Cool stuff! I wonder how it would change if you used median error instead of mean. I suspect that in the early part of the forecast, the mean might be skewed by a few really bad misses (particularly in the cool season).
Also, if you're using URMA for your truth, I think you're probably getting the max hourly temperature, not the high temperature. Those two things are close, but could be off by a degree or two. (The same thing applies for the low temperature, but it's probably worse for the low.)