From what I've been studying it the average position changes slowly over time...
So a reading you took yesterday at one spot might be 5m off the next day.
It seems like there are two kinds of problems you have to deal with:
1. Day to day drift, which can be fixed with an new offset from a known position on power up.
2. Noise coming in (this is the biggest part of the error) so that even if you are sitting still
your GPS signal might tell you you are moving around a bit.
I've been studying (among other things
) way of improving gps and combining it with IMU/Odometry,
and it seems workable but heavy on the maths (for me), but where is the fun if there is no challenge right?
Any ways here are two of the most suitable GPS recievers for mobile robotics that I've found:U-Blox demo kitETek GPS from SparkFun