Accounting for unpredictable spatial variability in plankton ecosystem models

Abstract

Limitations on our ability to predict fine-scale spatial variability in plankton ecosystems can
seriously compromise our ability to predict coarse-scale behaviour. Spatial variability which
is deterministically unpredictable may distort parameter estimates when the ecosystem model
is fitted to (or assimilates) ocean data, may compromise model validation, and may produce
mean-field ecosystem behaviour discrepant with that predicted by the model. New statistical
methods are investigated to mitigate these effects and thus improve understanding and prediction
of coarse-scale behaviour e.g. in response to climate change. First, the standard model
fitting technique is generalised to allow model-data ‘phase errors’ in the form of time lags,
as has been observed to approximate mesoscale plankton variability in the open ocean. The
resulting ‘variable lag fit’ is shown to enable ‘Lagrangian’ parameter recovery with artificial
ecosystem data. A second approach employs spatiotemporal averaging, fitting a ‘weak prior’
box model to suitably-averaged data from Georges Bank (as an example), allowing liberal
biological parameter adjustments to account for mean effects of unresolved variability. A
novel skill assessment technique is used to show that the extrapolative skill of the box model
fails to improve on a strictly empirical model. Third, plankton models where horizontal variability
is resolved ‘implicitly’ are investigated as an alternative to coarse or higher explicit
resolution. A simple simulation study suggests that the mean effects of fine-scale variability
on coarse-scale plankton dynamics can be serious, and that ‘spatial moment closure’ and
similar statistical modelling techniques may be profitably applied to account for them.

More information

Identifiers

Catalogue record

Export record