Professional Documents
Culture Documents
discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/234003415
DOWNLOADS VIEWS
54 104
6 AUTHORS, INCLUDING:
boxes and arrows], (3) by defining OACs in the form of optical properties in the initial state, T0,
these can be inputted into the ecosystem model based on those predictions which are
performed at T1 [see Fig. 1, ‘red’ oval, box and arrows], and (4) this OAC spectral prediction
(IOPs) is then later assessed [Forward RTE, HYDROLIGHT 4.2, see Fig. 1, ‘green’ boxes and
arrows] and the compared with satellite and in situ optical measurements.
The main difference in this approach is that all OAC concentration spectra are fed in as an initial
state to the ecosystem model, instead of only using a single chlorophyll a concentration as the
model startup. Instead of carbon or nitrogen as the ecosystem model currency, OPERA
proposes to input IOP spectra (absorption and backscattering). This has not been tried before
and is the reason for indicating in the title ‘A Revolutionary Approach’. From the near surface
fields, in conjunction with ocean model estimates of mixed layer depth, the vertical distribution
of OAC spectra will be estimated for each pixel or group of pixels viewed from space (Figs. 2 and
3). The model will also provide a more realistic representation of the light field, given by
Inherent Optical Properties (IOPs), illumination conditions, sea surface (focusing and
defocusing), and bottom reflectance, based on which light dependent biogeochemical processes
within the model will be forecasted.
Inverse Problem of RTE
Gordon and Boynton (1998) showed that the 3‐D field of Ed and Lu can be inverted to obtain
total absorption and backscattering spectra (Fig. 4a). In a review paper by Gordon in 2005,
several inverse problem modeling approaches were discussed and what can be retrieved as far
as IOPs from such an effort/RTE. STO CMRE currently has a copy of the
Figure 2. Graphical representation (model output, R. Arnone, NRL‐Stennis) of remotely sensed
diffuse attenuation coefficient with an OPERA model assimilation surface grid superimposed.
Depth
Figure 3. Representation of pixel assimilation of IOP spectra from surface measurements and
then extrapolation with depth.
Gordon and Boynton (1998) code and we are working on the implementation of the code on a
PC (originally run on a Mac) and optimizing it for speed and performance as this inverse needs
to be run many times to generate surface and vertical fields of absorption and backscattering
spectra. The results from a test run and comparison with in situ absorption spectra are shown
in Fig. 4b.
(a) (b)
Figure 4. Results from solving the inverse problem of the RTE and estimating IOPs from
hyperspectral in situ measurements of Ed and Lu [(a) Gordon and Boynton 1998 and (b)
preliminary results from STO CMRE efforts.
Denconvolution
Seawater subcomponents of phytoplankton, organic detrital material, inorganic minerals, and
dissolved material have been identified because each has a unique optical signature (thus it
should be theoretically possible to quantify each subcomponent using optical measurements)
as well as unique biogeochemical cycling characteristics (controlling processes and rates). Since
spectral backscattering and spectral absorption are the optical parameters we can expect to
readily retrieve from remote sensing reflectance (Gordon 2002; Gordon et al. 2009), we will
focus on deconvolving these properties.
Basis Vector Approach ‐ If the spectral shape of either backscattering or absorption was unique
and relatively consistent for each seawater subcomponent, then a solution could be readily
achieved with a basis vector analysis, finding the optimal combination of these spectral shapes
to limit the variance in the measured total spectrum. This approach has been attempted before
for a more limited set of subcomponents (Roesler et al. 1989; Roesler et al. 1995; Carder et al.
1999; Ciotti et al. 2002; Ciotti and Bricaud 2006). The problem is that many spectral shapes are
too similar. For example, since almost all the particulate subcomponents will have a broad
polydisperse size distribution, this will result in similar spectral shapes for backscattering,
monotonically decreasing with increasing wavelength. If there happens to be many more small
particles than large particles in a particular subpopulation – for example the phytoplankton
subfraction during a cyanobacteria bloom – then the spectral slope may be steeper. However,
thermodynamic equilibrium forces in nature tend to work toward distributing particle volume
equally in each size class (see discussion in Sullivan et al. 2005). These natural size distributions
result in spectral shapes modeled well with a simple power law (Boss et al. 2001; Twardowski et
al. 2001), with slopes typically falling between 0 and 1. While hyperspectral backscattering
spectra may be expected to provide unique structure around, for example, pigment absorption
peaks due to anomalous dispersion (Stramski et al. 2001), multi‐spectral remote sensing data
likely cannot provide the spectral resolution necessary for this type of analysis.
Considering absorption, one subcomponent can be expected to provide unique structure due to
pigmentation: phytoplankton. Minerals, detritus, and dissolved material all have the same
basic spectral shape however, monotonically decreasing with increasing wavelength (Roesler et
al. 1989; Tassan and Ferrari 2003; Twardowski et al. 2004). We thus cannot reasonably expect
to separate these subcomponents using absorption.
