stewart bernard (csir)

20
IOCS 2015, San Francisco Phytoplankton Biodiversity in the Coastal Zone using Hyperspectral Sensing Stewart Bernard 1,2 , Lisl Robertson 2 , Stefan Simis 3 , Hayley Evers-King 2,3 , Mark Matthews 2,4 1. CSIR; 2. UCT; 3. PML; 4. CyanoLakes

Upload: lythuy

Post on 01-Jan-2017

241 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Stewart Bernard (CSIR)

IOCS 2015, San Francisco

Phytoplankton Biodiversity in the Coastal Zone using Hyperspectral Sensing Stewart Bernard1,2, Lisl Robertson2, Stefan Simis3, Hayley Evers-King2,3, Mark Matthews2,4 1. CSIR; 2. UCT; 3. PML; 4. CyanoLakes

Page 2: Stewart Bernard (CSIR)

Hyperspectral radiometry contains considerable potential information on phytoplankton community characteristics, above the first order signal variability typically determined by gross constituent concentrations. Such information has significant value across a broad range of applications, ranging from water quality to long term understanding of ecosystem variability We anticipate the launch of several hyperspectral sensors in the next five – seven years, with high quality radiometry and useable repeat cycles. Such sensors require complex compromises between user needs, increased spectral information, quality of radiometry, swath width/revisit times and other engineering/cost constraints…. Arguably, the community needs a clear and quantitative case outlining the advantages (and disadvantages) of hyperspectral missions across a wide range of water types; typical assemblage variability/succession types and targeted phytoplankton groups, biogeochemistry etc; and the ability to meet user needs and demonstrate impacts of achieving an innovative sensing capability….

Hyper- vs Multispectral: What advantages for phytoplankton diversity applications?

Page 3: Stewart Bernard (CSIR)

Main advantages of hyperspectral radiometry: •  Increased number of spectral bands, i.e. equivalent of additional multi-

spectral bands, allowing the targeting of new phenomena e.g. phycoerythrin and phycocyanin absorption and fluorescence, longer wavelength scattering related features….

•  High spectral resolution allows new techniques such as derivative analyses, similarity indices, and targeting the precise wavelength location of shifting spectral peaks….

•  High spectral resolution allows new atmospheric correction/reflectance processing techniques targeting removal of gaseous absorption related features in the water leaving radiance….

Hyper- vs Multispectral: What advantages for phytoplankton diversity applications?

Page 4: Stewart Bernard (CSIR)

Hyperspectral Sensors: Emerging Opportunities…

atmospheric variability, air–water interface re!ections and refractionfrom diffuse and direct sky and sunlight (Brando & Dekker, 2003;Hochberg et al., 2011; Hu et al., 2012; Wettle, Brando, & Dekker,2004). Another equally important feature of sensor performance forsuccessful measurement of freshwater systems is suf"cient dynamicrange to be able to make sensitive measurements over low radiancepixels (water) while not saturating over neighboring bright pixels(land or sunglint).

The combined effects of SNR anddynamic range impact the accuracyof biophysical retrieval (e.g., Hu et al., 2012; Vanhellemont & Ruddick,2014). For example, when Giardino, Brando, Dekker, Strömbeck andCandiani (2007) used Hyperion to measure CHL and TR in Lake Gardain Northern Italy, they had to convolve a 5 ! 5 low pass "lter over theimage to reduce the effects of the sensor's poor SNR and environmentalnoise (Brando & Dekker, 2003; Wettle et al., 2004), effectively reducingthe spatial resolution from 30m to 150 m. Similarly, Vanhellemont andRuddick (2014) found it necessary to bin Landsat 7 ETM+ data to9 ! 9 pixels (270 m) to reach the noise equivalent of Landsat 8 OLI,and had to further bin the data to 11 ! 11 (330 m) due to the limiteddigitization (8 bits) of Landsat 7 ETM+. Freshwater ecosystems arespatially complex, and typically have both low (water) and high(land) radiance targets in a single scene, making simultaneous mea-surement of both problematic. The high SNR and large dynamic rangeproposed for the HyspIRI mission makes it uniquely well designed formeasuring freshwater ecosystems accurately andmoderate to high spa-tial resolution.

2.7. Current observation capabilities

For every type of measurement, there are tradeoffs in sensor resolu-tion. Fig. 5 shows some of the most common satellite sensors used forfreshwater ecosystem measurements and their relation in terms ofspectral (x-axis), temporal (y-axis), and spatial resolution (size of thebubble). HyspIRI's proposed spectral, temporal and spatial characteris-tics occupy an observation space shared with only a few other satellitemissions. However, HyspIRI's observational capabilities make it uniqueand necessary for freshwater ecosystem measurements, as it occupiesa unique niche in sampling space. Freshwater ecosystemmeasurementsfrom satellite remote sensing can be classi"ed based on the samplingstrategy and frequency. We categorize these different schemes into1) continuous samplers, 2) targeted mappers, and 3) global mappers.Continuous samplers are geostationary satellites that can image hightemporal frequency (e.g., Korea's Ocean Color Satellite GOCI thatmakes a measurement once an hour) of a speci"c location to providenear-continuous monitoring of dynamic processes such as harmfulalgal blooms and river plumes. Continuous samplers provide coarsespatial resolution over a speci"c, targeted region. Targeted mapperscan be considered pseudo global mappers. Also in a lower earth orbit(although not necessarily sun synchronous, e.g., the Hyperspectral Im-ager for the Coastal Ocean, HICO, onboard the International Space Sta-tion), targeted mappers acquire data over particular areas based ondata acquisition requests (e.g., NASA's EO-1 Hyperion or commercialmissions suitable for freshwater like Worldview 2 and 3; WV2, WV3),or regular acquisitions over a region of interest (e.g. the Italian SpaceAgency's proposed PRISMA mission, or the German EnvironmentalMapping and Analysis Program) thatwill providemapping-like capabil-ities over a speci"c region.

Fig. 5 shows the observation capabilities of common current andnear to launch sensors in terms of temporal, spectral, and spatial resolu-tions. Several missions, such as the soon to be launched Sentinel-2Mul-tispectral Instrument (S2-MSI) provide different spectral bands atdifferent pixel resolutions. Thus, while S2-MSI will have 13 spectralbands across the visible, near and shortwave infrared regions, it willonly have four broad “multispectral bands” in the visible and near infra-red regions at 10 meter pixel resolution.

