1. Computed the compass bearing between every consecutive pair of route points (2274 route points, spanning 70 kilometres and with median spacing ~20 m, denser on curves, sparser on straights.
2. Calculated the turning angle at each point (change in bearing).
3. Grouped consecutive turns in the same rotational direction (left or right) into a single ‘curve’, since one physical hairpin bend is usually represented by several GPS points each contributing a few degrees.
4. Retained only groups whose total angle exceeded a minimum threshold (10–30°, tested) to count as a genuine curve rather than GPS jitter.
5. Counted only cumulative turns of ≥15° (a reasonable cutoff for a ‘real’ curve) giving a result of ~330 changes of direction (roughly 165 left turns and 165 right turns)

Notes on Terminology

GPS Jitter

This is small, essentially random positional error in a GPS-recorded point. Even a stationary GPS receiver reports slightly different coordinates from one moment to the next — typically ±3–10 m under normal conditions — due to atmospheric signal delay, multipath reflection (signal bouncing off terrain, cliffs, or trees before reaching the receiver), and satellite geometry. On a mountain road like Route 2331, jitter is worse: dense tree canopy and steep valley walls (both present in this route) commonly degrade GPS accuracy further.

The practical effect: even if a vehicle is driving in a perfectly straight line, the recorded points won’t form a perfectly straight line — they’ll zigzag slightly. When the bearing between each pair of consecutive points is computed, that zigzag shows up as small, alternating left/right “turns” that didn’t actually happen on the road.

Noise Threshold

This is the cutoff used to decide whether a computed turn angle represents a real change in the road’s direction, or is just jitter. It was set at 1–2° per point: any bearing change smaller than that gets ignored (treated as “essentially straight”), while anything larger gets classified as a genuine left or right turn.

The threshold is a judgment call, not a measured fact — there is no independent data on this specific route’s actual GPS error characteristics, so  a small, conventional value was selected and tested against a slightly larger one (2°) to see if the final curve count changed much. It didn’t change dramatically (330 vs. 328 curves).