Geostatistical interpolation can reliably extend coverage of a very high‐resolution model of temperature‐dependent sex determination

Abstract

Aim Recognition that statistical models do not always reliably predict habitat suitability under future climate scenarios is leading increasingly to explicit incorporation of the physiological constraints that underlie species’ distributions into spatially explicit predictions. However, computational intensity constrains the use of high‐resolution, process‐explicit models. We examined whether geostatistical analysis can effectively interpolate a biophysical model, reducing the computational investment typically required for using mechanistic methods to inform physiological predictions. Location New Zealand [40°40′00″ S 174°00′00″ E]. Methods We used a spatially explicit, mechanistic microclimate model to predict hourly temperatures at five soil depths under two scenarios of climate warming. Using the predicted soil temperatures as input to a biophysical model of temperature‐dependent embryonic development, we estimated incubation temperatures and corresponding hatchling sex ratios for tuatara, a reptile with temperature‐dependent sex determination, at a submetre horizontal spatial resolution. We then applied ordinary kriging, a robust method of geostatistical interpolation, to estimate predictions throughout the full extent of our study location, an additional 480,000+ microsites, and validated the interpolation against an independent set of predictions. Results Ordinary kriging accurately predicted spatial variability in incubation temperatures. Mean predictions were similar between methods, and error in the geospatial model generally decreased with increasing soil depth. Error was higher for the geospatial model of the ‘maximum warming’, compared with the ‘minimum warming’, scenario of climate change. Main conclusions Our results show that ordinary kriging can be a reliable method for interpolating variability in high‐resolution predictions. However, the effects of error on the accuracy of interpolated predictions will become more severe as values approach a physiological threshold, such as the minimum and maximum incubation temperatures that result in extreme sex ratio bias. For distribution models, the widths of geographic areas predicted to be suitable for, in this case, maintaining balanced sex ratios, compared to those predicted to be unsuitable, may be narrower than in reality.

Spatiotemporal variation in ecophysiological traits align with high resolution niche modelling in the short-range banded ironstone endemicAluta quadrata

“…At each location, hourly values were summarized as daily averages for these 90-day periods were again summarized to a mean wettest and driest quarter average for each point location over a 10-year period. In order to rescale these data back to the native 1–arc-second resolution, we used an interpolation approach ( Carter et al., 2018 ), where the microclimatic data at our 20–arc-second resolution were fed into a generalized linear model (GLM) informed by the edaphic and geomorphological data for each location. We generated unique GLMs for each microclimatic parameter for the wettest and driest quarters using the ‘stats’ package.…”

Novel Conditions in Conservation Translocations: A Conservative-Extrapolative Strategic Framework

“…Comparisons of remnant populations with translocated populations are useful to understand changes in demography (e.g., sex ratios, mating systems, and populations trends), phenotypic plasticity and phenology (e.g., Mitchell et al, 2008;Miller et al, 2012;Rout et al, 2013). Further development of mechanistic models for tuatara should be undertaken to explore responses of populations to climate change in existing and novel environments (e.g., Carter et al, 2018). The resulting mechanistic models could also be compared to results from correlative SDMs, potentially providing key insights into processes shaping the species' range limits (Tingley et al, 2014;Briscoe et al, 2016).…”

Unveiling geographical gradients of species richness from scant occurrence data

“…A poorly explored but promising strategy to develop macroecological models of geographical variation of species richness and thus overcome knowledge gaps includes techniques of geostatistical interpolation. Interpolation models originated in the geological sciences for mining purposes (Cressie, ) and have been incorporated increasingly in other disciplines (Zhou et al, ), including ecology and biogeography (Carter, Kearney, Hartley, Porter, & Nelson, ; Kreft & Jetz, ). Spatial interpolation takes advantage of the spatial autocorrelation in geographical data, which is the similarity among observations as a function of their distance (Cliff & Ord, ).…”