Inaccuracy of the UTM to/from lat, lon conversion

When converting coordinates from UTM to lat, lon then back to UTM with utm.to_latlon and utm.from_latlon you don’t end up on the initial point. The error is about 20 to 30 centimeters, depending on the position of the initial point.

import utm
import pyproj
import numpy as np

def utm_latlon_utm_error_with_utm_package(x, y, number, letter):
    lat, lon = utm.to_latlon(x, y, number, letter)
    x1, y1 = utm.from_latlon(lat, lon, force_zone_number=number)[:2]
    return np.linalg.norm([x1 - x, y1 - y])

def utm_latlon_utm_error_with_pyproj_package(x, y, number, letter):
    proj = '{:02d}{}'.format(number, letter)
    lon, lat = pyproj.transform(pyproj.Proj('+proj=utm +zone={}'.format(proj)),
                                pyproj.Proj('+proj=latlong'), x, y)
    x1, y1 = pyproj.transform(pyproj.Proj('+proj=latlong'),
                              pyproj.Proj('+proj=utm +zone={}'.format(proj)),
                              lon, lat)
    return np.linalg.norm([x1 - x, y1 - y])

def lon_lat_to_utm(lon, lat):
    n = utm.latlon_to_zone_number(lat, lon)
    l = utm.latitude_to_zone_letter(lat)
    proj = '{:02d}{}'.format(n, l)
    x, y = pyproj.transform(pyproj.Proj('+proj=latlong'),
                            pyproj.Proj('+proj=utm +zone={}'.format(proj)),
                            lon, lat)
    return x, y, n, l

lat, lon = 2.339404, 48.857767
x, y, n, l = lon_lat_to_utm(lon, lat)
print('Distance after UTM --> (lon, lat) --> UTM conversion')
print('\twith the utm package: {:.6f} m'.format(utm_latlon_utm_error_with_utm_package(x, y, n, l)))
print('\twith the pyproj package: {:.6f} m'.format(utm_latlon_utm_error_with_pyproj_package(x, y, n, l)))

Running this Python snippet gives:

Distance after UTM --> (lon, lat) --> UTM conversion
	with the utm package: 0.280308 m
	with the pyproj package: 0.000000 m

Any idea on where this might come from?

[chlxt]

and

utm.from_latlon(*utm.to_latlon(193185, 1833395, 29, 'Q'))

now gives

(193185.00006130006, 1833395.009825141, 29, 'Q')

[tjwilli58]

Looks like PRs #46 and #47 have the same issue.