Landscape Ecology

, Volume 27, Issue 1, pp 59–71 | Cite as

Methodological, temporal and spatial factors affecting modeled occupancy of resident birds in the perennially cultivated landscape of Uttar Pradesh, India

Research Article


Biodiversity persistence in non-woody tropical farmlands is poorly explored, and multi-species assessments with robust landscape-scale designs are sparse. Modeled species occupancy in agricultural mosaics is affected by multiple factors including survey methods (convenience-based versus systematic), landscape-scale agriculture-related variables, and extent of remnant habitat. Changes in seasonal crops can additionally alter landscape and habitat conditions thereby influencing species occupancy. We investigated how these factors affect modeled occupancy of 56 resident bird species using a landscape-scale multi-season occupancy framework across 24 intensively cultivated and human-dominated districts in Uttar Pradesh state, north India. Convenience-based roadside observations provided considerable differences in occupancy estimates and associations with remnant habitat and intensity of cultivation relative to systematic transect counts, and appeared to bias results to roadside conditions. Modeled occupancy of only open-area species improved with increasing intensity of cultivation, while remnant habitat improved modeled occupancy of scrubland, wetland and woodland species. Strong seasonal differences in occupancy were apparent for most species across all habitat guilds. Further habitat loss will be most detrimental to resident scrubland, wetland and woodland species. Uttar Pradesh’s agricultural landscape has a high conservation value, but will require a landscape-level approach to maintain the observed high species richness. Obtaining ecological information from unexplored landscapes using robust landscape-scale surveys offers substantial advantages to understand factors affecting species occupancy, and is necessary for efficient conservation planning.


Farmland birds Intensity of cultivation Multi-season occupancy Remnant habitat 


Farmlands are increasingly being recognized for their ability to conserve a range of species, but landscape-scale multi-species assessments from tropical non-woody agricultural areas are sparse (Chazdon et al. 2009). Species occurrences on agricultural landscapes are affected by a range of factors at the landscape scale. Birds have been extensively studied and form the basis of much of the understanding. Two of the most important factors appear to be extent of remnant habitat and intensity of cultivation, particularly the amount of area under cultivation (Bennett et al. 2006; Radford and Bennett 2007). Additional agriculture-related variables such as chemical inputs, livestock, percent area cultivated and human density can also affect patterns of abundance and species persistence (Chamberlain and Fuller 2000; Chamberlain et al. 2000). Seasonal variation in species richness and abundance can also occur due to changing conditions caused by different crops (Amano et al. 2008; Elphick et al. 2010; King et al. 2010). Studies generally focus on one or few guilds, though complete species assessments are increasing (Elphick 2004; Amano et al. 2008). Finally, survey methodology is a key consideration. Relatively simple volunteer-based surveys repeated annually using convenience methods have proven invaluable to track bird populations in agricultural areas (Chamberlain and Fuller 2000; Chamberlain et al. 2000). Convenience-based methods such as roadside counts have the advantage of being relatively rapid and cheap, but their efficacy during one-time surveys of multiple species is less understood. Assessments of species persistence on agricultural landscapes rarely, if ever, consider all these elements together. Considering the growing importance of agricultural landscapes for biodiversity conservation, we assessed factors affecting occupancy of resident bird species in a human-dominated, perennially landscape in north India where systematic ornithological studies have been sparse.

Using an explicitly landscape-scale multi-season survey design, we assessed factors affecting occupancy of resident bird species in Uttar Pradesh state, north India. Uttar Pradesh is located along the Gangetic floodplains which is one of four of the most intensively cultivated landscapes in the world. The region has been converted to near-complete cultivation for over 300 years, and Uttar Pradesh has the highest human population of any state in India (Anonymous 2010; Ellis et al. 2010). Previous studies in Uttar Pradesh have focused on remnant forested reserves and on globally threatened species (Javed and Rahmani 1998; Iqubal et al. 2003; Sundar 2009). Even basic ecological aspects such as distribution ranges and habitat relationships that are vital for conservation planning are unknown for most species (see Lakshminarayan 2007). Rural landscape planning here centers on increasing agricultural intensification and ignores potential impacts on birds (Sundar 2011). Continued persistence of numerous bird species (>400 species) across multiple habitat guilds is known from anecdotal information (Sundar and Subramanya 2010), and convenience-based techniques could potentially rapidly provide the much-needed information on the large number of bird species. Large scale declines of habitat are suspected to have caused declines of species across multiple habitat guilds. These include francolins (grassland/scrublands; Madge and McGowan 2002), Ciconia episcopus (wetlands; Hancock et al. 1992), and Ocyceros birostris (woodland; Kinnaird and O’Brien 2007). To ascertain the importance of this landscape for bird conservation and understand determinants of species persistence, we provide a thorough evaluation of occupancy of 56 resident bird species representing various habitats (human habitation, grasslands, scrublands, wetlands and woodlands). We use the multi-season occupancy framework to (1) compare two methods (systematically-located transects versus convenience-based roadside observations) to assess if ecological understanding of species distribution and habitat associations using systematic methods are comparable using cheaper and more rapid survey methods, and (2) assess the impact of agricultural intensification and persisting habitat on occupancy of resident bird species.


Study area

This study was conducted in the state of Uttar Pradesh that has three distinct seasons based on rainfall and temperature regimes: summer (Mar–Jun), rainfall or monsoon (Jul–Oct), and winter (Nov–Feb). Crops were seasonal and primary crops were monsoonal flooded-paddy, winter wheat and mustard, and fields were largely left fallow in the summer. Many areas also have sugarcane and dry, uncultivated scrublands. The human population in Uttar Pradesh is 199 million with a density exceeding 800 people km−2, ca. 70% of the population are farmers, >75% of the land-holding is <1 ha, and the state accounts for 50% of the production of cereal food grains in India (Anonymous 2010; Office of the Registrar General of India 2011).