Global mappers are valuable for providing regular, repeated mea-surements of the globe over long periods of time. They typically arealso archivalmissions,meaning they provide a time series of regular ob-servations. Archival global mappers are the most important category ofmeasurement for addressing multiple end user goals of resource moni-toring and ecosystem science. Archival global mapping missions withfree and open data access policies have transformed scienti"c under-standing of earth surface processes (National Research Council, 2007;Wulder et al., 2012), and provide the most valuable datasets for moni-toring (e.g., McCullough, Loftin, & Sader, 2012), and understandingfreshwater ecosystem processes and change (e.g., Olmanson, Brezonik,& Bauer, 2014). While Fig. 5 depicts the observation capabilities of com-mon current and near-ready to launch satellite missions, it includescontinuous and targeted mappers, such as Worldview 2 & 3 and Hype-rion which may not be suited for ecosystem change measurements.Fig. 6 explicitly summarizes the global mapping capability current andnear future global mapping capability for freshwater ecosystem scienceand management. In comparison with current global mapping capabil-ities, HyspIRI occupies a unique measurement space in both its spatialresolution and temporal resolution, and provides signi"cantly morespectral information than any other global mapper (Fig. 6).

3. Case studies

The following case studies illustrate how the characteristics of ahyperspectral global mapping satellite mission, such as the plannedHyspIRI mission, address the needs of freshwater aquatic system scien-tists and managers. We use as our example for freshwater aquatic ecol-ogy the remote sensing of primary producers. In the following casestudies we highlight published data and existing methods, demonstrat-ing thematurity of the science. However, each case study demonstratesexisting gaps in the spatial, temporal, and spectral characteristics of theapplication, highlighting the need of a mission that will "ll these gaps.

3.1. Site description

TheMantua lake system is an important freshwater wetland systeminNorthern Italy that provides critical habitat for aquatic vegetation andwater birds in the region. TheMantua system is formed by the dammingof theMincio River, a tributary of the Po, and fed by Lake Garda, the larg-est lake and longest river of Italy, respectively. The lake waters are

Fig. 5. The spectral (x-axis), temporal (y-axis), and spatial (size of the bubble) character-istics of satellite sensors commonly used for freshwater ecosystem measurements. Note:sensors that provide different spatial resolutions are plotted separately, and sensorswith overlapping resolution characteristics are slightly jittered for graphing purposes.

7E.L. Hestir et al. / Remote Sensing of Environment xxx (2015) xxx–xxx

Please cite this article as: Hestir, E.L., et al., Measuring freshwater aquatic ecosystems: The need for a hyperspectral global mapping satellitemission, Remote Sensing of Environment (2015), http://dx.doi.org/10.1016/j.rse.2015.05.023

PACE  

Hes)r,  E.L.,  et  al.,  Measuring  freshwater  aqua)c  ecosystems:  The  need  for  a  hyperspectral  global  mapping  satellite  mission,  Remote  Sensing  of  Environment  (2015),  hLp://dx.doi.org/10.1016/j.rse.2015.05.023  

ENMAP  

Page 5: Stewart Bernard (CSIR)

Hyperspectral Sensors: Need for High Quality Radiometry

Moses,  W.J.;  Bowles,  J.H.;  Corson,  M.R.  Expected  Improvements  in  the  Quan)ta)ve  Remote  Sensing  of  Op)cally  Complex  Waters  with  the  Use  of  an  Op)cally  Fast  Hyperspectral  Spectrometer—A  Modeling  Study.  Sensors  2015,  15,  6152-­‐6173.  

!"#$%&$'!"#$!"()' %#$&""

"

#$%$&%'(")%"*)+$,$-.%/0"1$,'*"234"-5")-#")1'+$"634"-5")-#",'*$("%/)-"%/)%"'7"%/$"89:;"#$%$&%'(")%"*)+$,$-.%/0"1$%*$$-"234"-5")-#"634"-5<"=/$"0%)-#)(#"#$+>)%>'-"'7")"#)(?">5).$")&@A>($#"1B"89:;"*)0"%)?$-"%'"($C($0$-%"%/$"&'51>-)%>'-"

'7"#)(?"-'>0$!"($)#'A%"-'>0$!")-#"#>.>%>D)%>'-"-'>0$"7'("%/$",'*EFGH"0B0%$5<"I'("%/$"/>./EFGH"0B0%$5!"%/$"#)(?"-'>0$!"($)#'A%"-'>0$!")-#"#>.>%>D)%>'-"-'>0$"*$($"'1%)>-$#"7('5"%/$">-7'(5)%>'-"0/$$%"7'("%/$""IJKELML3N":O;F"#$%$&%'(<"9-"&/)-.>-." %/$"IE-A51$("7('5"P<3" %'"Q<4"*/>,$"5)>-%)>->-." %/$"7'&)," ,$-.%/!" %/$"#>)5$%$("'7" %/$"

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

"

'()*+,-!."=/$")+$().$"FGH"&),&A,)%$#"7('5"%/$")%E0$-0'("()#>)-&$0"T'7"),,"2S4">5).$0U"7'("%/$",'*E")-#"/>./EFGH"0B0%$50<"=/$"()%>'"'7"%/$")+$().$"FGH"7'("%/$"/>./EFGH"0B0%$5"%'"%/)%"7'("%/$",'*EFGH"0B0%$5">0"C,'%%$#"'-"%/$"0$&'-#)(B")W>0<""

L<P<P<"95C('+>-."X77$&%>+$"FGH"%/('A./"Y'0%EY('&$00>-."

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

!"#$%&$'!"#$!"()' %#%&""

"

"

'()*+,- ./" #$%" &'%(&)%" *%(+%,-" %((.(/" 0.(" -$%" 1.23" &,4" $5)$3678" /9/-%:/" 5," -$%"+.,+%,-(&-5.,/" .0" +$13*" &,4" 6;<" &,4"*=>?<@AABC" (%-(5%'%4" 0(.:" -$%" DB" &-:./*$%(5+&119"+.((%+-%4",.5/9"5:&)%/"E/5,)"-$%",.,315,%&("1%&/-"/FE&(%/"%((.(":5,5:5G&-5.,"&**(.&+$"

@55C"6%:53H,&19-5+&1"7I83(%4"H1).(5-$:J"

