210 likes | 367 Views
Réunion BIOCAREX 22 Février 2013 Villefranche sur Mer. Retrieval of the phytoplankton community structure at the BOUSSOLE site from hyper-spectral light absorption measurements. E manuele Organelli , Annick Bricaud , David Antoine and Julia Uitz TASK 7.
E N D
Réunion BIOCAREX 22 Février 2013 VillefranchesurMer Retrievalof the phytoplankton community structure at the BOUSSOLE site fromhyper-spectral light absorptionmeasurements EmanueleOrganelli, AnnickBricaud, David Antoine andJuliaUitz TASK 7
February 2012-presentPost-doc Laboratoire d’Océanographie de Villefranche GeneralObjective: Exploitinghyper-spectralmeasurementsofopticalpropertiestoidentifychanges in the phytoplankton community structure at the BOUSSOLE site TASK 7 Applydifferentexistingmethodstoretrievealgalsize and pigment information from IOP and AOP measurements Inter-comparisonofappliedmethodsand comparisonofresultswith HPLC data Analysisof the seasonal and interannualchanges in the phytoplankton community structure at the BOUSSOLE site
FUCO PERID Diatoms Dinoflagellates MICRO-PHYTOPLANKTON (>20µm) Markerpigments (MPs) and SizeClasses 19’-HF 19’-BF ALLO Coccolithophores Crysophytes, Prymnesiophytes Cryptophytes NANO-PHYTOPLANKTON (2-20µm) TChlb ZEA Prasinophytes Prokaryotes PICO-PHYTOPLANKTON (0.2-2µm)
Bidigareet al. (1990) Bricaudet al. (2004) Phytoplankton light absorption
SPECTRAL SIMILARITY ANALYSIS (Millie et al., 1997) • The procedure applied on the BOUSSOLE data set is the routine used in Organelli 2010 (PhD thesis) adapted for Mediterranean phytoplankton communities. AppliedMethods • PARTIAL LEAST SQUARES (PLS) regression technique (Martens and Næs, 1989) • Model development was necessary to be applied on the BOUSSOLE data set. Spectral-responsebasedapproaches (Brewinet al., 2011) Hyper-spectralabsorption data Fourth-derivativeanalysis (Bidigareet al., 1989)
(rx1, ry1) (x1, y1) (rx2, ry2) V1 (x2, y2) V2 SAM (Spectral Angle Mapper) (Sohn & Rebello, 2002) 0 < SI < 1 SpectralSimilarityAnalysis (Millieet al., 1997) Significantrelationshipsbetween SI values and the concentrationsofMPsof 8 taxonomicgroupswerefoundforMediterraneancommunititesand tested on a different data set.
PLS is a multivariate regression method especially used in chemistry for spectroscopy analysis. • Only few applications in the field of the optical detection of phytoplankton (Moberget al, 2002; Staehr and Cullen, 2003; Seppälä and Olli, 2008; Martìnez-Guijarroet al., 2009). PartialLeastSquaresregression (PLS) PLS is a multivariate analysis technique that relates a data matrix of predictor variables (X) to a data matrix of response variables (Y). PREDICTOR variables(X) RESPONSE variables(Y) Fourth-derivative absorption spectra (400-700 nm) MPs of the three phytoplankton size classes (Uitzet al., 2006) and TChla. (5 variables)
Only samples from the First Optical Depth PLS steps 1° 2° TRAINING TEST Organelli et al. (in press), Applied Optics
1/2 GLOCAL PLS MODELS (aphy) Samples=716
2/2 GLOCAL-TRAINED MODEL RESULTS (BOUSSOLE data) Samples=484 Good retrieval only of Tchla and DP
1/2 MedCAL PLS MODELS (apand aphy) aphy(λ) PLS models Samples=239
2/2 MedCAL-TRAINED MODEL RESULTS (BOUSSOLE data) aphy(λ) PLS models Samples=484 • Good retrieval of Tchla, DP, Micro, Nano and Pico. • ap and aphy modelscan be used interchangeably.
GLOCAL vs MedCAL-TRAINED MODELS • Distribution of 3 size classes at BOUSSOLE for a given [Tchla] more similar to that of Mediterranean samples than that of Atlantic and Pacific Oceans. • Amplitude and band center of the fourth –derivative spectra similar between BOUSSOLE and the other Mediterranean areas.
1/2 BOUSSOLE Time Series HPLC vs MedCAL PLS MODELS TIME SERIES
2/2 BOUSSOLE Time Series HPLC vs MedCAL PLS MODELS TIME SERIES
Retrieval of algal biomass and size structure from in vivo hyper-spectral absorption measurements can be achieved by PLS: • Good retrievals were obtained using models trained with a regional data set including only Mediterranean data (MedCAL dataset); • The dataset assembled from various locations in the World’s oceans (GLOCAL) gave satisfactory predictions of Tchla and DP only. PLS REMARKS • Prediction ability is very similar for ap(λ) and aphy(λ) PLS models. Retrieved temporal variations reproduce well those derived from HPLC data over the BOUSSOLE time series. The PLS technique is an encouraging method to use for the identification of phytoplankton size classes from absorption coefficients as inverted from AOPs (reflectance and remote sensing reflectance).
1/3 Kd R 1692 QC spectra (399-600 nm by 3 nm) 10:00 - 14:00 ~ 143 days Hyper-spectralAOPs and IOPsfrom the Buoy (Jan-Sep 2012) ap+ aCDOM QUALITY CONTROL 918 QC spectra 10:00 - 14:00 ~ 120 days ap???
2/3 aCDOM(440) vsatot(440) SCDOMvsatot(440) aCDOMfromap+ aCDOM y=0.2716*x+0.0099 R2=0.70 y=-0.0517*x+0.0202 R2=0.56
3/3 ap+ aCDOM aCDOM ap= (ap+ aCDOM) - aCDOM Retrievalofapforhigh-frequencymeasurements ap
To test and improve the quality of the AOPs inversion products (atot, aCDOM and ap). Works tobedone…. • To apply MedCAL PLS models toAOP-retrieved atotandap spectra. PLS models have to be adapted for spectral resolution of 3 nm. • To apply other methods to AOP-retrieved and in situ absorption spectra. Inter-comparison with PLS results and analysis of temporal changes over the BOUSSOLE time series. Merci!