That is very true. I developed a GPS application for the Palm and this is a very interesting issue to deal with.

If you calculate individual samples too often then the inherent GPS error and even floating point precision can introduce large errors over time. If you don't calculate samples often enough then obviously your calculations can lose accuracy since there may be non-linear movement between the two measurements that you will have missed.

The best approach I came up with was to base the sample frequency on the speed reported by the GPS unit. If the GPS unit is reporting a speed of 0.2mph then do a sample every 20 or 30 seconds since you're moving scarcely faster than the potential error of GPS so you're going to be measuring GPS error in precision if you sample too quickly. If the GPS unit is reporting a speed of over about 5mph (well beyond the GPS error in precision) then sample every 2 or 4 seconds.

The idea of tracking distance traveled by GPS depends completely on the accuracy of the GPS unit. A less accurate GPS unit is going to produce significantly worse results than a more accurate GPS unit since each individual position error will be summed over time.

Also, unless this is for operation in an aircraft I'm not sure it's a good idea to try to include changes in altitude in your calculation. Even with an 8% grade (which is severe and unusual in normal driving!) the total distance traveled will only vary by about a third of a percentage point. And the vertical precision of GPS is far less accurate than the horizontal precision. Try standing still with a GPS and you'll see that your horizontal position moves within an area of about a meter or two while your altitude can vary by as much as 4 or 5 meters on a second-by-second basis even when you are completely motionless. So the accuracy of your odometer is going to be negatively affected by the inclusion of the altitude component that, even if accurate, would only add about 0.3% to the total distance traveled. But considering the amount of altitude error that can exist second by second I'll bet you'd be adding more than 0.3% error to your calculation which makes the inclusion of altitude a bad idea.

Regards,
Craig Steiner