Whole-ecosystem test of early warnings for cyanobacteria blooms
Theory suggests that changes in time-series statistics can foreshadow an upcoming regime shift. Such approaches have rarely been tested in field experiments. As lakes are enriched with nutrients, theory predicts that both variance and autocorrelation should increase prior to the onset of cyanobacteria blooms. We enriched Peter (low color) and Tuesday (stained) lakes with N&P (3 mg P m-2 d-1, N:P ratio 15 by moles) to test this prediction. Paul Lake was unenriched as a reference ecosystem. Before and during enrichment, sondes measured chlorophyll, phycocyanin, dissolved oxygen, pH and temperature every 5 minutes. Pigment concentrations were measured manually every day, and a wider set of limnological variables were measured weekly. Standard deviations (SD), lag-1 autocorrelation coefficients (AC), and other potential indicators were computed for 28-day rolling windows. The enriched lakes were compared to the reference lake using the Quickest Detection method (Oikos (2014) 123: 290-297). The detection criterion was an increase in the test statistic (SD or AC) by more than two standard deviations above the reference lake.
A cyanobacterial bloom occurred in Peter Lake (low color) after 70 days of enrichment. Time series for chlorophyll and phycocyanin showed strongly significant increases in SD and AC more than 55 days before the onset of the bloom. Thus the time series of pigment concentrations changed as expected from theory. In Tuesday Lake (stained), chlorophyll increased but there was no bloom of cyanobacteria. AC showed significant increases about 14 days after enrichment started. However no significant increases in SD occurred until 80 days after enrichment. The increases in SD were associated with a modest increase in chlorophyll near the end of the enriched period.
We conclude that indicators correctly forecast the impending cyanobacteria bloom in low-color Peter Lake, and correctly gave no indication of impending bloom in stained Tuesday Lake. In this whole-lake experiment resilience indicators were accurate, in the sense that they gave two correct forecasts (one positive and one negative) and no incorrect forecasts. These findings suggest that time series of SD and AC, interpreted together, are robust indicators of resilience in enriched lakes that could potentially develop cyanobacterial blooms.