described above. Therefore, EPA recommends that the States continue to seek funds 
to support this type of data collection in order to directly generate the data supporting 
attainment assessment of the full array of applicable dissolved oxygen criteria. 
APPLICATION OF LOGISTIC REGRESSION TO ASSESS SHORT- 
DURATION DISSOLVED OXYGEN CRITERIA COMPONENTS 
In the prior sections, it was noted that the data collection frequency of the long term, 
fixed-station water quality monitoring program is inadequate to assess attainment of 
short-duration criteria components. However, the greater than 20-year record of 
dissolved oxygen measurements collected relatively synoptically throughout the 
mainstem Bay, tidal tributaries and embayments, and collected regularly throughout 
the annual cycle provides a very substantial data base from which to derive infer¬ 
ences and define quantitative relationships between seasonal and monthly mean 
dissolved oxygen concentrations and the frequency of observations above and below 
specified criterion concentrations. Where relationships are strong, the logistic regres¬ 
sion procedure produces models in the form of simple equations that 
estimate/predict the likelihood that the criterion threshold concentration was attained 
or violated during the period. 
This method was explored originally to measure attainment of the 1992 Chesapeake 
Bay dissolved oxygen restoration goal (Jordan et al. 1992) and was adapted for 
assessing attainment of the 2003 Chesapeake Bay dissolved oxygen instantaneous 
minimum (see Chapter 5, pages 27-62, in U.S. EPA 2004). The 2003 method modi¬ 
fications included spatial and temporal refinements to the predictive models, with 
consequent improvements to their goodness of fit. The early (1992) models esti¬ 
mated exceedance based on segment-specific seasonal means and whether the means 
were from depths above or below pycnocline. The 2003 method update was enriched 
with an additional decade of monitoring data (1990-2000) for the regression analysis 
and provided segment-specific models for individual months and depths. Recent 
progress on this work again includes several additional years of new fixed-station 
and continuous monitoring buoy data (2001-2005) and modifications to implemen¬ 
tation procedures that could provide results for attainment assessment through the 
CFD methodology in a format consistent with other dissolved oxygen criteria. 
In this latest iteration, logistic regression models for the individual instantaneous 
minima are developed for each station. The independent variables are, as before, 
mean dissolved oxygen, month and water depth. The addition of a depth-squared 
variable for deep stations is being tested, but not yet implemented. The dependent 
variable is an indicator that the minimum threshold (e.g., the instantaneous criterion 
concentration) is violated. (Since the CFD methodology is based on percent failure, 
the dependent variable is based on exceedance rather than attainment.) This model¬ 
building step currently uses the entire 1985-2005 water quality data record at each 
station. Over time, however, if trends in ambient dissolved oxygen indicate signifi¬ 
cant, sustained change in a segment, then the extent of the historical record to be 
included in this step should be re-examined. 
The collection of station models is used to estimate a predicted probability of 
exceedance for each station, for each month in the 3-year, multi-month seasonal 
assessment period, at each meter of depth. Then, for each month, the predicted prob- 
appendix e • Potential Methods for Assessing Shorter Duration Dissolved Oxygen Criteria 
