I need to conduct some soundscape analysis collected in bird colonies on remote islets. Seabirds are loud! I end up with files with regular audio saturation/clipping. The waveform is pretty obvious. We reduced the gain but even though the birds move around and sometimes get very close and saturate the mic anyway.
Not at all, be only carful on the results you present. For example, the frequency modulation should still be visible in your data. You only cannot determine the power level (only a lower limit). However, these type of signals are good for replica based filtering (i.e. classification by means of matched filter, cross correlation, etc). IMHO, clipped signals are much more useful than low SNR signals.
I don't think there's any value to relying on acoustic measurements from segments of the WAV file that are clipped. You can't trust any spectral info on a clipped segment, for example.
However, you can still work with the segments that are unclipped. The rest of this answer will depend on what your goals are with your soundscape analysis. If you are considering seabirds are part of the soundscape (this is appropriate, but depends on your project's goals), then run your third octave level (TOL) analysis or create your power spectral densities (PSDs) as you normally would, but only on the non-clipped segments.