phytoplankton within the current depth horizon and their growth rate. Finally, in
regions of surface macronutrient depletion, the model allows for a switch from
nutrient-limitation to light-limitation at depths below the mixed layer. While
additional work is needed to validate various aspects of this ‘carbon-based
approach’, it does provide an excellent framework for incorporating more
sophisticated descriptions of physiological variability (see Sect. 8.6).
8.4 Validation Efforts
Validation of satellite-based NPP estimates has largely been limited to matchup
comparisons with field
14 C uptake measurements. While considerable ambiguity
remains regarding exactly what the
14 C method measures, it is generally accepted
that reasonably long (i.e., [12 h) incubations yield carbon fixation rates that
approximate net primary production. The ‘ambiguity’ of the measurement includes
unconstrained artifacts of sample confinement in bottles, unnatural light conditions
during incubation (either on deck of the ship or in situ), and an incomplete
understanding of how respiratory and other metabolic pathways impact lifetimes
of newly formed carbon products. Alternative measures of photosynthetic primary
production (e.g., gross primary production) are less frequently used for validation
of satellite-based estimates. Both,
18 O 2 incubations and the more recently developed triple-oxygen isotope method (
17 DO 2 ) provide a measure of gross primary
production. Importantly, the latter method does not require sample incubations
(Luz and Barkan 2009). However, empirical conversions are required to equate the
different oxygen and carbon measurements and to characterize losses between
gross and net photosynthetic production. Hence, we limit the following discussion
to comparisons with measurements of
14 C uptake.
A series of blind, round-robin exercises were initiated by NASA in the mid1990’s in order to evaluate the performance of a wide variety of satellite NPP
models. This Primary Productivity Algorithm Round Robin (PPARR) exercise has
since evolved in its scope and expanded in the range of model types, numbers, and
field measurements represented (Campbell et al. 2002; Carr et al. 2006; Friedrichs
et al. 2009; Saba et al. 2010, 2011). Key findings of the PPARR activities have
been: (1) increasing model complexity does not equate to improved predictive
skill, (2) model skill varies regionally, and (3) reducing uncertainty in input
parameters to NPP models (e.g., PAR, Chl) can reduce average RMS errors by
[50 %. The first point above has been made on several occasions (Siegel et al.
2001; Behrenfeld and Falkowski 1997a), but is perhaps best demonstrated by Carr
et al. (2006). In their report, a cluster analysis was performed on the correlation
between NPP estimates for[30 models, spanning a wide range of complexity. The
analysis revealed that correlations between model NPP estimates were not grouped
according to model complexity, such that the simplest and most complex models
often showed the highest correlation. Instead, correlations between models were
largely determined by the underlying expression used to describe variability in
8 Oceanic Net Primary Production
211
regions of surface macronutrient depletion, the model allows for a switch from
nutrient-limitation to light-limitation at depths below the mixed layer. While
additional work is needed to validate various aspects of this ‘carbon-based
approach’, it does provide an excellent framework for incorporating more
sophisticated descriptions of physiological variability (see Sect. 8.6).
8.4 Validation Efforts
Validation of satellite-based NPP estimates has largely been limited to matchup
comparisons with field
14 C uptake measurements. While considerable ambiguity
remains regarding exactly what the
14 C method measures, it is generally accepted
that reasonably long (i.e., [12 h) incubations yield carbon fixation rates that
approximate net primary production. The ‘ambiguity’ of the measurement includes
unconstrained artifacts of sample confinement in bottles, unnatural light conditions
during incubation (either on deck of the ship or in situ), and an incomplete
understanding of how respiratory and other metabolic pathways impact lifetimes
of newly formed carbon products. Alternative measures of photosynthetic primary
production (e.g., gross primary production) are less frequently used for validation
of satellite-based estimates. Both,
18 O 2 incubations and the more recently developed triple-oxygen isotope method (
17 DO 2 ) provide a measure of gross primary
production. Importantly, the latter method does not require sample incubations
(Luz and Barkan 2009). However, empirical conversions are required to equate the
different oxygen and carbon measurements and to characterize losses between
gross and net photosynthetic production. Hence, we limit the following discussion
to comparisons with measurements of
14 C uptake.
A series of blind, round-robin exercises were initiated by NASA in the mid1990’s in order to evaluate the performance of a wide variety of satellite NPP
models. This Primary Productivity Algorithm Round Robin (PPARR) exercise has
since evolved in its scope and expanded in the range of model types, numbers, and
field measurements represented (Campbell et al. 2002; Carr et al. 2006; Friedrichs
et al. 2009; Saba et al. 2010, 2011). Key findings of the PPARR activities have
been: (1) increasing model complexity does not equate to improved predictive
skill, (2) model skill varies regionally, and (3) reducing uncertainty in input
parameters to NPP models (e.g., PAR, Chl) can reduce average RMS errors by
[50 %. The first point above has been made on several occasions (Siegel et al.
2001; Behrenfeld and Falkowski 1997a), but is perhaps best demonstrated by Carr
et al. (2006). In their report, a cluster analysis was performed on the correlation
between NPP estimates for[30 models, spanning a wide range of complexity. The
analysis revealed that correlations between model NPP estimates were not grouped
according to model complexity, such that the simplest and most complex models
often showed the highest correlation. Instead, correlations between models were
largely determined by the underlying expression used to describe variability in
8 Oceanic Net Primary Production
211