NOAA GFDL - Geophysical Fluid Dynamics LaboratoryGFDL - Geophysical Fluid Dynamics Laboratory

Estuarine forecasts at daily weather to subseasonal time scales

October 21st, 2020

Key Findings

  • Forecasts of temperature and salinity in Chesapeake Bay are skillful beyond the few days of current forecasts
  • Taking the mean of multiple forecasts driven by different atmospheric ensemble members improves the estuarine forecast skill
  • Dissolved oxygen remains challenging to forecast.

Andrew C. Ross, Charles A. Stock, Keith W. Dixon, Marjorie A. M. Friedrichs, Raleigh R. Hood, Ming Li, Kathleen Pegion, Vincent Saba, Gabriel A. Vecchi. Earth and Space Science. DOI: 10.1029/2020EA001179

Estuary and coastal ocean forecasts based on predictions of temperature, salinity, currents, and storm surge have been shown to be able to protect lives and property from storm surge, assist search and rescue operations, and protect public health. These forecasts enable predictions of harmful algal blooms and the dispersion of oil spills. Many estuaries are also home to ecologically and economically important ecosystems and fisheries, and forecasts may also be useful for improving the management of fisheries and locating ideal fishing spots. This research demonstrates that temperature and salinity can be skillfully forecast at both the bottom and surface levels of an estuary up to two weeks in advance.

NOAA currently maintains a number of operational forecast systems for estuaries, including in Chesapeake Bay, Delaware Bay, and San Francisco Bay, which forecast conditions for the next 48 hours. However, realizing many of the possible benefits from these systems, such as improved management of fisheries, will require forecasts with longer lead times. Because estuaries are generally shallow coastal bodies of water, they respond strongly to atmospheric forcing, and since skillful atmospheric weather forecasts are currently possible out to around 10 days, longer lead forecasts of estuaries should be possible. Estuaries are also driven by regular, predictable tidal cycles of velocity and elevation, which should enhance predictive skill. Furthermore, estuarine salinity and circulation respond slowly to changes in river discharge, which implies a degree of predictive skill from previously observed river discharge and current conditions.

To test whether extended lead forecasts for estuaries are possible, the authors developed and tested a model system for producing 35-day forecasts of temperature, salinity, and dissolved oxygen in Chesapeake Bay. The model system consists of an ocean model with routine use in the Chesapeake Bay research and forecasting community, forced using 35-day atmospheric forecasts from NOAA’s Global Ensemble Forecast System produced as part of the Subseasonal Experiment. Using this system, the authors produce a total of 425 forecast simulations that cover the warm seasons (April through August) of 1999 to 2015. The skill of the forecasts was evaluated by comparing the forecast predictions with observations and with the results from a numerical model hindcast simulation designed to reproduce historical conditions.

Overall, the authors find that temperature and salinity can be skillfully forecast at both the bottom and surface levels of the estuary up to two weeks in advance. Forecast skill is higher for salinity and for bottom variables. The skill of the forecasts can be significantly improved by producing multiple ocean model forecasts, each driven by a different atmospheric forecast ensemble member, and averaging the results. Bottom dissolved oxygen concentrations can be forecast skillfully when the results are compared with the hindcast, but not when the forecasts are compared with the observations. This suggests that improvements to the dissolved oxygen model may be needed to obtain skillful forecasts.

The authors also closely evaluated the skill of the forecasts during two extreme events: Hurricane Irene in August 2011 and a heat wave in June 2008. The model system forecasted both events well, and heat waves and tropical cyclones are two types of extreme events with promising predictability at longer time scales, which suggests the potential for long-lead warning of potential impacts of these events on estuaries.

Skill of the model forecasts of surface temperature, surface salinity, and bottom dissolved oxygen at various forecast lead times ranging from the first day after the initial forecast day (Lead 1) to three weeks after the initial forecast day (Lead 21). A skill score of 100% indicates that the forecasts were a perfect match to the observations (which are obtained from the model hindcast for this figure), and skill scores above 0% correspond to skillful forecasts with a mean square error lower than the average square error of the long-term climatology, while skill scores below 0% indicate unskillful forecasts with high mean square error. These figures show that surface temperature skill is uniformly high across the bay during the first week of the forecast and declines to unskillful levels after the second week. Bottom oxygen skill declines over a similar timescale, although skill is primarily concentrated in the deep central channel. Surface salinity skill remains high for the entire 35-day forecast in regions away from the tributary rivers.

 

(a) Observed surface air temperature anomaly from an atmospheric reanalysis on June 9, 2008 and (b) surface air temperature anomaly on June 9 predicted by the ensemble mean of the GEFS forecasts that were used in this study and were initialized on June 4. These two subfigures show that the heatwave observed over the eastern US on June 9 was correctly predicted 6 days in advance by the atmospheric model. (c) shows the impact of the heatwave on sea surface temperature anomaly in Chesapeake Bay; the ensemble mean forecast (dashed blue line) correctly predicted water temperatures that were nearly 4 °C above normal. (d) shows the impact of the heatwave on dissolved oxygen concentration; in this case, the ensemble mean forecast (dashed blue line) correctly predicted a switch from a below-normal to an above-normal amount of water with oxygen concentration below the hypoxic threshold (i.e. high hypoxic volume) following the heatwave.