Seeing the woods for the trees – when is microclimate important in species distribution models?

Potter et al. (2013) highlight the challenges and provide recommendations for progress in representing microclimate in Species Distribution Models (SDMs), which are widely used to predict distributions by establishing relationships between climatic variables and species presence. They show that the grid lengths of published SDMs are typically four orders of magnitude larger than the length of animals under study, and three orders of magnitude larger than plants. They conclude that the mismatch between the length scales of climate data and the species themselves is a major barrier for progress, and that the ideal spatial resolution for climate data in SDMs, notwithstanding practical constraints, is between 1 and 10 times the length of the organism. At face value, this statement is worrying for ecologists concerned with the practical challenges of modelling current or future climatic constraints on animal and plant distributions. The need for predictive models is pressing, but a spatially explicit and accurate representation of global or regional climate at a scale of a few metres, centimetres or even millimetres is impractical, even with rapid advances in computer processing power and high resolution remote sensing data. While agreeing with many of the recommendations for progress proposed by Potter et al. (2013), we question whether aiming for such ultrafine spatial resolution in all circumstances is a useful goal. The apparent mismatch between the scale of climate data and the size of organisms is bridged implicitly in most SDMs with a ‘mean field approximation’, by assuming that grid-cell average climatic variables are statistically meaningful predictors of the probability of species persistence. This assumption does not necessarily require that the gridded climate data accurately represent the microclimate experienced by any individual within that grid cell, or as Potter et al. (2013) suggest, that all species within a grid cell experience the same climate. The mean field approach simply requires that macroclimate is a good predictor of the aggregated population-level effect of many individual responses to the spatial and temporal variation in microclimate that influence individual performance. Often this is a reasonable assumption; while it is true that ‘current climate data are coarse because it is difficult and expensive to sample simultaneously at fine scales and large geographical extents’ (Potter et al., 2013), gridded climate data only exist at all because they are useful. Statistical correlations between large-scale climate and a range of physical and biological phenomena (not limited to species distributions) are well established and frequently robust without downscaling to the scale of the mechanistic processes involved. When studying thermal physiology, it is clearly desirable to take temperature measurements as close as possible to the organism (or organ) of interest. However, the aim of a SDM was not to predict physiological rates of an individual, but the presence or absence of populations, where the response variable is usually mapped at a similar resolution to the climate data itself. Linking SDMs to physiological mechanisms is appealing as it may aid our understanding of the mechanisms underlying species responses (Kearney & Porter, 2004; Crozier & Dwyer, 2006; Buckley et al., 2010), but downscaling coarse resolution climate data to an appropriate spatial and temporal scale (Kearney et al., 2012) is only half of the challenge. Physiological rates must also be scaled, either explicitly or implicitly, up to higher levels of ecological organization – to individual fitness, survival and reproductive rates, population dynamics (Buckley, 2008) and possibly metapopulation dynamics and dispersal (Bennie et al., 2013) before they translate into distribution maps at the desired ecological scale (Fig. 1). Such ecological models are themselves sensitive to mean field approximations (Morozov & Poggiale, 2012) – for example, scaling from individuals to populations or local populations to metapopulations. In practice, greater accuracy in describing the operative temperature of organisms (Bakken, 1992) will only improve SDM predictions if coupled with appropriate ecological models to scale from physiology to distribution. Representing microclimate in SDMs is likely to be desirable in at least three types of circumstance – when predictions of distribution are needed at a high resolution; when mechanistic insights into the physiology that drives species responses to climate are required; and when the mean field approximation of classic SDMs is unsatisfactory and nonlinear responses to microclimate availability become critical in determining the distribution of species at a large scale. We advocate a focus on identifying the conditions under which mean field approximations break down, and the spatial scales at which such breakdowns occur, in different systems, tested where possible against field observations. In this way, we may strengthen general recommendations for the appropriate scales at which best to model the responses of species distributions to climate change.