bee trees in forest.
The Manual (traditional) mapping method to measure and inventories the vegetation coverage of the Melaleuca forest will take a long time and cost. Better means are needed for land use inventory. In many of the developed countries and some of developing countries, Remote Sensing (RS) and Geographic Information System (GIS) are widely used to provide up-to-date information (Luney and Dill, 1970). Recent advances in RS and GIS technology have become cost effective and affordable, by virtue of the following reasons: (a) satellite images are sufficiently accurate and reliable, (b) changes over time can be identified, (c) computers have the power to rapidly process large quantities of data and (d) object-oriented GIS provide enormous flexibility in storing and analyzing any type of data, providing decision support modeling for effective management (Buchan, 1997).
RS technology is particularly useful tool to produce a broadly consistent database at spatial, spectral and temporal resolution which is useful for forest. As well, RS data can be made effective and economical for periodic preparation of accurate inventories and also for managing and monitoring forests. Different techniques are available for differentiating and mapping forest units. Studying large area, data can be processed in an automated way; for thematic mapping, images are visually interpreted, enabling the forestry experts to use their knowledge and experience (Lillesand and Kiefer, 1994) The majority of studies used satellite RS to map forest types defined on the basis of structural and bioclimatic attributes related to the degree of canopy closure. Roy et al. (1991); Saxena et al. (1992) and Liu et al. (2008) explored forest type identification using remotely sensed data by both visual interpretation and supervised classification techniques. An analysis of the cost effectiveness of these techniques revealed the latter to be more economical. Other studies have concentrated on purely digital classification techniques with refinements to increase the accuracy of classifications (Sudhakar et al., 1996;Tuia and Camps-Valls, 2008;Dixon and Candade, 2008). Classification accuracy may also be increased through the use of contextual and ancillary information, such as historical land use information, site characteristics (i.e., topography) and geobotanical relationships (Brondizio et al., 1996;Paradella et al., 1994;Trivedi et al., 2006;Tuomisto et al., 1994). The additional technique to improve accuracy and also replace information which missed by clouds and shadow of clouds is fusion technique (Hall, 1992;Chen et al., 2007).
Some studies have been conducted to use RS in forest and forest management. Sulong et al. (2002) was used Landsat and aerial photography to mapping mangrove in Kemaman district, Terengganu state and Lim et al. (2009) explored RS to mapping high spatial land cover in Mebok Estuary, Penang, Malaysia. Rebelo et al. (2007) investigated RS and GIS for inventory of wetlands by mapping and change detection in Seri Lanka. Jussof and Pathan (2009) utilized airborne hyperspectral sensing for mapping individual oil palm trees with high accuracy RS technique which can guide researchers to make plan for oil palms forest as one of the economical sources in Malaysia (Rebelo et al., 2007).
GIS is associated with basic term, geography and information system. The literal interpretation of geography is "writing about earth". Write about the Earth, geographic deals with the spatial relationship of land with man (Chandra and Ghosh, 2006). A key tool in studying the spatial relationships is the map which is a graphical property of spatial relationships and phenomena over a small segment of the Earth or the entire Earth. On the other hand, an information system is a chain of operations which consists of from planning the observation to using the observation-derived information in some decision making process (Brondizio et al., 1996).
In that regard, it is important to know the geographic information such as vegetation distribution area of Melaleuca cajuputi and Acacia auriculiformis as two main sources of nectar and pollen for Apis dorsata colonies in Marang district. Also the location and distribution of the colonies in order to determine the density of colonies were used to create a map which can be helpful to understand nesting behavior and predict high possibility locations of nesting sites by Apis dorsata to produce more and matured honey at appropriate time for improving honey products on either the quality on the quantity in Melaleuca forest.
This study sheds light on using remote sensing and geographic information system to understand and predict Apis dorsata nesting behavior in Marang district, Terengganu state of Malaysia.

Study area:
This study was conducted in Marang district, located in the state of Terengganu at the northern east of Peninsular Malaysia (Fig. 1); between upper left of 5°01'N, 103°11'E and lower right of 4° 50' N, 103° 24'E. The district's topography consists of the South China Sea coast, peat swamps, hills and plains. The sub-districts of Rusila, Pulau Kerengga and Merchang lie on the coast with sandy plains gradually giving way to hills in the interior. The Jerong sub-district is mainly hilly, while the sub-districts of Bukit Payong and Alor Limbat consist of rolling plains and peat swamps. The dominant trees species in this area are Mangrove, Melaleuca, Acacia, Rubber and Coconut trees.

Fig. 1: Study area
Image data: Spot-5 image in Multispectral mode with 10 meter resolution and 10×10 m pixel size was used for this current inventory. The image had been taken at 03/05/2007 that was near to field truthing so as to minimize the influence of time on landscape elements (Pohl and Van Genderen, 1998).
Preprocessing: Before image processing based on Richard and Jia (2006) bands 3, 2 and 1 (NIR, Red and Green) which known as False Color Composition (FCC) in order to extract useful information from the optimal band combination with linear enhancement was selected as suitable preprocessing to classify and identify various land cover map in this study area.

