ABRUPT CHANGE IN LONG-TERM LAKE DATA: DETECTION AND PATTERNS

Understanding mechanisms of abrupt ecological change is increasingly important as climate change places novel pressures on system drivers. However, the effectiveness of detection techniques and the patterns of abrupt change detection across and within ecosystems are unknown. We simulated abrupt changes in time-series to test the sensitivity of detection methods and applied those methods to lake monitoring data from the North Temperate Lakes (NTL) Long-Term Ecological Research site. In 1000 simulations of Gaussian time-series data with constant mean and standard deviation (0 and 1 respectively), Pettitt’s test for change-point detection found a change (p < 0.05) in 2% of the datasets, less than expected by chance. When the mean increased from 0 to 1 or 2 halfway through the time-series, simulating abrupt change, detection rose to 68% and 100% respectively. Detection of abrupt change was slightly lower when the mean changed gradually. When fitting simulations to abrupt and gradual change curves, the gradual model rarely outperformed the abrupt model, even when the change was gradual. Of 169 seasonal average time-series collected by NTL (8 variables, 7 Wisconsin lakes, 1 - 4 seasons), 49 had significant change points, with 24 fitting the abrupt model and 9 fitting the gradual change model. There was no distinct pattern in changes among biological, physical, and chemical lake variables. Testing abrupt change detection methods with simulations that reflect the types of data collected by the NTL will strengthen our ability to quantify patterns of abrupt change in lake ecosystems.

Primary Presenter: Allison Kneisel, University of Wisconsin - Madison (ankneisel@wisc.edu)

Authors:

Allison Kneisel, University of Wisconsin - Madison  (ankneisel@wisc.edu)

Monica Turner, University of Wisconsin - Madison  (turnermg@wisc.edu)