MERIS Case 1 Validation ->

Transcription

1 MAVT meeting March 2006 MERIS Case 1 Validation -> Performance of the NN case 2 water algorithm for case 1 water Presenter: Roland Doerffer GKSS Forschungszentrum, Institute for Coastal Research doerffer@gkss.de

2 Case 2 Water Algorithm COASTLOOC COLORS MAPP MAPP Bio-optical measurements a(λ), a(λ), b (λ), (λ), Pig, Pig, TSM, TSM, C, C, Y Lu/Ed(λ,z), RLw RLw Bio-optical Model Model Incl. Incl. variability Set Set up up radiance transfer model model Monte Monte Carlo, Carlo, Hydrolight fann, fann, bann bann MERIS MERIS Processor 3 angles 8 RLw ann Simulation of of RLw(λ,θ v,θ v,θ s,φ s,φ vs vs > Spectra Training and and Test Test of of ann ann TSM Pig Gelb Conf. Flag MERIS MERIS Product

3 Comparison between algal_1 and algal_2 algorithm algal_1 algorithm based on reflectance ratios using 4 bands: 442, 490, 510, 560 nm Algal_1 takes absorption and scattering of phytoplankton into account Based on a large worldwide data base of mainly case 1 water attenuation and reflectance measurements Algal_2 neural network uses bands 1-7, 9 range nm Trained to determine IOPs (absorption of pigments, yellow substance + bleached particles, scattering of all particles) NN is trained with simulated bi-directional water leaving radiance reflectances Bio-optical model for reflectance simulations is mainly based on North Sea measurements

4 MERIS Algal_1 Algorithmus

5 The MERIS case II water NN algorithm r, r log of reflectances c log of concentrations g geometry information q quality indicator NN input NN output r g r g invnn c forwnn r q c q c C: log(b_tot) log(a_ys+a_btsm) log(a_pig) All at 443 nm flag PCD1_16,17 q true if sum (r(i) / r (i)) > 4.0 If RLw < , RLw = ( ρ 0.003)

6 Scheme of a bio-optical model: optical components for MERIS Water sample In situ AC-9 BB-4 particle particle scattering scattering backscattering backscattering tsm Gelbstoff yellow substance particle total absorption Absorption Absorption of of Total Total --bleached bleached fraction fraction = phytoplankton phytoplankton absorption absorption Absorption Absorption of of bleached bleached fraction Absorption fraction Absorption of of = spm spmabsorption bleached bleached Fraction Fraction + gelbstoff gelbstoff gelbstoff gelbstoff absorption absorptionspectrum = spectrum total total gelbstoff gelbstoff spectral spectralexponent exponent Optical properties at 442 nm algal_2 a_ys

7 Changes in MEGS 7.4 compared to IPF 4.10 Training of NN based on larger data set (GKSS MAVT cruises, data of Kai Sorensen, Marcel Babin, Herbert Siegel) Conversion for algal_2 now: algal_2 = 21 * a_pig_442^1.04 Algal_2, tsm, ys flag threshold set to 4.0 Cut-off reflectance rho = NN net based on log reflectances a_ys on log scale in product Why cut-off? To avoid that neg. or uncertainly low reflectances (log!) go into the NN NN is trained with the cut-off Problem when this happens for band1 and band2: chlorophyll and ys is unreliable These conditions are presently not flagged!!

8 Bio-optical model Based on: MAVT North Sea / German Bight (GKSS), Norwegian waters (NIVA, Uni Oslo, NERSC), Baltic Sea (IOW), recommendations by M. Babin Gelbstoff absorption exponent: Bleached particle absorption exponent: Particle scattering exponent: White particles scattering exponent: 0.0 Phytoplankton pigment absorption: > 221 spectra from different areas and seasons Gelbstoff absorption ays(443): m-1 Particle scattering bp(443): m-1 White particle scattering: m-1 Phytoplankton pigment abs. apig(443): m-1 Minimum particle scattering bp(443): 0.25*a_pig(443) Bleached particle absorption abp(443): 0.1*bp(443)+ran_gauss*0.03*bp(443)

9 Pigment absorption spectra H187, Norway different locations Norway all stations a (m -1 )) norm a (m -1 )) norm wavelength (nm) wavelength (nm)

10 Pigment absorption Chl. a, H Heincke187 a443 pigment n=95 chlorophyll a [mg/l] a(443) (m -1 )) Conversions: Chl. a [mg m-3] = 21 * a_pig_442 ^1.04

11 Validation of scaling factor for Chl2.hplc versus Chl2-apig(442) chlorophyll a [µg/l] Chl2.hplc= x apig(442) 0,77135 Chl2.hplc a(442) (m -1 )) Chl2.hplc=21,848 x apig(442) 1,044 0,1 0,01 0,1 1 apig(442)

