This research examines how spatially mediated uncertainty impacts the predictive value of ecological models that are used to assess habitat and species vulnerability. A major challenge of relying on models to identify ecological consequences is spatial mismatch – the area represented by a grid square is larger than the scale at which microclimate conditions vary and most organisms experience their environments. As a result, models capture little of the fine-scale environmental variation that drives critical physiological and ecological processes. However, the understanding of how model spatial resolution affects predictive models is limited, impairing the ability to make evidence-based management decisions.

The broad scientific objective of this research is to understand how model resolution drives the ability to detect variation in microclimate conditions over space and time and how that variation drives organisms’ vulnerability to changing conditions. The technical objective of this project was proof-of-concept for a novel application of unoccupied aerial systems (UAS): examining how the spatial resolution of a model impacts the capacity to assess microclimate-scale heterogeneity at ultra-fine resolutions. The driving hypothesis of this research is that sub-meter scale variation in microclimate conditions can impact model predictions, effects that may be moderated by terrain complexity. If this hypothesis is true and testable using current modeling approaches, this project allows the team to (i) quantify that variation and (ii) identify the spatial resolution(s) at which sampling should be targeted or whether ultra-fine scale variation can be modeled at all.

Technical Approach

The project team used emerging methods in UAS remote-sensing to quantify the effects of spatial resolution on the ability to detect and predict the microclimate-scale effects of broad-scale environmental variation. UAS-collected terrain data were used to drive a process-based model of hourly substrate temperatures over one year. The values and spatial distributions of model predictions at spatial resolutions of 1 meter (m) and 13 centimeters (cm) were compared to determine whether cm-scale variation in microclimate conditions could be detected.


Substrate temperatures generally did not vary by spatial resolution but were spatially differentiated by resolution; the finer-resolution model showed finer-scale heterogeneity in microclimate conditions. The implications of the results are that, at least at fine and ultra-fine spatial resolutions, model accuracy is independent of resolution. Even without direct validation, the project team can be reasonably confident in models’ relative accuracy. Likewise, increasing the spatial resolution of a model above 1m is unlikely to benefit research without an explicit, ultra-fine spatial component. However, the findings confirm the assumption that spatial mismatch can mask substantial heterogeneity in microclimate conditions, without which ecological assumptions and management decisions may be incorrect. Finally, the findings illustrate that microclimate heterogeneity is detectable at ultra-fine, cm-scales without direct measurement.


Continuation of this research is critical for validating the findings at the continental scale, with the goal of linking spatial and temporal resolution to model uncertainty within a reliable statistical framework. By addressing a key issue in using gridded data in ecological modeling, the results will provide United States Department of Defense resource managers with insight into the drivers of microclimate landscapes. The modeling approach is spatially and temporally transferable. It has the potential to increase the precision of management decisions and reduce the costs associated with conservation-focused data collection.