A Mechanistic Species Distribution Model for Monarch Butterflies

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

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

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: Host-plant distribution data Lab data on physiological tolerances + + Climate data

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

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

Talk outline 1. Development of predictor layers (host plant and temperature models) 2. Citizen-science data sources used to test the model 3. Relationships between predictor layers and monarch distributions

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

Building Milkweed Prediction Maps with Niche Models � Collected observation records (GBIF, on-line herbaria, i. Naturalist, and Journey North) with location and date Thinned to eliminate observations <12 km 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.

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

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 Zalucki 1982 45 DD 120 DD Total GDD required: 351 DD +45 DD 67 DD ? 32 DD 28 DD 24 DD 35 DD Plus 45 DD before egg-laying begins DZmin = 11. 5°C (52. 7°F)

Most DD formulas do not account for lethal and sub-lethal effects of high temperature � Laboratory results (Batalden et 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 ? ?

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

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

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

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

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

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

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

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.

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 center of milkweed diversity – but recall that their primary host (A. syriaca) is distributed throughout. Monarch distributions north of where the maximum number of generations are predicted, but south of where multiple generations aren’t possible.

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

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

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 Voelker for 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

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

Milkweed species modeled Season Sp Summer AS_AS Start 916 Thinned species 125 asperula Summer AS_CURA Summer AS_EX 488 329 146 curassavica 90 exaltata Summer AS_FA 273 82 fascicularis Summer AS_GL Summer AS_HI Summer AS_INC 181 377 2309 75 glaucescens 62 hirtella 244 incarnata Summer AS_INV 279 53 involucrata Summer AS_LANU Summer AS_LAT Summer AS_LINA 121 253 461 62 lanuginosa 80 latifolia 107 linaria Summer AS_OE AS_OV AS_PER AS_PUM 214 138 161 282 90 oenotheroides 54 ovalifolia 58 perennis 59 pumila Summer AS_PUR 379 91 purpurascens Summer AS_QUAD Summer AS_SPEC 409 1138 Summer AS_STEN 250 74 stenophylla 855 314 1457 1818 183 108 subverticillata 67 sullivantii 184 syriaca 255 tuberosa 76 variegata 1398 1088 376 195 verticillata 192 viridiflora 86 viridis Summer Summer AS_SUBV AS_SUL AS_SYR AS_TUB AS_VAR Summer AS_VERT Summer AS_VIRIDF Summer AS_VIRIDI 104 quadrifolia 180 speciosa Spring Spring AS_AS AS_CURA AS_GL AS_LINA AS_SUBU AS_VIRIDI 113 338 102 153 86 72 94 asperula 250 curassavica 76 glaucescens 121 linaria 74 subulata 52 viridis 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 monthly temperature (12 variables), mean monthly precipitation (12 variables), elevation, latitude, and longitude.