[OSM-talk] GPX trackpoint stacks

[Nick Hill]

Robert Hart wrote:

> In most cases the difference between two adjacent latitude values in the
> list is 0.000021458 degrees. 360/0.00002146 = 16776959 = 2^24 (or near
> enough to make no difference).
> 
> So my garmin is quantising to equivalent of 2.4m if my math is correct.... 

2^24 would be the significant accuracy of a single float.

I have pushed the double values into a single float then compared the 
original double to the single precision float value for lat and lon of 
all GPX points in the database to see the error this generates.

Tabulated showing number of points with error on either lat/lon in 
entire GPX point database of 29.7m points:

Note uD is microdegree or 10E-6 degrees.

Latitude
Error	Number
 >1uD	9421352
 >2uD	162528
 >3uD	77111
 >4uD	0

Longitude:
Error	Number
 >1uD	1430291
 >2uD	870796
 >3uD	474829
 >4uD	138523
 >5uD	100227
 >6uD	62230
 >7uD	24304
 >8uD	0

Either Lat or Lon
Error	Number
 >1uD	10223626
 >2uD	1033323
 >3uD	551940
 >4uD	138523
 >5uD	100227
 >6uD	62230
 >7uD	24304
 >8uD	0

The error pattern is very different between lat and lon.