Do note that realistic "warm" biomes do have occasional freezes; every few years the orange crops in Florida are badly damaged by a hard freeze, for instance (the record low for Orlando, FL is 19 F (-7 C)). Some DF players seem to expect that the temperature ranges will be a lot narrower than is the case in much of the real world; for instance, my town, with authoritative records only since 1952, has in only those 60 years had a historic high of 99 F (37 C) and a historic low of -18 F (-28 C); this while being a hair on the cool side of "temperate" (officially just barely "mountain temperate", with an overall year-round meta-average of about 51 F (11 C)). Interior locations on large land masses can have much wider temperature swings than that.
The real problem however is that while DF has temperature, what it models far more poorly is thermal inertia and specific heat. As I understand it, in DF, if something drops below its freezing point or above its melting or boiling point, it freezes, melts, or vaporizes as appropriate pretty much on the spot; with no consideration of the energy required or released. Additionally, there isn't yet a system for handling multiple levels of frozen. Even given the current limitations, we could hypothetically have "thin ice" handled as a contaminant ("... a bucket of water laced with thin ice"), and "thick ice" handled as a brook tile, which can be both walked on and used for water, before hitting the current "solid ice" which is a full square. Eventually, when we get mixed liquid handling, perhaps we could have freezing 1/7 at a time, with results like a cube with 5/7 water and 2/7 ice. (Perhaps every day with average temperatures below freezing has a 1-in-N chance of increasing the ice of fresh water by 1/7, and a much smaller chance for salt water.)