Yes, lilkely an interpolation problem. What is you default interpolation set to? When you scale down, you can have artefacts due to spatial-frequency folding. How/when these artefacts appear is a combination of the image characteristics and the frequency response of the interpolation algorithm. Some algorithms look better/worse on some images but you can't draw conclusions, which is why you can safely ignore most blanket statements about such and such algorithm being better. There are also people who advocate a stepped algorithm (ie scale down in several steps).
A general cure is to pre-blur the image before scaling it down to remove all the details that won't be visible anyway in the scaled down version. This acts as a low-pass filter, and removes the high frequencies that as a consequence won't be folded over (the stepped algorithm is just the accumulated cutoffs of each step acting as a global low-pass filter).
For instance, staring with this (click for full size)(path rendering already created a small moiré effect):
If you scale down 4x, you get this:
The top result had no pre-blur, the bottom one had a 1.5 blur applied. It doesn't look less sharp, but the moiré is much less visible.
For some information on the LoHalo/NoHalo methods in Gimp 2.10 (that are enhanced version of the sinc/Lanczos in 2.8) see here and here.
A general cure is to pre-blur the image before scaling it down to remove all the details that won't be visible anyway in the scaled down version. This acts as a low-pass filter, and removes the high frequencies that as a consequence won't be folded over (the stepped algorithm is just the accumulated cutoffs of each step acting as a global low-pass filter).
For instance, staring with this (click for full size)(path rendering already created a small moiré effect):
If you scale down 4x, you get this:
The top result had no pre-blur, the bottom one had a 1.5 blur applied. It doesn't look less sharp, but the moiré is much less visible.
For some information on the LoHalo/NoHalo methods in Gimp 2.10 (that are enhanced version of the sinc/Lanczos in 2.8) see here and here.