Estuarine, Coastal and Shelf Science (2002) 55, 437–463 doi:10.1006/ecss.2001.0917, available online at http://www.idealibrary.com on Toward a Predictive Understanding of Primary Productivity in a Temperate, Partially Stratiﬁed Estuary L. W. Harding, Jr. a,b , M. E. Mallonee a and E. S. Perry c a Horn Point Laboratory, University of Maryland, Center for Environmental Science, Box 775, Cambridge, Maryland 21613, U.S.A. b Maryland Sea Grant College, 0112 Skinner Hall, University of Maryland, College Park, Maryland 20742, U.S.A. c 2000 Kings Landing Road, Huntingtown, Maryland 20639, U.S.A. Received 17 January 2001 and accepted in revised form 2 October 2001 This paper addresses the development of a predictive understanding of phytoplankton primary productivity (PP) in estuaries, drawing on an extensive set of observations in Chesapeake Bay and contemporary modeling approaches coupled to empirical data. PP was measured at 575 stations on 68 cruises, 1982–2000. Descriptions presented here are based on PP data from 455 stations occupied 1982–1998, and from 120 stations occupied 1999–2000 that were used for model validation. The nearly two-decade sampling period encompassed a broad range of freshwater ﬂow and nutrient loading to the Bay. Mean net 14 C-PP for 1982–1998 was 1055 mg C m 2 d 1 ( SE=46·0) in the main stem Bay for all seasons and regions. Phytoplankton dynamics included a spring maximum of biomass, expressed as euphotic-layer chlorophyll (chl a), and a summer maximum of net 14 C-PP displaced by approximately 4 months from the biomass maximum. The annual maximum of euphotic-layer chl a exceeded 100 mg m 2 in mesohaline and polyhaline regions of the Bay during April and May, whereas highest net 14 C-PP of 1700–2500 mg C m 2 d 1 occurred in the mesohaline Bay in July and August. Euphotic-layer chl a and net 14 C-PP were much lower in the light-limited oligohaline Bay and no time lag was observed. Mean gross 14 C-PP was 1564 mg C m 2 d 1 ( SE=85·0) for 1995–1998. Concurrent measure- ments using 14 C and O 2 methods generated estimates of the photosynthetic quotient (PQ) to conﬁrm our interpretation of net and gross PP from 14 C uptake in full- and partial-day incubations, respectively, and allowed us to reconcile apparent diﬀerences in 14 C- and O 2 -determined PP in recent studies. PQ values (=O 2 produced/CO 2 ﬁxed) were estimated as 1·48 from regression of stoichiometrically converted net O 2 -PP on net 14 C-PP, and as 1·38 from the regression of gross O 2 -PP on gross 14 C-PP. PQ values in this range typically correspond to phytoplankton that are using oxidized nitrogen (NO  3 -N) as the main N source, consistent with the view that N-limitation occurs on an annual scale in the Bay. We used empirical data for net and gross 14 C-PP to estimate annual integrated production (AIP) of 282 to 538 g C m 2 yr 1 (net) and 347 to 662 g C m 2 yr 1 (gross). Simple, linear regression of net AIP on annual, mean euphotic-layer chl a was signiﬁcant and explained 62% of the variance. Inter-annual variability of net AIP was related to the volume of freshwater ﬂow and to total nitrogen (TN) and total phosphorus (TP) loading during February and March. We tested the performance of published models to estimate PP in Chesapeake Bay. The Vertically Generalized Productivity Model (VGPM) overestimated net and gross PP, and adjusted forms of VGPM termed VGPM-A for net and gross PP gave signiﬁcantly improved performance for Chesapeake Bay. We explored an alternative approach to VGPM that allowed us to obtain non-unity exponents for independent variables, using step-wise and multiple regressions in log-space ﬁrst to identify independent variables that predicted net and gross PP, and subsequently to determine coeﬃcients of the terms. This approach resulted in the Chesapeake Bay Productivity Model (CBPM-1) that estimated net and gross PP with root mean square error (RMSE) of 28·3% and 34·9%, respectively. We then developed models of the physiological input, P B opt , a variable that expresses optimal photosynthesis in the water column normalized to chl a, using several independent variables and measured values of P B opt . Substitution of the models of P B opt in CBPM-1 resulted in CBPM-2 that required no explicit input of P B opt and estimated net and gross PP with RMSE of 120% and 49·8%, respectively. Lastly, we validated CBPM-1 and CBPM-2 using observations from 1999–2000 not included in development of the models. CBPM-1 estimated net and gross PP with RMSE of 17·8% and 35·8%, respectively, and CBPM-2 estimated net and gross PP with RMSE of 80·3% and 47·6%, respectively. We believe the use of contemporary PP models that require simple inputs amenable to remote sensing, such as the adjusted VGPM, CBPM-1, and CBPM-2, should permit us to resolve spatial and temporal diﬀerences of net and gross PP in estuaries and give improved estimates of AIP, an essential measure of ecosystem health and productivity.  2002 Elsevier Science Ltd. All rights reserved. Keywords: primary productivity; time-series; prediction; models; Chesapeake Bay Introduction Estuaries are among the most productive of marine ecosystems, supporting an abundant and diverse fauna and ﬂora (cf. Odum, 1971; Boynton et al., 1982; Nixon et al., 1986; Houde & Rutherford, 1993). The middle Atlantic bight of the East coast of the United 0272–7714/02/090437+27 $35.00/0  2002 Elsevier Science Ltd. All rights reserved.