#$%"-2.3K&,4"7I83(%4"&1).(5-$:"5/",.-"+.:*E-&-5.,&119"5,-%,/5'%"&,4"2&/"-$%(%0.(%"&**15%4"-."&11"ALB"5:&)%/M"N5)E(%"OB"/$.2/"-$%"&'%(&)%"*%(+%,-"%((.(/".0"-$%"%/-5:&-%4"+$13*"+.,+%,-(&-5.,/"0.("-$%"1.23" &,4" $5)$3678" /9/-%:/" 25-$" &,4" 25-$.E-" /*&-5&1" &'%(&)5,)" &-" 7I8" 2&'%1%,)-$/M" #$%" &'%(&)%"*%(+%,-" %((.(" 0.(" -$%" 1.23678" /9/-%:"2&/" '%(9" $5)$" @AOBMODPCM" H1).(5-$:/" /E+$" &/" -$%" -2.3K&,4"7I83(%4" &1).(5-$:" -$&-" &(%" %:*5(5+&119" *&(&:%-%(5G%4" &(%" '%(9" /%,/5-5'%" -." -$%" +.%005+5%,-/" .0" -$%"(%)(%//5.," %FE&-5.," ).'%(,5,)" -$%" &1).(5-$:M" #$%/%" &1).(5-$:/" &(%" KE51-" .," -$%" &//E:*-5.," -$&-"'&(5&-5.,/"5,"-$%"(%01%+-&,+%/"&-"-$%"2&'%1%,)-$/"5,+1E4%4"5,"-$%"&1).(5-$:"&(%"*(5:&(519!"50",.-"/.1%19!"4E%"-."'&(5&-5.,/"5,"-$%"2&-%("FE&15-9"*&(&:%-%(".0"5,-%(%/-M"H/"/E+$!"-$%/%"&1).(5-$:/"&(%"'%(9"/%,/5-5'%"-."'&(5&-5.,/"5,"-$%"(%01%+-&,+%/"4E%"-."&,9".-$%("0&+-.("/E+$"&/"/%,/.(",.5/%M"#$5/"5/"5:*.(-&,-"-."K%&("5,":5,4"2$%,"&//%//5,)"-$%"7I83(%4"&1).(5-$:"K%+&E/%"-$%"%00%+-/".0"-$%"/%,/.(",.5/%"&(%"%Q*%+-%4"-."K%"$5)$%(" 5," -$%" 1.,)%("2&'%1%,)-$/M"#$%",E:%(5+&1":%-$.4!".," -$%".-$%("$&,4!" +.,/54%(/" -$%" %,-5(%"/*%+-(E:"2$51%"(%-(5%'5,)"2&-%("FE&15-9"*&(&:%-%(/"@ABBRSDT",:"5,".E("+&/%C"&,4"+&,"-%,4"-."&'%(&)%".E-" -$%" ,.5/%" &+(.//" -$%" /*%+-(E:" 25-$.E-" K%5,)" .'%(19" /%,/5-5'%" -." ,.5/%" 5," .,%" .(" -2." *&(-5+E1&("2&'%1%,)-$/M"#$%"&K/.1E-%"%((.("95%14%4"K9"&"/%:53&,&19-5+&1"&1).(5-$:"+&,"K%":5-5)&-%4"K9"&4UE/-5,)"-$%"+.%005+5%,-/".0"-$%"(%)(%//5.,"%FE&-5.,M"#$%"%:*$&/5/"$%(%"5/!"-$%(%0.(%!",.-".,"-$%"&K/.1E-%"%((.("5,"-$%" +$13*" +.,+%,-(&-5.,/" (%-(5%'%4" 0(.:" -$%"7I83(%4" &1).(5-$:"KE-" .," -$%" 5:*(.'%:%,-" &+$5%'%4" K9"/*&-5&119"&'%(&)5,)"-$%"7I8"(%01%+-&,+%/"&,4"+$&,)5,)"-$%"N3,E:K%(M"V$%,"7I8"(%01%+-&,+%/"0(.:"-$%"1.23678"/9/-%:"2%(%"/*&-5&119"&'%(&)%4!"-$%"*%(+%,-"%((.("5,"-$%"(%-(5%'%4"+$13*"+.,+%,-(&-5.,"(%4E+%4"-."OLDMWLPM"#$%"*%(+%,-" %((.(" 0.(" -$%"$5)$3678"/9/-%:"2&/"XDMBYP"25-$.E-"7I8"/*&-5&1" &'%(&)5,)"&,4"OYMWYP"25-$"/*&-5&1"&'%(&)5,)"&-" -$%"7I8"/*%+-(&1"K&,4/M" I,"/E::&(9!" -$%"%((.(" 5," -$%" (%-(5%'%4"+$13*" +.,+%,-(&-5.," (%4E+%4" K9" LDMOWP"2$%," -$%" N3,E:K%("2&/" +$&,)%4" 0(.:"XMT" -." OMBM" N.(" K.-$"/9/-%:/!"/*&-5&1"&'%(&)5,)"5,"-$%"7I8"2&'%1%,)-$/"(%/E1-%4"5,"&**(.Q5:&-%19"TBP"(%4E+-5.,"5,"%((.("5,"-$%"(%-(5%'%4"+$13*"+.,+%,-(&-5.,M"

Page 6: Stewart Bernard (CSIR)

In Situ Measurements: Need for Improved Capability

HPLC & CHEMTAX ≠ necessary biophysical phytoplankton assemblage data We cannot use satellites to determine phytoplankton diversity if we do not routinely make sufficiently detailed measurements of that diversity. There needs to be a community and agency push towards broader adoption, further development and standardised protocols for emerging sensors such as imaging flow cytometers, holographic microscopes and other technologies allowing routine measurements of detailed cellular information such as size and taxonomy.

Report on IOCCG workshop: “Phytoplankton Composition from Space: towards a validation strategy for satellite algorithms”, October 2014 Consideration of the measurements required for validating PFT algorithms produced the following list. • Size-fractionated measurements of both HPLC pigments and particulate light absorption. • Measurements of phycobilin concentrations, equally in size fractions, to the suite of pigments (for Synechococcus, cryptophytes, Trichodesmium) • FlowCytobot/FlowCAM/flow cytometry (both traditional and imaging) • Radiometry, both above water and in-water, hyperspectral • Inherent optical properties (absorption, backscattering, VSF) • Particle size distribution (PSD, e.g. via LISST) • Size-fractionated measurements • Genetics/-omics

Page 7: Stewart Bernard (CSIR)

From  100963  spectra  collected  in  the  Bal)c  Sea…  

…289  (0.29%)  show  extreme  phycoerythrin  absorp)on    and  a  dis)nct  orange    peak  at  590-­‐600  nm  

PE  

Possibly  fluorescence  by  PE  

Simis  et  al.  in  prep  

www.ferryscope.org  

Hyperspectral Approaches: Examples of Additional Spectral Information

Page 8: Stewart Bernard (CSIR)

No  590  nm  band  from  current  sensors  

Hyperspectral Approaches: Examples of Additional Spectral Information

Page 9: Stewart Bernard (CSIR)

Hyperspectral Approaches: Examples of Additional Spectral Information

400 450 500 550 600 650 700 750 800 850 9000

2

4

6

8x 10−3

wavelength (nm)

R rs (s

r−1 )

1 mg.m−3

10 mg.m−3

200 mg.m−3

100 101 1020

0.5

1

1.5

B/G ratio R ratio B/G: dy/dx R: dy/dx

Fig. 6. Modelled EAP Rrs (above) for typical dinoflagellate assemblage with effective di-ameter of 16 µm. Chl a concentrations are 1, 2, 5, 10, 20, 30, 50, 100, 150, 200 mg.m!3.Below the shift frommaximum peak reflectance height in the blue/green to the red is shown(dotted lines), for increasing Chl a. The first derivatives of these slopes (solid lines) crossat a Chl a of around 15 mg.m!3, the point at which the red features of high biomass re-flectance spectra start to dominate.