We employed a double-stratification procedure to select survey sites. First, using a spatial principal component analysis, we simplified the landscape at the district-scale using natural-log transformed agricultural and human presence variables (both known to affect species occurrence); data were obtained from state records for 2003 ( Appendix 1 provides the list of variables, and the variation explained by the first three principal components. Principal component 1 primarily explained variation due to variables that described intensity of cultivation and degree of human presence, and we identified four distinct land-use zones based on this component (Fig. 1b). These corresponded to sugarcane-dominated areas, a zone covering the Himalayan foothills with the distinct terai landscape, areas with the rice–wheat cropping cycle, and an arid scrubland area contiguous with the central Indian highlands (numbers 1–4 in Fig. 1b). Second, because we were interested to assess the impacts of intensity of cultivation (here onwards, “IOC”), we selected the rice–wheat stratum covering 24 districts and further recognized four strata based on percent of land under rice, the predominant crop in the monsoon. Rice fields were used to cultivate wheat during the winter allowing strata to be valid in the winter as well. IOC strata represented increasing levels of homogenization, and were also correlated to human density. We overlaid these districts with 10 × 10 km grids, removed incomplete grids (cut by district borders), and those dominated by large towns or rivers. Bird communities in towns and riverine areas are of conservation interest, but sampling was precluded by the very high human population in towns and monsoonal flooding of rivers, and we focused on the cultivated component of the landscape. We randomly selected (without replacement) 10 grids in each of four degrees of IOC for a total of 40 sampling units. Such stratification procedures using predominant elements of the landscape have assisted to identify areas with relatively consistent characteristics and improve parameter precision during large-scale monitoring (Stillman and Brown 1995; Jongman et al. 2006).
Fig. 1

Sampling design to survey occupancy of birds in Uttar Pradesh, India. See Methods for details. Scale bar and legend (intensity of rice cultivation) are for (c)

Extent of habitat

We classified satellite images (LISS III; 24 × 24 m resolution) corresponding to the closest dates when bird surveys were carried out during winter 2008–2009 to obtain broad land use categories. Cloud cover was present in nine of the 551 grids, constituted <0.1% of land cover and was discarded. We carried out unsupervised classification of the images using ERDAS Imagine version 8.5 with a convergence threshold of 0.95 to extract 90 clusters. Each cluster was then assigned to broad land cover classes of interest using ground-truthed points and Google Earth. Land use classes with similar reflectance (e.g. wetlands and wet fields; fallow fields and habitation) overlapped, and were misclassified in several clusters. These were identified visually, reclassified using on-screen digitization, Area of Interest and recode functionality. Final land cover maps included categories that combined similar habitats corresponding to broad habitat classes used by birds (Ali and Ripley 1989; Grimmett et al. 1999). Total classification accuracy was 82%; the full error matrix is provided in Appendix 2. Total area (ha) of each class was extracted for all 551 grids (Table 1). Grassland and scrublands patches were combined as “scrublands” (see Table 1). In the full complement of grids, the dominant land use was crops (67%). Extents of classified habitats in the four IOC strata varied significantly (multi-response permutation procedure using package “vegan” in R (Oksanen et al. 2010), significance of delta = 0.001) suggesting that human presence and agricultural parameters were useful to obtain ecologically meaningful strata. Extent of agriculture was significantly negatively correlated to the extents of all the other habitats (Pearson’s correlation, P < 0.05), wooded areas were significantly positively correlated to habitation (P = 0.01) and negatively to open areas (P = 0.04).
Table 1

Habitat classes and their extents (rounded values) extracted from winter 2008–2009 LISS III satellite imageries for 551 focal 10 × 10 km grids in Uttar Pradesh, India

Habitat class

Habitats merged for classes

Grid-level extent in ha [Average ± SD (range)]

% Contribution to entire landscape


Standing crop, fallow land

6714 ± 1046 (1739–9083)


Human habitation


654 ± 324 (5–1937)



Uncultivated fallow land, saline wastelands

863 ± 593 (32–3380)



Grass patches, stands of Prosopis juliflora, denuded woodlands.

1010 ± 775 (31–7526)



Lakes, ponds, rivers, seasonal marshes with vegetation including reedbeds, water chestnut and water hyacinth

200 ± 202 (7–1678)



Bamboo groves, fruit groves, plantations, trees

544 ± 553 (0–4979)


