Here are a few reasons I've seen for errors:
1) mixed up weather stations.
- For example: The "Kamloops" data from Jan/Feb 1895 is clearly not Kamloops, both in terms of the mild temperatures and the excessive rainfall. I suspect that this data is actually for Chilliwack, but somehow got mixed up when the guy was entering the data into the database.
2) Missed negative signs.
- We see these all the time. Just last couple months I can find several +28 readings in the middle of some god forsaken frozen hinterland.
3) Weather stations that moved around without documentation.
- Many weather stations, especially 100+ years ago moved around, but they did not have different station ID numbers. One example would be Big Creek, BC. It started in the 1800s, stopped for a few years, and then restarted until it ended in the 1990s. The earlier data was much warmer than the latter. I suspect that this is because it was moved up in elevation, perhaps from 500m to 1100m.
4) Glitches due to human or instrument error.
- Examples would be a +25 in December in Grand Forks, or a +41 in Labrador that is sandwiched between +16 days. I have seen the actual paper copies of these two stations, and they do indeed say 77F and 106F respectively.
5) Faulty readings due to station automation.
- This is an extremely common error these days as many stations are now automated, especially on the precipitation side of things. Automated stations can go months without getting checked while debris skews the measurements. Other stations record precipitation everyday even when there is none. Global Warming theories predicted that precipitation would increase, but we've seen a 30 year drop in precipitation. One has to wonder if instrument error has anything to do with this.