Signals of local bioclimate-driven ecomorphological changes in wild birds
Study areaWe conducted field studies in both regions from August to March, each year from 2012 to 2016. In north India, we selected the two traditional breeding colonies of the Painted Storks, viz., the Delhi Zoo (28° 36′ N 77° 14′ E) and Keoladeo National Park (KNP) (27° 17′ N 77° 52′ E), Bharatpur, Rajasthan (Fig. 1). In the Delhi Zoo, close to the river Yamuna, the Painted Storks nest in the traditional heronries with other colonial nesters, Little Cormorant, Indian Cormorant, Black-headed Ibis, and Night Heron38. The KNP, a Ramsar site spread over 29 km2, situated at the confluence of the rivers Gambhir and Banganga on the western edge of the Gangetic basin, supports diverse fauna, flora, and a mosaic of habitats, wetlands, woodlands, scrub forests, grasslands, and heronries39. In 2013, we recorded 680 adults and 310 nests in the Delhi Zoo and 1584 adults and 430 nests of Painted Storks in the KNP.We selected the Vedanthangal Bird Sanctuary (VBS), the nesting colonies at Melmaruvathur Lake, and Koonthankulam Bird Sanctuary (KBS). The KBS & VBS are the newly declared Ramsar sites in Tamil Nadu, south India. The VBS (12° 32′ 02″ N and 79° 52′ 29″ E) is a 40.3-hectare community reserve effectively protected by the state Forest Department, Tamil Nadu, and Vedanthangal villagers40. It is the oldest breeding waterbird reserve in south India, located 85 km southwest of Chennai. More than 40 species of waterbirds, both residents and migrants, live here. Along with the other 17 heronry species, the Painted Storks build nests every year from November to April during its breeding season. The Painted Stork nesting colonies at Melmaruvathur Lake (12° 25′ 53″ N and 79° 49′ 36″ E) are about 20 km away from the VBS. Here, the Painted Storks build nests at 1.8–5 m above the water level, on trees of Acacia nilotica and Barringtonia acutangula on mounds surrounded by water41. In 2012, we recorded a total of 3185 nests in the VBS, with a maximum number of nests belonging to Spot-billed Pelican (1050 nests) followed by Painted Stork (550 nests), Asian Open-bill (770 nests), and others.Birds have been breeding in Melmaruvathur Lake since 2013, and we counted 80 nests of Spot-billed pelican, 45 nests of Oriental White Ibis, and 56 nests of Painted Stork during the winter of the year 2014. The Lake is spread over 0.19 km2 with islets (mounds) with four clusters of Acacia nilotica and Barringtonia acutangula trees. Rainwater and domestic sewage from the neighboring residential complex are the primary water source, and recreational boating attracts a large crowd visiting the Melmaruvathur temple41. KBS (8° 29′ 44″ N and 77° 45′ 30″ E) is about a 1.3 km2 protected area, declared a bird sanctuary in 1994 and an Important Bird Area40. It comprises Koonthankulam and Kadankulam irrigation tanks actively protected and managed by the local community. We noticed the frequent failures of breeding events due to water shortages related to monsoon failures in VBS and KBS. In 2015, we also observed Painted Storks’ breeding failure across northern India for unknown reasons; therefore, data could not be collected for those periods.Bioclimatic variablesWe obtained the bioclimatic variable, particularly temperature at 2 m height for all the four study sites, from the National Aeronautics and Space Administration (NASA) Langley Research Center (LaRC) Prediction of Worldwide Energy Resource (POWER) Project funded through the NASA Earth Science/Applied Science Program. The monthly average data from 2010 to 2020 was downloaded from the POWER Project’s Hourly 2.0.0 version on 2022/01/04.Digital images of Painted Storks collected under field conditionsUsing Binoculars (Olympus 10X50), Digital Cameras (Canon 5D Mark III and Sony handy-cam), we monitored and recorded all active nests with juveniles and adult Painted Storks twice a week. The nests were on trees, 3–7 m in height, and chicks and adults were visible, which aided the photography. Nests were numbered for our records by taking note of tree branching patterns, the nest’s position on the tree, and other local identification marks. Numbering the nests helped us identify the individuals associated with a given nest and avoided re-recording the same individual (pseudoreplication). Storks show site fidelity42,43, and hence we assumed the same breeding pairs occupied the same nesting site.During the initial months of the breeding seasons, pairing