Trend coefficients and AUC statistics across four control populations (“Pop’n 1” to “Pop’n 4”) and four warming populations (“Pop’n 5” to “Pop’n 8”), as calculated from empirical time series during the approach to criticality. Each statistical metric was calculated within fifteen-day sliding windows, throughout the pre-critical interval. Given that the temperature of the system increased to 12°C on day sixty, we also considered three pre-critical intervals: Days 1 to 60, Days 20 to 60, and Days 30 to 60. Here, we show analyses of empirical time series performed within the sixty-day pre-critical interval. See analyses within the forty- and thirty-day pre-critical intervals in S7 Fig. To evaluate trends in these metrics, we calculated Kendall’s rank correlation coefficient during the pre-critical interval. Negative values indicate a decreasing trend prior to local bifurcation, while positiv

Trend coefficients and AUC statistics across four control populations (“Pop’n 1” to “Pop’n 4”) and four warming populations (“Pop’n 5” to “Pop’n 8”), as calculated from empirical time series during the approach to criticality. Each statistical metric was calculated within fifteen-day sliding windows, throughout the pre-critical interval. Given that the temperature of the system increased to 12°C on day sixty, we also considered three pre-critical intervals: Days 1 to 60, Days 20 to 60, and Days 30 to 60. Here, we show analyses of empirical time series performed within the sixty-day pre-critical interval. See analyses within the forty- and thirty-day pre-critical intervals in S7 Fig. To evaluate trends in these metrics, we calculated Kendall’s rank correlation coefficient during the pre-critical interval. Negative values indicate a decreasing trend prior to local bifurcation, while positiv