Abstract
Aquatic ecosystems are subject to many stressors including eutrophication, hydrologic alteration, invasive species, and climate change. These stressors alter ecosystems and their resilience – how an ecosystem responds to disturbances like storms, droughts, and other discrete events. Disturbance impacts can vary widely in magnitude, duration, and the changes they induce on ecosystem structure, function, and services. Understanding and predicting change has motivated development of theories and frameworks for several resilience concepts. However, applying resilience methods in practice is often challenged by data limitations and the inherent complexity of ecosystems.
The goal of this dissertation was to advance the application of resilience concepts to real world ecosystems using data intensive methods. I focused specifically on two ecosystem changes and resilience concepts: the prediction of algal blooms in lakes and understanding patterns and controls of disturbance in estuaries following tropical cyclones. I first evaluated if spatial early warning statistics (EWS), based on theory that generic changes in system dynamics are reflected in statistical properties, are expected to change prior to algal blooms. Using a spatial model incorporating physical forces that control transport in aquatic systems, I found that spatial standard deviation and autocorrelation distinguished between bloom states and changed predictably near thresholds. I then tested those findings and compared spatial EWS to previously studied EWS in time series data using a whole-lake nutrient addition experiment. Spatial EWS did not change consistently before the bloom, while temporal standard deviation did for 3 out of 4 variables. I then utilized high frequency time series from 18 lake-years of both experimental and non-experimental conditions to quantify temporal EWS performance at separating low from high resilience states, a necessary step for the method’s potential use for bloom management. Using high frequency data from the same lake fertilization experiments, I also explored the ability of near-term forecasting to accurately predict bloom initiation timing, a short but critical period for taking management action. Accurately forecasting bloom timing was difficult but possible and depended on both model initial conditions and flexibly adjusting parameters as new observations were collected. Finally, I used a new algorithm for detecting disturbance and recovery in high frequency data to quantify disturbance occurrence, timing, length, and severity in salinity and dissolved oxygen across 19 estuaries and 59 tropical cyclones in the eastern United States. Most estuaries recovered from hurricane-initiated disturbances within days, but some lasted weeks or months, and properties of both storms and the sites they impacted were related to disturbance characteristics.
This dissertation shows that resilience concepts can be operationalized to measurable properties, which can be used to understand and predict change with possibilities for application to ecosystem management. Operationalizing resilience is crucial to maintaining ecosystem services such as clean water, fisheries, and carbon sequestration into a future where stresses on aquatic systems are projected to intensify. My findings also demonstrate the power of ecosystem scale experiments as well as high-frequency and long-term data to test and advance understanding and management.