6, as per the SeaWiFs OC4 Chl a retrieval algorithm [21]) shows the importance of signal inthe red for Chl a estimation in waters with concentrations over about 25 mg.m!3.

Other notable features of Rrs with increasing eutrophication are convergence at around 660nm, as well as around 560 nm, and the well constrained region between these points. Thedecrease in Rrs in the blue is indicative of decreased influence of gelbstoff absorption withrespect to Chl a absorption as the latter becomes more dominant spectrally.

4.5. EAP Rrs validation at very high biomass

Some very high biomass TSRB measured Rrs are presented here. The measurements (N=4,6 and 5 respectively) represent a range of assemblage types and size distributions, which theEAP forward modelled Rrs do not consider here. This brief validation exercise shows that achosen effective diameter of 12 µm most accurately matches all three high biomass blooms(for Chl a of 110, 150 and 180 mg.m!3 respectively). Most of the Rrs measurements werefrom a Prorocentrum triestinum-dominated bloom in 2005, with varying lesser proportions ofDinophysis acuminata, D. fortii and P. reticulatum [22]. P. triestinum is a small dinoflagellate

#203293 - $15.00 USD Received 17 Dec 2013; accepted 29 Jan 2014; published 1 Jul 2014(C) 2014 OSA 14 July 2014 | Vol. 22, No. 14 | DOI:10.1364/OE.22.016745 | OPTICS EXPRESS 16755

Robertson  Lain,  L.,  Bernard,  S.,  and  Evers-­‐King,  H.  (2014)  Biophysical  modelling  of  phytoplankton  communi)es  from  first  principles  using  two-­‐layered  spheres:  Equivalent  Algal  Popula)ons  (EAP)  model.  Op)cs  Express,  22(14),  16745–16758..  

Eutrophic and hypertrophic waters contain considerable phytoplankton scattering related signal in the 570 – 650 nm and above 680 nm. Additional hyperspectral related signal in these regions would be very valuable….

Page 10: Stewart Bernard (CSIR)

Hyperspectral Approaches: Examples of Phytoplankton Diversity Applications….Multiple Groups

multispectral (TIR) resolution, at a spatial scale of 60 m GSD, with a 19-day return rate. Competing interests necessarily in!uence the develop-ment of this satellite sensor suite to optimally image the Earth to ad-dress the science questions put forth by the Decadal Survey.

The Case 2waters of the coastal zonehave special needswith respectto sensor sensitivity, sensor calibration (Kohler, Bissett, Steward, &Davis, 2004), dark pixel constraints (Gao & Davis, 1997), and atmo-spheric correction (Gao & Goetz, 1990). The ocean is a radiometricallydark target with a surface albedo in the range of 5–10% (Kirk, 1994).

As a result, image sensors collecting light emitted from optically deepocean targets must have the appropriate sensitivity, or SNR, for suchan environment. Historically, imager SNR is optimized in the green(~550 nm) part of the visible spectrum with decreasing SNR towardsthe blue and UV and also towards the NIR (Moses et al., 2012), as isthe case with the AVIRIS sensor (Green et al., 1998). Sensor design fordark ocean targets must reconcile the competing need for high SNRthroughout the visible range and high dynamic range (or saturation ra-diance) in theNIR in order to both quantify small magnitude differences

36.8

36.60 2 4

BiomassChl-a (mg/m3)

37.0-122.2 -122.0 -121.8

A

36.8

36.6

-122.2 -122.0 -121.8

0 3 6

BiomassChl-a (mg/m3)

0 2 4

BiomassChl-a (mg/m3)

36.8

36.60 2 4

-122.2 -122.0 -121.8

BiomassChl-a (mg/m3)

0 3 6

BiomassChl-a (mg/m3)

36.8

36.6-122.2 -122.0 -121.8

0 3 6

BiomassChl-a (mg/m3)

37.0

36.8

36.6

36.8

36.6

B

C

D

E

F

Fig. 5. Phytoplankton biodiversity— PHYDOTax estimates. Taxon-speci"c biomass estimated for dino!agellates, diatoms, and cyanobacteria using PHYDOTax applied to the ATREM+ at-mospheric corrected imagery. 10 Apr 2013: A) dino!agellates, B) diatoms, C) cyanobacteria. 31 Oct 2013: D) dino!agellates, E) diatoms, F) cyanobacteria.

9S.L. Palacios et al. / Remote Sensing of Environment xxx (2015) xxx–xxx

Please cite this article as: Palacios, S.L., et al., Remote sensing of phytoplankton functional types in the coastal ocean from the HyspIRI PreparatoryFlight Campaign, Remote Sensing of Environment (2015), http://dx.doi.org/10.1016/j.rse.2015.05.014

Palacios,  S.L.,  et  al.,  Remote  sensing  of  phytoplankton  func)onal  types  in  the  coastal  ocean  from  the  HyspIRI  Preparatory  Flight  Campaign,  Remote  Sensing  of  Environment  (2015),  hLp://dx.doi.org/10.1016/j.rse.2015.05.014  

Fig. 5. Phytoplankton biodiversity — PHYDOTax estimates. Taxon-specific biomass estimated for dinoflagellates, diatoms, and cyanobacteria using PHYDOTax applied to the ATREM+ at- mospheric corrected imagery. 10 Apr 2013: A) dinoflagellates, B) diatoms, C) cyanobacteria, D) dinoflagellates, E) diatoms, F) cyanobacteria.

HyspIRI simulations from airborne data using the PHYDOTax spectral library/inversion approach

Page 11: Stewart Bernard (CSIR)

Hyperspectral Approaches: Examples of Phytoplankton Diversity Applications….Cyanobacteria

Lin  Li  ,  Rebecca  E.  Sengpiel  ,  Denise  L.  Pascual  ,  Lenore  P.  Tedesco  ,  Jeffrey  S.  Wilson  &  Emmanuel  Soyeux  (2010)  Using  hyperspectral  remote  sensing  to  es)mate  chlorophyll-­‐a  and  phycocyanin  in  a  mesotrophic  reservoir,  Interna)onal  Journal  of  Remote  Sensing,  31:15,  4147-­‐4162  

Spectral ratio based approach

Figure 6. Correlation between estimated and measured phycocyanin (PC) concentration.

