CGCS2000 and WGS84 mixed: 0.4 m error at large scale

• Control point 108.9401740, 34.3415680 in CGCS2000 and an RTK point with the same printed coordinates overlap 0.4 m apart on a 1:500 map, 0.8 mm on paper.
• Projected to EPSG:4546 the two differ by x 0.31 m and y 0.26 m, a combined 0.40 m toward the south-east.
• proj4js reports zero difference because it applies no datum shift between 4490 and 4326.

1. Confirm the EPSG code of both datasets; many systems store 4490 data tagged as 4326.
2. Take the same physical location from both sources, project to one plane, and measure the planar distance.
3. Divide the difference by the number of years; about 3 cm/year indicates epoch drift rather than a projection error.
4. Check whether any seven- or four-parameter transformation was applied at all.
5. Test with proj4js and note that towgs84 defaults to 0, so it will show no difference.

Store everything as CGCS2000 and run a datum normalisation on WGS84 input; without official parameters, solve a four-parameter transform from three or more common points and keep residuals under 5 cm.

const a = proj4('EPSG:4490', 'EPSG:4546', [108.9401740, 34.3415680]);
const b = proj4('EPSG:4326', 'EPSG:4546', [108.9401740, 34.3415680]);
console.log(a, b);
// 两者在本例中相同：proj4 默认把 4490 与 4326 当作同一基准（towgs84 全 0）

const pairs = [
  { from: [308611.42, 3804888.31], to: [308611.73, 3804888.05] },
  { from: [312004.18, 3807215.66], to: [312004.51, 3807215.41] },
  { from: [305887.90, 3802333.07], to: [305888.19, 3802332.83] }
];
function apply4p(x, y, p) {
  const c = Math.cos(p.rot), s = Math.sin(p.rot);
  return [
    p.dx + p.k * (x * c - y * s),
    p.dy + p.k * (x * s + y * c)
  ];
}

• Quoting 'CGCS2000 equals WGS84' as a universal rule; it only holds at small scales or centimetre tolerances.
• Applying a fabricated seven-parameter set without check points, which can move the data further than the 0.4 m it was meant to remove.