Bird surveys

All 40 sampling grids were mapped in detail using 1970 Survey of India topographical sheets, were sub-gridded into four 5 × 5 km units, and 1-km transects were located systematically in the center of each sub-grid (Fig. 1d). Bird surveys were carried out during winter 2008–2009, and summer and monsoon 2009. Road routes traversed to reach transects were kept constant each season, and constituted survey routes for the convenience-based road surveys. On average, 3 km (±4.9 SD) of road was traversed in each sub-grid driving at <20 km h−1. Within a grid, length of road traversed was similar across each sub-grid, but varied between grids due to the existing road network. Entry and exit points varied with each grid depending on the availability of roads. As much as possible, we chose routes that traversed across sub-grids and avoided routes bordering sampling areas. This was more systematic compared to roadside surveys frequently used in the region that employ arbitrary routes (see references in Urfi et al. 2005). Occurrence of a sub-set of 10 bird species covering all the habitat guilds was noted along road routes to compare with data from systematic transects. Systematic survey data were collected from 1-km transect walks. Transects were walked by the same two people at a constant speed using a compass and range finder to maintain a straight route and cover 1 km. It took 1–3 days to complete each grid (mode = 2). Birds seen in a width of 30 m either side were noted, and a width of 150 m was maintained for larger species like herons, storks, cranes and hornbills. All observations were made within 3 h of sunrise avoiding foggy, rainy and windy days, and each grid was surveyed once each season. To avoid systematic directional bias each season in coverage of grids, three sections with similar number of grids were made in the entire study area (Fig. 1c) and order of survey in sections was determined each season randomly without replacement. Factors known to bias detection of birds were minimized by using only presence/absence data, using the same observers during all surveys, surveying during consistent times of the day, temporal stratification by season, and using only visual detections. Observations from both observers were pooled before analyses. Data were collected for 10 species using both methods and an additional 46 species were enumerated during transect walks (see online Appendix 3; nomenclature follows Gill et al. 2009). In both methods presence/absence information for each species was recorded in sub-grids. A full data set from one grid for a species across the three seasons (winter–summer–monsoon) was 110001001110. Each four-digit record represents a season, and “1” represented sub-grids where the species was sighted.

Occupancy modeling

We used the robust multi-season occupancy model (MacKenzie et al. 2003) in Program MARK to estimate seasonal occupancy of resident birds in grids. Estimated occupancy ranges from 0 (species never sighted) to 1 (species seen in all sub-grids of all sample grids). The multi-season framework does not require assuming detection is perfect during visits (detection probability = 1, especially where the species is sighted) and between time periods (here, season). The multi-season model explicitly estimates vital rates of occupancy (\( \varepsilon \): rate of extinction; \( \gamma \): rate of emigration). For the purposes of this study, these represent seasonal variations in occupancy due to changes in crops, and allow estimation of improved total and seasonal occupancy \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \)without requiring the assumption of constant occupancy across seasons. Visits in each sub-grid provided a space-for-time replication with four data points each season per grid. Observations were made within 5 × 5 km sub-grids to estimate parameters across a 10 × 10 km scale. This resolution is large enough to encompass numerous individuals of each species in each sub-grid, yet small enough to model landscape-level species occupancy providing a robust design (see MacKenzie et al. 2006).

We constructed six basic models to assess variations in detection probability (\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{p} \)) and seasonal variations (\( \varepsilon \) and \( \gamma \)). The models assumed variations in detection probability occurring at three temporal levels: at each sampling occasion (\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{p} \)Full), in every season (\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{p} \)S), and no variation (\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{p} \).). These models also assumed occupancy parameters to vary either seasonally (\( \varepsilon \)S, \( \gamma \)S), or no variation (\( \varepsilon \)., \( \gamma \).). The six basic models therefore included the simplest model (\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \).\( \varepsilon \).\( \gamma \). \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{p} \).), the full model (\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \).\( \varepsilon \)S\( \gamma \)S\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{p} \)Full), and four additional models with differing assumptions for, and combinations of, detection probability and occupancy parameters.

Models were ranked in order of parsimony, and we used multi-model selection with Akaike Information Criteria for small sample sizes (AICc) to determine best models (Burnham and Anderson 2002). We used model weights (w; that sum to 1 for all candidate models) and AICc to determine model uncertainty (multiple top models within 2 AICc units) and if the top model amongst the six basic models had relatively high support (w ≥ 0.65). To account for IOC and association with extent of favored habitat, we modeled occupancy as a function of these covariates (\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \)IOC+Hab) using logit link functions expressed as: logit \( (\psi_{i} ) = \beta_{0} + \beta_{IOC} \times (IOC) + \beta_{Hab} \times (\% Hab) \). We assume only one preferred habitat for each species, and the non-crop habitat class used with each species was based on field observations and published information (see online Appendix 3). For example, extent of wetlands was used to estimate habitat coefficients for birds of the wetland guild (e.g., Grus antigone, Amaurornis phoenicurus), extent of woodlands were used for species of the woodlands guild (e.g. Ocyceros bicornis) and so on. If model uncertainty was evident covariates were added to all the basic models before estimating parameters using model averaging. Else, covariates were added to the top model alone and parameters of interest (seasonal \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \), βIOC and βHab) were derived from the top model with covariates. We only retained models and species for which models converged and parameters of interest could be estimated. We avoided overparametrization by only testing the effect of covariates on occupancy, and make no assumptions on their effect on vital rates or detection probability since they were not of immediate interest.

Mapping distributions using estimated occupancy

We developed distribution maps for all species across the 24 districts with grid-level values representing the probability of occupancy \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) using grid-level values of IOC and extent of preferred habitat obtained using satellite image classification with estimated β-values for each species (White and Burnham 1999). To facilitate comparisons between methods, maps were created separately for species that had estimable parameters from both methods. We only used satellite maps from winter 2008–2009, and derived estimates of βHab for this season only. Mapped distributions may therefore vary seasonally.


A total of 229 resident and migratory bird species were recorded. Data was sufficient to estimate seasonal \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) for 56 resident species, and coefficients could be estimated for 48 of these. Modeling yielded strong evidence of seasonal variations (\( \varepsilon \) and \( \gamma \) ≫ 0) for all species notwithstanding survey method.

