Fig 1.
This figure provides a comprehensive flow diagram outlining the methodology of the study, including each major step and process leading to the derivation of the risk index.
BRT: boosted regression tree; GLM: generalized linear model; MARS: multivariate adaptive regression spines; MaxEnt: Maximum entropy; RF: Random Forest; GRDI: Global Gridded Relative Deprivation Index; Ai: overlapping area of cropland (in sq. km.); Aj: overlapping area of built-up (in sq. km.). The administrative layer of the map was obtained from the DIVA-GIS website (https://diva-gis.org/data.html). The base layer of the maps is an elevation raster sourced from the SRTM website (http://srtm.csi.cgiar.org/srtmdata/) and was created using ArcGIS software. The species photographs were provided by Anirban Chaudhuri and Raju Vyas through personal communication and were used with their prior permission.
Table 1.
The ensembled model fit metrics for the five participating modelling methods and for the final ensemble model for estimation of habitat suitability of the Big Four species. A total of five model algorithms were used with the threshold of < 0.75 AUC score. AUC: Area under Curve, ΔAUC: Change in Area under curve (Training – Cross Validation), PCC: Proportion Correctly Classified, TSS: True Skill Statistic.
Table 2.
Summary of the highest contributing variable from each category for each snake species, along with its mean (μ) percentage contribution to the model. This table highlights the most influential predictors across categories, while the full list of all final covariates and their contributions for the Big Four species is provided in S3 Table.
Fig 2.
Model evaluation plot, showing the average training ROC of both training and cross-validation (CV) and variables selected by the models for the replicate runs under five models of B. caeruleus.
(A) showing ROC plot of boosted regression tree (BRT), (B) generalized linear model (GLM), (C) multivariate adaptive regression spines (MARS), (D) Maximum entropy (MaxEnt), (E) Random Forest (RF), (F) The suitable habitat extent in the present scenario where the ‘class 5’ determines the extremely suitable habitat extent, and (G) The overlapping cropland and built-up/urban areas with the suitable extent (Class 4 and Class 5) in the present scenario. The administrative layer of the map was obtained from the DIVA-GIS website (https://diva-gis.org/data.html). The base layer of the maps is an elevation raster sourced from the SRTM website (http://srtm.csi.cgiar.org/srtmdata/) and was created using ArcGIS software. The species photograph was provided by Anirban Chaudhuri through personal communication and were used with their prior permission.
Fig 3.
Model evaluation plot, showing the average training ROC of both training and cross-validation (CV) and variables selected by the models for the replicate runs under five models of D. russelii.
(A) showing ROC plot of boosted regression tree (BRT), (B) generalized linear model (GLM), (C) multivariate adaptive regression spines (MARS), (D) Maximum entropy (MaxEnt), (E) Random Forest (RF), (F) The suitable habitat extent in the present scenario where the ‘class 5’ determines the extremely suitable habitat extent, and (G) The overlapping cropland and built-up/urban areas with the suitable extent (Class 4 and Class 5) in the present scenario. The administrative layer of the map was obtained from the DIVA-GIS website (https://diva-gis.org/data.html). The base layer of the maps is an elevation raster sourced from the SRTM website (http://srtm.csi.cgiar.org/srtmdata/) and was created using ArcGIS software. The species photograph was provided by Anirban Chaudhuri through personal communication and were used with their prior permission.
Fig 4.
Model evaluation plot, showing the average training ROC of both training and cross-validation (CV) and variables selected by the models for the replicate runs under five models of E. carinatus.
(A) showing ROC plot of boosted regression tree (BRT), (B) generalized linear model (GLM), (C) multivariate adaptive regression spines (MARS), (D) Maximum entropy (MaxEnt), (E) Random Forest (RF), (F) The suitable habitat extent in the present scenario where the ‘class 5’ determines the extremely suitable habitat extent, and (G) The overlapping cropland and built-up/urban areas with the suitable extent (Class 4 and Class 5) in the present scenario. The administrative layer of the map was obtained from the DIVA-GIS website (https://diva-gis.org/data.html). The base layer of the maps is an elevation raster sourced from the SRTM website (http://srtm.csi.cgiar.org/srtmdata/) and was created using ArcGIS software. The species photograph was provided by Raju Vyas through personal communication and were used with their prior permission.
Fig 5.
Model evaluation plot, showing the average training ROC of both training and cross-validation (CV) and variables selected by the models for the replicate runs under five models of N. naja.
(A) showing ROC plot of boosted regression tree (BRT), (B) generalized linear model (GLM), (C) multivariate adaptive regression spines (MARS), (D) Maximum entropy (MaxEnt), (E) Random Forest (RF), (F) The suitable habitat extent in the present scenario where the ‘class 5’ determines the extremely suitable habitat extent, and (G) The overlapping cropland and built-up/urban areas with the suitable extent (Class 4 and Class 5) in the present scenario. The administrative layer of the map was obtained from the DIVA-GIS website (https://diva-gis.org/data.html). The base layer of the maps is an elevation raster sourced from the SRTM website (http://srtm.csi.cgiar.org/srtmdata/) and was created using ArcGIS software. The species photograph was provided by Anirban Chaudhuri through personal communication and were used with their prior permission.
Fig 6.
This figure illustrates the variation in risk indices of Big Four species across different Indian states, with color coding to distinguish risk levels in present and future climatic scenarios.
Fig 7.
The maps indicate the risk index scores for prioritization of each district in the (A) present, (B) SSP245 (2041-2060), (C)SSP245 (2061-2080), (D) SSP585 (2041-2060), and (E) SSP585 (2061-2080) future scenarios for B. caeruleus.
The administrative layer of the map was obtained from the DIVA-GIS website (https://diva-gis.org/data.html). The species photograph was provided by Anirban Chaudhuri through personal communication and were used with their prior permission.
Fig 8.
The maps indicate the risk index scores for prioritization of each district in the (A) present, (B) SSP245 (2041-2060), (C)SSP245 (2061-2080), (D) SSP585 (2041-2060), and (E) SSP585 (2061-2080) future scenarios for D. russelii.
The administrative layer of the map was obtained from the DIVA-GIS website (https://diva-gis.org/data.html). The species photograph was provided by Anirban Chaudhuri through personal communication and were used with their prior permission.
Fig 9.
The maps indicate the risk index scores for prioritization of each district in the (A) present, (B) SSP245 (2041-2060), (C)SSP245 (2061-2080), (D) SSP585 (2041-2060), and (E) SSP585 (2061-2080) future scenarios for E. carinatus.
The administrative layer of the map was obtained from the DIVA-GIS website (https://diva-gis.org/data.html). The species photograph was provided by Raju Vyas through personal communication and were used with their prior permission.
Fig 10.
The maps indicate the risk index scores for prioritization of each district in the (A) present, (B) SSP245 (2041-2060), (C)SSP245 (2061-2080), (D) SSP585 (2041-2060), and (E) SSP585 (2061-2080) future scenarios for N. naja.
The administrative layer of the map was obtained from the DIVA-GIS website (https://diva-gis.org/data.html). The species photograph was provided by Anirban Chaudhuri through personal communication and were used with their prior permission.