Leslie Ries (SESYNC, University of MD) Cameron Scott ( NatureServe )

1. Leslie Ries (SESYNC, University of MD) Cameron Scott (NatureServe) Timothy Howard (New York Natural Heritage Program) Tanja Schuster (Norton-Brown Herbarium, University of MD) Rick Reeves (Foxgrove Solutions) Karen Oberhauser (University of MN) A Mechanistic Species Distribution Model for Monarch Butterflies:Towards a general platform for understanding large-scale butterfly distributions

2. Correlative vs. Mechanistic Species Distribution Models (SDMs) • Correlative (“Niche”) SDMs use occurrence data to infer ranges • BENEFITS: Long history, broad applicability • DRAWBACKS: Weak basis for causation, lack of test data • Mechanistic (“process”) models use knowledge of species’ responses to abiotic or biotic conditions to predict ranges • BENEFITS: A priori predictions of causal mechanisms can be tested with independent data • DRAWBACKS: Species-specific Banks et al. 2008

3. A simple mechanistic model for butterflies • Limited by host plant distribution • Limited by physiological constraints • General process-based model would combine host-plant distributions, temperature tolerances, and climate data to predict distributions Our key data sources: Lab data on physiological tolerances Climate data Host-plant distribution data + +

4. Goal: build a mechanistic SDM for the monarch butterfly • Well-understood biology • Data to test model predictions at large scales, thanks to 1000’s of citizen scientist volunteers • A model that works for species with complex annual cycle could be broadly applicable across species, thus meeting a principle challenge of building mechanistic SDMs

5. The monarch butterfly annual cycle Summer expansion and breeding (May – Aug) Fall migration (Sept – Oct) Spring migration and breeding (Mar – Apr) Today, focus on the eastern migratory population in North America during spring and summer Overwintering (Nov – Feb)

6. Talk outline • Development of predictor layers (host plant and temperature models) • Citizen-science data sources used to test the model • Relationships between predictor layers and monarch distributions

7. Modeling host-plant resources • Multiple niche models to predict distributions of monarch host plants (most in genus Asclepias, Apocynacaea) • ~100 species in North America, ~50 with records of monarch use

8. Building Milkweed Prediction Maps with Niche Models • Collected observation records (GBIF, on-line herbaria, iNaturalist, and Journey North) with location and date • Thinned to eliminate observations <12km apart and <50 records after thinning • 19,101 observations downloaded, 8,053 were left after grouping into seasonal bins and thinning on minimum separation distance • 36 environmental layers used to inform niche model • Random Forests in R to provide a consensus map based on 1000’s of individual regression trees • Output maps for individual species compiled into single seasonal maps showing number of modeled species.

9. Example for Asclepiassyriaca, most common milkweed and primary host Observation records Diversity index Summer “niche” map Species modeled: 7 spring 27 summer

10. Modeling physiological responses to temperature using Degree Days (DD) • Determine temperature at which growth can begin (DZmin), each degree above that over 24 hrs is considered a “degree day” • Often, maximum temperature is set (DZmax) after which degree days are no longer accumulated ? DZmin = 11.5°C (52.7°F) Zalucki 1982 45 DD 32DD Total GDD required: 351DD +45DD 120DD 28DD 67DD 24DD 35DD Plus 45DD before egg-laying begins

11. Most DD formulas do not account for lethal and sub-lethal effects of high temperature • Laboratory results (Bataldenet al. in press) show that for monarchs: • No growth at 38°C (100.4°F) • Some lethal effects at 40°C (104°F) • Only 20% survivorship at 42°C (107.6°F) • 100% mortality at 44°C (111.2°F) • Model distinguishes Growing Degree Days (GDD: energy is accumulated) and Lethal Degree Days (LDD: slow growth or cause death) Sub-lethal and lethal effects ? ? DZmin = 11.5°C

12. Mapping GDD and LDD • Temperature data from NOAA temperature stations • Used ordinary kriging to interpolate temperatures between stations every day from 1990-2009. • GDD and LDD were accumulated by season for spring (Mar-Apr) and summer (May-Aug) and converted to number of generations Predicted generations 3105 weather stations