Comparison of methods

All parameters for both methods could be estimated for eight bird species. Systematic transects yielded inadequate data for C. episcopus to model seasonal occupancy, and habitat coefficients could not be modeled for Francolinus francolinus using roadside observations. Relative to estimates from systematic transects, modeled occupancy using roadside observations was consistently higher and more precise (lower SEs) between seasons for woodland species, and yielded lower estimates for scrubland and wetland species (Fig. 2). Neophron percnoterus was the only species associated with human habitation for which methods could be compared. Transect data for this species suggested high seasonal variation in estimated \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) while roadside observations suggested no seasonal difference. Both methods suggested that N. percnopterus had a weak negative relationship with extent of human habitatation, and the direction of the relationship with IOC varied by method. Grus antigone was the only species with similar direction of association with IOC (negative) and extent of habitat (positive) in both methods, but transect data showed a stronger relationship with both variables (Table 2; Fig. 3). Seasonal estimates of \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) were consistent with method only for the two most widespread species: Ardeola grayii (wetland) and Streptopelia decaocto (woodland). For other species, seasonal occupancy estimates varied by method. Roadside observations, for example, suggested much lower winter \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) estimates for Francolinus pondicerianus, while transect data suggested lower winter estimates for O. birostris (Table 2). Congruent with estimated parameters, mapped distributions using grid-level \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) estimates were similar across methods only for the two most widespread species (Fig. 3). Bias in predicted distribution due to method was most obvious for O. birostris, but considerable differences were apparent for all species (Fig. 3).
Fig. 2

Comparisons of average (+SE) seasonal occupancy estimates \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) using data from systematically located transects (“Transect”) versus convenience-based roadside observations (“Road”) for eight bird species in Uttar Pradesh, India. Species are pooled by habitat guild, and data for one species associated with human habitation is not included here

Table 2

Parameter estimates (with SE) derived from multi-season occupancy modeling for 10 bird species comparing two survey methods (data from systematically located transects, “Transect”, versus convenience-based roadside observations, “Road”) in 24 districts in Uttar Pradesh, India




\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \)Win

\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \)Sum

\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \)Mon



\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \)Win

\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \)Sum

\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \)Mon



F. francolinus

0.02 (0.07)

0.33 (0.07)

0.55 (0.1)

−3.79 (2.44)

0.002 (0.003)

0.04 (0.04)

0.4 (0.08)

0.63 (0.1)



F. pondicerianus

0.9 (0.11)

0.92 (0.07)

0.95 (0.04)

−1.33 (0.84)

0.003 (0.003)

0.45 (0.09)

0.72 (0.07)

0.77 (0.07)

0.25 (0.32)

0.0001 (0.0009)

P. cristatus

0.91 (0.1)

0.75 (0.2)

0.94 (0.05)

−1.26 (1)

−0.002 (0.001)

0.5 (0.09)

0.66 (0.06)

0.66 (0.07)

−0.47 (0.35)

0.0008 (0.0008)

C. episcopus






0.77 (0.07)

0.75 (0.08)

0.72 (0.1)

0.01 (0.4)

0.002 (0.002)

A. grayii

0.86 (0.1)

0.97 (0.04)

0.99 (0.01)

0.05 (0.7)

0.003 (0.007)

0.99 (0.03)

0.98 (0.02)

0.98 (0.02)

−0.66 (1)

−0.006 (0.005)

N. percnopterus

0.35 (0.26)

0.61 (0.28)

0.34 (0.22)

1.85 (2)

−0.008 (0.006)

0.39 (0.09)

0.32 (0.09)

0.27 (0.11)

−0.4 (0.33)

−0.0003 (0.001)

G. antigone

0.8 (0.1)

0.79 (0.12)

0.52 (0.09)

−1.14 (0.53)

0.01 (0.006)

0.59 (0.09)

0.46 (0.08)

0.36 (0.08)

−0.57 (0.35)

0.002 (0.002)

S. decaocto

1 (0)

0.92 (0.12)

0.92 (0.1)

−2.04 (2)

0.009 (0.009)

0.91 (0.06)

0.91 (0.04)

0.91 (0.05)

−1.41 (0.64)

−0.0002 (0.001)

O. birostris

0.18 (0.09)

0.55 (0.12)

0.74 (0.12)

0.33 (0.48)

−0.001 (0.001)

0.63 (0.1)

0.62 (0.09)

0.62 (0.09)

0.18 (0.35)

0.0006 (0.0009)

D. vagabunda

0.35 (0.21)

0.71 (0.11)

0.78 (0.11)

−0.43 (0.88)

0.01 (0.004)

0.66 (0.1)

0.76 (0.09)

0.69 (0.08)

0.71 (0.41)

−0.0004 (0.0009)

Occupancy was derived for each season (Winter 2008-09—\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \)Win; Summer 2009—\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \)Sum; Monsoon 2009—\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \)Mon). Coefficients show relationship of species occupancy with landscape-level intensity of cultivation (βIOC) and extent of habitat (βHab). Inestimable parameters are indicated as “IE”

Fig. 3

Species distribution maps in 24 districts of Uttar Pradesh, India comparing two methods: systematically located transects (left) versus convenience-based roadside observations (right). Distributions are grid-level back-transformed \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \)estimates using coefficients for intensity of cultivation and extent of preferred habitat

Transect-based occupancy estimates and habitat associations

Winter \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) was very low for some species (e.g. F. francolinus, Eudynamys scolopaceus, Merops orientalis, O. birostris) suggestive of seasonal movements (See online Appendix 4). Coefficients (βIOC and βHab) could be estimated for 48 species. Open-area bird species, e.g. Mirafra assamica and Galerida cristata, on average, had positive associations with IOC. Species associated with human habitation, scrublands, and woodlands had negative associations with IOC on average with greater variation in parameter estimates (wider SEs; Fig. 4a). On average, wetland species showed no apparent association with IOC (Fig. 4a). Occupancy of scrubland, wetland and woodland species improved strongly with increasing extent of habitat (Fig. 4b). Open-area species showed no relationship to extent of open habitat, while species associated with human habitation showed a decline in occupancy with increasing extent of habitation (Fig. 4b). Seasonal variations in \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) estimates were apparent for scrubland, wetland and woodland habitat guilds that improved occupancy during the monsoon with flooded rice paddies (Fig. 4c). Seasonal \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) reduced during the monsoon for birds of open-areas and remained similar across seasons only for birds associated with human habitation (Fig. 4c). On average, woodland and scrubland species, in that order, had the lowest seasonal \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) estimates compared to species in other habitat guilds (Fig. 4c; see online Appendix 4).
Fig. 4

