THIS PAGE IS NOW OUT OF DATE, FOR CURRENT INFORMATION ON THIS PROJECT, VISIT THE PUBLICATIONS LINK AT THE TOP AND LOOK FOR A PREPRINT ASSOCIATED WITH A PAPER TITLED "Assessing risk for butterflies in the context of climate change,
demographic uncertainty, and heterogenous data sources."
updated 30 Jan 2022
Prioritizing western butterflies for conservation concern
In our 2021 paper on western butterflies and climate change, we reported on widespread reductions in the density of butterflies observed across the 11 states of the western US. While we reported relative rates of species-specific changes in density, most of the results were focused on aggregate indices (the total number of butterflies seen on a particular day) and spatial results (among locations rather than among species). The effort reported below is a followup to that work in which we gather and organize data in a way that highlights individual species with the goal of facilitating conservation attention and hopefully furthering our understanding of butterflies in the Anthropocene.
__________________________________
OVERVIEW AND MAIN RESULTS
We combine different lines of information (many of which are new and were not included in the previous publication mentioned above) into a “risk index” for each of 396 species. These results reflect a work in progress and should be treated as such; they will be part of a manuscript submitted for peer review in early 2022.
In the meantime, reader beware. At this point, errors in data processing and analyses are likely, and I am in the process of checking and double checking everything to find those errors over the next couple of months. Also, our final product will encompass subspecies throughout the west in a more qualitative assessment, coming soon.
Fig. 1. (Click on the image to the left for a full-size version.)
Western butterflies organized into two groups (A on the left and B on the right) and ordered by decreasing risk index values (the most at-risk species are at the top). The A group species on the left are those for which we have enough observational data from NABA counts or the Shapiro monitoring project to be used in population models; B group species are those without those data (but see below, end of legend). The risk index values are shown on the right hand side of each panel, and include 80% highest density intervals for the A group species (with error propagated from Bayesian analyses of annual trends). The risk values for the A group species are based 47% on NABA projected 50 year occupancies, 47% on Shapiro historical trajectories, and 6% on changes in range size reflected in iNaturalist observations. The columns (with colored circles) correspond to different variables quantified for each species, with larger circles indicating greater risk associated with a particular variable; for example, a large circle under host breadth reflects a highly specialized species. The letters next to each species name (N, S, E, and W) signify where the bulk of the species range lies (“W” for example indicates a species with most of the range within the 11 western states). On the right panel, the asterisks next to the risk index values mark species with some data from the NABA counts, but not enough to be used in full PVA models; blue asterisks reflect an 80% or greater chance of increasing, red asterisks decreasing, and black indicates somewhere in between.
Fig. 2. (Click on the image to the left for a full-size version.)
Phylogeny of 394 western butterflies, based on Zhang et al. (2019). The outer ring identifies species in three different risk categories: darkest purple corresponds to the upper quantile of risk above 0.95; the medium purple between 0.75 and 0.95 quantiles; the lightest color less than 0.75. Species names in bold indicate A group species (with risk values derived from monitoring data); all others are B group species.
Zhang, J., Cong, Q., Shen, J., Opler, P.A. and Grishin, N.V., 2019. Genomics of a complete butterfly continent. BioRxiv, p.829887.
Fig. 3. (Click on the image to the left for a full-size version.)
Comparison of A group and B group species, including some of the variables (in panels A through F) also shown in Fig. 1. Panel G shows weighted latitudinal midpoints, and panel H is a proportional breakdown of the geography of species ranges (corresponding to the same range designations shown with letters next to species names in Fig. 1).
Fig. 4. (Click on the image to the left for a full-size version.)
A geographic summary of risk values across all species, with cumulative risk in A and mean risk in B. For example, darker values in B indicate that on average species in a certain area have higher risk indices (from Fig. 1).
Fig. 5. (Click on the image to the left for a full-size version.)
Examples of exposure maps for two species: Hesperia nevada (in A and B) and Lycaena xanthoides (in C and D). Development (urban and agricultural) exposure is shown in the top panels, with climate change exposure in the bottom two, where darker colors indicate greater multivariate departure from baseline conditions.
Fig. 6. (Click on the image to the left for a full-size version.)
Relationship between historical rates of change (summarized as geometric population growth rates) and projected 50 year occupancies for species with sufficient data in the NABA counts. Intervals are 80% Bayesian highest density intervals.
