Fig 1.
Model validation using AUC and partial ROC.
AUC and partial ROC vary across models with different number of underlying occurrence records: (A) Test AUCs of 10-fold, 30:70% subsampled, cross-validated Maxent models (black circles) show how well models trained on 70% of the data predicted the 30% of data set aside for testing; final model training AUCs (grey crosses) show the performance of the final models using 100% of the data. The models with performance below an AUC of 0.85 are indicated in bold. (B) Average partial ROC ratio shows which models are significantly different from 1 (black circles; better prediction than by chance) and which are not (bold circles).
Fig 2.
Trends in model commission, congruence, and omission with across number of points used in models.
Boxplots showing how model commission (A), congruence (B), and omission (C) varied across number of occurrence records used for models. Commission and congruence were measured as percentage of total sum of cells predicted as suitable: high commission (false positives) means a high percentage of cells predicted as suitable were outside the EDR for the species, i.e., models predicted additional suitable habitat to what experts believe, while high congruence (true positives) means a high percentage of suitable cells were within the EDR, i.e., agreed with expert opinion. Omission (false negatives) was measured as percentage of EDR predicted as unsuitable by statistical models, i.e., areas where experts believe the species to occur, but models did not confirm it.
Fig 3.
Predicted changes in habitat suitability for the Black Mamba.
Current (A), 2050 (B-D), and 2090 (E-G) predicted habitat suitability (A, B, and E), change in habitat suitability from current (C and F), and uncertainty in prediction (D and G) for an example species, the Black Mamba, Dendroaspis polylepis. Predictions are overlayed with the current EDR for the species as shaded blue area and with recorded reliable observations as yellow points. Grey shaded grid cells in the background show the raw predicted habitat suitability before areas at high cost-distance to known occurrences were cut out. These areas are likely unoccupied due to patchy connection to main habitat or high cost-distance to known areas of occurrence (light grey to dark grey) but are technically suitable and may be shown to be occupied pending further targeted data collection. Maps were created in ESRI ArcPro 3.1.0 [52]. Basemap shows WHO admin 0 country boundaries 2024 (CC BY 4.0). The designations employed and the presentation of the material in this publication do not imply the expression of any opinion whatsoever on the part of WHO concerning the legal status of any country, territory, city or area or of is authorities, or concerning the delimitation of its frontiers or boundaries.
Fig 4.
Examples of different responses to climate change predicted for different snakes.
Examples of current habitat suitability (D-F) and predicted changes in suitability by 2050 (G-I), and 2090 (J-L) for three example species (photos of species shown in A-C) of high medical importance. Overall suitability is predicted to increase greatly for the Black-necked Spitting Cobra, Naja nigricollis across East Africa (A, D, G, J), remain relatively stable for the Northeast African Carpet Viper, Echis pyramidum (B, E, H, K), and decrease for the Puff Adder, Bitis arietans (C; F; I; L). Maps were created in ESRI ArcPro 3.1.0 [52]. Basemap shows WHO admin 0 country boundaries 2024 (CC BY 4.0). The designations employed and the presentation of the material in this publication do not imply the expression of any opinion whatsoever on the part of WHO concerning the legal status of any country, territory, city or area or of is authorities, or concerning the delimitation of its frontiers or boundaries.
Fig 5.
Distribution shifts and SHOI changes in the snake species with highest predicted increase in overlap with people by 2090.
Trends in habitat suitability and SHOI for the four species with the highest significant (i.e., positive and different from 1) predicted increase in overlap with humans by 2090: Naja nigricollis (A-C), Bungarus multicinctus (D-F), Agkistrodon piscivorus (G-I), and Agkistrodon contortrix (J-L). A, D, G, and J show the number of new grid cells that are predicted to become suitable (range expansion; red), and the number of cells that will stop being suitable (contractions; blue). The change in total number of suitable grid cells (grey), may be identical to range expansion in species where no contractions occur (G and J), remain constant for species that show similar amounts of expansions and contractions (A) or show an intermediate response (D). Despite some of the species experiencing both expansions and contractions, all show positive trends in total cumulative suitability (green) and SHOI (yellow; B, E, H, and K), either because increased suitability in their existing range or because of range shifts towards more populated areas.) C, F, I, and L show maps of predicted exposure decreases and increases across each species distribution. Maps were created in ESRI ArcPro 3.1.0 [52]. Basemap shows WHO admin 0 country boundaries 2024 (CC BY 4.0). The designations employed and the presentation of the material in this publication do not imply the expression of any opinion whatsoever on the part of WHO concerning the legal status of any country, territory, city or area or of is authorities, or concerning the delimitation of its frontiers or boundaries.
Fig 6.
Direction of predicted shifts in distributions for venomous snakes around the World.
Predicted range shifts for each individual species (small black arrows) as well as median (large red arrows) of all species’ predicted range shifts for each biogeographic region (black outlines) for 2050 (A) and 2090 (B). Regional summary arrows are multiplied by factor 20 for better visibility. Maps were created in ESRI ArcPro 3.1.0 [52]. Basemap shows WHO admin 0 country boundaries 2024 (CC BY 4.0). The designations employed and the presentation of the material in this publication do not imply the expression of any opinion whatsoever on the part of WHO concerning the legal status of any country, territory, city or area or of is authorities, or concerning the delimitation of its frontiers or boundaries.
Fig 7.
Predicted changes in cumulative SHOI globally and in example regions.
Maps of total global SHOI (product of cumulative habitat suitability and log transformed human population density; A) and median predicted changes in total SHOI by 2050 (D) and 2090 (G). Insets show zoomed in maps for example regions in West Africa (red border; B, E, H) and North-Eastern South America (black border; C, F; I). The most widespread predicted increases in overlap between medically important snakes and people can be seen in Eastern North America, across the Indian subcontinent, and Eastern Asia. Decreases are much more widespread. Within example regions, increases and decreases are patchier, with increases predicted along the Andes, Northern, and Western West Africa. Basemap shows WHO admin 0 country boundaries 2024 (CC BY 4.0). Maps were created in ESRI ArcPro 3.1.0 [52].
Fig 8.
Predicted changes in species richness globally and in example regions.
Maps of current global species richness for all snakes of medical importance (A) and predicted number of individual species gained (D) or lost (G) in each grid cell by 2090 according to median model projections. Insets show zoomed in maps for example regions West Africa (red border; B, E, H) and North-Eastern South America (black border; C, F; I) and mostly show species gained across the Andes and northern West Africa. Maps were created in ESRI ArcPro 3.1.0 [52]. Basemap shows WHO admin 0 country boundaries 2024 (CC BY 4.0). The designations employed and the presentation of the material in this publication do not imply the expression of any opinion whatsoever on the part of WHO concerning the legal status of any country, territory, city or area or of is authorities, or concerning the delimitation of its frontiers or boundaries.
Fig 9.
Most important region-specific snake species based on their SHOI.
Fig 10.
Predicted increases and decreases in SHOI per region.