Look at the same problem from the wavenumber domain instead of counting samples directly.
Step 1: Nyquist wavenumber.
For a sampling interval \(\Delta x\), the Nyquist wavenumber is \(k_N = \dfrac{1}{2\Delta x}\), corresponding to the shortest recoverable spatial wavelength \(\lambda_{min} = 2\Delta x = 10\) m.
Step 2: Relate wavelength to target width.
A localized magnetic anomaly from a buried body of width \(w\) behaves, along the profile, like roughly half a spatial wavelength of the signal it produces, so the smallest body whose anomaly is not aliased or smeared out corresponds to \(w \approx \lambda_{min} = 2\Delta x\).
Step 3: Substitute.
\(w_{opt} = 2 \times 5\ \text{m} = 10\) m - the same figure obtained purely from the frequency-domain (Nyquist) argument, confirming \(\boxed{10\ \text{m}}\).