Methods

Microcystis Blooms - HAB

Microcystis Blooms - HAB

The main factors that determine HAB occurrence and characteristics in the Potomac River are nutrient availability (primarily phosphorus), salinity, water temperature, and light availability. For blooms to occur, the water temperatures have to be above 15 °C (59 °F) and salinity below 5 ppt. More intense blooms are also likely to occur if conditions are still (little wind mixing) and cloud free (higher sunlight). An overriding influence on bloom occurrences is river flow rates, most likely due to its effect on nutrient availability. As a result, monitoring has shown HAB variability associated with annual and seasonal weather patterns affecting nutrient delivery to the estuary.

Forecast approach: This year's forecast is based on a linear relationship between total Potomac River flow and the percent of water quality samples containing bloom levels of Microcystis. Note that these methods are slightly different than those previously used.

The algal community samples for assessing cyanobacteria levels are collected from the region of the Potomac between the Washington DC/Maryland border down to the Route 301 bridge. MD DNR typically collects 8 sets of 9 samples from June to September for a total 72 samples each year. An algal sample in this case is considered at bloom levels if it contains greater than 10,000 cells per milliliter of the cyanobacteria Microcystis.

River flow data are collected at the Point of Rocks location on the Potomac River. The 2008 portion of the data set is final flow data while January through May 2009 data are provisional.

Data used to develop the relationship are from 1989-2006, however, they exclude 1997 and 2004. The bloom conditions in those two years are much lower than expected by the trajectory of the model and may be a function of conditions related to conditions associated with extremely wet conditions (2003) to flood level flow events (January 1996) impacting the region. When cumulative flow over the Point of Rocks on the Potomac River is greater than 5.9 x 10¹¹ cubic feet over the 17 months preceding the summer as is the case with these two years, our linear model fails. Further model developments are needed to better capture the conditions involving such high flow years to more accurately model the bloom patterns.

Data are available for the earlier period 1985-1988. These data are not included in this cyanobacteria forecast model, however, as phosphate bans were going into effect among the partner States and Washington DC in the Potomac River basin over multiple years.  The post-1989 P-ban implementation era represents the time beyond this important and significant nutrient control period. The forecast could benefit from the flow model being expanded to include nutrient loads or a similar nutrient index.

The worst year in our 1989-2006 time series was 1994 and it showed 25% of all summer samples containing bloom levels of Microcystis. Since that time we have experienced widespread blooms (2004) but that bloom expanded in July and died out in August limiting its duration. Recently, bloom conditions have been less than levels forecast with our basic river flow model. In 2007 MD DNR samples showed no bloom detections and in 2008, the model predicted 9% of samples would detect bloom conditions when only 1.3% of samples detected blooms. Once again this year based on the flow conditions leading up to this summer we predict 9% of samples will show Microcystis bloom conditions for summer 2009, a moderately severe cyanobacterial bloom year for the Potomac River.

Prorocentrum Blooms - HAB

Prorocentrum Blooms - HAB

The Prorocentrum forecast is based on a model that relates river flow and water quality (temperature, salinity, and clarity) to Prorocentrum bloom conditions. Seven routine water quality monitoring stations were chosen to represent distinct regions in the Maryland section of Chesapeake Bay that typically see high concentrations of Prorocentrum (Figure 1). Water quality data for the model was supplied by the Chesapeake Bay Program and Maryland Department of Natural Resources, and river loads data was supplied by USGS. Data from January 1985 to May 2009 was used in the model.

The predictive model uses a classification and regression tree method to partition "bloom" and "no bloom" conditions based on above-pycnocline water temperature, salinity, water clarity, and mean river flow. A fuzzy logic membership curve was implemented that incorporates the uncertainty of how to define a bloom concentration of Prorocentrum (Figure 2). The membership curve defines how to classify a particular concentration as "bloom" or "no bloom". A lower boundary is set for defining a "bloom" that reflects the certainty among regional harmful algal bloom experts that concentrations below 1000 cells·mL^-1 are not indicative of "bloom" conditions. Conversely, it is generally agreed that concentrations greater than 5000 cells·mL^-1 are indicative of "bloom" conditions. Concentrations between 1000 and 5000 cells·mL^-1 are allowed to have membership in both "no bloom" and "bloom" designations, depending on their proximity to either the lower or upper thresholds. For example, a Prorocentrum concentration of 4000 cells·mL^-1 would be classified as 75% "bloom" and 25% "no bloom". Fuzzy memberships are used to weight the classification and regression tree decision-making, such that higher weights are given to data that are closer to either extreme (i.e., weights are proportional to the certainty of the bloom designation).