Estimated occupancy parameters for 56 bird species in Uttar Pradesh, India using multi-season occupancy modeling. Graphs show point estimates of coefficients affecting species occupancy (averaged across habitat guilds). Coefficients are for (a) intensity of cultivation and (b) extent of preferred habitat (see Methods and online Appendix 1), and (c) seasonal occupancy \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \)

Predicting distributions using estimated occupancy

Using back-transformed values to model \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) over the entire study area, the five most widely distributed species over the entire study area were from different habitat guilds (modeled \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) averaged over 551 grids; Acridotheres tristis, \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) = 1, associated with human habitation; Prinia inornata, \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) = 1, scrubland species; S. decaocto, \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) = 0.97, woodland species; Bubulcus ibis, \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) = 0.96, wetland species; Vanellus indicus, \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) = 0.94, open-area species). Patchiness and areas of highest occurrence varied greatly across species (Fig. 3, see online Appendix 5). Seasonally, 19-25% of the species occurred sparsely (\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) < 0.5) across the entire landscape (e.g., Pycnonotus cafer, O. birostris, Gynmoris xanthocollis). Restricted or very patchy distribution was apparent for some species (e.g., F. francolinus, Ardea purpurea, N. percnopterus; Fig. 3; see online Appendix 5). More than 50% of the species, however, occurred at relatively high levels of occupancy (\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) > 0.75) with widespread predicted distributions (Appendices 4, 5).


We used multi-season occupancy modeling to (1) assess utility of the convenience-based roadside observations to survey species, and (2) determine the current status of 56 bird species and their associations with extent of remnant habitat and intensity of cultivation. Estimates from roadside sampling on a subset of 10 species were consistently different relative to estimates from systematic transects, and appeared to reflect conditions along the road. Occupancy estimates and species associations were therefore not pertinent to the entire landscape. Occupancy modeling suggested that habitat associations and detrimental effects of intensity of cultivation were strongest for species of scrublands, wetlands and woodlands. This suggests that species from these guilds have incurred the greatest losses so far, and are likely to suffer the steepest declines from additional simplification of the landscape to agriculture. These results are consistent with findings in other agricultural areas (Chamberlain and Fuller 2000; Chamberlain et al. 2000; Fuller et al. 2005; Radford and Bennett 2007; Amano et al. 2008). Using data from systematically placed transects on a complement of 56 species, we assessed that 19–25% of the species here are not very widespread (\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) < 0.5). The current IUCN nomination for all of these is “Least Concern” (see online Appendix 3). Common species may therefore not be faring well throughout their distribution range, and the results show the need for habitat improvements for many such species. The results underscore ongoing calls to undertake monitoring efforts to determine status of common and widespread species (Gaston 2010). These findings provide the first evidence of landscape-scale and seasonal effects on resident bird species occupancy in agricultural areas in South Asia.

Comparison of methods

Roadside observations appeared to reflect roadside conditions and this was particularly obvious with woodland bird species. Trees to shade travelers were a common feature along roads in Uttar Pradesh (personal observation). Compatible with this setting, estimated \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) for all three woodland species showed negligible seasonal variation and higher estimated occupancy using roadside observations compared to data from systematic transects. The results, however, provide strong evidence of the conservation value of roadside trees in simplified landscapes such as Uttar Pradesh. Lower estimates of seasonal \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) from roadside counts for other habitat guilds were consistent with smaller extent of other habitats (scrublands, wetlands and woodlands) along roadsides (Table 2; Fig. 3). Roadside observations therefore appear to be prone to inconsistent biases for species from different habitat guilds. For example, the globally-threatened G. antigone would appear to be much rarer, and O. birostris would appear to be far commoner, than they are in the larger landscape. Diametrically opposite conclusions were drawn regarding species associations with IOC and habitat with differing method for the majority of species. Temporal variations in bird occupancy using roadside methods may therefore reflect changes in roadside conditions and provide erroneous understanding of species status relative to the entire landscape. We therefore suggest that the additional effort to design and use systematic methods, especially during one-time surveys in previously unexplored areas, will be worth it since ecological information pertinent to the larger landscape can be obtained.

Species occupancy and habitat associations

Scrubland and woodland species had the lowest average \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) estimates, were strongly negatively associated with IOC, and positively with extent of habitat (Fig. 4). Birds from these two habitat guilds may have experienced the greatest losses in Uttar Pradesh, and are likely to be most sensitive to agricultural intensification. Open-area species were positively related to IOC and had relatively high estimated \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) suggesting that future increases in agriculture may benefit species of this guild. Wetland species had reasonably high estimated \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) despite a very low occurrence of wetlands (2% of the entire landscape; Table 1). This suggests that many wetland species here are able to use croplands. However, their strong positive relationship with extent of wetlands suggests that habitat losses have been, and will be, detrimental to species from this habitat guild as well. Despite intensive and long-term perennial cultivation, and negligible effort to improve conditions for bird species, more than 50% of the species were still relatively widespread (\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) > 0.75) suggestive of a high conservation value of this landscape. Widespread species include the globally-threatened G. antigone that was unambiguously associated negatively with increased IOC but positively with extent of wetlands (Table 2). These findings confirm suspicions that wetland attrition resulting from agricultural intensification may be responsible for population declines of this species (BirdLife International 2003). Other species with relatively high occupancy rates like Passer domesticus are suspected to be declining, or have declined, in other parts of their distribution range (Gaston 2010). Occupancy of P. domesticus was strongly negatively associated with IOC in Uttar Pradesh. This space-for-time replication design suggests that agricultural intensification may have caused declines of this species.

