Using biogeochemical data assimilation to assess the relative skill of multiple ecosystem models in the Mid-Atlantic Bight: effects of increasing the complexity of the planktonic food web

Abstract

Abstract. Now that regional circulation patterns can be reasonably well reproduced by ocean circulation models, significant effort is being directed toward incorporating complex food webs into these models, many of which now routinely include multiple phytoplankton (P) and zooplankton (Z) compartments. This study quantitatively assesses how the number of phytoplankton and zooplankton compartments affects the ability of a lower-trophic-level ecosystem model to reproduce and predict observed patterns in surface chlorophyll and particulate organic carbon. Five ecosystem model variants are implemented in a one-dimensional assimilative (variational adjoint) model testbed in the Mid-Atlantic Bight. The five models are identical except for variations in the level of complexity included in the lower trophic levels, which range from a simple 1P1Z food web to a considerably more complex 3P2Z food web. The five models assimilated satellite-derived chlorophyll and particulate organic carbon concentrations at four continental shelf sites, and the resulting optimal parameters were tested at five independent sites in a cross-validation experiment. Although all five models showed improvements in model-data misfits after assimilation, overall the moderately complex 2P2Z model was associated with the highest model skill. Additional experiments were conducted in which 20 % random noise was added to the satellite data prior to assimilation. The 1P and 2P models successfully reproduced nearly identical optimal parameters regardless of whether or not noise was added to the assimilated data, suggesting that random noise inherent in satellitederived data does not pose a significant problem to the assimilation of satellite data into these models. However, the most complex model tested (3P2Z) was sensitive to the level of random noise added to the data prior to assimilation, highlighting the potential danger of over-tuning inherent in such complex models.

Reviews and syntheses: parameter identification in marine planktonic ecosystem modelling

“…Xiao and Friedrichs (2014b), for example, found that while the optimization of a range of NPZD models to satellite data tended to reduce model-data misfit, this was often achieved through the adoption of extremely unrealistic parameter estimates, sometimes being multiple orders of magnitude higher or lower than their best a priori estimates. The same authors (Xiao and Friedrichs, 2014a) showed that adding synthetic noise to assimilated satellite data led to the introduction of similar errors, and a significant deterioration of one model's predictive skill. The extreme parameter estimates were not representative for the system and the model performance turned out to be poor when the model was tested against independent data that were not used during the optimization procedure.…”

“…Adjoint versions of plankton ecosystem models have been constructed primarily to compute ∇J for an efficient search with gradient descent algorithms in the parameter-cost function manifold (e.g. Lawson et al, 1996;Fennel et al, 2001;Schartau et al, 2001;Spitz et al, 2001;Friedrichs, 2002;Faugeras et al, 2003;Zhao et al, 2005;Friedrichs et al, 2007;Xiao and Friedrichs, 2014a). To elucidate the nature of adjoint model developments is beyond the scope of this paper, but a brief summary about adjoint model developments is given in the Appendix C. The advantage is that all elements of the Hessian can be approximated with finite differences of adjoint model results (e.g.…”

“…A notable counterpoint to those studies is given by Oschlies and Schartau (2005), who found that simultaneous optimization of an NPZD model at three time-series sites (Schartau and Oschlies, 2003) led to improved performance when the model was applied within a 3-D simulation of the North Atlantic. On the one hand, it seems likely that this improvement was dependent on assimilating data from three highly dissimilar North Atlantic locations, which prevented the inclusion of compensatory errors that were highly specific to any one site (see also Xiao and Friedrichs, 2014a). On the other hand, in Schartau and Oschlies (2003) and in Oschlies and Schartau (2005) it is also stressed that the apparent improvement is associated with some ambiguous rapid nitrogen remineralization pathway in their simple NPZD model, which can be incorrect in either simulations (1-D and 3-D), but with the same positive effect on primary production rates in the central North Atlantic.…”

“…Similarly, for small changes in the assimilated data an extra flexibility may lead to very different model solutions, also leading to increased uncertainty in model predictions (e.g. Xiao and Friedrichs, 2014a).…”

“…7.10 for an excellent discussion in a general statistical context). By looking at the effects of adding noise to assimilated remote sensing data, Xiao and Friedrichs (2014a) found that the most complex model they evaluated was also the most sensitive to the introduction of synthetic errors in the assimilated data (Fig. 3).…”