I don’t see iNat supporting different location accuracy thresholds for different taxa, and it’s unlikely we’ll support a location accuracy threshold for Research Grade status in general, but I can certainly bring it up with the team.
https://forum.inaturalist.org/t/create-filter-by-location-accuracy/298/6
And one should only vote yes for the “Location is accurate” part of the Data Quality Assessment if you think the observer put the wrong location for the observation (eg a wild porpoise observation in the middle of the Atacama). It is not related to positional accuracy.
Not to go too off-topic here, but in my experience this happens when you open your phone’s camera app and immediately snap a photo. The phone doesn’t have time to connect with satellites and other signals to get a precise location, so I usually try to wait 5 seconds or so before snapping a photo, especially if I’m in an area without cell coverage. But if you’re using a DSLR I suggest geotagging your photos on a computer if possible.