Several species assumed to be adapted to human habitation showed negative associations with extent of habitation (e.g., M. migrans, Acridotheres ginginianus, Gracupica contra; see online Appendix 4). However, seasonal estimates of \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) remained unchanged on average only for this habitat guild congruent with the unchanged extent of habitation (Fig. 4c). This result suggests that while these species are strongly associated with human habitation, they likely interact with multiple components of the landscape as birds in farmlands often do (Fuller et al. 2005). While scrubland birds on average showed a positive association with habitat, strong negative associations were apparent for some scrubland species (e.g., F. francolinus, Centropus sinensis, Turdoides malcolmi, Saxicola caprata; see online Appendix 4). The prevalence of a number of sparsely available habitat classes necessitated their clumping into “scrublands” (see Table 1), and producer’s accuracy of the classified imagery was low (see online Appendix 2) likely clouding habitat associations of individual species. Alternatively, these species may have positive associations with extent of croplands (that was negatively correlated with the extents of all habitat types), and more detailed studies are required to clarify this. Occupancy estimates of most habitat guilds improved during the monsoon (Fig. 4c) when flooded rice paddies dominate the landscape. This result is congruent with findings that flooded rice supports the largest number of bird species compared to other crops in South and South-east Asia (Amano et al. 2008; Sundar and Subramanya 2010). Despite the apparent positive effects of flooded-rice paddies, retaining natural habitat will be necessary to maximize species persistence. These results are similar to findings from other agricultural landscapes where extent of habitat was a strong driver of species persistence and abundance across different habitat guilds (Richardson and Taylor 2003; Elphick 2004; Radford and Bennett 2007; Amano et al. 2008).

Distribution mapping

Ideally, a second-stage of field sampling in a new selection of grids to validate modeled occupancy would increase the precision and utility of results and maps. Some patterns and trends, however, are strongly apparent. The study provided support for suspicions of declines for F. francolinus, C. episcopus, N. percnopterus, and O. birostris. These species have very wide distribution ranges in South Asia (Ali and Ripley 1989; Grimmett et al. 1999), but estimated occupancy (\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) < 0.4) and mapped distributions in Uttar Pradesh (Fig. 3) reflected sparse and patchy distributions that render them prone to further declines. Suspected reasons for decline of these species, respectively, include hunting (Madge and McGowan 2002), decline of wetlands (Hancock et al. 1992), possible effect of the veterinary drug Diclofenac (Cuthbert et al. 2006), and the deterioration (fragmentation and reduction) of woodlands (Kinnaird and O’Brien 2007). Mapped distributions indicated high spatial variability among species, and no single part of the state appeared ideal for all the focal species (see Fig. 3; online Appendix 5). This suggests that while improvement of habitat at individual sites will undoubtedly be useful, an explicit landscape approach will be necessary to enable continued and improved species persistence. Predictive mapping using reliable habitat relationships are missing for South Asian birds (Lakshminarayan 2007). This study suggests that a thorough understanding of distributions and habitat associations based on robust field sampling is important to assess species status. Long-term temporally replicated monitoring assessments of birds are largely absent in South Asia and we may be missing early opportunities to identify species and habitats that require conservation attention.

Methodological caveats

Estimated bias for seasonal \( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) met “good precision” standards (SE\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \)/\( \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{\psi } \) < 0.3; Bailey et al. 2004). However, for few species, bias in estimated coefficients describing species-habitat associations was wide, and many more species had low precision for coefficients of species-IOC associations. Such uncertainty can be indicative of low sampling frequency (MacKenzie et al. 2006). It could also reflect differences in relative effects of habitat and IOC on species occurrence. This is being investigated in a separate paper.

Using space-for-time replication can provide biased estimates of occupancy for species especially when detection probabilities are low (Kendall and White 2009). Biases due the sampling design in our study are, however, likely minimal due to the following reasons. (1) The simplified nature of the landscape with open crops dominating all sampled areas renders sub-units similar. (2) The small sampling time-frame (sampling of each grid was completed within 3 days) reduced bias due to non-closure considerably. (3) Biases due to the selection of sample grids without replacement were minimized by the sequential visiting of sub-grids, because birds are mobile (allowing detection during each sampling occasions), and because all sub-units were sampled. (4) Spatial dependence of sub-units was likely low due to the size of sub-grids (5 × 5 km). (5) Finally, though decomposing the detection process given spatial replication can be of biological interest (Hines et al. 2010), it was not of interest in this study. To assess if problems due to space-for-time sampling were obvious in the analyses, we compared modeled occupancy estimates for pairs of similar species with comparable detection probabilities across different habitat guilds that had obviously differing occupancies in the field. For example, Psittacula krameri (woodland) was the more widespread species compared to P. cyanocephala, and modeled occupancy reflected the same pattern. Similarly, occupancy of Bubulcus ibis > Egretta garzetta (wetland), Accipiter badius > Butastur teesa (woodland), Corvus splendens > C. macrorhynchos (human habitation), P. inornata > P. socialis (scrubland), and occupancy of S. caprata was > Saxicoloides fulicatus (scrubland; see online Appendix 4). Results appear generally useful to provide an understanding of factors affecting species occupancy.


