Shallow lakes are known for sudden shifts between a desired clear and an undesired turbid state despite only incremental changes in the underlying drivers. Such sudden shifts are a major challenge for lake managers who can be confronted with abrupt losses of desired ecosystem services without easily observable warning signals. Predictive tools for the loss of ecosystem resilience are vital to respond with timely mitigation measures and avert a shift to the undesired state. Early-warning indicators (EWIs) have faithfully preceded critical transitions in minimal models but have proven more elusive in real-world data, suggesting a mismatch between measurement strategy and the detectability of EWIs. Here, we capitalize on data simulated using the aquatic ecosystem model PCLake+ which represents real systems more closely than reductionistic models and which allows the generation of critical transitions in response to gradual changes in phosphorus load. We tested the effect of different sampling intervals (daily to yearly) on the detection of three often-used EWIs across a range of food web and nutrient-related variables. Moreover, we included one integrated sampling interval (yearly average of daily measurements) to represent time-integrated measurements. EWIs generally performed better at shorter intervals (daily, weekly) but integrated measurements over the year also proved suitable to detect oncoming state shifts. We propose that lake managers should aim for high-frequency measurements of variables that can be easily and cheaply measured (e.g. oxygen, Secchi) or, alternatively, focus on integrated approaches using passive samplers or sedimented material.
