I would think a purely technical approach to this is flooding the plant every hour, and then measuring the differential of how much water had flooded the roots vs. how much water was pumped out.

If the differential is tending to 0 then it means that not only the soil is saturated, but the plant is as well. This is _quite_ a basic concept, since you do not want a fully saturated soil all of the time.

A proper biological approach is to estimate how much water exactly the plant needs per unit of time, and then make sure the differential is always exactly that amount. This is not an exact science, especially when you're only working with a single plant and your conditions are probably far from ideal.

This approach would easily take into account other hidden variables, such as the rate of evaporation of water from the soil that depends on the ambient temperature. It also scales to hydroponics.

This is basically hydroponics. And by "basically" I mean "exactly". Replace the soil with an inert, porous medium like hardened clay. And if you aerate the water with say, an aquarium pump/stone, you don't even need to drain it...

The drain back into the reservoir is so you don't have to worry adding too much water in one go as it'll just drain back out right away. The soil will only hold a certain amount. By not having the soil soak water back up from the pan you don't need to worry about controlling the amount of water you just need to time the period between waterings. The drier the soil the more water it'll absorb every watering period so it's reasonably self correcting.

A rough ballpark on watering cycles is usually good enough. I'd watch the leaves to give you an idea. I'd stop watering and wait until the leaves show signs of under-watering and then use slightly less time as my watering period. I'd guesstimate based on your setup that the period will be measured in days.

IMHO if you're going to have that much set-up you might as well go hydro.

And, as I'm assuming the real aim is to build cool things perhaps you could use Deep Learning to do leaf classification (Over Watering | Under Watering | OK). That way you could use a webcam to control the watering instead of the sensors. Knowing your watering times and regular classification samples you could use a fourier transform to help identify the optimum watering period. Perhaps someone could do this as an API service. I do Deep Learning on images as my job so if you want I could tell you how to create the training data and once you have that I could train a classifier for you.