The study showed that despite high human density, long history of cultivation, extensive simplification to non-woody agriculture, and absence of specific efforts to conserve birds, agricultural landscapes can have high conservation values. The results underscore the need for regional planning to explicitly include improvement of non-crop habitats and consider requirements of species occurring outside of formal protected. Carefully designed surveys can aid to identify species that have the potential to be used as flagship or umbrella species to achieve and monitor impacts of habitat improvements. In Uttar Pradesh, for example, widely occurring charismatic species represented varied habitat guilds: F. pondicerianus (scrubland), P. cristatus (woodland; officially recognized as India’s national bird), G. antigone (wetland; officially recognized as the state bird of Uttar Pradesh), and O. birostris (woodland). Finally, human-dominated and agricultural areas tend to be dominated by common and widespread species, but absence of research attention on such species may lead to assumptions that the species are faring well. This may be incorrect as we demonstrate, for example, that at least three species appear to require immediate conservation intervention. Globally, vast tracts of agricultural areas, many similar to Uttar Pradesh in having high human densities and non-woody cereal crops, remain poorly surveyed for biodiversity (Chazdon et al. 2009; King et al. 2010). Improving knowledge on the requirements of persisting biodiversity on these agricultural landscapes is necessary. Using robust landscape-scale surveys offer considerable advantages, and information from such studies can aid in developing efficient conservation plans.



K.S.G.S. thanks the following for their funding support: the Bell Museum via the University of Minnesota (UMN; Avian Conservation Fellowship, Dayton-Wilkie Fund for Natural History), the International Crane Foundation (ICF), the Kushlan Research Award in Ciconiiform Research and Conservation via the Waterbirds Society, The National Geographic Society’s Conservation Trust Grant, and graduate student fellowships via UMN. For support during field work we thank N. Rothman, B. Wright, J. Zajicek, ICF, the Uttar Pradesh Forest Department, the Wildlife Protection Society of India, and field assistants. We are very grateful to the farmers who permitted field work on their lands. For helpful discussions during the design phase we thank T.W. Arnold, F.J. Cuthbert, and N. Jordan. We thank S. Bagchi, S. Galatowitsch, J. Harris, and two anonymous reviewers for helpful comments on the manuscript.

Supplementary material

10980_2011_9666_MOESM1_ESM.pdf (1.3 mb)
Supplementary material 1 (PDF 1291 kb)


