The question is which is the most accurate Total Distance?
If you are getting regular GPS readings and the distance and speed are of driving-scale rather than walking-scale, then the most accurate will be the Latitude and Longitude method, since most of the error will come from the GPS position error at the start and end of the trip, which for a run-of-the-mill consumer device nowadays is within 5 metres,
regardless of the distance traveled.
Whilst you are moving, there will also be some error in the positions of the intermediate points, but they mostly cancel out because:
(i) in the direction of travel, what you lose/gain on this reading, you gain/lose on the next reading, eg 15 + 25 = 25 + 15
(ii) laterally to direction of travel, since you're presumably travelling at car-like speeds, and because the errors tend to change slowly, those sideways movements tend to be negligible*.
The tripmeter distance primarily depends on the radius of the wheel from the axle centre to the road-tyre interface, which varies with tyre size and wear and pressure, which also varies with vehicle load (more weight = more pressure) and temperature. Distance also depends on the slip/grip between the tyre and the road, eg road surface material, and to a lesser degree, rain and incline. Odometers usually read off an undriven wheel, but you still get slip due to mechanical friction on the vehicle side of the tyre, and braking.
And then you have the issue that the odometer reading usually comes off one wheel, and thus one side of the car, so that if you are measuring distance on a track, then the outside wheel will travel slightly further than the inside wheel - for a 2 metre wide car, there would be a 12 metre difference between the left wheel and right wheel distances travelled for each lap.
Lastly, the GPS Latitude Longitude method is quietly ignoring altitude. But, happily, most driving happens on flat or gentle inclines, and GPS is less accurate for altitude anyway, thus I have not yet worried about it for my uses.
* or perhaps this extra distance due to random lateral error noise is mostly counteracted by the "lost" distance due to the path of line segments on a curve being shorter than the path of the curve, eg, if you did a 90 degree turn on a unit-radius circle, then the curve circumference distance is pi / 2 = 1.55 but the straight line distance is sqr(2) = 1.41