Figure 7. Phycocyanin (PC) concentration derived by applying the relationship shown infigure 5 to the calibrated AISA imagery of Geist Reservoir (see figure 1 for its location). Notethat a PC ‘hot spot’ is located on the left side of the figure, and the high PC value along the shoreof this reservoir is due to the effect of the bottom reflection, which cannot be accounted for bythe relationship in figure 5.

Airborne imaging spectrometry 4157

Dow

nloa

ded

by [U

nive

rsity

of C

ape

Tow

n Li

brar

ies]

at 1

4:55

16

June

201

5

The results for the spectral ratios (e.g. R699–705 / R670–677 and R698–716 / R671–684)presented in this study are not as good as previously published studies where highcorrelations and low RMSEs were obtained (e.g. Kallio et al. 2001, 2003), indicatingthat these spectral-band ratios are not optimal spectral indices for Geist Reservoir.

3.3 Estimation of PC concentration

The results for estimating PC are presented in table 3, indicating that PC concentra-tion correlated best with the spectral index R628. The relationship between R628 andPC concentration is shown in figure 5 where two variables are linked via an exponen-tial relation. The correlation between measured and estimated PC values is shown infigure 6, indicating that the relationship between R628 and PC concentration resultedin a RMSE of 25.52 mg l–1 when compared to the measured PC concentration.Figure 7 shows the PC concentration map of Geist Reservoir derived by applyingthe relationship shown in figure 5 to the calibrated AISA image. A PC ‘hot spot’ isevident on the left-hand side of figure 7, but the high PC value along the shore of this

Table 3. Spectral indices (x) derived from AISA reflectance spectra and their relationships tocorrelation to measured phycocyanin concentration (y).

Spectral index Model R2 RMSE(mg l–1)

[1]0.5(R600 ! R647) – R628 log10(y) " 0.008x–0.93 0.69 30.42[2]R647 / R628 log10(y) " –99.61x2 ! 218.5x – 117.8 0.62 30.92[3]R704 / R628 log10(y) " –3.137(log10(x))

2

! 10.21log10(x) – 6.2560.47 32.46

[4]R628 log10(y) " 0.2896e–1.104log10(x) 0.80 25.52

[1] Dekker (1993); [2] Schalles and Yacobi (2000); [3] Simis et al. (2005); [4] Millie et al. (1992).

2.40

2.20

2.00

1.80

1.60

1.40

1.20

1.00–1.85 –1.75 –1.65 –1.55 –1.45 –1.35

Log10 R628

y = 0.2896e–1.104x

R2 = 0.8005

Lo

g10

PC

(µg

l–1)

Figure 5. Relationship between phycocyanin (PC) and the AISA reflectance at 628 nm (R628).

4156 L. Li et al.

Dow

nloa

ded

by [U

nive

rsity

of C

ape

Tow

n Li

brar

ies]

at 1

4:55

16

June

201

5

Page 12: Stewart Bernard (CSIR)

Hyperspectral Approaches: Examples of Phytoplankton Diversity Applications….Cyanobacteria

Kudela,  R.M.,  et  al.,  Applica)on  of  hyperspectral  remote  sensing  to  cyanobacterial  blooms  in  inland  waters,  Remote  Sens-­‐  ing  of  Environment  (2015),  hLp://dx.doi.org/10.1016/j.rse.2015.01.025  

Scattering based approach

from Microcystis to Aphanizomenon. Low AMI values (~2) were empiri-cally determined to be dominated byMicrocystiswhile high values (~5)were dominated by Aphanizomenon. A caveat of this approach is that theAMI doesn't provide useful information if there is no cyanobacterialbloom.

3.4. Field application of indices

Fig. 7 shows the two biomass indices, the CI and SLH, plotted versusthe corresponding microcystin concentration from the same dates forPinto Lake, and the AMI plotted versus microcystin concentration for

A

B

C

5

10

15

20

25

30

Wat

er T

empe

ratu

re (°

C)

050

100150200250300

1000104

105

Chl

orop

hyll

(µg/

L)P

hcoc

yani

n (µ

g/L)

0

2

4

6

8

10

1001000

104

2010/4 2010/10 2011/4 2011/10 2012/4 2012/10 2013/4 2013/10 2014/4

Mic

rocy

stin

LR

(ppb

)

Date

a b c d e 1 2345 6 7 f 89 g h i 10 11

Fig. 3. Time series of approximately weekly water samples collected at Pinto Lake, CA showing (A) water temperature, (B) chlorophyll (open circles) and phycocyanin (shaded circles)concentration, and (C) microcystin LR concentrations. The yellow, orange, and red shading on the timeline denotes periods of dominance by Aphanizomenon, a mixed community, anddominance by Microcystis, respectively. Black/gray triangles indicate data collection for above-water spectra (black) and HICO imagery (gray). The numbers/letters with the triangles inthe bottom panel indicate annotations for the corresponding spectra in Fig. 5. (For interpretation of the references to color in this !gure legend, the reader is referred to the web versionof this article.).

bb /b = 0.003, C = 20 µg/Lbb /b =0.003, C = 10 µg/Lbb /b = 0.003, C = 5 µg/Lbb /b = 0.0150, C = 20 µg/Lbb /b = 0.0150, C = 10 µg/Lbb /b = 0.0150, C = 50 µg/L

Fig. 4. Results from the SLH algorithm derived from Hydrolight simulations of cyanobacteria, with varying concentrations of chlorophyll and backscattering ratios. SLH is relatively insen-sitive to changes in chlorophyll. Vertical gray lines indicate MASTER bands used for SLH.

6 R.M. Kudela et al. / Remote Sensing of Environment xxx (2015) xxx–xxx

Please cite this article as: Kudela, R.M., et al., Application of hyperspectral remote sensing to cyanobacterial blooms in inlandwaters, Remote Sens-ing of Environment (2015), http://dx.doi.org/10.1016/j.rse.2015.01.025

Page 13: Stewart Bernard (CSIR)

Hyperspectral Approaches: Examples of Phytoplankton Diversity Applications….Phaeocystis

Lubac,  B.,  H.  Loisel,  N.  Guiselin,  R.  Astoreca,  L.  Felipe  Ar)gas,  and  X.  Mériaux  (2008),  Hyperspectral  and  mul)spectral  ocean  color  inversions  to  detect  Phaeocys)s  globosa  blooms  in  coastal  waters,  J.  Geophys.  Res.,  113,    

dl2Rrs is observed around 480 nm, and when Cp tendstoward 90%, this minimum is observed around 510 nm.[20] The method based on the shifts of the extremes in the

dl2Rrs spectra is chosen to identify the spectra correspondingto a P. globosa bloom in our whole Rrs(l) data set. On thebasis of the results presented above, we define two criteriato discriminate the P. globosa blooms from diatoms blooms(Cp higher than 60%): the maximum of dl2Rrs in the 460–480 nm spectral range has to be observed beyond 471 nm,and the minimum of dl2Rrs in the 480–510 nm spectralrange has to be observed beyond 499 nm. The application ofthese two criteria to our whole Rrs(l) data set provides 21spectra associated with P. globosa blooms. This is inagreement with a qualitative analysis based on our in situobservations.

