Predictive capability of U.S. OECD phosphorus loading- eutrophication response models

Mathematical models are being used with increasing frequency to estimate the responses of water bodies to changes in nutrient loadings, and to assist in developing and assessing the water quality benefits to be accrued from various eutrophication management options. Although the current models have various forms and varying degrees of complexity, two basic types of mathematical models have been developed recently for eutrophication modeling: models and models. Dynamic models consist basically of a series of interrelated differential equations which attempt to describe the biological, chemical, and physical reactions and interactions governing aquatic plant growth (usually algae), including nutrient loads and other driving forces, such as light and temperature. These equations are solved simultaneously, usually by computer, to give a quantitative expression of algal growth dynamics and other related lake responses (such as in-lake nutrient levels) over a given period. Such models are calibrated (tuned) f o r a given water body by adjusting the values of the coefficients in the various equations to reflect the specific in-lake conditions for the various parameters included in the equations to the extent necessary to make model output match measured characteristics of the water body. Therefore, such models, because they were system dependent (that is, they must be readjusted for each water body), are not widely applicable to other water bodies having substantially different characteristics, or to the same water body if the driving force load or other characteristics are substantially changed. In addition, the data requirements for these models are usually very extensive. Because of such factors, the use of dynamic models has thus far generally been limited to sitespecific examples. No dynamic model has been demonstrated to have appreciable predictive capability beyond the situations for which they were designed. Prominent examples of dynamic models that have been developed for the North American Great Lakes include the phytoplankton model (Lake I) developed for Lake Ontario by Thomann e t a l . ; 1 , 2 t he dissolved oxygen (DO) model developed for lake Erie by DiToro and Connolly; and the phytoplankton class succession model developed for Saginaw Bay, Lake Huron, by Bierman. A statistical or empirical eutrophication model, by contrast, is basically a statistical regression that quantifies a basic cause-and-effect relationship, and does not attempt to account for or characterize every component involved in the eutrophication process. The statistical models discussed in this paper were developed by plotting the value of the chosen measure of effect (that is, algal production as measured by a response parameter such as chlorophyll concentration) as a function of the factor exerting primary control over algal growth (P load suitably normalized), for each in a group of water bodies, and determining the line of best fit or regression through the individual points. Such models can he viewed, therefore, as statistical approaches relating nutrient (P) inputs and resultant eutrophication-related water quality responses to these inputs. They will not provide a detailed description of in-lake nutrient or algal dynamics, but rather, are formulated to describe the steady-state or equilibrium response. The data requirements are usually much less extensive for statistical models than for dynamic models.