Model predictive control sounds like specialist vocabulary, and the underlying idea is something people do intuitively. You have a goal, a forecast, and a set of actions. You plan a sequence, take the first step, then re-plan with updated information.
The driving analogy
Approaching a curve, you do not choose one steering angle and hold it. You look ahead, predict where the car will go, adjust, and repeat continuously. That loop of predict, act, observe, and re-plan is the whole concept.
Applied to a field
The goal is keeping soil moisture in a range that supports the crop while using as little water as possible. The forecast covers weather and expected plant uptake. The actions are irrigation events over the coming days.
The controller plans a watering schedule across a horizon, applies the first decision, then recomputes once new soil moisture and weather data arrive. If rain arrives unexpectedly, the plan absorbs it rather than continuing on a stale schedule.
Why it beats fixed schedules
A calendar-based schedule cannot respond to conditions. A purely reactive system waters after stress appears, which is already too late. Predictive control looks forward while continuously correcting, which is why it handles variable weather better than either alternative.
The catch is that predictive control is only as good as its forecast and its model. Garbage inputs still produce garbage schedules, which is why data quality work is inseparable from control design.