3.3. Comparison of the RRS(l)-Hyperspectral AnalysisWith the af (l)-Similarity Index Analysis

[21] The classification of the Rrs(l) spectra obtained fromthe two criteria presented above, is compared to that

achieved by means of a similarity index analysis asdescribed by Millie et al. [1997]. This last method wassuccessfully applied for the detection and assessment of theharmful alga, Karenia Brevis [Craig et al., 2006]. In thepresent study, the similarity index (SI) is computed betweenthe second derivative of an in situ reference P. globosaabsorption spectrum (aphaeo-in situ

ref (l)) and the second deriv-ative of an in situ unknown phytoplankton populationabsorption spectrum (af-in situ(l)) in the 400–540 nmspectral range. Then, based on a relationship between SIand the P. globosa biomass, the absorption spectra identifiedas P. globosa are selected. This study is performed in 400–540 nm spectral range because, the spectral variability ofaf(l) is sufficiently high to discriminate a phytoplanktoncommunity dominated by P. globosa from a phytoplanktoncommunity dominated by diatoms. Indeed, the SI valuecomputed between the second derivative of the P. globosaabsorption spectrum and the second derivative of thediatoms absorption spectrum, both obtained from culture,

Figure 4. (a) Mean and (b) second derivatives spectra of Rrs(l) associated with the group 1 (P. globosa),group 2 (mixed), and group 3 (diatoms) (see Table 1).

C06026 LUBAC ET AL.: ALGAL BLOOMS DETECTION IN COASTAL WATERS

8 of 17

C06026

is around 0.59 in the 400–540 nm spectral range, versus0.78 in the 400–700 nm spectral range.[22] Because of the lack of in situ phytoplankton absorp-

tion measurements, af-in situ(l) is derived from Rrs(l) usingthe quasi-analytical algorithm of Lee et al. [2002]. Figure 6ashows the spectra of af(l) retrieved from our whole Rrs(l)data set. aphaeo-in situ

ref (l) is then computed as the meanspectrum of the retrieved af-in situ(l) associated with P.globosa blooms (group 1 in Table 2). To verify the

consistency of the retrieval, the spectrum of aphaeo-in situref (l)

and its second derivative are compared to thoseobtained from P. globosa cultures by Astoreca [2007],(aphaeo-culture(l)) (Figures 6b–6c). A relatively good agreementis observed between the field retrieved and laboratorymeasuredabsorption spectra (Figure 6b). Importantly, the maxima andminima of the second derivative of aphaeo-in situ

ref (l) andaphaeo-culture(l) are observed at the same wavelengths(Figure 6c). The main sources of discrepancies between the

Figure 5. Variation of the P. globosa biomass (Cp, expressed in %) as a function of (a) the position ofthe maximum of the second derivative of Rrs(l) in the range 460–480 nm (l(max(d2Rrs))), and (b) theposition of the minimum of the second derivative of Rrs(l) in the range 480–510 nm (l(min(d2Rrs))). Theblack lines represent the linear regressions between Figure 5a Cp and l(max(d2Rrs)), and Figure 5b Cp

and l(min(d2Rrs)). The regression equations, the number of observations (N), and the squared correlationcoefficient (r2) are given.

Figure 6. (a) Phytoplankton absorption spectra (N = 93) retrieved from Rrs(l) using the quasi-analytical algorithmdeveloped by Lee et al. [2002]. (b) Field retrieved (aphaeo-in situ

ref (l)) and laboratory measured (aphaeo-culture(l)) normalizedabsorption spectra of P. globosa (see text for details), and (c) their second derivatives.

C06026 LUBAC ET AL.: ALGAL BLOOMS DETECTION IN COASTAL WATERS

9 of 17

C06026

Page 14: Stewart Bernard (CSIR)

400 600 8000

50

100

% aphy

of total absorption

%

100 mg/m3

30 mg/m3

10 mg/m3

5 mg/m3

1 mg/m3

0.1 mg/m3

400 600 8000

50

100

% bbphy

of total bb

%

400 600 8000

2

4

m1

Downwelling diffuse attenuation: Kd

400 600 8000

0.005

0.01

Chl a in Rrs

: 0.1 mg/m3

Rrs

(sr

1)

400 600 8000

0.005

0.01

Chl a in Rrs

: 1 mg/m3

400 600 8000

0.005

0.01

Chl a in Rrs

: 5 mg/m3

400 600 8000

0.005

0.01

Chl a in Rrs

: 10 mg/m3

Wavelength (nm)

Rrs

(sr

1)

400 600 8000

0.005

0.01

Chl a in Rrs

: 30 mg/m3

Wavelength (nm)

400 600 8000

0.005

0.01

Chl a in Rrs

: 100 mg/m3

Wavelength (nm)

Phytoplankton component from EAP two layered sphere model. Other components:

bbs

( )=1.2

and agd

( )=agd

(400)exp[ 0.012*( 400)] where agd

(400)=0.15*log(Chl)+0.25

IOP Budget Based Assemblage Contribution to the Ocean Colour Signal

Assemblage related sensitivity in reflectance is highly dependent on algal biomass, as determined by the relative algal contribution to the IOP budget…

a(er  Robertson  Lain,  L.,  Bernard,  S.,  and  Evers-­‐King,  H.  (2014)  Biophysical  modelling  of  phytoplankton  communi)es  from  first  principles  using  two-­‐layered  spheres:  Equivalent  Algal  Popula)ons  (EAP)  model.  Op)cs  Express,  22(14),  16745–16758..  

Page 15: Stewart Bernard (CSIR)

Phytoplankton size and the changing hyperspectral signal with increasing biomass

An Equivalent Algal Population model coupled to Ecolight allows description of the reflectance range associated with assemblage effective diameters from 4 to 40 µm. Case 1 waters only, i.e. the best case scenario….

a(er  Evers-­‐King,  H.,  Bernard,  S.,  Robertson  Lain,  L.,  and  Probyn,  T.  A.  (2014)  Sensi)vity  in  reflectance  aLributed  to  phytoplankton  cell  size:  forward  and  inverse  modelling  approaches.  Op)cs  Express,  22(10),  11536–11551.  

reflectance  over  modelled  size  range  (suitable  for  biomass  based  on  allometry)  at  a)  0.1,  1  b)  3  c)  10  and  d)  100  mg/m3  [Chl  a].  The  central  solid  line  represents  the  modelled  value  plus  expected  measurement  uncertainFes  at  effecFve  diameter  =  10  microns  

Page 16: Stewart Bernard (CSIR)

Note that in highly scattering waters the size related signal is very much reduced…..

