2022 journal article
Temporally resolved coastal hypoxia forecasting and uncertainty assessment via Bayesian mechanistic modeling
Hydrology and Earth System Sciences, 26(4), 1131–1143.
Abstract. Low bottom water dissolved oxygen conditions (hypoxia) occur almost every summer in the northern Gulf of Mexico due to a combination of nutrient loadings and water column stratification. Several statistical and mechanistic models have been used to forecast the midsummer hypoxic area, based on spring nitrogen loading from major rivers. However, sub-seasonal forecasts are needed to fully characterize the dynamics of hypoxia over the summer season, which is important for informing fisheries and ecosystem management. Here, we present an approach to forecasting hypoxic conditions at a daily resolution through Bayesian mechanistic modeling that allows for rigorous uncertainty quantification. Within this framework, we develop and test different representations and projections of hydrometeorological model inputs. We find that May precipitation over the Mississippi River basin is a key predictor of summer discharge and loading that substantially improves forecast performance. Accounting for spring wind conditions also improves forecast performance, though to a lesser extent. The proposed approach generates forecasts for two different sections of the Louisiana–Texas shelf (east and west), and it explains about 50 % of the variability in the total hypoxic area when tested against historical observations (1985–2016). Results also show how forecast uncertainties build over the summer season, with longer lead times from the nominal forecast release date of 1 June, due to increasing stochasticity in riverine and meteorological inputs. Consequently, the portion of overall forecast variance associated with uncertainties in data inputs increases from 26 % to 41 % from June–July to August–September, respectively. Overall, the study demonstrates a unique approach to assessing and reducing uncertainties in temporally resolved hypoxia forecasting.