22 DWYER, NIXON, OVIATT, PEREZ, AND SMAYDA 



require a quantitative compartment model for the ecosystem. Thus i 

 is a "black-box" approach and minimizes the effects of thi 

 investigator's biases about the structure and function of the ecosys 

 tem under study. 



Since spectral frequency response analysis of ecosystem behavio 

 in response to year-to-ye£ir fluctuations provides little informatioi 

 about its linearity or stability in response to fluctuations at highe 

 amplitudes or frequencies of input, ecosystem experiments usinj 

 artificicil inputs of the desired characteristics are required. Eisne 

 (1971) proposed this approach to examine the dynamios of ai 

 ecosystem perturbed far from a stable point and outlined criteria fo 

 input design which optimize the detection of response abov( 

 background noise. 



Applying a time-varying input to a whole ecosystem usualh 

 presents significant methodological problems. Using controlled 

 replicated microcosms to simulate an ecosystem alleviates some o 

 the difficulties. Potentially, microcosm "models" can be used t( 

 delineate the boundaries of the ecosystem's stable domain, h 

 addition to spectral analysis of the Narragansett Bay time series, W( 

 describe the results of a first experiment using a sewage input tc 

 perturb 150-liter microcosms simulating Narragansett Bay beyonc 

 their nominal dynamics. 



MATERIALS AND METHODS 



Description of Data 



The dynamics of phytoplankton abundance in Narragansett Bay 

 have been described in detail (Smayda, 1976, and the literature cited 

 therein). Smayda and his students have sampled a station in the West 

 Passage of the bay at weekly intervals since 1959. Using this long 

 time series of physical and biological data, we discuss solar radiation, 

 surface water temperature, surface ammonia concentration, cind total 

 phytoplankton abundance (Fig. 1). Solar-radiation and water- 

 temperature values are for the day of sampling and represent ar 

 essentially complete record for the period from 1959 to 1975 

 Averages of solar radiation and temperature weighted over the days 

 preceding sampling would probably model phytoplankton growth 

 processes more realistically. Complete temperature data and informa 

 tion to generate the appropriate weighting functions were not readily 

 available, however, so unweighted grab measurements were used in 

 these analyses. Surface nutrient sampling was begun in 1972. 

 Ammonia, nitrate, nitrite, and phosphate all show the same 

 qualitative pattern. Ammonia alone is used because its fluctuations 