Image processing:
In order to extract the information from RS data, especially from multispectral RS data, land use/land cover classification based on statistical pattern recognition techniques which applied to multispectral RS data is one of the most often used methods (Narumalani et al., 2002). Digital image classification uses the spectral information represented by the digital numbers in one or more spectral bands and attempts to classify each individual pixel based on this spectral information.
In either case, classification approaches try to assign all pixels in the image to particular classes or themes. In this research 5 classes were determined based on objectives of study. The classes were (i) Melaleuca cajuputi trees (ii) Acacia auriculiform trees (iii) Water bodies (iv) Non-vegetation areas (v) Cloud/Haze/Shadow. Some parts of maps were covered by cloud in SPOT-5 images that caused cloud be the last class in this part.
For mapping study area into 5 classes, Maximum Likelihood Classification (MLC) algorithm was chosen and performed as one of the powerful algorithms among different supervised classification. Supervised classification and delineation of the study area were based on the name assigned to class according to identification done by ground check, visual interpretation and the homogeneous signature represented by the classes.
Field survey: Acquisition of ground data for this study was conducted in two different periods based on the study objectives. First field data was conducted in August 2008. In situ data was collected from 25 sites from which 15 sites were dominated by Melaleuca cajuputi (Gelam) and 10 sites were selected from Acacia auriculiformis (Acacia) distribution. The locations were determined and recorded by GPS Juno™ ST handheld which is one of the products of Trimble Company with accuracy of 2.5-5 m to collect coordinate point.
The second field data acquisition was conducted from January to April 2009 to determine Apis dorsata nesting sites. In this part all coordinates from nesting sites of Apis dorsata colonies were procured by honey hunters using the GPS Garmin 60CX with an accuracy of 7-10 m. All coordinate data of latitudes and longitudes (WGS84) were then converted to Malaysian Rectified Skew Orthomorphic (RSO) using GPS Pathfinder Office 2.90.
Spatial map: Melaleuca cajuputi and Acacia auriculiformis distribution which derived from SPOT-5 classified image was converted from raster to vector and transferred as a layer to Arc/GIS 9.2. Other layers were integrated inclusive distribution of Apis dorsata nesting sites, access paths such as roads and rivers which derived from digital cartographicmaps with 1:25000 scales were gathered and overlaid each other on ArcGIS v 9.2.
To determine forage range and searching ability, buffer function in ArcGIS v 9.2 was used in two distances, 1 km for forage range and 1 kilometer for searching ability as base on motorbike parking points. Figure 2 shows the output of MLC classification with 91.98% overall accuracy and 0.8635 Kappa coefficient. The performance of MLC technique on SPOT-5 image based on ground truthing was analyzed statistically, visually and graphically (Table 1). The statistical results show that Melaleuca cajuputi is the most dominant species in Marang district in Terengganu state.

RESULTS
Thirty three points were collected as nesting sites by using GPS hand held in study area. All coordinate positions were converted and transferred as a layer to the map (Fig. 3). Based on Apis dorsata nesting sites and Spot-5 image classification, proportion of Apis dorsata nesting sites and locations was determined (Table 2). From total 32 nesting sites, about 17 nesting sites were located in Acacia auriculiformis trees which have witnessed 185 combs. In Melaleuca cajuputi only 10 nesting sites with 158 combs was observed in field survey. To add more, 6 nesting sites were located in other vegetation areas with about 7 combs (Fig. 4). According to Dyer and Seeley (1991) the foraging range of Apis dorsata is within 500-1000 m consequently, buffer with 1000 m was created for each nesting sites to show the forage rang of Apis dorsata in Marang district (Fig. 5).

DISCUSSION
Based on the map and field survey, it is clear that almost all Apis dorsata nesting sites were located in the Acacia auriculiformis areas. It means that Apis dorsata construct their nests on the branches of Acacia auriculiformis trees branches for nesting support. The branch and trunk of Melaleuca cajuputi are too upright to serve as nesting support.   It can be concluded that Apis dorsata cannot construct the nest on Melaleuca cajuputi trees, so it has to nest on the available branches of Acacia auriculiformis. It was assumed that it is related to shape and form of this plant species. Based this study it can be understood that Melaleuca cajuputi is mostly using by Apis dorsata as source of nectar and pollen in Marang district, Malaysia. Similar to the studies by Mardan (2007), Apis dorsata prefer to use Melaleuca cajuputi trees as source of pollen and nectar.

CONCLUSION
This study was conducted to examine application of remote sensing and geographic information system to understand nesting behavior of Apis dorsata in Marang district, Malaysia. Also, it is going to prepare map to indicate possible locations of Apis dorsata nesting sites for honey hunting in study area based on basic requirements of Apis dorsata for nesting such as Acacia auriculiformis as nesting trees and Melaleuca cajuputi as source of nectar and pollen. Our results depicted that the Apis dorsata is prefer nesting on Acacia auriculiformis rather than Melaleuca cajuputi.