24 DWYER, NIXON, OVIATT, PEREZ, AND SMAYDA 



are most pronounced and because there is evidence that it cycles 

 more rapidly than other nutrients in the bay. Phytoplankton 

 abundcinces, by species, were counted beginning in 1960. Since 

 samples for 1975 have yet to be processed and digitized, they are not 

 included in the cell-count time series. 



Since spectral analysis requires complete time-series data, it was 

 necessary to interpolate missing data points linearly. For each time 

 series the total number of interpolated points was less than 10% of 

 the total number of data points. 



Solar-radiation and temperature time series show strong deter- 

 ministic oscillations at 1 cycle/year, but the periodicity for ammonia 

 is less clear, and that for total phytoplankton abundance is 

 undistinguishable. The frequency structure of these time series is 

 manifest only after transformation to the frequency domain by 

 spectral analysis. 



As part of a multiphase approach to studying the Narragansett 

 Bay ecosystem, a set of 150-liter, scaled, benthic— pelagic micro- 

 cosms were developed and are being used in perturbation experi- 

 ments (Nixon et al., 1978; Oviatt, Perez, and Nixon, 1977; Perez 

 et al., 1977). Data used here are from a sewage stress— relaxation 

 experiment described by Oviatt, Perez, and Nixon (Fig. 2). We chose 

 ammonia and chlorophyll time series from a microcosm receiving a 

 high sewage step input (microcosm 2, Fig. 2). Since the microcosms 

 were not sampled at regular intervals (there was sometimes a day's 

 delay in sampling), it was necessary to approximate the complete 

 time series by use of linearly interpolated points (11% of the total 

 time series). These microcosm time series are much shorter than 

 those from Narragansett Bay, and the ammonia series was sampled 

 twice per week. 



Systems and Holistic Responses 



The trend toward developing holistic measures of natural 

 ecosystem dynamics has led to perturbation experiments on ecosys- 

 tem models (Child and Shugart, 1972; Waide etal., 1974; Harwell, 

 Cropper, and Ragsdale, 1977) in which one or more model 

 compartments or inputs are artificially displaced. Harwell, Cropper, 

 and Ragsdale showed that estimated model stability (e.g., in terms of 

 natural frequency and damping ratio) may depend crucially on what 

 is perturbed, how large the perturbation amplitude is, and what is 

 measured as the response. Perturbations applied to real ecosystems 

 may generate responses that are functions of the timing of 

 application as well as of system state. This may be especially 



