I have 20 years of historic hourly temperature data in Omaha. I want, given just a seed, to be able to generate a pretty realistic year of hourly temperatures. (I have a program that takes such temperature series as an input, and the ability to randomly generate these would help in analyzing this program's performance.)
I don't need these generated temperature signals to depend on solar information, climate change, or anything like that (i.e., I'm not forecasting weather). I just want it to be a believable year-long Omaha temperature signal.
This problem is new to me, but I've tried a few (totally uninformed) things:
- Random linear combinations of each historic signal (this blends out longer-term anomalies).
- Random walk that trends towards the average of the historical signals (perhaps there is hope yet in this method, but it's hard to tune to look realistic).
Does anyone have any ideas or perhaps resources on this topic that they could link me to?