important enough to deserve inclusion in our models, and which can be ignored? These are incredibly deep questions—they represent some of the most difficult practical challenges that working scientists in *any* discipline face in designing their models—and giving a general answer to them is beyond the scope of our project here. Still, it is worth our time to briefly examine the plethora of climate models that have sprung up in the last few decades, and to think about the conceptual underpinnings of this highly diverse collection of scientific tools. Perhaps we can at least suggest the *shape* of an answer to these questions with respect to climate science in particular.

In practice, climate scientists employ a large family of models for different purposes. Zero-dimensional energy balance models like the one we just constructed are the most basic models actually used in the real world, and form what can be thought of as a the “lowest level” of a kind of “model pyramid.” The logic of energy balance models is sound, and more sophisticated energy balance models add more detail to account for some of the factors we just enumerated; with every addition of detail, the model becomes capable of generating more accurate predictions but also becomes more difficult to work with. For instance, we might move from the ZDEBM to a one-dimensional energy balance model, modeling the Earth not as a point but as a *line*, and expressing the parameters of the model (like albedo) not as single terms, but as differential equations whose value depends on where we are on the line. This allows us to take the latitudinal variation of incoming solar energy into account, for example: in general, areas

*better* models, it is also the case that more sophisticated models generally leave more room for failure, either as a result of measurement error, because the model accounts for only half of an important feedback loop, or for some other reason. Recall the characterization of models as *artifacts*—in some ways, they are very like mechanical artifacts, and the old engineering adage that “anything that moves can break” applies here as well. We will revisit this point in **Chapter Five** when we discuss the special difficulties of modeling complex systems.

131