Our proposal is to combine the spectral basis vector analysis for both absorption and
backscattering. This approach should work because even though the spectral shapes for
absorption and backscattering are similar between, for example, minerals and detritus, the
relative magnitudes are not. Minerals have relatively low absorption and high backscattering.
Organic detrital material has relatively low absorption and low backscattering. Phytoplankton
have relatively high, spectrally unique absorption and low backscattering. Dissolved material
absorbs but does not scatter at all. These relationships may allow deconvolution of remote
sensing reflectance into absorption (a) and backscattering (bb) of each subcomponent. Writing
out the relationships,
at ( ) aˆ m ( ) m aˆ d ( ) d aˆ ( ) aˆ g ( ) g a w ( ) , (1)
bbt ( ) bˆbm ( ) m bˆbd ( ) d bˆb ( ) 0 bbw ( ) (2)
"aggregate" bas is v ec tor
2
1 aˆ g ( )
aˆ m ( )
0
bˆb ( ) X aˆ ( )
-1 m
bˆ ( ) X aˆ m ( )
bm
-2
400 450 500 550 600 650 700
wavelength (nm)
Figure 5. Aggregate basis vectors from Eq. 3. The two phytoplankton basis vectors have been
‐3
combined here. [chl] was assumed to be 1 mg m
‐1 ‐1 ‐1
. Exponential slopes for, and were 0.002
nm , 0.011 nm , and 0.014 nm , respectively. A power law slope of 1 was used for all
subcomponents. The dashed black curve is the aggregate CDOM basis vector using a spectral
‐1
slope of 0.011 nm for, same as for. Note that even with similar (and in one case identical)
exponential absorption spectra shapes for minerals, detritus and CDOM, and identical bb basis
vectors for all subcomponents, the aggregate basis vectors are entirely unique.
bb/a Approach ‐ Interpretation of ocean colour remote sensing data in Case 2 waters is
particularly challenging due to the wide dynamic range of constituent concentrations, variability
in relative concentrations of constituents and variability of specific inherent optical properties
(SIOPs) for any given constituent. Spectral deconvolution offers a physics‐based approach to
extracting constituent information from remote sensing signals and also presents an
opportunity to analyse sensitivity to measurement uncertainties and variability in SIOPs.
Results from an initial sensitivity study for a new spectral deconvolution approach are presented
here and offer a new insight into the impact of measurement uncertainties in IOP
measurements on both forward modelling and spectral deconvolution algorithm performance.
Let’s assume we have 4 non‐water constituents: phytoplankton (), minerals (m), detritus (d)
and coloured dissolved organic material (g) and
a a* . am* .m ad* .d a *g .g aw
bb bb* . bb m .m bb d .d 0 bb w
* *
also assume we have [bb/]() e.g. from Rrs
bb* . bb m .m bb d .d 0 bb w
* *
bb
a .
a* . am* .m ad* .d a *g .g aw
We can rearrange to get
b b b
0 a* . b bb* . am* . b bb m .m ad* . b bb d .d
* *
and then establish a general model for n constituents
n
b b
0 ai* b bbi* .xi aw b bbw ,
i 1 a a
where , m, d, and g found by minimizing least squares for N wavelengths where N ≥ n. If we
assume an 5% uncertainty on bb and independently and then run 1,000 bootstrap iterations
for each combination of Chl, minerals, CDOM using a range of these values typical of Case I
Waters (max Chl = 1mg m‐3, max CDOM = 0.02 m‐1 and max MSS = 01. g m‐3), like the Ligurian
Sea, the following spectral deconvolution performances are found (Figs 6a, b and c)
(a) (b)
(c)
Figure 6. Solution spread errors for (a) Chl, (b) MSS & (c) CDOM expressed as half range
percentage error for uniformly distributed percentage input errors (bb ± 5%, a ± 5%). The
nature and magnitude of input errors both strongly influence algorithm performance.
Ecosystem‐Physical Model – Currently, the Harvard Ocean Prediction System is being used that
is an interdisciplinary model system with (1) primitive equations, biogeochemical model (P, Z,
NO3, NH4, Chl and Det) with OI assimilation. The compatible physical/biological assimilation is
Physical data assimilation only – adjustment of the physical fields leads to misalignment
between physical and biological fronts, causing spurious cross‐frontal fluxes and
consequently spurious biological responses (e.g. enhanced productivity).
Biological data assimilation only – little or no feedback to the physics. Physical and
biological fronts become misaligned, causing spurious cross‐frontal fluxes and
consequently spurious biological responses (e.g. enhanced productivity).
Six‐step method:
o initial estimation of synoptic physical features
o melding physical data into these fields to obtain the best real‐time estimates
o physical dynamical adjustment to generate vertical velocities
o initial estimation of mesoscale biological fields based on Physical‐biological
correlations
o melding biological data into these fields, and
o biological dynamical adjustment with frozen physical fields to balance the
biological fields with each other, the model parameters, and the 3‐D physical
transports.
The generation of these fields is done in “adjustment space”, outside of the simulation of
interest (“simulation space”).