__________________________________
DATASETS AND OBSERVATIONS
The information that contributes to our risk index includes the following variables, with a brief description of each:
NABA occupancy. This is the result of population modeling and forecasting based on “4th of July” community scientist butterfly counts managed by the North American Butterfly Association. In the 2021 paper we modeled historical rates of change. Here we are using a new population model to forecast (for each species) the fraction of current locations expected to be extant 50 years into the future. The relationship between forecast occupancy and population growth rate (estimated from historical counts) is shown in Fig. 6.
Shapiro monitoring. Species-specific rates of change over recent decades come from North America’s longest consistently-maintained butterfly monitoring program of 10 sites across Northern California. The project was initiated by Dr Art Shapiro (UC Davis) in the 1970s and is now run in collaboration between Art and my lab at the University of Nevada, Reno. These data complement the NABA data in being less geographically extensive but more temporally intensive (observations every two weeks). The modeling of the Shapiro data used here (generating historical rates of change) is similar to what has been done with this data in the 2021 paper and other papers by myself and colleagues (e.g., Nice et al 2019).
iNaturalist. This public resource complements both NABA and Shapiro data in being the most geographically dispersed, but also the least temporally extensive. In the 2021 paper, we used iNaturalist records in conventional time series analyses. Here we are reporting a novel approach that we think makes even better use of the unique (extremely dispersed) nature of iNaturalist data. Specifically, we use lat/long coordinates from iNat to generate estimates of the spatial extent of observation over the last 15 years for any species with at least 100 records (most species have hundreds or thousands of records). Then (across species) we save residuals from the relationship between that spatial estimate and an expert-derived estimate of geographic range (see below). Those residuals tell us something about the space over which a particular species has been seen recently relative to the space over which we might expect it to be seen (based on the expert-derived range).
Geographic range. These are values for geographic areal extent (range) based on maps in the Swift Guide to Butterflies of North America (2nd ed) by Jeffrey Glassberg. Values in square kilometers were generated for each species from KML files for the 11 western states.
Development. This variable quantifies current “exposure” to anthropogenic land use derived from publicly-available geographic information databases (example in Fig. 5). For each species, the outline of the expert-derived geographic range (see previous) was used to calculate the percentage of the area encompassing spatial cells in various categories associated with both urban and suburban development as well as agriculture (but not including pasture lands).
Climate departure. Similar to the previous, we calculated exposure to recent climate change throughout the range of each species. Our climate data come from the TerraClimate system, and estimates of change were generated as rates of change in Mahalanobis distances from climatic baselines, with rates measured using Sen’s slope estimators that are relatively robust to outliers.
Precipitation. This is simply a characterization of average annual precipitation within the range of a species based on 30-year normals, again from the TerraClimate system. The inclusion of this static climate variable (in contrast to the previous index of climate change) was inspired by our results from the 2021 paper in which average precipitation was an important predictor (along with climate change) of rates of change in butterfly density (driers areas seem to be faring worse than wetter areas).
Voltinism. The expert-derived ranges (see above) were processed in a way that allowed us to calculate the fraction of the range for each species that is univoltine (one generation per year, as opposed to two or more generations). Previous work with both NABA and Shapiro datasets has suggested univoltinism as a risk factor for western butterflies, especially in the context of climate change.
Wingspan. This variable is a proxy for overall body size, and we expect larger species to be more resilient to various stressors. This expectation is based on previous and current work with western butterflies (e.g., Forister et al. 2016) and likely reflects greater dispersal of larger species. Data used here were taken originally (and with permission) from Butterflies and Moths of North America, and have been supplemented ad hoc for species not in the original data.
Host breadth. Finally, dietary specialization is frequently considered a risk factor for butterflies and other animals. The values studied here come originally from Scott’s Butterflies of North America, supplemented with ad hoc literature searches. We calculated an index of taxonomic diet breadth as the number of host families plus the natural log of the number of host genera.
__________________________________
additional METHODS
Each of those variables described above was processed through transformations (details to be included in a manuscript) that put them on the same scale (between zero and one) which in turn allowed us to add up values across variables to come up with a risk index for each species. As part of the transformations, variables were flipped (or not, depending on the variable) so that higher values correspond to greater threat. Greater threat is reflected in Fig. 1 with larger circles. For example, a smaller geographic range has a larger “threat circle”, and more severe rates of climate change also have a larger circle. The adding up of the transformed variables for each species included a weighting scheme such that maximum values on all indices would result in a score of one.
The 394 species that we studied divide naturally into two groups: those with data in either the NABA or Shapiro monitoring programs (183 species) and those without a sufficient presence in those programs for population models (211 species). For no particular reason, I’m calling these the A group (183 species) and the B group (211), and the division is not arbitrary: species not encompassed by monitoring programs tend to have smaller and more southern ranges; those species (the B group) are also on average more exposed to climate change (because of their more southern ranges) and slightly less exposed to development (see Fig. 3). These differences are biologically meaningful and relevant to interpretation, see Discussion below.
