Global PFTs retrieved by AGIOP applied to monthly OLCI Rrs data (GlobColour level 3 product) in January, April, June and October 2017. Panels from left to right: diatoms, haptophytes, cyanobacteria, and total Chl-a.
Global Retrieval Algorithms for Phytoplankton Functional Types (PFTs):
toward the Applications to OLCI and GlobColour Merged Products
Hongyan Xi 1 , Svetlana N. Losa 1 , Antoine Mangin 3 , Mariana A. Soppa 1 , Philippe Garnesson 3 , Julien Demaria 3 , Odile Hembise Fanton d’Andon 3 , Astrid Bracher 1,2
1
Alfred Wegener Institute, Helmholtz Centre for Polar and Marine Research, Bremerhaven, Germany;
2Institute of Environmental Physics, University of Bremen, Bremen, Germany
3
ACRI-ST, 06904 Sophia Antipolis Cedex, France
Objectives Data sets
Results
Conclusions
Adapted Generalized IOP (AGIOP) Empirical Orthogonal Functions (EOF) based algorithm)
We focus on PFT retrieval algorithms that are then applied to Sentinel-3 (S3) OLCI data and merged ocean colour (OC) products from CMEMS GlobColour archive:
• Two algorithms were investigated for their capability in PFT retrievals, namely the adapted generalized IOP (AGIOP) and the empirical orthogonal function (EOF)-based algorithm, using in situ measurements, matchups between in situ and satellite data, and satellite OC products.
• The retrieved PFTs (mainly the diatoms, haptophytes, prokaryotic phytoplankton (cyanobacteria)) based on in situ data sets are compared with the in situ diagnostic pigment analysis (DPA) based PFTs.
• The two algorithms are also preliminarily applied to the GlobColour merged OC products and OLCI data.
In-situ datasets
• 15 cruises from 2007 to 2018 worldwide
• 208 collocated R
rs, HPLC pigments and absorption
Satellite data and matchups
• GlobColour merged ocean colour (OC) R
rsproducts
• Sentinel 3A OLCI L3 R
rsproduct
• Matchups between GlobColour merged R
rsand in situ global HPLC data collected in 1997–2012
Campaigns and stations with collocated Rrs, HPLC pigments and absorption data.
Sensors
Number of Matchups
Available Wavebands (nm) Number of bands 1x1 3x3
5x5 412 443 490 510 531 547 551 555 560 620 670 678
SeaW 1223 609 X X X X X X 6
SeaW+MO+ME 408 266 X X X X X X X X X 9
SeaW+ME 502 129 X X X X X X X X 8
SeaW+MO+ME 212 64 X X X X X X X X X X X 11
MO+ME+V 3 2 X X X X X X X X X X X X 12
SeaW+MO+ME 766 516 X X X X X X X X 8
MO+V 25 27 X X X X X X X X X 9
SeaW 1596 880 X X X X X 5
Numbers of available matchups between HPLC pigments and Rs (1x1 pixel, 3x3 and 5x5 pixels) with different band combinations from the merged OC productsr. Blue highlights the matchups used in the EOF based algorithm (SeaW = SeaWiFS, MO = MODIS, ME = MERIS, V = VIIRS).
Schematic overview of EOF steps in model building and prediction. Multiple linear regression models are fit to log- transformed pigment concentrations, yp, as the response variable and EOFs derived from a spectral (reflectance) data set, X, as predictor variables. Model building (left) is used for “full-fit” models to all data samples (M) or to a training subset of samples for cross-validation. Prediction (right) is used for the assessment of the model error on a validation subset of samples (I) for cross-validation or in the extrapolation of model predictions to GlobColour
merged Rrs data sets (adapted from Bracher et al. 2015).
DPA based PFTs – Y[M,P]
PFTs
o Global in situ HPLC data 1988-2012 (left) o Rrs Matchups (right)
IOPs
Linear or nonlinear least squares inversion
• Adapted from GIOP by Werdell et al. (2013), assuming that af(l) is a linear sum of subcomponents with unique spectral dependencies.
• af(l) is decomposed into absorption by 3 PFTs – diatoms, haptophytes, and cyanobacteria. Specific absorption of the three PFTs were obtained from natural waters where one PFT was dominating.
• Using Rrs at different wavebands and the spectral shapes of the IOPs as input, eigenvalues (the Chl-a
concentrations of the 3 PFTs, adg(440), and bbp(440)) can be derived via linear or nonlinear least squares inversions.
Methods
Regressions between observed (obs.) based on DPA and predicted (pred.) PFTs using EOF modes derived from GlobColour merged Rrs products at 9 bands. Top panel: using Rrs 1x1 pixel, bottom panel: Rrs 3x3 pixels.
AGIOP
EOF-based algorithm
Global PFTs retrieved by EOF-based algorithm (9 bands) applied to GlobColour merged monthly Rrs products in January, April, June and October 2011.
Panels from left to right: diatoms, haptophytes, cyanobacteria, and total Chl-a.
• By both algorithms (especially EOF) total Chl-a were generally overestimated typically at small values therefore in oligotrophic regions for global retrievals.
• AGIOP generally works but is unstable when using different minimization inversion methods. Inversion by LMI is robust for global data but less coverage of valid retrievals.
