Grappling with uncertainty in ecological projections: a case study using the migratory monarch butterfly

Abstract

Projecting species' responses to future climate conditions is critical for anticipating conservation challenges and informing proactive policy and management decisions. However, best practices for choosing climate models for projection ensembles are currently in flux. We compared including a maximum number of models against trimming ensembles based on model validation. This was done within the emerging practice of ensemble building using an increasingly larger number of global climate models (GCMs) for future projections. We used recently reported estimates on primary drivers of population fluctuations for the migratory monarch butterfly (Danaus plexippus) to examine how multiple sources of uncertainty impact population forecasts for a well-studied species. We compared mean spring temperature and precipitation observed in Texas from 1980 to 2005 with predictions from 16 GCMs to determine which of the models performed best. We then built tailored climate projections accumulating both temperature (in the form of growing degree days) and rainfall using both "complete" (all 16) and "trimmed" (bestperforming) ensembles based on three emission scenarios. We built the tailored projections of spring growing conditions to assess the range of possible climate outcomes and their potential impacts on monarch development. Results were similar when mean predictions were compared between trimmed and complete ensembles. However, when daily projections and uncertainty were accumulated over the entire spring season, we showed substantial differences between ensembles in terms of possible ecological outcomes. Ensembles that used all 16 GCMs included so much uncertainty that projections for future spring conditions ranged from being too cold to too hot for monarch development. GCMs based on bestperforming metrics provided much more useful information, projecting higher spring temperatures for developing monarch larvae in the future which could lead to larger summer populations but also suggesting risk from excessive heat. When there is a strong basis for identifying mechanistic drivers of population dynamics, our results support using a smaller subset of validated GCMs to bracket a range of the most defensible future environmental conditions tailored to the species of interest. Yet understating uncertainty remains a risk, and we recommend clearly articulating the rationale and consequences of selecting GCMs for long-term projections.

Multi‐season climate projections forecast declines in migratory monarch butterflies

“…As forecast lead time increases, however, so does the relative contribution of climate uncertainty, primarily due to large variations in climate projections across emissions scenarios during 2080–2100 (Hawkins & Sutton, 2009). Uncertainty in population forecasts will always increase with forecast lead time, and there are limited options for reducing climate uncertainty other than selecting a subset of available GCMs for projections based on their ability to accurately model historical values of relevant climate variables within the geographic region of interest (Neupane et al, 2022). However, parameter uncertainty can be reduced, over the near and long term, by collecting targeted data to better understand mechanistic links between breeding‐season temperatures and precipitation and local monarch abundance (Iles & Jenouvrier, 2019).…”

“…We used coupled atmosphere–ocean general circulation models (GCMs), under a range of emissions scenarios, to project spring and summer climate variables into three future periods: early (2023–2043), middle (2050–2070), and end (2080–2100) of the twenty‐first century. We used a systematic approach to select an ensemble of GCMs from a set of candidate models, with the goal of excluding models that were not well‐suited for the region of interest, while retaining a sufficient number of models to adequately characterize present and future climate conditions and uncertainty (Cavanagh et al, 2017; Harris et al, 2014; Karmalkar et al, 2019; Neupane et al, 2022). For this, we compared observed temperatures and precipitation (data from Daymet) with modeled values from 23 GCMs acquired from the sixth iteration of the Coupled Model Intercomparison Project (CMIP6; Eyring et al, 2016; Table S1) for each year in a validation period that spanned from 1980 (the first year Daymet data are available) to 2014 (the latest year hindcasts are available for GCMs from CMIP6).…”

Pathways to global-change effects on biodiversity: new opportunities for dynamically forecasting demography and species interactions

“…When this complexity of human impacts meets complex natural systems, where different interacting species are differently affected by environmental drivers, it becomes imperative to understand key pathways through which environmental change can alter natural communities [3–5]. Understanding these pathways allows us to define more nuanced ecological forecasting, proposing different scenarios under which populations remain viable in the future, when they go locally extinct, or when they invade new habitats [6–8].…”

“…imperative to understand key pathways through which environmental change can alter natural communities [3][4][5]. Understanding these pathways allows us to define more nuanced ecological forecasting, proposing different scenarios under which populations remain viable in the future, when they go locally extinct, or when they invade new habitats [6][7][8].…”

Integrated Population Models: Achieving Their Potential

“…High precision on estimates of covariate effects is critical for prediction, since forecast errors will magnify over time. Thus, when parameter precision is poor, forecasts will be rendered effectively useless ( [89] demonstrate this point with climate predictions). However, by increasing the amount of data available for inference via integrated modeling techniques, it is possible to improve the precision of demographic parameters and their relationships to covariates and thus alleviate some uncertainty in species' projections [3,90].…”