and copulations of the breeding pairs could be readily noticeable. We took consecutive photographs when they were copulating at the nest. After disengagement following the copulation, the birds (male and female) standing side by side at the nest were also photographed. The first author noted all the relevant spatial orientations of males and females during each copulation event in the field notes. Thus nearly 100 copulations involving different individuals of the Painted Storks pair were photographed. To minimize measurement errors, we selected for further analysis only the images of males and females standing parallel and close to each other, perpendicular to the camera. Since we used the digital images of the free-living Storks, we did not have the freedom to choose all morphological features resulting in some missing values. Therefore, we selected a hundred and forty-eight individuals for the analysis from nearly 1500 localized adults. The technique has an efficiency of less than 10% of the population, more efficient than the traditional capture, measure, and release of individuals. Though many individuals were recorded, only a few were subjected to the analyses. Moreover from the digital images, not all the morphological characters of the individuals were measured. The birds’ orientation towards the camera assumes importance because the correct direction ensures maximum exposure of body parts in the picture. In many pictures, correct orientation was missing as the birds were behind other individuals or branches of the trees or leaves. Therefore, selecting the right digital image becomes crucial. Keeping all the above criteria, we filtered images that were later included in the analysis.Calibrations of subject-distance using Exif MetadataWe extracted the EXIF metadata from each JPEG image of Painted Stork. EXIF metadata includes the filename, type, date, and time of the image captured, image width and height in pixels, camera model, lens information, field of view, focal length, and subject-distance. The subject-distance (Painted Stork distance from the camera) being a critical variable and its Exif metadata were standardized with the following equation.$${text{Subject{-}distance}} = 0.7864 times {text{(EXIF subject{-}distance)}}^{{1.0301}}$$
(1)
Using the Eq. (1) derived from an earlier study5, we regressed actual subject-distance with the Exif subject-distance from the images. Then multiplying with the field of view, available as Exif metadata (angle of view) with standardized subject-distance (Eq. 1), the total image size (length and width) in metric units was estimated. We excluded the cropped or manipulated images because Image (size) estimation is possible only for the images coming straight from the camera with EXIF tags. The methodological details for calibration and estimation of in-situ measurements of the morphological variables are given in Mahendiran et al.5.Measurements of the morphological variablesWe created a TPS file for JPEG images of Painted Storks with the TPSUtility Program44. Using the TPS file in the TPSDig (v. 2.17) program44, we measured the selected characters (morphological variables) in pixels. Later, it was used along with the total image size to estimate the size of the specific morphological features in metric units, following Mahendiran et al.5. Ten different morphological variables were measured: Bill length (upper and lower mandible), tibia & tarsus length of both legs, distances among the ear, nostril and corners of the mouth, and body length. We estimated the dimensions of the rigid body parts, viz., bill length, tibia, and tarsus using the given methodology13,15,21. Bill length is the distance from the tip of the upper mandible to the beginning of skin corners near nostrils, the proximal end of the beak (marked as ‘a’ in Fig. 3); Tibia length is the distance from the joint of the tibia-tarsus to the feathers (marked as ‘b’ in Fig. 3); Tarsus length is the distance between the tibia-tarsus joint and foot (marked as ‘c’ in Fig. 3). We took measurements of each individual’s right and left legs and other characters, viz., inter-distances among the nostril, corner of the eye, corner of the mouth on each side (marked as ‘d’, ‘e’, ‘f’ in Fig. 3). Body depth is the distance from the base of the neck near the breast to the tip of the tail (marked as ‘g’ in Fig. 3).Data analysisWe performed the statistical analysis in R45, primarily through the nlme, ggbiplot, nnet, tidyverse, devtools