  1. Ali S, Ripley SD (1989) Handbook of the birds of India and Pakistan. Oxford University Press, Bombay, IndiaGoogle Scholar
  2. Amano T, Kusumoto Y, Tokuoka Y, Yamada S, Kim E-Y, Yamamoto S (2008) Spatial and temporal variations in the use of rice-paddy dominated landscapes by birds in Japan. Biol Conserv 141:1704–1716CrossRefGoogle Scholar
  3. Anonymous (2010) Agricultural statistics at a glance 2010. Directorate of Economics and Statistics, Department of Agriculture and Cooperation, Ministry of Agriculture, Government of India. Accessed Apr 2011
  4. Bailey LL, Simons TR, Pollock KH (2004) Estimating site occupancy and species detection probability parameters for terrestrial salamanders. Ecol Appl 14:692–702CrossRefGoogle Scholar
  5. Bennett AF, Radford JQ, Haslem A (2006) Properties of land mosaics: implications for nature conservation in agricultural environments. Biol Conserv 133:250–264CrossRefGoogle Scholar
  6. BirdLife International (2003) Saving Asia’s threatened birds: a guide for government and civil society. BirdLife International, CambridgeGoogle Scholar
  7. Burnham KP, Anderson DR (2002) Model selection and multimodel inference. A practical information-theoretic approach, 2nd edn. Springer, New YorkGoogle Scholar
  8. Chamberlain DE, Fuller RJ (2000) Local extinctions and changes in species richness of lowland farmland birds in England and Wales in relation to recent changes in agricultural land-use. Agric Ecosyst Environ 78:1–17CrossRefGoogle Scholar
  9. Chamberlain DE, Fuller RJ, Bunce RGH, Duckworth JC, Shrubb M (2000) Changes in the abundance of farmland birds in relation to the timing of agricultural intensification in England and Wales. J Appl Ecol 37:771–788CrossRefGoogle Scholar
  10. Chazdon RL, Harvey CA, Komar O, Griffith DM, Ferguson BG, Martínez-Ramos M, Morales H, Nigh R, Soto-Pinto L, van Breugel M, Philipott SM (2009) Beyond reserves: a research agenda for conserving biodiversity in human-modified tropical landscapes. Biotropica 41:142–153CrossRefGoogle Scholar
  11. Cuthbert R, Green RE, Ranade S, Saravanan S, Pain DJ, Prakash V, Cunningham AA (2006) Rapid population declines of Egyptian vulture (Neophron percnopterus) and red-headed vulture (Sarcogyps calvus) in India. Anim Conserv 9:349–354CrossRefGoogle Scholar
  12. Ellis CE, Goldewijk KK, Siebert S, Lightman D, Ramankutty N (2010) Anthropogenic transformation of the biomes, 1700 to 2000. Glob Ecol Biogeogr 19:589–606Google Scholar
  13. Elphick CS (2004) Assessing conservation trade-offs: identifying the effects of flooding rice fields for waterbirds on non-target bird species. Biol Conserv 117:105–110CrossRefGoogle Scholar
  14. Elphick CS, Taft O, Lourenço PM (2010) Management of rice fields during the non-growing season. Waterbirds 33(Special publication 1):181–192CrossRefGoogle Scholar
  15. Fuller RM, Devereux BJ, Gillings S, Amable GS, Hill RA (2005) Indices of bird-habitat preference from field surveys of birds and remote sensing of land cover: a study of south-eastern England with wider implications for conservation and biodiversity assessment. Glob Ecol Biogeogr 14:223–239CrossRefGoogle Scholar
  16. Gaston KJ (2010) Valuing common species. Science 327:154–155PubMedCrossRefGoogle Scholar
  17. Gill F, Wright M, Donsker D (2009) IOC World Bird Names (version 2.9). Accessed Dec 2010
  18. Grimmett R, Inskipp C, Inskipp T (1999) A guide to the birds of India, Pakistan, Nepal, Bangladesh, Bhutan, Sri Lanka, and the Maldives. Butler & Tanner Limited, Frome, Somerset, United KingdomGoogle Scholar
  19. Hancock JA, Kushlan JA, Kahl MP (1992) Storks, ibises and spoonbills of the world. Academic Press, UKGoogle Scholar
  20. Hines JE, Nichols JD, Royle JA, MacKenzie DI, Gopalaswamy AM, Kumar SN, Karanth KU (2010) Tigers on trails: occupancy modeling for cluster sampling. Ecol Appl 20:1456–1466PubMedCrossRefGoogle Scholar
  21. Iqubal P, McGowan PJK, Carroll JP, Rahmani AR (2003) Home range size, habitat use and nesting success of swamp francolin Francolinus gularis on agricultural land in northern India. Bird Conserv Int 13:127–138CrossRefGoogle Scholar
  22. Javed S, Rahmani AR (1998) Conservation of the avifauna of Dudwa National Park, India. Forktail 14:57–66Google Scholar
  23. Jongman RHG, Bunce RGH, Metzger MJ, Mücher CA, Howard DC, Mateus VL (2006) Objectives and applications of a statistical environmental stratification of Europe. Landscape Ecol 21:409–419CrossRefGoogle Scholar
  24. Kendall WL, White GC (2009) A cautionary note on substituting spatial subunits for repeated temporal sampling in studies of site occupancy. J Appl Ecol 46:1182–1188Google Scholar
  25. King S, Elphick CS, Guadagnin D, Taft O, Amano T (2010) Effects of landscape features on waterbird use of rice fields. Waterbirds 33(Special publication 1):151–159CrossRefGoogle Scholar
  26. Kinnaird MF, O’Brien TG (2007) The ecology and conservation of Asian hornbills: farmers of the forest. The University of Chicago Press, Chicago & LondonGoogle Scholar
  27. Lakshminarayan S (2007) Taking Indian ornithology into the information age. Indian Birds 3(4):122–137Google Scholar
  28. MacKenzie DI, Nichols JD, Hines JE, Knutson MG, Franklin AB (2003) Estimating site occupancy, colonization, and local extinction when a species is detected imperfectly. Ecology 84:2200–2207CrossRefGoogle Scholar
  29. MacKenzie DI, Nichols JD, Royle JA, Pollock KH, Bailey LA, Hines JE (2006) Occupancy modeling and estimation. Academic Press, San DiegoGoogle Scholar
  30. Madge S, McGowan P (2002) Pheasants, partridges and grouse. A guide to the pheasants, partridges, quails, grouse, guineafowl, buttonquails, and sandgrouse of the world. Princeton University Press, Princeton and OxfordGoogle Scholar
  31. Office of the Registrar General of India (2011) Population of States/Union Territories by sex and percentage share of population in total population: 2011. Accessed Apr 2011
  32. Oksanen J, Blanchet FG, Kindt R, Legendre P, O’Hara RB, Simpson GL, Solymos P, Stevens HH, Wagner H (2010) Vegan: community ecology package. R package version 1.17-4. Accessed Feb 2011
  33. Radford JQ, Bennett AF (2007) The relative importance of landscape properties for woodland birds in agricultural environments. J Appl Ecol 44:737–747CrossRefGoogle Scholar
  34. Richardson AJ, Taylor IR (2003) Are rice fields in Southeastern Australia an adequate substitute for natural wetlands as foraging areas for egrets? Waterbirds 26:353–363CrossRefGoogle Scholar
  35. Stillman RA, Brown AF (1995) Minimizing effort in large-scale surveys of terrestrial birds: an example from the English uplands. J Avian Biol 26:124–134CrossRefGoogle Scholar
  36. Sundar KSG (2009) Are rice paddies suboptimal breeding habitat for Sarus Cranes in Uttar Pradesh, India? Condor 111:611–623CrossRefGoogle Scholar
  37. Sundar KSG (2011) Agricultural intensification, rainfall patterns, and large waterbird breeding success in the extensively cultivated landscape of Uttar Pradesh, India. Biol Conserv. doi:10.1016/j.biocon.2011.09.012
  38. Sundar KSG, Subramanya S (2010) Bird use of rice fields in the Indian subcontinent. Waterbirds 33(Special Publication 1):44–70CrossRefGoogle Scholar
  39. Urfi AJ, Sen M, Kalam A, Meganathan T (2005) Counting birds in India: methodologies and trends. Curr Sci 89:1997–2003Google Scholar
  40. White GC, Burnham KP (1999) Program MARK: survival estimation from populations of marked animals. Bird Study 46(Supplement):120–138CrossRefGoogle Scholar

Copyright information

© Springer Science+Business Media B.V. 2011

Authors and Affiliations

  1. 1.Conservation Biology Program, University of MinnesotaSt. PaulUSA
  2. 2.International Crane FoundationBarabooUSA

Personalised recommendations