1 Virginia Institute of Marine Science, The College of William and Mary, USA

ESTIMATING OCEANIC PRIMARY PRODUCTIVITY: AN EVALUATION OF ESTIMATING OCEANIC PRIMARY PRODUCTIVITY: AN EVALUATION OF OCEAN COLOR ALGORITHMS AND GENERAL CIRCULATIONOCEAN COLOR ALGORITHMS AND GENERAL CIRCULATION MODELSMODELS

11 Virginia Institute of Marine Science, The College of William and Mary, USA Virginia Institute of Marine Science, The College of William and Mary, USA 2 2 Columbia Climate Center, Columbia University, USAColumbia Climate Center, Columbia University, USA

Vincent S. Saba,Vincent S. Saba,11 Marjorie A.M. Friedrichs, Marjorie A.M. Friedrichs,1 1 Mary-Elena Carr Mary-Elena Carr 22, and the PPARR4 Team, and the PPARR4 Team

Figure 1. Locations, sample size, and temporal range of PP measurements. Only BATS and HOT represent a time-series at a single station. Integrated PP to the 1% light-level depth was measured in situ at BATS, NABE, Arabian Sea, HOT, and some of the Ross Sea stations. The other regions represent on-deck PP measurements.

Model # Contributer Type Details for models with variants Referencechl SST PAR MLD

1 Saba SAT,DI,WI x - Eppley et al., 1985

2 Saba SAT,DI,WI x x x x - Howard and Yoder, 1997

3 Saba SAT,DI,WI x x x - Carr, 2002

4 Ciotti SAT,DI,WI x x x - Morel and Maritorena, 2001

5 Dowell SAT,DI,WI x x x x -

6 Kameda SAT,DI,WI x x x - Kameda and Ishizaka, 2005

7 Scardi SAT,DI,WI x x x x - Scardi, 2001

8 Westberry SAT,DI,WI x x x Standard VGPM Behrenfeld and Falkowski, 1997

9 Westberry SAT,DI,WI x x x VGPM but with Eppley, 1972 Pbopt Behrenfeld and Falkowski, 1997

10 Westberry SAT,DI,WI x x x CbPM Behrenfeld et al., 2005

11 O'Malley SAT,DI,WI x x x -

12 Armstrong SAT,DR,WI x x x - Armstrong. 2006

13 Asanuma SAT,DR,WI x x x - Asanuma et al., 2006

14 Tang SAT,DR,WI x x x Uses Pbopt from BF Tang et al., 2008

15 Tang SAT,DR,WI x x x Uses Pbopt from SVM model

16 Marra SAT,DR,WI x x x -

17 Antoine SAT,DR,WR x x x x - Antoine and Morel, 1996

18 Uitz SAT,DR,WR x x x -

19 Mˇlin SAT,DR,WR x x - Mˇlin, 2003

20 Smyth SAT,DR,WR x x x - Smyth et al., 2005

21 Waters SAT,DR,WR x x x x Uses SST Ondrusek et al., 2001

22 Waters SAT,DR,WR x x x No SST Ondrusek et al., 2001

23 Westberry SAT,DR,WR x x x CbPM improved Westberry et al., 2008

24 Bennington BOGCM -

25 Bopp BOGCM -

26 Lima BOGCM - Moore et al., 2004

27 Dutkiewicz BOGCM -

28 Gregg BOGCM Assimilated Gregg, 2007

29 Gregg BOGCM NOBM free-run Gregg and Casey, 2007

30 Tjiputra BOGCM - Bentsen et al., 2004; Wetzel et al., 2005

31 Vichi BOGCM -

32 Yool BOGCM Daily

33 Yool BOGCM Monthly

34 Buitenhuis BOGCM - LeQuere et al., 2007

35 Dunne BOGCM - Dunne et al., 2006

Input variables used:

Table 1. Basic details of the 35 PP models used in the analysis. DI = Depth Integrated, DR = Depth Resolved, WI = Wavelength Integrated, WR = Wavelength Resolved.

SUMMARY - We assessed the skill of 35 primary productivity (PP) models (Table 1) by comparing their output to measured PP data at 9 different regions (Fig. 1) that represent various marine ecosystem types.

- In 7 of 9 regions, ocean color (SAT) models outperformed general circulation models (BOGCMs) (Fig. 4a). Among most models, skill was highest in the pelagic North Atlantic (NABE), the Arabian Sea, and the Antarctic Polar Frontal Zone (APFZ) (Fig. 2,4a). Models had weak skill in the Mediterranean Sea (MED) and the Ross Sea (Fig. 2,4a).

- Among SAT models, PP was typically over-estimated at high surface chlorophyll concentrations and under-estimated at low concentrations (Fig. 4b).

- All models typically under-estimated PP in most pelagic regions (BATS, NABE, Arabian Sea, HOT) (Fig. 3). Among coastal areas [Northeast Atlantic (NEA), Black Sea, MED, Ross Sea], SAT models had a tendency to over-estimate PP (Fig. 3).

