GCPs and fieldwork

Why this matters

A sharp map can still be shifted from its true position as a whole. An RTK drone improves the camera positions, but the final map still passes through camera estimation, surface reconstruction and coordinate transformation. To tell a user the map is “accurate”, you need points on the ground surveyed in a systematic way. Field mistakes such as forgetting to subtract the pole height, or mixing height systems, shift the whole job without anyone noticing.

Image positions, GCPs and checkpoints

Separate the position evidence into three groups. Each must have a source, units, a reference system and a quality status.

RTK (real-time kinematic) corrects positions during the flight using data from a base station or network. PPK (post-processed kinematic) records raw data and computes positions afterwards. A “fixed” status on the screen describes the quality of the signal solution, but it does not cover errors from a wrong base-station coordinate, camera set-up or image processing.

Placing and surveying points

Targets must be clearly visible and identify the same point in several images. Choose stable ground, avoid leaves and vehicles, and never use a shadow as a reference. Spread the points to cover the edges and centre and the high and low ground, not clustered near the take-off point on one side. The OpenDroneMap documentation suggests at least 5 GCPs (8–10 for larger jobs), each visible in at least 3 images, but there is no universal number for every job. The number and distribution must come from the site size, terrain and the agreed criteria.

Height systems

GNSS gives ellipsoidal height ($h$), above a mathematical surface, but surveying and engineering use orthometric height ($H$), referred to mean sea level. The two differ by the geoid height ($N$), as explained by NOAA NGS:

$H=h-N$

Thailand has the TGM2017 geoid model developed by the Royal Thai Survey Department with Chiang Mai University, used to convert GNSS heights to orthometric heights. Use a model compatible with the job’s datum, and never apply the $N$ value from the example to a real site.

points = [  # id, role, E, N, antenna h (m), pole height p (m), pole already subtracted by receiver
    ("G1", "control", 500.00, 1000.00, 52.430, 2.000, False),
    ("G2", "control", 700.00, 1010.00, 53.105, 2.000, False),
    ("C1", "check", 540.00, 1030.00, 50.610, 2.000, True),
    ("C2", "check", 660.00, 1120.00, 52.880, 1.800, False),
]
GEOID_N = -30.000   # assumed training value; read from TGM2017 at the real location

for pid, role, e, n, h_ant, pole, corrected in points:
    h_ground = h_ant if corrected else h_ant - pole
    ortho = h_ground - GEOID_N
    print(f"{pid} {role:<7} h_ground {h_ground:.3f} m  H {ortho:.3f} m")

controls = [p[0] for p in points if p[1] == "control"]
checks = [p[0] for p in points if p[1] == "check"]
print("used for adjustment:", controls, "| kept for checking:", checks)

G1 control h_ground 50.430 m  H 80.430 m
G2 control h_ground 51.105 m  H 81.105 m
C1 check   h_ground 50.610 m  H 80.610 m
C2 check   h_ground 51.080 m  H 81.080 m
used for adjustment: ['G1', 'G2'] | kept for checking: ['C1', 'C2']

Checking before you leave

Before packing up, check that all images open, the time and camera are correct, the targets are clear in the images, the edges of the area are covered and the data is backed up. Coming back to fly again costs many times more than checking in the field.

import math

checks = {  # id: (surveyed E, N, Z), (model E, N, Z)
    "C1": ((500.00, 1000.00, 10.00), (500.03, 999.96, 10.08)),
    "C2": ((600.00, 1000.00, 10.00), (600.00, 1000.06, 9.98)),
    "C3": ((500.00, 1100.00, 11.00), (499.92, 1100.06, 11.04)),
}
for pid, (ref, model) in checks.items():
    de, dn, dz = (round(m - r, 3) for m, r in zip(model, ref))
    print(f"{pid}: dE {de:+.2f}  dN {dn:+.2f}  horizontal {math.hypot(de, dn):.2f} m  dZ {dz:+.2f} m")

C1: dE +0.03  dN -0.04  horizontal 0.05 m  dZ +0.08 m
C2: dE +0.00  dN +0.06  horizontal 0.06 m  dZ -0.02 m
C3: dE -0.08  dN +0.06  horizontal 0.10 m  dZ +0.04 m

Module lab

Common mistakes

Summary

• Image positions help processing, GCPs adjust the model and checkpoints are kept aside for an independent check
• Points must be clearly visible, spread across the site and kept in a traceable register
• Subtract the pole height once, then convert with $H=h-N$ from a suitable geoid model such as TGM2017
• Check the data before leaving the site and decide on re-flights from evidence

Check your understanding

1. The antenna height is 45.620 m, the pole is 1.800 m and the receiver has not subtracted it. What is the ground height?
2. $h$ = 43.820 m and $N$ = −28.500 m (assumed). What is the orthometric height?
3. A checkpoint has ΔE = −0.06 m and ΔN = +0.08 m. What is the horizontal error?
4. What happens to the accuracy assessment if checkpoints are used to adjust the model?
5. Does an RTK fixed status certify map accuracy? Why?