Illustration of the complexity of the learned decision rule during the period of intensive irrigation in the replicate corresponding to Fig 3.

Each p(A) indicates the probability for applying A amount of water (0, 10, 20, 30 or 40 mm) prescribed based on values of nine state variables (only six shown for the space restriction). ESW1 and ESW2 are similar on Day 240 and 258, but the probabilities for no irrigation p(0) are significantly different. After 10 mm of irrigation, p(0) falls by 7 percentage points on Day 249, remains unchanged on Day 252, and rises by 18 percentage points on Day 265.