Authors: Donald McKenzieUnderstanding future ecosystem dynamics under climate change relies on model projections, not only the predicted values of important variables per se, but also on estimates of the uncertainties associated with them. There are practical limitations to model projections, but in this talk I focus on intrinsic limitations that are less tractable to brute-force improvements in modeling and data processing. I identify four, and then focus on the last three, as the first has been explored in depth for decades now. (1) chaos, manifest as sensitivity to initial conditions (I won’t go over this ground). (2) the “coarse-graining” problem: how to aggregate fine-scale processes to larger scales in an unbiased manner. (3) the “middle-number” problem: related to #2, but in particular affecting systems with enough elements to be intractable to brute-force analysis, but too few and too varied to be subject to global averaging. (4) the “stationarity” problem: relationships that are valid in one environment, such as bioclimatic-envelope or other empirical models, or parameter choices in “process-based” models, change when projected onto future environments, such as a warming climate. I give examples from climate-fire projections and their outcomes on future landscapes.