Bio-optical modeling of primary production on regional scales: the Bermuda BioOptics project

Regional to global scale estimates of primary production must rely on remotely sensed quantities. Here, we characterize in situ light–primary production relationships and assess the predictive capability of several global primary production models using a 6-yr time series collected as part of the US JGOFS Bermuda Atlantic Time Series (BATS). The consistency and longevity of this data set provide an excellent opportunity to evaluate bio-optical modeling methodologies and their predictive capabilities for estimating rates of water-column-integrated primary production, ∫PP, for use with satellite ocean-color observations. We find that existing and regionally tuned parameterizations for vertically integrated chlorophyll content and euphotic zone depth do not explain much of the observed variability at this site. Fortunately, the use of these parameterizations for light availability and harvesting capacity has little influence upon modeled rates of ∫PP. Site-specific and previously published global models of primary production both perform poorly and account for less than 40% of the variance in ∫PP. A sensitivity analysis is performed to demonstrate the importance of light-saturated rates of primary production, Psat∗, compared with other photophysiological parameters. This is because nearly one-half of ∫PP occurs under light-saturated conditions. Unfortunately, we were unable to derive a simple parameterization for Psat∗that significantly improves prediction of ∫PP. The failure of global ∫PP models to encapsulate a major portion of the observed variance is due in part to the restricted range of ∫PP observations for this site. A similar result is found comparing global chlorophyll-reflectance algorithms to the present observations. More importantly, we demonstrate that there exists a time-scale (roughly 200 d) above which the modeled distributions of ∫PP are consistent with the observational data. By low-pass filtering the observed and modeled ∫PP time series, the modelʹs predictive skill levels increase substantially. We believe that the assumptions of steady state and balanced growth used in bio-optical models of ∫PP are inconsistent with observational data. Most of the observed variance in ∫PP is driven by a variety of ecosystem disturbance processes that are simply not accounted for in bio-optical models. This puts important bounds on how ∫PP models should be developed, validated and applied.