Students Research Presentations: Predicting anoxic volumes of Chesapeake Bay: Utilizing bootstrapping to improve forecasts

Year:

2017

Source:

Coastal and Estuarine Research Federation, Providence, Rhode Island

Abstract:

The health of Chesapeake Bay suffers from anoxic events, which occur seasonally in this stratified estuary. Over the past several decades, the frequency and severity of anoxic conditions have varied greatly. Given the deleterious effects on estuarine biology and chemistry, efforts to forecast the annual volumes of anoxic water have been made for nearly a decade in Chesapeake Bay. Several variables, including nitrogen loads and freshwater flow from contributing rivers, correlate with the observed anoxic volume and can be used for its predictive modeling. However, residuals of such models often violate the assumption of normality, therefore, normality-based predictive intervals of anoxic volume become unreliable. We relax the parametric assumption by employing a sieve bootstrap for calculating the predictive intervals. We demonstrate the performance of the sieve bootstrap approach by comparing it with a traditional parametric approach in a leave-one-out cross-validation study. Both types of intervals are assessed by their observed coverage probability (measure of calibration) and width (measure of sharpness). Using information from 1985-2016, we forecasted the anoxic volumes for 2017 and evaluated the success of the prediction. Additionally, we apply a sliding window approach to study temporal changes in the relationships between predictors and anoxic volumes in Chesapeake Bay.

Mentors:

Vyacheslav Lyubchich, Ph.D.

Students:

Robert Jarrett, Rutgers University