- High model skill in the APFZ was likely due to the very short time-series of measured PP data (1 month) (Fig. 1) thus seasonal variability was not represented in the analysis.

- SAT models performed considerably better in HOT than in BATS (Fig. 1,4a). We surmise that this was due to the strong mesoscale eddy activity that affects biological production at BATS. However, we also note that this performance difference was greatly reduced when SeaWiFS chlorophyll was used in place of in situ chlorophyll (Fig. 5).

- In 5 of 6 regions, measured chlorophyll, as opposed to SeaWiFS, produced higher model skill whereas SeaWiFS PAR, as opposed to NCEP PAR, consistently produced higher model skill (Fig. 5).

- Future efforts will be directed toward understanding why certain models perform better than others and to understand why models have lower skill in coastal areas.

Model Skill - Mean and Variability of PPSAT,DI,WI SAT,DR,WI SAT,DR,WR BOGCM

Standard deviation of observed PP

Error of PP measurement

- For all regions and stations, we provided BOGCMs with date, location, and day-length. In addition to these parameters, SAT models were provided with ship-measured surface chlorophyll, sea-surface temperature, NCEP modeled photosynthetically active radiation (PAR), and modeled mixed-layer depth (MLD) (region-specific from various sources). We also provided SAT models with SeaWiFS surface chlorophyll and PAR for regions that had PP data after September, 1997. The SeaWiFS data was analyzed separately (Fig. 5).

- All models estimated integrated PP (mg C m-2 day) to the 1% light-level depth. We compared model output to integrated PP that was measured to the 1% light-level both on-deck and in situ depending on the region (Fig. 1).

- Model skill was assessed using root mean square difference (RMSD).

- Target diagrams were created based on bias (B) and RMSDCP. These statistics assess how well models estimate the mean and variability of PP (Fig. 3).

METHODS

Individual Model Skill at each Region

Figure 2. Total RMSD for each model at each region. Lower RMSD = higher model skill. Models performed with the highest skill in NABE, the Arabian Sea, and APFZ. Most models had poor skill in MED and the Ross Sea. BOGCMs typically did not perform well in the polar regions.

Figure 3. Target diagrams for each region. The distance from the origin to each symbol is total RMSD. Models falling within the solid-circle provide better instantaneous estimates of PP than the mean of the observed PP. Models falling within the dashed-circle are indistinguishable in terms of skill be because their bias and variability are less than the inherent error of the PP measurements (we used 50% error for low PP and 20% error for high). If bias > 0, PP is over-estimated; as RMSDCP approaches 0, the variability of PP is more accurately estimated. HOT has no dashed-circle because the standard deviation of PP is equal to the error of the PP measurement.

Pelagic Regions

Typical PP estimate (+/-)

Typical model skill

BATS - Moderate to WeakNABE - Good

Arabian Sea - GoodHOT - Moderate to GoodAPFZ +/- Good

Coastal Regions

Typical PP estimate (+/-)

Typical model skill

NEA +/- Moderate to Good

Black Sea + Moderate to Good

MED + Weak

Ross Sea +/- Weak

The majority of the PPARR4 team is listed in Table 1. This research was funded by the NASA - Ocean Biology and Biogeochemistry Program. Corresponding author e-mail: [email protected]

Figure 4. A) Mean RMSD for all models at each region. SAT DR,WI and SAT DR,WR typically outperformed all other models. B) Log of in situ surface chlorophyll vs. mean SAT model error (log(PPm)-log(PPd) for all regions. SAT models typically under-estimated PP at low chlorophyll concentrations and over-estimated PP at high concentrations.

A B

Figure 5. Mean RMSD for SAT models at regions with post September 1997 PP data. In all regions except BATS, measured chlorophyll (in situ or on-deck) produced higher model skill. SeaWiFS PAR also produced higher model skill. SeaWiFS PAR was not available for the Ross Sea. Error bars are standard error.

-1.0

-0.8

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

-1.00 -0.12 0.76

APFZ

Bias

RMSDCP

-1.0

-0.8

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0

Ross Sea in situ

Bias

RMSDCP

-1.0

-0.8

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0

HOT

Bias

RMSDCP

-1.0

-0.8

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0

Arabian Sea

Bias

RMSDCP

-1.10

-0.80

-0.50

-0.20

0.10

0.40

0.70

1.00

-1.10 -0.80 -0.50 -0.20 0.10 0.40 0.70 1.00

MED

Bias

RMSDCP

-1.0

-0.8

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0

NABE

Bias

RMSDCP

-1.0

-0.8

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0

NEA

Bias

RMSDCP

-1.0

-0.8

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0

Black Sea

Bias

RMSDCP

-1.0

-0.8

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0

BATS

Bias

RMSDCP

-1.0

-0.8

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0

Ross Sea on deck

Bias

RMSDCP