https://books.google.com/ngrams/graph?content=physically+bas...
(Tweak the patameters to end in 1995 to see what I mean.)
I poked around papers from siggraph 1986-1988 and the closest I could find to the use of the term was a panel discussion in 1998:
"STEVE FEINER: Our second speaker will be Professor Don Greenberg. ... Don's group has been at the forefront of computer graphics for some twenty years now, most recently doing pioneering work on physically based modeling and radiosity approaches." ( https://dl.acm.org/doi/10.1145/1402242.1402254 )
I did run across a book using the term in 1994, "The State of the Art in Physically-based Rendering and its Impact on Future Applications" (to your point using the dash between physically and rendered): https://link.springer.com/chapter/10.1007/978-3-642-57963-9_...
The term around then was basically a catch-all for raytracing, radiosity and BRDF-oriented rendering techniques as opposed to the triangle-texture-shading techniques. The former took way more compute-per-pixel/"visual impact" than the latter and thus was done in software while the latter was starting to fit into hardware transistor counts available at the time. There were a variety of sytheses of the various techniques, so some kind of shorthand was definitely needed.
My guess, being mildly involved in the field in the mid-90s, is that while one could have in 1987 said "physics-based" rendering, the researchers, some of whom actually trained in physics, knew that even the techniques they were exploring (raytracing, radiosity, BRDF, light fields) were approximations of physics from the physics and optics literature and not actually grounded in the cycle of hypothesis and experimentation that would constitute "physics-based rendering".
Later the term came to have more specific connotations. One good summary of it from a practical point of view when this book came out was: https://news.ycombinator.com/item?id=38110448
https://books.google.com/books?id=Vw1DAQAAIAAJ&q=%22physical...
I see the term without the dash used in 1994 in Eurographics Computer Forum "A Theoretical Framework for Physically Based Rendering": https://www.cs.rpi.edu/~cutler/classes/advancedgraphics/S11/...
In 1994 siggraph paper by Greg Ward, https://radsite.lbl.gov/radiance/papers/sg94.1/Siggraph1994a..., the term is explicitly described in a footnote:
'The term "physically-based rendering" is used throughout the paper to refer to rendering techniques based on physical principles of light behavior for local and global illumination.'
I defer of course to mattpharr if he cares to comment! :)