Skip to main content

Data from: Factors influencing herpetofauna abundance and diversity in a tropical agricultural landscape mosaic

Cite this dataset

Basu, Parthiba; Ghosh, Deyatima (2020). Data from: Factors influencing herpetofauna abundance and diversity in a tropical agricultural landscape mosaic [Dataset]. Dryad.


Agricultural intensification and the associated factors, including land transformation, are among the major global threats affecting biodiversity especially herpetofauna. However, little information is available about how different factors shape herpetofauna species assemblages in agricultural landscape at different spatial scales from patch (125 – 250m) to the landscape (500 – 1000m).

We assessed the diversity of amphibians and reptiles in areas under low and high degrees of agricultural intensification and explored different factors regulating diversity at different spatial scales using four sampling methods.

Diversity and abundance of amphibians varied significantly between the two zones, but not for reptiles. Agricultural intensification Index (AII), calculated based on agrochemical use and area under agriculture at 250m scale, seemed to affect amphibians both at patch as well as at 500m and 1000m landscape scales. The AII influenced reptilian diversity only at patch and 500m scales. Vicinity of natural forest had a stronger influence on reptilian abundance. Semi-natural vegetation impacted herpetofauna diversities at larger spatial scales. The extent of water bodies influenced the reptilian abundance at 250m patch scale and amphibian abundance both at 250m and 1000m scale.  Fallow lands affected only reptilian diversity at all spatial scales. Plantation affected amphibian at all scales but reptiles only at the landscape scale. Habitat heterogeneity regulated only amphibian diversity.

These results highlight the fact that different patch and landscape-scale factors regulate the diversity of reptiles and amphibians differentially. Such scale specific information will crucially inform future conservation action for the herpetofauna in the agricultural landscape.



The study was conducted in Balasore District, Odisha state in Eastern India (20.9517° N, 85.0985° E). The average altitude of the district is 19.8m. Balasore district covers an area of 3,634 km2. Broadly the district can be divided into three geographical regions- the Coastal belt, the inner alluvial plain and the North-Western hills. It is bounded by Midnapore district of West Bengal in its North, the Bay of Bengal in the east, Bhadrak district in the South and Mayurbhanj and Keonjhar districts on its western side. The district receives an average rainfall of 1583 mm and the average temperature varies from 22°C to 32°C. (Balasore Official Website -

The study area was primarily a lowland paddy growing area and except areas of high agricultural intensity, the cultivation is rainfed. Areas of low agricultural intensity were flanked by the Kuldiha Wildlife Sanctuary and had substantial natural vegetation in the landscape. The high agricultural intensity areas had little or no natural vegetation. Natural vegetation in the landscape were either shrub or occasional trees e,g,Shorearobusta, Madhucaindica, Croton roxburghii, Holarrhenaantidysenterica, Gmelina arborea etc.

Pesticide application is very high in the intensive agricultural areas and as revealed from our own survey farmers mostly apply Cypermethrine in the rice paddy fields. Frequency of pesticide application is either negligible or low in low cropping intensity rain fed areas. Fertilizer is applied two or three times in both high and low intensity cropping areas.

We classified the agricultural intensification zones based on- 1) An agricultural intensification index (Flohre et al., 2011) formed with three parameters- i) Pesticide input per unit area ii) Fertiliser input per unit area and iii) area under agriculture at patch scale and 2) Extent of natural vegetation at landscape scale (1000 m).


Sampling was carried out from March 2016 to March 2017 in a total of thirteen sites. Each site was separated with a minimum aerial distance of 5 km, such that each sites were independent sampling points and there was no spatial replicate. Eight sites were classified as low (Low zone) and the remaining fiveas high intensification zones (High zone).

Four sampling methods – active, passive, area constrained active search and transect walking were used for detecting herpetofauna. A combination of survey methods was used to increase the efficiency for sampling these cryptic species (Michael et al., 2012, Carpio,Cabrera& Tortosa,2015; Sung,Karraker& Hau,2011).

Passive sampling included drift fence, pitfall and double-ended funnelsestablished within a 10m X 10m quadrat at three sub-sites located 100m apart in the form of a triangle i,e each site had three of these passive trapping arrays. Four drift fences each of length 7.6m (Cassani et al., 2015) and height 91cm were installed in an “X” shape (Enge, 1997; Ryan et al., 2002; Semlitsch&Moran, 1984) with 4 pitfall traps (made from 20L of bucket, Greenberg et al., 1994) buried at each corner of the “X” and one at the centre of the trapping unit, making a total of 5 pitfall traps for each passive sampling unit. We established 8 double-ended funnels at each trap station, one on either side of each drift fence.

