For the time being, it probably doesn't make good economic sense to dedicate sophisticated machinery to managing a patch of ground that's unprotected from the elements, when it might just as well be working inside a greenhouse, where it actually can operate 24/365, and where it won't need the structural strength to stand up to gale force winds.
On the other hand, greenhouses are most useful when combined with regular gardens and used seasonally to start plants earlier than they could be started in the open, or when the shade from other plants would impede sprouting or development. Even a relatively frail machine, better adapted to spending its time indoors, might venture out in calm weather, long enough to set out plants it had started in trays and peat pots, provided it was sufficiently mobile.
This scenario, a machine that does the tedious work of planting seeds and tending plants in a greenhouse, moving them out to open ground when conditions allow, is what might be termed a natural starting point for the development of such machines, a more limited, more surmountable engineering problem than a machine intended to perform all aspects of land management. A machine applied this way need not be able to perform absolutely every horticultural operation to be useful, nor would it need to be able to deal with a completely uncontrolled environment.
It's very likely that there are other such natural starting points for the development of cultibots. Collectively, these natural starting points represent a threshold of minimal investment before a return on that investment can be forthcoming. Once that threshold has been crossed, at any point, incremental improvements should be adequate to insure that machines which can handle the whole job are eventually produced, and the return on that initial investment could be very sweet indeed.
There's another, equally important threshold to consider, the automation of the production of these machines. So long as they are hand-crafted prototypes, they have no chance of competing economically with hand labor, or, as is more likely, with the transportation of produce from milder climates. Mass production will get them into the game.
Self-reconfiguring factories that not only build such machines but which can also replicate, by building the equipment for new factories and the machines to assemble them, will drive down the cost to the point where the logic behind it all becomes inexorable, but by that point we're no longer talking only about machines for land management, and there had best be very solid safeguards in place. The point of mentioning this scenario in the context of cultibotics at all is that land management may be the one application of robotics where the size of the potential market could justify the investment to cross this threshold. Once crossed, of course, the technology would be generally applicable.