12 TSM scattering, H H187 ac-9 scattering all ok stations 10 0 Heincke 187 all b at 440 nm (m -1 )) bb4 440 (m -1 )) tsm [mg/l] tsm (mg/l) AC-9 Conversions: TSM [ g m-3] = 1.72 * b_tsm_442 BB-4

13 Relationship between yellow substance (ays 442) and bleached particle absorption (bpa 442) Wappen c01- c bpa 442 bpa ays 442

14 Relationship between TSM scattering (b 440) and bleached particle absorption (abp 440) 0.5 y = x R 2 = abp Reihe1 Linear (Reihe1) b 440

15 MEGS 7.2 algal_1 / algal_2=21*a_pig^1.04 Red dots indicate pixels localized in the map (next pages), these dots have been selected by hand, because they form a second cloud in the scatter plot. As seen in the map they belong mainly to case 2 water. The regression line was also determined by hand only for the upper data cloud (case 1 water) Algal_2=21*a_pig^1.04

16 World map of MERIS data of MEGS 7.2 processing, red dots indicate data which form a second cloud in the scatter plot Algal_2 computed

17 Comparison Case 1 / Case 2 Water Algorithm for Pigment Adriatic Sea, May 3, 2002 Algal_1 Algal_2

18 Cascade Algal_1 Algal_2

19 West Canda_ferry algal_1

20 West Canda_ferry algal_2

21 Transect reflec_1 reflec_2 reflec_

22 Transect chlorophyll chl. a µg/l pixel coordinate x algal_1 algal_2

23 West Canda_ferry algal_1 with non-confidential algal_1 flag

24 Rio de la Plata Algal_

25 Algal_2 Rio de la Plata

26 Algal Transect 1 Algal_2 transect plata concentration mg m algal_ algal_2 reflec_ longitude deg

27 Algal_1 algal_2 band 1 reflectance threshold Algal_2 transect plata concentration mg m algal_1 algal_2 reflec_ longitude deg

28 algal_2 vs algal_1 Plata transect algal_2 20 algal_ algal_1

29 Rio de la Plata transect algal_2 vs algal_1 Plata transect algal_2 1 algal_ algal_1

30 Amazone Spectra

31 Algal transect 1 Amazone transect algal mg m-3 1 algal_1 algal_ latitude deg

32 Algal_2, TSM. YS a442 Amazone transect 1 case2 conc transect conc mg m-3, g m ys a442 m-1 algal_2 total_susp yellow_subs latitude 0.001

33 Hawai Algal_1 Algal_2

34 Hawai Yellow substance and TSM

35 Hawai , unflagged area Algal_1 : , mean: Algal_2 : , mean: Yellow s : , mean: TSM : , mean:

36 Algal 1 and 2 Mean algal_1_mean algal_2_mean 1.E+02 1.E+01 1.E+00 mg/m³ 1.E-01 1.E-02 1.E product id

37 Algal 1 and 2 Stdev algal_1_stddev algal_2_stddev 1.E+02 1.E+01 1.E+00 mg/m³ 1.E-01 1.E-02 1.E product id

38 MERIS Case 2 algorithm test against IOCCG simulated data set 10 1 a_tot sun TSM scatter 10 0 a_tot_442 NN GKSS [m -1 ] b_tot_442 NN GKSS [m -1 ] a_tot_442 test_case [m -1 ] bb_tot_442 test_case [m -1 ] Total absorption at 442 nm Total scattering at 442 nm

39 How to compute total scattering and absorption from case 2 products Algal_2 Conversion from a_pig_443 (NN output) to chlorophyll concentration algal_2=21*a_pig_443^1.04 Inverse to compute absorption again: a_pig_443=exp(log(algal_2/21)/1.04) TSM Total_susp = 1.73*b_tot_443 Inverse to compute total scattering again: b_tot_443=total_susp/1.73 Total absorption: a_tot_443= a_pig_443 + yellow_subs

40 Med algal_1 algal_2

41 Med TSM and YS

42 Med b_tot and a_tot at 443 nm

43 Summary and conclusions After improvements in MEGS 7.4 compared to IPF 4.0: NN based case 2 products can be used also in case 1 conditions Algal_1 and algal_2 similar values Algal_2 larger range compared to algal_1 Algal_1 more robust against errors in atmospheric correction Algal_2 and YS fail when reflectances are below cut-off threshold YS and TSM can be used also for case 1 conditions Values look reasonable, but are not validated From NN b_tot and a_tot at 443 nm can be easily computed Robust products according to experience of IOCCG working group

44 MEGS 7.2 relationship, log ref. input Red line: alg1=21*apig^1.04 MEGS 7.3 coefficients