First of all, Tolkien was trying to create a missing Anglo-Saxon (or, as some argued, Celtic), mythos. For details, see Q1 and Q2
If you recall your Norse mythology, which strongly influenced Tolkien (Midgard==Middle Earth), the geography of the Norse mythos' world bears very little resemblance to Earth (same goes to many ancient myth systems):
Secondly, Arda indeed completely does not make sense from a geologist's perspective (at least, one dealing with biology, ecology and paleontology).
This comes from an actual geologist and paleontologist (senior researcher at the Paleontological Institute of the Academy of Sciences of Russian Federation), Kirill Yeskov, who was so dis-satisfied with Arda's planetary issues, that he wrote an entire book - "The Last Ring-Bearer" - which was based entirely on a two premises: (1) "The way Arda should have been geologically to make LOTR scientifically plausible" and (2) "History is written by the winners, not those interested in truth".
Of course, just like LOTR arising from a language design by a linguist Tolkien, the Arda which arose out of paleontologist Yeskov turned out into a wonderful fantasy place, and a great book to read.
But now, to list the actual geographical musings of Yeskov, quoted from an article by him explaining the origins of his book.
Emphasis mine, and if you prefer not to read through the entire essay (translated from Russian, so the style is a bit heavy), the very last paragraph is the most critical.
Tolkien was a practicing scientist, too, but a linguist rather than a natural scientist like Yefremov1, so the foundation of professional knowledge he had used to erect Middle Earth was different. It is fairly obvious to me that the Game the Oxford professor decided to play with nature began, in essence, with the creation of imaginary languages, with their own alphabets and grammar. Then he created the epic tales to match those languages, then the peoples who wrote those tales, and only then the steppes, mountains, and forests for those people to pasture their herds, build citadels, and battle the “Dark from the East.” This, precisely, was the sequence: “In the beginning was the Word” – Ainur’s music, pure and simple. Truly an excellent model of the Act of Creation!
However, Tolkien the philologist had obviously had a very weak interest in this last, non-living component of Middle Earth – its physical geography – and created it only because he had to, with predictable results. It is a well-known fact that the Professor had painstakingly verified, to the day, the lunar phases during his heroes’ long quest. I believe that, but the problem is that he had overlooked some much more significant elements of the local natural history background.
The Middle Earth has several built-in physical defects, and there’s no getting away from that. In his well-known essay Must Fantasy Be Stupid? Pereslegin provides a detailed classification of errors commonly committed by fantasy authors. He uses Tolkien’s work as an example of one of them, an “irreversible professional error”: “It occurs in a geologically unstable world. Tolkien, being a professor of English Literature, knew nothing of plate tectonics, while the topography of Beleriand and Eriador are highly important to the story; therefore, it seems impossible to fix the author’s mistake.”
(To explain: if a planet has a single continent – Middle Earth – it means that the convection currents in the planet’s mantle form a single cell, meaning that the entire “light” part of the continental crust has gathered over the point where the mantle material sinks toward the core, like foam gathers over the bathtub drain. (This had happened on Earth at least twice, in mid-Proterozoic and late Paleozoic, which is when two super-continents of Megagea and Pangea formed.) When subcontinents collide, they bunch up into folds (e.g., the Himalayas that arose at the collision of the Indian subcontinent with the Eurasian plate). This means that there ought to be a huge mountain plateau like Tibet smack in the center of Middle Earth; where is it?
Pay attention, now – strictly speaking, such errors are trifles. In Pereslegin’s litany of sins an “irreversible professional error” is classified under tolerable errors, being one of the minor ones. It’s obvious that one person can’t be equally proficient in linguistics and geology (I suspect that Yefremov had committed no fewer errors creating Tormansian languages than Tolkien had in Middle Earth tectonics). So we can pardon the Professor – the infraction he had committed was not particularly dangerous to society; The Lord of the Rings can go free. This will acknowledge it to be a regular fantasy text – I mean, a real good one, easily in the top five…
Do you like this option? Me neither. Because The Lord of the Rings is not a good, or even the best, fantasy text. It is sui generis, the only one of its kind; therefore, we will not settle for anything less than a full exoneration.
We will assume that Middle Earth is as real as our world, so if some of the details do not fit our concepts, it’s our problem. On the other hand, we will adhere scrupulously to the laws of nature. As Tolkien himself wrote, it’s easy to imagine a green sun, but “To make a Secondary World inside which the green sun will be credible, commanding Secondary Belief, will probably require labour and thought, and will certainly demand a special skill, a kind of elvish craft.” Well, the sun has its usual color in Middle Earth (and probably belongs to G-2 spectral class), its surface gravity and geochemistry do not seem different from ours, and even the lunar month is 28 days. Therefore we have to approach the task wielding Occam’s Razor (as is customary to the European intellectual tradition): we will appeal to magic and suchlike only when out of all other options.
It turns out that all the seeming contradictions of Middle Earth’s natural history can be resolved with a single assumption: that Tolkien is describing only the northwestern part of the local landmass, rather than the whole thing. Actually, it’s not even an assumption: Tolkien’s map is obviously intentionally cut off in the south and the east; why should we assume that his world ends there? There’s enough room there for the hypothetical central plateau or even other continents and archipelagos.
 Ivan Yefremov was one of the classics of Soviet Science Fiction, who happened to be a paleontologist by profession