Evers-­‐King,  H.,  Bernard,  S.,  Robertson  Lain,  L.,  and  Probyn,  T.  A.  (2014)  Sensi)vity  in  reflectance  aLributed  to  phytoplankton  cell  size:  forward  and  inverse  modelling  approaches.  Op)cs  Express,  22(10),  11536–11551.  

Fig. 2. Ranges of modelled Rrs to variations in De f f and [Chl a], under high bbs and highagd conditions. Note the differences in scale, where (b - ES) shows much higher Rrs valuesthan (a - REFA). Dots indicate Rrs associated with smallest cells. c) Shows example rangesof spectral Rrs at selected [Chl a] across the modelled size range using ES.

dominant against non-algal backscattering. Similarly, additional absorption in the presence ofhigh agd , suppresses size related variability in the blue. In these “case two” type waters, partic-ularly the highly scattering case, the differences in the two radiative transfer schemes becomesmost pronounced. The Rrs values generated by ES under high bbs (Fig. 2(b)), are significantlyhigher than those from the REFA approach. Also, different trends in size related variabilitywith biomass are observed, particularly in the red, as a result of the propagation of the highbbs influence in to the bidirectional effects, which in the REFA approach are constrained by thef/Q parameterisation both in terms of magnitude and spectral shape.In interpreting these results in the context of ocean colour data, it should be noted that the

substantial sensitivity observed in Fig. 1 likely represents a “best case” scenario. The simplifiedsize parameterisation may overstate the available signal (see discussion below (section 3.5). Itis likely that IOP covariance is often more extreme (i.e. high biomass is often associated withcase 2 coastal waters (Fig. 2) and measurement and/or atmospheric correction errors are likelyto add substantial uncertainty.

3.2. Inversion errors in cell size retrievals

Comprehensive discussion and estimation of the ambiguity and minimum errors associatedwith the physical problem of ocean colour inversion was conducted by Defoin-Platel and Chami[20]. An approximation of this approach, using the simulated data, can provide insight in to boththe role of phytoplankton size in this ambiguity, and the minimum errors one may expect fromthe inversion approach used here prior to application to in situ data where measurement errorwill also factor. In addition to the Nelder-Mead simplex, several mathematical techniques in-cluding Levenberg-Marquardt optimisation and an evolutionary algorithm were explored, how-ever the Nelder-Mead simplex provided the most consistent and optimal results for both thein situ and simulated data inversion.Figure 3 shows the root mean squared errors in De f f prediction over the simulated data set,

for four generalised conditions: low agd and low bbs (“case one”/the “Benguela type” watersabove), high agd and low bbs, low agd and high bbs, and high agd and high bbs (case two/gelbstoffand sediment influenced waters). Lowest errors occur in the context of Benguela type waters,with low agd and low bbs. High error and substantial scatter in RMSE values across biomasslevels for the high bbs scenarios suggests significant ambiguity may be introduced under highlyscattering conditions. As a result of this presumably more accurate handling of bidirectional-

#203303 - $15.00 USD Received 17 Dec 2013; revised 6 Feb 2014; accepted 6 Feb 2014; published 6 May 2014(C) 2014 OSA 19 May 2014 | Vol. 22, No. 10 | DOI:10.1364/OE.22.011536 | OPTICS EXPRESS 11545

Phytoplankton size and the changing hyperspectral signal with increasing biomass

Page 17: Stewart Bernard (CSIR)

Phytoplankton ultrastructure and the changing hyperspectral signal with increasing biomass

Vacuoles greatly increase the scattering for equivalent cells – this is a clear and easily used signal, at least for high biomass inland/coastal waters…

MaLhews,  M.  W.  and  Bernard,  S.:  Using  a  two-­‐layered  sphere  model  to  inves)gate  the  impact  of  gas  vacuoles  on  the  inherent  op)cal  proper)es  of  Microcys)s  aeruginosa,  Biogeosciences,  10,  8139-­‐8157,  doi:10.5194/bg-­‐10-­‐8139-­‐2013,  2013.  

8152 M. W. Matthews and S. Bernard: Impact of gas vacuoles on inherent optical properties

0.05 0.1 0.5 1

0.05

0.1

0.5

1

b bp 7

00 n

m (m

−1)

bbp 420 nm (m−1)

A

0 0.2 0.4 0.6 0.8−5

−4

−3

−2

−1

0

bbp (m−1)

Dep

th (m

)

B

bbp 420 nmbbp 700 nm

Fig. 8. (A) bbp(700) vs. bbp(420) from 11 profiles. Note exponen-tial relationship with log-scaled axes. (B) Five selected depth pro-files with minimum depth of 0.8m. Note that bbp(420)> bbp(700)at depth.

order of magnitude as those estimated from Rrs. The b∗bp val-

ues are likely underestimates since Chl a was measured atthe surface and the blooms were floating. There is some dis-agreement in the spectral shape between Zhou et al. (2012)(Chl a approximately 5mgm−3) and the two-layered sphereestimates (Fig. 7f). Interestingly, bbp is positively sloped to-wards the red near the surface and negatively sloped at depth(Fig. 8b). Therefore there is some evidence for an upward-sloping bbφ for M. aeruginosa, in contrast to those of Zhouet al. (2012), which are downward-sloping. The explana-tion for the depth-variable slope is not known, but seems tobe related to biomass; if biomass affects the spectral slope,then this finding has has important implications for bbp mea-sured using cultures. In summary, the b∗

bφ values are typi-cally larger or in the upper range of those determined experi-mentally for cultured algae (e.g. Whitmire et al., 2010; Vail-lancourt, 2004), and are in good agreement with the limitedmeasurements made on vacuolate M. aeruginosa.The modelled Rrs spectra correspond to chromatoplasm