The modeling geometry has 125 x 115 grid points, 0.016o (1.5 km) resolution, areal extent is
200 x 200 km and there are 30 levels. Model initialization is with Seadatanet climatology
(Version 1.3.0, 0.125o horizontal resolution, 33 vertical levels, http://gher‐
diva.phys.ulg.ac.be/web‐vis/clim.html) for temperature and salinity, NO3, Chlorophyll and NH4.
Model adjustment takes 3‐4 days with the chlorophyll and surface velocity fields shown in
Figure 7.
CONCLUSIONS
This is a three year project (Ocean Strategic Services Beyond 2015) funded under the EU FP7
Programme starting on 1 Nov 2011. To successfully complete the OPERA model, the various
components on this modeling effort have been divided into individual groups. Below is a listing
of those groups and the tasks and research that will be required. As this is a ‘Revolutionary
Approach’, there are many issues and questions that will still need to be addressed. As to the
specifics of each step and approach, the OPERA science team may not be able to answer them
at this time. This is a very exciting and innovative program (OSS2015).
C. Trees (NATO STO CMRE) – Vertical extrapolation of the diffuse attenuation
coefficient, optical proxies for nutrients and grazing for the ecosystem‐physical
model and overall management of the components for the OPERA model.
V Sanjuan‐Calzado (NATO STO CMRE) – Inverse problem for the RTE to estimate IOPs
from AOPs and forward RTE solution for converting predicted IOPs from the
OPERA to AOPs and comparison with remote sensing data.
M. Twardowski (WET Labs, Inc) – Basis vector deconvolution for OACs.
D. McKee (Strathclyde University) – bb/a deconvolution for OACs.
S. Sukru, M. Gunduz, F. Bengil (DEM‐IMST) – Ecosystem‐physical modeling.
Day 5 Day 10
Day 15 Day 20
Figure 7. Spin‐up of the ecosystem‐physical model (HOPS) showing the chlorophyll and
surface velocity fields at 5 day intervals up to a 20 day period.
References
Boss, E., and W.S. Pegau. 2001. Relationship of light scattered at an angle in the backward
direction to the backscattering coefficient. Applied Optics 40:5503‐5507.
Carder, K.L., F.R. Chen, J.P. Cannizzaro, J.W. Campbell, and B.G. Mitchell. 2004. Performance of
the MODIS semi‐analytical ocean color algorithm for chlorophyll‐a. Advances in Space
Research 33:1152‐1159.
Ciotti, A.M. and A. Bricaud (2006). Retrievals of a size parameter for phytoplankton and spectral
light absorption by Colored Detrital Matter from water‐leaving radiances at SeaWiFS
channels in a continental shelf region off Brazil. Limnology and Oceanography Methods, 4,
237–253.
Ciotti, A.M., Cullen, J.J., and Lewis, M.R. 1999. A semi‐analytical model of the influence of
phytoplankton community structure on the relationship between light attenuation and
ocean color. Journal of Geophysical Research 104:1559‐1578.
Gordon, H.R. 2002. Inverse methods in hydrologic optics. Oceanologia, 44(1): 9‐58.
Gordon, H.R., and G.C. Boynton. 1998. A radiance‐irradiance inversion algorithm for estimating
the absorption and backscattering coefficients of natural waters: Stratified water
bodies. Applied Optics. 37, 886‐896.
Roesler, C.S., M.J. Perry, and K.L. Carder. 1989. Modeling in situ phytoplankton absorption from
total absorption spectra. Limnology and Oceanography 34:1512‐1525.
Roesler, C.S., and M.J. Perry. 1995. In situ phytoplankton absorption, fluorescence emission,
and particulate backscattering spectra determined from reflectance. J. Geophys. Res.
100:13,279‐13,294.
Stramski, D., Bricaud, A. and Morel, A. 2001. Modeling the inherent optical properties of the
ocean based on detailed composition of the plankton community. Appl. Opt., 40, 2929–
2945.
Sullivan, J.M., M.S. Twardowski, Donaghay, P.L., and Freeman, S.A. 2005. Using scattering
characteristics to discriminate particle types in US coastal waters, Applied Optics
44:1667‐1680.Tassan, S, and G.M. Ferrari. 2003. Variability of light absorption by aquatic
particles in the near‐infrared spectral region. Applied Optics 42:4802‐4810.
Twardowski, M.S., E. Boss, J.B. Macdonald, W.S. Pegau, A.H. Barnard, and J.R.V. Zaneveld. 2001.
A model for estimating bulk refractive index from the optical backscattering ratio and
the implications for understanding particle composition in Case I and Case II waters. J.
Geophys. Res., 106(C7):14,129‐14,142.
Twardowski, M.S., E. Boss, J.M. Sullivan, and P.L. Donaghay. 2004. Modeling spectral absorption
by chromophoric dissolved organic matter (CDOM). Mar. Chem., 89: 69‐88,
doi:10.1016/j.marchem.2004.02.008.
Zaneveld, J.R.V., A.H. Barnard, and E. Boss. 2005. Theoretical derivation of the depth average
of remotely sensed optical parameters. Optics Express 13:9,052–9,061.