Reciprocal levelling — which source of error is NOT completely eliminated by reciprocal levelling across an obstacle? Choose the correct option.

Introduction / Context:
Reciprocal levelling is performed across rivers, valleys, or other obstacles where equal backsight and foresight distances from one setup are not possible. By observing from both banks (stations) and combining readings, several systematic errors cancel out. However, one atmospheric effect is variable and not always fully removed.

Given Data / Assumptions:

• Two setups on opposite sides of an obstacle.
• Readings taken reciprocally on staves at each end.
• Aim is to eliminate systematic errors depending on sight length and instrument imperfections.

Concept / Approach:
By exchanging the instrument position, the effects of Earth’s curvature and collimation error (line of sight not truly horizontal) tend to cancel because they act with opposite signs over equal long sights. Refraction, however, depends on instantaneous atmospheric gradients along the line of sight and can differ between observations, so it is not completely eliminated in practice, only reduced on average.

Step-by-Step Solution:

Verification / Alternative check:
Field practice often repeats reciprocal readings and averages results to further reduce refraction effects.

Why Other Options Are Wrong:

• Earth’s curvature: cancels with reciprocal observations.
• Non-adjusted collimation: cancels as readings are swapped.
• Bubble tube non-adjustment: largely manifests as a collimation tilt and is mitigated similarly.

Common Pitfalls:
Assuming a single reciprocal set guarantees perfect cancellation; strong temperature gradients can still bias results.

Final Answer:
Refraction