__________________________________
RESULTS
These results are a first pass and are likely to evolve, especially for the B group (for which we don't have monitoring data).
In Fig. 1, the species with the highest risk values are at the top of each of those lists (A group on the left, B group on the right). Those are on page 1, and the other pages (with lower-ranked species) are included mainly for interest and comparison. Note that the quantitative risk value is shown to the right of each species. Those values include 80% credible intervals for the A list because they incorporate uncertainty from Bayesian analyses of NABA and Shapiro data. Not surprisingly, the uncertainty is large. The same risk values, but broken into high, medium and low categories are shown in a phylogenetic context in Fig. 2; those values are significantly clustered (statistical details coming soon). More details on the A and B group species are shown in Fig. 3, and then Fig. 4 provides two perspectives on the spatial distribution of risk (across both groups of species). Finally, Fig. 5 is an illustration of exposure metrics for two species, and Fig. 6 is a methodological detail relating to parameters estimated from the NABA count data. See sections above for figure legends and additional details.
Also note (reading Fig. 1) that the uppercase letters near the name of each species reflect a qualitative judgement (based on my reading of field guides) of where the bulk of the distribution lies for each species: E, S, N or W (east, south, north or west). Species with a “W” have most of their range encompassed by the 11 western states, species with a "N" range mostly across Canada and only dip down a bit into the western states; etc. Please take these qualitative assessments with a big grain of salt; they need refinement.
__________________________________
DICUSSION
When looking at the ranking of species (Fig. 1 and Fig. 2), it is important to keep in mind a core uncertainty that we face. We are confident that many butterflies in the west are experiencing reduced population densities, and we are equally confident that habitat loss, climate change and habitat degradation (e.g., pesticides) are the main drivers of decline (based largely on spatial variation in faunal-level indices); in contrast, we have had less success identifying species-specific traits that predict historical rates of change. There are interesting and spatially-limited exceptions: for example, during the mega-drought years of 2011 to 2015, multivoltine species were more resilient at low elevations in California. But when we look broadly across the west we have less predictive power, and it is my hypothesis that this is a result of both the severity and ubiquity of threats (climate change in particular) but also the diversity of threats (death by a thousand cuts) and heterogeneity of species-specific responses. So, we're confident in the big picture, but where does the species-specific uncertainty leave us with respect to identifying at-risk taxa? The answer is different for the two groups (A and B).
For the A group, we have monitoring data and I suggest that should be our best (but not exclusive) guide, especially when viewed in conjunction with all of the other variables described above. For example, two species might have similar projected occupancies over the next 50 years, but if one of those species has a high exposure to development we might focus on that species first. For the B group, we lack monitoring data so we must rely solely on the other variables and a priori expectations. Lacking other information, it’s reasonable to expect that a small-ranged, host specialist exposed to severe climate change is more at risk over the coming decades than a species without those properties.
A final note on the challenge of identifying at-risk species as well as an apparent paradox. Given previous publications (the 2021 paper and others) highlighting the importance of climate change, you might expect to look at the A-list species and see a simple association between monitoring results and exposure to climate change. For a number of reasons (methodological and biological), the world is not that simple, and a key thing to remember is that our climate change results in the 2021 paper were derived from aggregate counts (total numbers of butterflies seen on individual days) and focused on variation among locations, not among individual species. When it comes to individual species, most have ranges that encompass areas that are more or less severely impacted by climate change, but our climate change exposure index is generated from the entire geographic range (we had to start somewhere!). Furthermore, the species with the most severe climate impacts across their entire range are species found more towards the southwestern deserts, and these are more often than not in the B group (Fig. 3), for which we lack monitoring data! Thus we have much to learn, and all of these results are presented with that in mind.
__________________________________
CONTRIBUTIONS
Many have contributed to the results reported on this page, including the authors of our 2021 paper. Direct contributions to the ranking and prioritization effort have been made by the following: Eliza Grames, Thomas Riecke, Kevin Burls, Chris Halsch, Cas Carroll, Kate Bell, Josh Jahner, Art Shapiro and Jeffrey Glassberg.
__________________________________
Please contact me (Matt Forister, forister@gmail.com) with ideas, comments or requests for data.
The image at the top of the screen includes Vanessa annabella, the west coast lady, one of our most severely declining and widespread species (photo credit Chris Halsch).