Predicting tomorrow night's low is a lot more complicated than you might think. Even if you guess the high temperature perfectly, there are dozens of parameters that influence night time cooling. Humidity, sky cover, surface winds, etc. And each of these is, in turn, influenced by larger scale effects. You'll need to accurately forecast when those high clouds will evaporate, how much surface wind the approaching low will generate, and how strongly that will mix the surface air with (usually drier and warmer) air aloft. Etc. Etc. Etc....
While people are still quite good at teasing subtle details out of the data (see
http://www.theweatherprediction.com/habyhints/ for an idea of the parameters to be considered), most news/teevee forecasting is done via computer models. Between the various weather service branches, dozens of colleges, and international sources, there are literally hundreds of models in various stages of tune. All are wildly complex ,receiving data from all manner of
non-standardized sources. Even if two models are very closely related, if one is getting a specific data set from a different source than another, their output will naturally vary. News/TV don't use anywhere near all of these models, but even the 'top 5' computer predictions will vary from one another.
Predicting tomorrow night's low is relatively easy, and forecasts are generally good to within a few degrees. But now try to predict the same temperature a week in advance. Chaos! As the fore-cast time increases, the models start to diverge rapidly. Minor differences in data or computing results rapidly compound in an exponential manner. By two weeks, the programs are pretty much 'just guessing.'