13. # Generations that could be produced based on available GDDs Spring prediction map Summer prediction map Predicted generations Predicted generations

14. Number of LDD (degrees over 38°C) accumulated during summer Average # accumulated LDD

15. Butterfly distribution data from 2 Citizen Science Projects Spring data: Journey North Summer data: North American Butterfly Association No. Years

16. Spring: host-plant and climate resources both associated with monarch distributions MILKWEED DISTRIBUTIONS # observations Modeled species predicted present The center of milkweed diversity in TX is associated with the greatest number of spring monarch sightings

17. Spring: host-plant and climate resources both associated with monarch distributions MILKWEED DISTRIBUTIONS GROWING DEGREE DAYS # observations Modeled species predicted present Predicted generations The center of milkweed diversity in TX is associated with the greatest number of spring monarch sightings Monarch sightings in spring reaches their northern-most distribution within a zone where there is warmth for growth, but not enough for a full spring generation.

18. Are host-plant and climate resources strongly associated with summer monarch distributions? MILKWEED DIVERSITY Monarchs/PH Modeled species predicted present Monarch distributions north of center of milkweed diversity

19. Are host-plant and climate resources associated with summer monarch distributions? MILKWEED DISTRIBUTIONS Monarchs/PH Modeled species predicted present Monarch distributions north of center of milkweed diversity – but recall that their primary host (A. syriaca) is distributed throughout.

20. Are host-plant and climate resources strongly associated with summer monarch distributions? MILKWEED DISTRIBUTIONS GROWING DEGREE DAYS Predicted generations Monarchs/PH Modeled species predicted present Monarch distributions north of where the maximum number of generations are predicted, but south of where multiple generations aren’t possible. Monarch distributions north of center of milkweed diversity – but recall that their primary host (A. syriaca) is distributed throughout.

21. Are monarchs avoiding excessive heat? Monarchs seem to be found where they are least likely to encounter temperatures above 38°C. Average number of accumulated LDD Monarchs/PH

22. Conclusions • Built models of milkweed distributions and GDD/LDD • Spring: Northward migration limited by energy for growth, seems concentrated near the center of milkweed availability • Summer: Southern limits driven by stressful temperatures, northern by host-plant availability and sufficient energy for multiple generations

23. Acknowledgements • Monarch Citizen Scientists for documenting monarch distributions • Elizabeth Howard and Journey North Staff, Jeff Glassberg and NABA Staff, Xerces Society for starting and maintaining Journey North and Fourth of July Butterfly Counts • Emily Voelkerfor helping compile the milkweed database • NSF # DBI-1052875 to SESYNC, ABI-1147049 to SESYNC and UMD for providing funding • USGS’s John Wesley Powell Center for Analysis and Synthesis working group, Animal Migration and Spatial Subsidies: Establishing a Framework for Conservation Markets, for good conversations Photo by Tony Gomez

24. Towards a modeling platform for monarchs and other butterflies • Our goal is to develop a modeling framework that can account for both climate and host-plant resources • Host-plant distributions and climate expressed as GDD and LDD may prove to be a useful modeling framework for many species of butterflies (and potentially other invertebrate herbivores) – meaning this approach could provide a general mechanistic model for understanding butterfly range dynamics • Species interactions may also be critical for many species, and that may require more species-specific approaches • For the monarch, we want to be able to use this platform to explore many issues of conservation concern: • Loss of milkweed habitat in the midwest due to Roundup-Ready crops • Increase in winter breeding in the southern US • Track population trends and try to pinpoint their cause or causes

25. Milkweed species modeled predictor layers created for 36 different variables: percent forest, percent cropland, percent water, percent wetland, percent urban/barren land, population density, presence of railroads, mean annual temperature, mean annual temperature, mean monthly temperature (12 variables), mean monthly precipitation (12 variables), elevation, latitude, and longitude.