Reckoning with complexity: robust time series and spatial analyses are critical when selecting GCM models for biophysical modeling studies

Modelers often use “off the shelf” climate projections from downscaled Global Climate Models (GCMs) to simulate the effects of climate change on biophysical processes such as wildfire regimes. Many downscaled GCMs are available at the scales relevant for biophysical modeling (e.g., 4-km resolution). When it is too computationally intensive to run biophysical models using all GCMs, modelers may select a subset of GCMs to represent different climate futures. These models are often chosen to bookend a range of climate changes. This “model selection” process typically focuses on a limited number of future climate characteristics (e.g., temperature and precipitation trends) while ignoring others, such as the timing of drought. An equally important concern when simulating multiple study areas, is that model selection is conducted at the encompassing regional scale and then applied to smaller landscapes within the region. However, if time series characteristics vary among GCMs and/or spatially within regions, then the drivers of biophysical projections may be misattributed. To investigate the extent and effects of these concerns, we quantified how multiple time series characteristics vary among 20 downscaled GCM projections from the statistically downscaled Multivariate Adaptive Constructed Analog (MACA) dataset for four watersheds in the Sierra Nevada Ecoregion, and assessed how each GCM’s time series characteristics vary between watershed and regional scales. We then simulated how each of the 20 GCMs influenced fire regimes in one of the watersheds using the biophysical, fire regime model RHESSys-WMFire. Finally, investigated how different time series characteristics influenced fire size, number of fires, and the timing of fires.

            We found that in some GCMs, periodic events occurred at the regional scale but not in all of the watersheds, whereas in others the inverse was true. When analyzing how different GCMs influenced fire regime projections, we found that even when two GCMs had similar temperature and precipitation trends, they could still produce very different fire regimes due to differences in other time series characteristics, such as precipitation variability. Our study demonstrates that it is essential for biophysical modelers to incorporate robust time series and spatial analyses into their GCM model selection approach in order to confidently interpret the mechanisms driving their climate change projections.