MATERIALS AND METHODS

Land cover analyses
Land cover analysis was done to delineate the areas under vegetation and soil (non- vegetation). The land use analysis was carried out for the region to identify the use of land, emphasising the functional role of the land in economic activities. Land use patterns reflect the characters of a society’s interaction with the physical environment, while land cover in its narrowest sense often designates only the vegetation cover extent on the Earth’s surface, as the spectral reflectance in green band (visible range of EM spectrum) and near-infrared would represent the photosynthetically active vegetation. A cloud free IRS-1C satellite multi spectral imagery of LISS III sensors with spatial resolution of 23.5m was used for the analyses. A land cover analysis was done through computation of vegetation indices, which was either Slope based or distance based (depending on the landscape type - hilly, arid, etc.). The Normalised Difference Vegetation Index (NDVI) was used to delineate vegetative from the non-vegetative features. High values in a vegetation index identify pixels covered by substantial proportions of healthy vegetation. Due to inverse relationship between vegetation brightness in the red (R) and near infrared (IR) region, a normalised difference ratioing strategy can be very effective.  Once the extent of vegetation cover was determined, further analyses were done to classify image with field data. FCC image was generated by merging all 3 bands, which aided in identifying the heterogeneous patches for selection of training sites. Attribute information corresponding to these training sites were collected from field using Global Positioning System (GPS). In addition to this, GPS was used to map the tree species, which were sampled along 200 m transect with 20 x 20m quadrant. Gaussian maximum likelihood classifier was used to classify the data based on the attribute data collected from field. The statistical probability of a given pixel present in a specific land use category was calculated and each pixel in the data was categorised into the land use class it most closely resembled or the probability of occurrence.

Sampling strategy
Sampling was done along four transects (Figure 1a) moving away from the reservoir towards the end of the river basin. Undulating topography in the area resulted in non-uniform length of the transects varying from 4 to 32 km. Three sampling plots of 30 sq. m each were located at every 4 km distance along the transects. These plots were placed in a mini-transect of 400 m at a 200 m interval, perpendicular to the main transect resulting in a total of 78 samples, distributed as 31 samples in moist deciduous forests, 16 in semievergreen, 11 in scrub jungles, 10 in acacia plantations, 7 in semievergreen forests, 2 in evergreen forests and one in a pine plantation. At each of these plots the entire package of baits (described below) were laid for collecting ants. Pine plantations, Pinus roxburghii being extremely few in the study area, resulted in only one sample. Predominance of deciduous forests in the study area resulted in the highest number of samples. The Northern region being very close to the dam, the catchment area was very small resulting in only 2x3 samples. A large number of samples towards the east and north were under extreme human pressure.

Sampling methods
Ants were collected from each sample using bait traps, pitfall traps, leaf litter techniques and visual collections. Terrestrial and arboreal baits were placed using 70% honey, tuna fish and fried coconut. Terrestrial baits (T) were placed on the ground and the arboreal baits were tied to a tree at a height of 6 ft from the ground. The terrestrial baits were placed in an equilateral triangle formation maintaining a distance of 30m between two traps. The arboreal baits (A) were alternated such that between a terrestrial honey and tuna fish bait, a coconut arboreal bait trap was laid. A distance of 30m was maintained between arboreal traps. The traps were laid at 0700 hrs and were picked up at 1700 hrs. A 10 cm long cylindrical tube with a mouth-opening diameter of 5 cm was used as a pitfall trap (P). A pit was dug with a mallet and the tubes were placed such that their mouth was flush with the ground. Five such pitfall traps were laid in every sample for a 24h period, 4 in the corners of each plot and one in the center. Each of these traps had 90% ethyl alcohol with 3 drops of glycerin (to prevent quick evaporation). The Burlese leaf litter technique was used in extracting ants from the leaf litter in four, 1x1 m quadrants at each sample. The quadrants were placed in the four smaller quadrants (L), between the pitfall traps and baits in four directions. The entire litter in the quadrant was pushed from the sides to the centre and the litter was picked up and put in a funnel. The funnel was fitted with a vial with ethyl acetate. A powerful light of 60 watts was shone on the litter in the funnel over a 12h period, so that all insects would drop down into the funnel. The insects were then sorted and ants were separated for identification. A visual collection was also done which involved sweep net method, checking in barks, rotting logs and on leaves. For the purpose of quantification, visual sampling was restricted for a period of an hour at each sample. Ants collected from all traps were sorted, cleaned and preserved in 70% ethyl alcohol. The vials were labeled with the place, date of collection and sample codes.

Ants were cleaned with saltwater solution, mounted and identified using the keys in Bingham (1975) and Bolton (1994). Only the workers were counted, disregarding the males and queens. A reference collection is maintained by TMM in the Entomology Laboratory at the University of Agricultural Sciences.

The maps of the study area were digitised using Mapinfo 5.0 (creation of vector layers). IDRISI was used to determine the NDVI index and also to arrive at a second level classification using a supervised classification approach - Maximum Likelihood Classifier (raster analyses). Geographic Resource Analysis Support System (GRASS) was used to build the 3-D image depicting high and low species richness of ants for the entire study area.