Fig. 6

Model of non-informative versus informative environment. We used our model to analyze the evolution of memory in two types of environments denoted informative and non-informative. Both environments were generated with the same overall probabilities of warning (0.5%) and stress (0.4%) periods to occur. Only 2’000 out of 100’000 time steps are shown. In a events were distributed randomly without temporal correlation between warning and stress periods. In such an environment the occurrence of a warning period had no informative value to whether a stress period will follow. In b a stress period followed a warning event with a probability of 80%. The distance between the warning and the stress event followed an exponential distribution with λ = 0.2 (see Additional file 5: S9.1 and S9.2 for a sensitivity analysis of lambda). Panel c shows the first 250 time steps of panel b. Panel d visualizes how the non-informative environment was constructed. Starting with a favorable period, the probability that the next period was warning was 0.5%, while the probability that the next period was ‘stress’ was 0.4%. Note that ‘warning’ and ‘stress’ periods were independent of each other (except for the fact that stress and warning could not immediately follow each other). Panel e visualizes how the environment with informative value was constructed. The informative value lies in the order of periods. A stress period followed a warning period with a probability of 80%