Since the objective is to create a predictive model of Prorocentrum blooms, a temporal offset was imposed, where water quality data from prior dates were fit to observations of Prorocentrum in the future. Twelve different offsets were applied, ranging from 0.5 months to 6 months prior, resulting in twelve distinct predictions for a particular date. Figure 3 shows the model with temporal offset of 0.5 months for station CB3.3C.

Each model's performance was assessed by fitting the model's predictions of 2005-2007 to observations of that same time period. Relative performances of each of the twelve models were calculated from Akaike Information Criteria (AIC) and evidence ratios calculated from equations defined in Burnham and Anderson (2002). For each prediction, the model with the highest evidence ratio was chosen. However, if water quality data was not available at the "best" model's designated offset, the second best model was used.



 

Figure 1. Map of Chesapeake Bay Program monitoring stations (red dots) used to develop the predictive model. Shaded regions represent the areas in which predictions apply. Green dots represent USGS river discharge gauges used in the model.




Figure 2. Fuzzy membership curve describing how a Prorocentrum "bloom" is defined. Cell concentrations below 1000 cells·mL^-1 are generally not indicative of bloom conditions. Conversely, cell concentrations above 5000 cells·mL^-1 are considered a reasonable "bloom". The overlap between 1000 and 5000 cells·mL^-1 conveys the uncertainty of defining a bloom.




Figure 3. A sample decision tree for water quality station CB3.3C ("Bay Bridge" region) using water quality conditions that occur 0.5 months prior (a).

Literature Cited:

Burnham, K.P., and D.R. Anderson. (2002) Model selection and multimodel inference. 2nd Edition, Springer-Verlag, New York, NY. 488pp.

DO - anoxia

DO - anoxia

There are many factors that determine the dissolved oxygen content of the tidal waters of Chesapeake Bay. Nutrient loading, water column stratification, wind and tidal mixing, and water temperatures are but a few of these factors. The two most important determining factors are water column stratification and nutrient loading.



Water column stratification is caused by density differences between the surface and deeper waters of the Bay. Cooler, saltier (more dense) water from the ocean flows underneath the warmer, fresher (less dense) water from the rivers that flow into the Bay. Between the lighter surface water and heavier deeper water is a boundary called the pycnocline. Oxygen consumed beneath the pycnocline cannot be replenished from above, and this leads to lower dissolved oxygen concentrations below the pycnocline. The pycnocline is typically strongest in spring and early summer when fresh water flows are usually at their highest.

Nutrient inputs to the Bay from the land are directly related to precipitation and therefore river flow. Nutrient loads from land-based sources (agriculture, urban runoff, etc.) are higher in the spring when river flows are typically at their highest. Nutrients that flow directly into the Bay from a pipe (sewage treatment plants, industry, etc.) are generally less sensitive to flow and are more consistent through the year. There is a direct relationship between the magnitude of these nutrient loads and the severity of low DO the Bay experiences. Nutrients – nitrogen and phosphorus – fuel the growth of the phytoplankton that make up the base of the Bay's food web. Unconsumed phytoplankton settle below the pycnocline and are decomposed by oxygen–consuming bacteria living in the mud on the bottom of the Bay. Since this is occurring below the pycnocline, this oxygen is not replenished from surface waters. This process occurs every year in Chesapeake Bay, fueled by spring flows that wash large amounts of nutrients into the Bay. Examination of the Chesapeake Bay Program's 20 year data set has shown that the severity of summertime low DO is directly related to the magnitude of spring nutrient loads.



The DO forecast was developed based on this relationship which is expressed as mean June to September anoxic volume (<0.2 mg L^-1) versus nutrient loads to the northern Chesapeake Bay. Nutrient loads are the combined total nitrogen (TN) and total phosphorus (TP) loads from the Susquehanna River and point sources on the upper western shore, upper eastern shore and the Potomac River.

The anoxic forecast released in 2009 is based on the January to April nutrient loads.

DO - hypoxia

DO - hypoxia

The hypoxic forecast model predicts oxygen concentration downstream from point sources of organic matter loads using two mass balance equations for oxygen-consuming organic matter, in oxygen equivalents (i.e., BOD), and dissolved oxygen deficit. This approach to modeling coastal and estuarine hypoxia has also been used successfully for Gulf of Mexico hypoxia (Scavia et al. 2003, 2004). The original model was calibrated and tested against 1950-2003 nitrogen load and hypoxic volume estimates assembled by Hagy (2002). The Chesapeake Bay Program provided load and hypoxic volume updates for 1986-2008, and even though the new estimates varied little from the original ones; the model was recalibrated for this application to the new 1986-2008 estimates. The summer hypoxic volume forecast was generated using the following relationship.

Additional Info

Summer ecological forecast technical documentation
Dave Jasinski, Peter Tango, Michael Williams, Ben Longstaff

This document describes the rationale and methods for determining the summer ecological forecasts for dissolved oxygen, harmful algal blooms and submerged aquatic vegetation for 2005.