We have touched on a number of key variables including tx power and terrain, but some additional attention should be paid to the antenna system as well. It is common in omnidirectionial applications to stack elements into vertically stacked, driven arrays to maximize gain towards the horizon. This additional gain helps a great deal, but there is a limit to how far you can go with this. For example, it is not unusual to achieve array factor gains of 9 dB, most often done by stacking dipoles. Properly phased, going from one to two dipoles will double the gain, hence will add 3 dB. Doubling the gain again by going to four dipoles will deliver a total array gain of 6dB. Another four dipoles, for total of eight will give you 9dB of gain over and above the gain of each individual element. If dipoles are used, your element gain will be roughly 2 dBi, so total gain will be about 11 dBi if losses are minimized. We run into practical limitations when trying to double this again to 16 elements so many stations stop at fewer than 16 elements in practice. The key point here is that it is much cheaper to add antenna gain than it is to add transmitter power. So, you should be maximizing your antenna gain first, plus your antenna placement (location and height above average terrain) before deciding on what your transmit power should be.
It is common practice to use circular, vertical and horizontal polarizations in FM broadcast. Circular has been popular because many stationary FM receivers have horizontally polarized antennas, while most car receivers use vertically polarized antennas. To service all of the listeners, circular polarization is a good compromise as it costs everyone -3dB of link budget but suffers no polarization mismatch losses other than that. However, vertical polarization is lower cost and services car radios well so may be the more common setup. You get to pick which polarization makes the most sense for your listeners.
I also want to mention that in all broadcast sitations of this magnitude, we are talking about a significant support structure (tower) to hold the antennas up. Towers of this size require proper engineering of anchors and foundation, structure, wind loading, earthquake resistance, corrosion protection (rust and electrolytic problems), lightning surge management, and aviation anti-collision lighting to name a few that I could remember. It would be wise to involve an experienced structural engineer with communications tower experience to deal with these issues. I have only been involved with these issues in designing grounding and lightning protection and even this one little topic is surprisingly complex. Get help if you are not competent in these areas.
Back on the topic of transmitter power, I had been looking at examples of FM stations in the USA, which have publicly available coverage contours and transmitter power specs. What they don't mention is that I think their transmitter power, for example 50,000 watts, is not the transmitter output power but rather the effective radiated power or effective isotropic radiated power. So, if we find through examples that 50,000 watts has provided good coverage for many FM stations in America out to at least 80 Km, and if we plan to use an antenna system with 10dB of gain, this would suggest that a transmitter with 5000 watts of output power is enough. Using a propagation analysis tool, like Radio Mobile, is a good way to test these examples because it will force you to develop a complete link budget including transmitter power, antenna gain, transmission line losses, and terrain losses too. We may find that 10,000 watts is overkill. Focus your attention and some money on the antenna system. Beware that in broadcast applications, the exact receiver antenna gain is not know, so broadcast engineers design to a goal of a particular field strength. In the US, for example, the 70 and 60 dBu contours define very good reception by most listeners, while field strengths of 54 dBu and lower may provide less than clear reception on the average consumer radio. You get to choose what level of field strength is good enough for your listeners. Alternatively, you could assume a receiver antenna gain (I suggest -6dBi) and receiver antenna height (suggested: 1m) for your analysis.
By the way, here is another estimation tool to check radius of coverage for flat topography:
https://www.fcc.gov/mb/audio/bickel/fmpower.html
It is, unfortunately, somewhat USA -centric but it might be helpful as it is very easy to use.
This company may be helpful:
https://www.electro-tech-online.com/custompdfs/2010/12/hpbcast.pdf