packages. We did not have the freedom to measure a few morphological variables due to the problems mentioned above, which led to missing values in the datasets. We filled the missing values with the impute function using the R Core team45 through mice & VIM packages. When the missing values are high in numbers, we discard the data rather than use the impute function. Since almost about 70% of the lower mandible values were missing, we discarded them and ended up having only nine morphological variables in the final analysis. Moreover, the lower mandible is movable, with the mouth being open and closed, producing a considerable variation in measurements.We designed the matrix (Individuals × Region × Sex) representing the intraspecific variations concerning the region and sexes of Painted Storks46. The individuals are in rows (R), their region in column (C1), and sex in column (C2). We considered the regional variations as a sequence of the latitudinal gradient of the study sites. The values of the individuals (R) were the selected morphological variables. This matrix helped us investigate the critical questions relating to eco-geographic variations and sexual dimorphism.To determine whether temperature varied between study sites, we conducted a two-way ANOVA to analyse the effect of study sites (between North India (DZ & KNP) and South India (VBS & KBS)) and months of the year on the temperature at 2 m. For each character, Dimorphism Index (DI) was calculated as a mean value of female divided by the mean male, multiplied by 100, following the method of Urfi and Kalam15. We estimated the general body size of Painted Storks from the selected morphological variables through Principal Component Analysis (PCA) and tested hypotheses on Eco-geographic variations (Bergmann’s or Allen’s rules)2,47 and the sexual dimorphism15,48. The dimension reduction through PCA was carried out after the imputation as there were a few missing values. Body depth was omitted only for the principal component analysis due to many missing values. However, the values of all the characters are presented in the summary statistics in Table 1. The first principal component is characterized as a measure of size, and subsequent components describe various aspects of shape; therefore, it is considered a measure of general body size15,48,49. The PC1 indicated the body size variation, and PC2 revealed leg length variation (tibia and tarsus). We used nested ANOVA to test their body size variation between regions and sexes. The sexes nested within the region explained the eco-geographic rules and sexual selection patterns.Using a multinomial logistic regression model, we compared the Painted Storks’ northern male (NM), southern male (SM), and female (SF) with the reference category, northern female (NF). Then, we classified the data through multinomial log-linear and feed-forward neural network models. We predicted the Painted Stork’s region and sex using the Machine Learning (ML) algorithms through open-source software Waikato Environment for Knowledge Analysis (WEKA.3.9.5) implemented in Java50. WEKA has standard Machine learning/data-mining algorithms with pre-processing tools generating insightful knowledge from the Painted Storks’ morphological data.Using the R and Python interfaces, we used different ML software frameworks, libraries, and computer programs, viz., TensorFlow and Keras, and extensively explored the WEKA workbench environment to predict the sex and region of the Painted Stork. We used the k-fold cross-validation (k = 10) to avoid overlapping test sets, including splitting the data into k subsets of equal size, using each subset for testing and the remainder for training. We analyzed using the WEKA on a Lenovo ThinkPad P53s Mobile Workstation with the 8th Gen Intel® Core i7 @ 1.80 GHz processor, 48 GB DDR4 Memory, NVIDIA® Quadro® P520 with 2 GB GDDR5 Graphics. The performance criteria for all the eight models were assessed by using the Precision (TP/(TP + FP)), Recall (TP/(TP + FN)), Area under Curve (AUC) = (Sensitivity + Specificity)/2, Accuracy = (TP + TN)/(TP + TN + FP + FN), where TP, TN, FN and FP are the acronyms of true positive, true negative, false negative and false positive, respectively. We used the WEKA experimenter environment to test the statistical significance of the selected Machine Learning algorithms. We performed the Paired T-tester based on the number of correctly classified instances and areas under the curve. More