1+ � values varying between 1.104 and 1.138 with a meanvalue of 1.12. The 1+� values estimated using Rrs are likelyto be justified for the following reasons: the overwhelmingoptical dominance of phytoplankton relative to other wa-ter components, meaning that from an optical perspective,the measurements were effectively performed on “cultures”;aside from Vg and some small uncertainty related to the es-timated PSD, the primary factor controlling phytoplanktonbackscatter is the chromatoplasm 1+ � value, which was theonly variable parameter used to determine appropriate val-ues for Rrs; the range of 1+ � values fall in the expectedrange of eukaryotic chloroplasts of 1.09 to 1.19, mean of 1.14(Bernard et al., 2009), and result in an overall homogeneousn of 0.97 by volume equivalence; and finally, the resultingvalues determined for b∗

bφ are in close agreement with mea-surements made on vacuolate cultures and natural blooms ofM. aeruginosa.Furthermore, the chromatoplasm 1+ � values accord with

previous two-layered modelling efforts: Quinby-Hunt et al.(1989) used a shell layer n= 1.13 and core n= 1.08 to best

Fig. 9. Rrs modelled at various concentrations of Chl a showingthe difference between vacuolate (solid lines) and non-vacuolate(dotted lines) populations of M. aeruginosa. Vacuolate cells weremodelled with shell 1+� of 1.12 and Vg = 50%. Non-vacuolate cellswere homogeneous cells with 1+� of 1.080. The values for atr(440)and ag(440) were constant at 0.5 and 1.5m−1, respectively. Formore details, see the text.

reproduced the scattering matrix of Chlorella, a species withsimilar morphology to M. aeruginosa, and the typical 1+ �

value for a shell layer proxy for the cell wall is 1.2 (e.g.Svensen et al., 2007). Given these considerations, the result-ing 1+ � values as determined by Rrs are likely to be ap-propriate, although assumptions are made regarding severalcomponents affecting the Rrs (e.g. the VSF for the triptonparticles estimated using a Mie model).

Q factors and Chl a-specific volume coefficients deter-mined using the two-layered sphere with the mean chro-matoplasm 1+ � value of 1.12 are shown in comparison tothose measured by Zhou et al. (2012) in Table 4. The val-ues for Qb(510) and Qbb(510) compare well, but measuredb∗φ values are substantially higher than those from the two-layered sphere. It appears that the two-layered model has alower total scattering due to the slightly lower value of Qb.The backscattering probability is also elevated for the two-layered model, although b∗

bφ(510) is very close.In comparison with eukaryotic speciesQbb is in the upper

range of values measured on cultures: from 0.0018 to 0.064at 510 nm (Zhou et al., 2012; Vaillancourt, 2004) and from0.006 to 0.061 at 442 nm (Whitmire et al., 2010). Surpris-ingly, the largest ofQbb values are from large dinoflagellatescontaining high intracellular carbon concentrations and un-usual chromosome morphology and internal structures. Thisinternal structure is used as an explanation for the higher-than-expectedQbb. Using similar reasoning, intracellular gasvacuoles in M. aeruginosa are responsible for the high Qbbvalues.

Biogeosciences, 10, 8139–8157, 2013 www.biogeosciences.net/10/8139/2013/

Page 18: Stewart Bernard (CSIR)

400 450 500 550 600 650 700 750 800 850 9000

0.5

1

1.5

2

2.5

3

wavelength (nm)

Abso

rptio

n (m

1 )

0.3 1 3 10 30100

400 450 500 550 600 650 700 750 800 850 9000

0.01

0.02

0.03

0.04

0.05

0.06

0.07

wavelength (nm)

Back

scat

terin

g (m

1 )

0.3 1 3 10 30100

Phytoplankton accessory pigments and the hyperspectral signal with increasing biomass: a brief experiment…

How does the accessory pigment related component of the reflectance signal change with increasing biomass – and increasing algal contribution to the IOP budget? A small experiment – modelling changing reflectance based on the Equivalent Algal Population/Hydrolight model, comparing dinoflagellate- and cryptophyte (Mesodinium rubrum) based IOPs. Average cell size and all other IOPs are the same – the only difference comes from the spectral refractive indices i.e. the different pigment content of the assemblage, and obviously the increasing biomass. Case 1 waters only, i.e. the best case scenario…. Total absorption and backscattering on the left, thick lines represent Mesodinium i.e. phycoerythrin containing cells, spectrally variable phase functions based on the backscattering probability used… Robertson, unpublished

Increasing  Chl  

Page 19: Stewart Bernard (CSIR)

400 450 500 550 600 650 700 750 800 850 9000

0.002

0.004

0.006

0.008

0.01

0.012

0.014

wavelength (nm)

Rrs

(sr

1 )

0.3 1 3 10 30100

Quantitatively useful pigment related differences – in this fairly extreme case – are apparent from ± 2 mg m-3 Chl a in the blue. As the biomass increase the higher Δ signals shift to longer wavelengths…. Robertson, unpublished

Increasing  Chl  

ΔRrs |R(Dino) – R(Meso)|

High Δ around 470 nm absorption peak (Chl c/

carotenoid) from ± 2 to 20 mg m-3 Chl a

High Δ around 520 nm absorption peak

(Phycoerythrin) from ± 4 mg m-3 Chl a

High Δ around shifting 580 – 590 nm reflectance

peak (Phycoerythrin) from ± 10 mg m-3 Chl a

Dino and Meso (thick lines) dominated reflectance with 0.3 -> 100 mg m-3 Chl a

Phytoplankton accessory pigments and the hyperspectral signal with increasing biomass: a brief experiment…

Page 20: Stewart Bernard (CSIR)

Phytoplankton Biodiversity in the Coastal Zone using Hyperspectral Sensing

1) What do you think are the driving science questions in your sub-discipline that will guide your community in the coming decade? Better understanding of the drivers and effects of variable primary production across oceanic and aquatic systems, and the importance of resolving phytoplankton community structure, preferably at the submeso- and event scale… 2) How will hyperspectral data help to address those questions? Offering more information on phytoplankton community structure, if we can better understand the causali 3) How does ‘scale’ (e.g., spectral, spatial, and/or temporal) affect your ability to address these science questions? What is the smallest measurement ‘scale’ needed to address your science? The most challenging scale is the the temporal/spatial, probably the daily or less revisit /300m spatial resolution boundary i.e. constellation/geostationary approaches 4) What are the common challenges across sub-disciplines in working with hyperspectral data? Engineering: quality of radiometry & spatial/temporal aspects. Science : better understanding of signal variability and constraints, robust error handling needed. 5) How do we coordinate and integrate common algorithm development efforts? Community platforms for collaboration/comparison; shared measured/synthetic data sets; realise that the atmospheric correction algorithms currently more of a constraint than the in-water algorithms 6) Are there any observational or programmatic gaps across the planned hyperspectral missions? Yes, routine and well constrained phytoplankton community structure measurements 7) What other space-based measurements or modeled data, done in parallel to hyperspectral measurements, would you like to have to obtain more out of ocean color? Hydrodynamic/biogeochemical/particle models using the same bio-optical models to allow convergence at Lw level