The Cumulative Frequency Diagram Method for Determining Water Quality Attainment

In accordance with the Chesapeake 2000 Agreement, the Chesapeake Bay Program has recently imple mented important modifications to (1) ambient water quality criteria for living resources and, (2) the procedures to determine attainment of those criteria. A novel statistical tool for attainment, termed the Cumulative Frequency Diagram (CFD) approach, was developed as a substantial revision of previous attainment procedures, which relied upon a simple statistical summary of observed samples. The approach was viewed as advantageous in its capacity to represent degrees of attainment in both time and space. In particular, it was recognized that the CFD could represent spatial data in a synoptic way: data that is extensively collected across diverse platforms by the Chesapeake Bay Program Water Quality Monitoring Program. Because the CFD approach is ne w to Bay Program applications, underlying statistical properties need to be fully established. Such properties are critical if the CFD approach is to be used to rigorously define regional attainments in the Chesapeake Bay. Advisory Committee charged our working group to provide review and recommendations on the CFD attainment approach. As terms of reference we used guidelines of Best Available Science recently published by the American Fishe ries Society and the Estuarine Research Federation. Statistical issues that we reviewed included, 1. What are the specific analytical/statistical steps entailed in constructing CFD attainment curves and how are CFDs currently implemented? (Section 2) 2. How rigorous is the spatial interpolation process that feeds into the CFD approach? Would alternative spatial modeling procedures (e.g., kriging) substantially improve estimation of water quality attainment? (Section 3) 3. What are the specific analytical/statistical steps entailed in constructing CFD reference curves? (Section 4) 4. What are the statistical properties of CFD curves? How does sampling density, levels of attainment, and spatial covariance affect the shape of CFD curves? What procedures are reliable for estimating error bounds for CFD curves? (Section 5) 5. From a statistical viewpoint, does the CFD approach qualify as best available science? (Section 6) 6. What are the most important remaining issues and what course of directed research will lead to a more statistically rigorous CFD approach over the next three years? (Section 7) The central element of our work was a series of exercises on simulated datasets undertaken by Dr. Perry to better evaluate 1) sample densities in time and space, 2) varying leve ls of attainment, and 3) varying degrees of spatial and temporal covariance. …