that the dissolved oxygen concentration at that depth in a segment is at or above a 
user-specified concentration, e.g., an instantaneous minimum of 1.7 mg liter 1 (see 
Appendix I for more details). 
Application of the Logistic Regression Approach. The method can be applied 
using the three-dimensional baywide interpolations of monthly average dissolved 
oxygen, as described for the determination of 30-day duration criteria. The monthly 
average concentrations at each fixed station at each half-meter are interpolated hori¬ 
zontally by the Chesapeake Bay interpolator to yield a basinwide grid of 
concentrations for each month. A comparable reference grid or a table of grid coor¬ 
dinates and depths relate the monthly cell concentrations to be evaluated with the 
correct designated use and corresponding criteria concentration (e.g, instantaneous 
minimum of 1.7 mg liter' 1 ). In the data processing step, a segment- and criterion 
level-specific prediction model uses the cell’s monthly average concentration, depth 
and month as factors in predicting the percent of the time that particular cell is at or 
above the criterion. The cell is scored as passing or failing the criterion level 
depending on the model results. The cell volume is accumulated in the pool of 
passing or failing totals for each designated use in each segment. Like the method 
for assessing the 30-day mean, the spatial extent of nonattainment, i.e., the 
percentage of the total volume exceeding the criterion in each designated use in each 
segment, is tallied for each month in the assessment period (most recent three years). 
The cumulative frequency distribution attainment and reference curves can then be 
derived, and the same statistical test for determining attainment as described for the 
direct assessment method can be applied. 
Strengths and Current Limitations. The logistic models are based on conditions 
represented by the fixed stations in the current monitoring program, which in most 
tributaries are sited in the main channel. Until more data are collected, the similarity 
of shallow areas to the midchannel in the same segment is not known. This approach 
would assume, in the absence of other data, that the main channel data are represen¬ 
tative of similar depths in the shallows. If salinity or other physical data from the 
shallows indicate that all or part of the shallow water column is more similar to a 
different depth in the midchannel (as is sometimes the case for various reasons), then 
the more representative depth would be used to estimate percent attainment. For 
example, the pycnocline typically is deeper in the portion of the Chesapeake Bay 
than on the flanks, and the depth of the pycnocline on one flank is typically deeper 
than the other. Thus a 4-meter-deep, above-pycnocline water parcel on one flank 
may be most similar to the 4-meter-above-pycnocline depth in the midchannel 
profile, while the 4-meter-deep, subpycnocline parcel on the opposite flank is more 
similar to the 5-meter depth in the midchannel profile. 
To date, dissolved oxygen concentrations have shown little significant trend in most 
areas of the Chesapeake Bay and its tidal tributaries and, therefore, history-based 
estimation models are reasonable. However, where significant trends are detected, it 
would be important to review the models and their basis in light of new, emerging 
empirical relationships at those locations. This approach provides an estimate of the 
chapter vi • Recommended Implementation Procedures 
