For a LEO ground station I would start from the link budget, not from a favourite array size, and in most of the cases I have worked the sweet spot has been a modest electronically steered array rather than a single large dish or a very large array.
The reason is the trade LunaWei already named. G/T sets the minimum aperture: work out the required G/T for your band, the satellite EIRP, the data rate and the rain margin, then translate that into an aperture and an element count. For a typical Ku-band downlink at a few hundred megabits per second, that usually lands in the region of a 64 to 256 element array, or an equivalent aperture of roughly 0.6 to 1.2 metres, depending on the element gain and the LNA noise figure. Going much larger buys you G/T, but it also buys calibration pain: every extra element adds phase and amplitude error, and at low elevation the scan loss and the mutual coupling change the pattern in ways a simple array factor does not predict. Going much smaller is cheaper, but then you are spending the saving on a bigger LNA, a lower noise temperature, or accepting a lower data rate on the worst passes.
Tracking is the second constraint, and it is often the binding one. A LEO pass moves fast, and a mechanically steered dish has to slew and settle while the satellite is already moving to the next position. An electronically steered array can switch beams in milliseconds and can even track two satellites in sequence without a mechanical move, which matters if you want handovers without dropping a pass. The cost of that is calibration: you need a stable reference, regular phase and amplitude checks, and a way to detect a failed element without taking the array down.
My practical rule is to price two options, as LunaWei suggested: the smallest array that meets the G/T target with a 2 to 3 dB margin, and the same array with one extra 3 dB of margin. If the second option costs more than about 30 to 40 percent extra for that 3 dB, I usually stay with the first and spend the difference on a better LNA, a cleaner feed, or a more robust calibration routine, because those buy back more link margin per unit cost than more elements. And I would ask the same question back to you: which band and data rate are you targeting, and is your bottleneck G/T, tracking speed, or cost?