Along each 100 metres distance between the three passive sampling stations, we deployed cover boards (made from Palmyra leaves) of 1m X 1m strating from the boundary of the passive trap station and then installed at an interval of 20m making a total of 18 cover boards for active sampling (Fig2). Each trap was left open for 72 hours making a total trap effort of 171 for each site and 2,223 for all 13 sites. We conducted a sampling session of 3 days per field where every day we performed, active search, transect walk. Passive traps and active traps were left open for 72 hours and every day traps were checked and data gathered. Since we could not afford to perform night sampling, deploying passive traps was an alternative to increase the sample strength. Sampling was performed from  7.00till 15.00hr and this time was maintained to minimise theeffect on observation.

The centroid of the triangle was used to establish a buffer ring of radius 125m demarcating an area of 5 hectares where we undertook area constrained active searches. We searched all the potential habitats by bush beating or turning up rocks.

We conducted a transect walk along the 250m diameter of the circle with two persons walking simultaneous checking for any herpetofauna within a distance of 2meters on either side of each other. All samples observed were collected, measured for snout-vent length, weighed, and photographed for all profiles at the end of each sampling period and were released at a distance of ~3km from their site of capture (Piatti, Souzab& Filho, 2010) in and around similar paddy fields to avoid overestimation of abundance. A representation of the sampling method is provided in Figure 1.

Landscape data

We considered 125 to 250m radius around the sampling plots as patch scale and 500 to 1000m as the landscape scale.

To determine the effect of existing land uses at different spatial scales, the area of land use was estimated using a Landsat-image through ArcGIS ver. 10.2.2. We calculated the proportion of seven landscape features - natural forest, agricultural lands, plantation, semi-natural vegetation, human settlement, fallow lands and water body within 125 meters, 250 meters, 500 meter and 1000 meter buffer sectors. Cases where landscape attributes could not be appropriately assigned to the above mentioned land-use types owing to the low resolution of these maps, we used Google Earth Pro image to overlay on the map for correctly categorising the land-use types. Landscape heterogeneity was calculated using the land use proportion datawith Simpson’s D index (McMahon, Purvis & Whelan, 2008).Figures 2,3,4 and 5show our study area and the supervised classification image prepared from ArcGIS along with magnified views of one site under High and Low intensification respectively,displaying the landscape buffer scales.

Statistical Analysis

We used Shanon-Weiner index to compute the diversity of herpetofauna in the two intensification zones. Diversity index between low and high agricultural intensification zones was compared using Hutcheson’s T-test (Gardener, 2017). Species abundance of herpetofauna and for reptiles and amphibians was compared using Man-Whitney U test. We performed a sample-based rarefaction to compare species richness between the two intensification zones while controlling for unequal sampling. Species richness estimation using Jackknife and Chao1 estimators were used to detect the actual species richness of the community as observed species is believed to represent a biased subsample of the real community (Colwell, 2009).

We calculated Agricultural Intensification Index (AII) for each site with three indicators- pesticide input, fertiliser input and agricultural land use at patch scale (data obtained from ArcGis - 250m radius) adapted from Andreas et al. 2011. The method of calculating AII was the following:

AIpesticide = {(Yi-Ymin)/(Ymax-Ymin)}/n*100

Yi= the pesticide input of a particular field

Ymin= the minimum pesticide input among all the 13 sites

Ymax = the minimum pesticide input among all the 13 sites

AIfertiliser and AIagriculture(Agricuturallanduse) were calculated in the same way.


(3 as a multiplier in the denominator is for the three indicators used in calculating the AII)

Amount and the frequency of application of pesticide and fertiliser were collected from five farmers in each study site. This data was standardised as pesticide and fertiliser inputs per acre of the field and were further used for estimating the AII.  This AII valueswerefurtherused in the GLM models along with various land-use elements as a predictor for abundance and richness of reptiles and amphibians at four different spatial scales e.g., 125m and 250m radius (patch) and 500m and 1000m (landscape).Generalised linear models (GLMs) were generated with negative binomial error distribution due to overdispersion in data (McCullagh &Nelder, 1989, White &Bennetts, 1996, Alexander,Moyeed& Stander, 2000;Lindén&Mäntyniemi, 2011).

All variables for the statistical models were checked for multicollinearity, and those with variation inflation factor values >10 (Hair, Anderson & Tatham, 1995) were dropped from the model (James et al., 2014, Bruce and Bruce, 2017).  Heteroskedasticity was detected using Bausch pagan test. Data were analysed in R (version 3.5.2) using “vegan” (version 2.4-2), “MASS”, “car”, “rich” packages.


Council of Scientific and Industrial Research

Rufford Foundation, Award: 18506-1