• EOF outperformed the AGIOP from both prediction accuracy of in-situ matchups and valid retrieval coverage for global data.
• Both AGIOP and EOF based PFT retrieval algorithms can well retrieve diatoms and haptophytes but perform less accurate for cyanobacteria mainly due to their general low concentration resulting weak signal in the reflectance spectra.
Acknowledgements
This work is supported by a collaborative project between ACRI-ST and AWI, OLCI- PFT. We express our gratitude to Jeremy Werdell for the original GIOP algorithm, to Marc Taylor for the EOF-based algorithm. We thank Sonja Wiegmann for her dedication in measuring and processing the absorption spectra and HPLC data for most of our past cruises, and all the previous and current PhytoOptics team members who participated in the past cruises for data collection and analysis. We are thankful to all the scientists and crew involved in the global HPLC data collection and analyses for providing their pigment data. We also thank the SeaWiFS, MODIS, MERIS, VIIRS, and OLCI data, and specially the GlobColour team for providing the OLCI and merged OC L3 products.
Key References
Bracher, A., Taylor, B.B., Taylor, M., et al. (2015). Using empirical orthogonal functions derived from remote sensing reflectance for the prediction of concentrations of phytoplankton pigments. Ocean Science 11: 139-158.
Werdell, P. J., Franz, B. A., Bailey, S. W., et al. (2013). Generalized ocean color inversion model for retrieving marine inherent optical properties, Appl. Opt., 52, 2019–2037.
Losa, S.N., Soppa M. A., Dinter T., et al. (2017). Synergistic exploitation of hyper- and multispectral precursor Sentinel measurements to determine Phytoplankton Functional Types at best spatial and temporal resolution (SynSenPFT). Front. Mar. Sci. 4: 203.
Contact: Hongyan Xi, hongyan.xi@awi.de
10-2 100
In situ Chla (mg m- 3)
10-2 10-1 100 101
GIOP-derived Chla (mg m-3 ) (a) Total Chl-a
10-2 100
HPLC based diatom (mg m- 3)
10-2 10-1 100 101
GIOP-derived diatom (mg m-3) (b) Diatoms
10-2 100
HPLC based hapto. (mg m- 3)
10-2 10-1 100 101
GIOP-derived hapto. (mg m-3 ) (c) Haptophytes
10-2 100
HPLC based cyano. (mg m- 3)
10-2 10-1 100 101
GIOP-derived cyano. (mg m-3) (d) Cyanobacteria ANT24-1 ANT24-4 ANT25-1 ANT26-4 ANT28-3 PS76 PS93 PS99 PS107 SO218 SO243
PFT retrievals from in situ Rrs data sets.
AGIOP conducted with NLSQ, Rrs at OLCI bands
R2= 0.50 MPD = 60.5%
RMSE = 0.90 mg m-3
R2= 0.36 MPD = 470%
RMSE = 1.06 mg m-3
R2= 0.44 MPD = 60.1%
RMSE = 0.55 mg m-3
R2= 0.11 MPD = 523%
RMSE = 0.72 mg m-3
10-2 100
In situ Chla ( g/l) 10-2
10-1 100 101
Chla-GIOP (g/l)
(a) Total Chl-a
10-2 100
Pigment based diatom conc. ( g/l) 10-2
10-1 100 101
GIOP-derived diatom conc. (g/l)
(b) Diatoms
10-2 100
Pigment based hapto conc. ( g/l) 10-2
10-1 100 101
GIOP-derived haptophytes conc. (g/l) (c) Haptophytes
10-2 100
Pigment based cyano conc. ( g/l) 10-2
10-1 100 101
GIOP-derived cyanobacteria conc. (g/l)
(d) Cyanobacteria
10-2 100
In situ Chla ( g/l) 10-2
10-1 100 101
Chla-GIOP (g/l)
(a) Total Chl-a
10-2 100
Pigment based diatom conc. ( g/l) 10-2
10-1 100 101
GIOP-derived diatom conc. (g/l) (b) Diatoms
10-2 100
Pigment based hapto conc. ( g/l) 10-2
10-1 100 101
GIOP-derived haptophytes conc. (g/l) (c) Haptophytes
10-2 100
Pigment based cyano conc. ( g/l) 10-2
10-1 100 101
GIOP-derived cyanobacteria conc. (g/l)
(d) Cyanobacteria
PFT retrievals from matchups between GlobColour merged OC Rrs at 8 bands and in situ HPLC data.
Left: Nonlinear least squares (NLSQ); Right : Non-negative linear matrix inversion (Nonneg-LMI).
R2= 0.60 MPD = 58.5%
RMSE = 1.75 mg m-3 N=231
R2= 0.35 MPD = 264.8%
RMSE = 4.4 mg m-3 N=157
R2= 0.07 MPD = 253.8%
RMSE = 0.7 mg m-3 N=300
R2= 0.07 MPD =668.1%
RMSE = 0.54 mg m-3 N=363
R2= 0.62 MPD = 48.1%
RMSE = 1.19 mg m-3 N=369
R2= 0.69 MPD = 49%
RMSE = 0.84 mg m-3 N=25
R2= 0.30 MPD = 234%
RMSE = 0.44 mg m-3 N=253
R2= 0.05 MPD = 436%
RMSE = 0.51 mg m-3 N=146
Losa et al. (2017)
