igths of 665
Crss, was
1e amount of
3-105 ?C for
casured after
action Cspom
(ESS method
ter (CDOM,
ıred with a
1ge 350—750
er, measured
corrected for
Vant (1987),
> better than
sured with a
1ge 400—800
e (Tassan &
ntation with
ich separates
bsorption by
nd Sundman
ance spectra.
it with some
properties of
ic absorption
Metsamaa et
lena species
ds.
lled spectral
chlorophyll,
> waters. The
chlorophyll-
n; CDOM
10 m’; and
1.
re the water
downwelling
with in situ
lance sensor
1g irradiance
1odelled and
int.
the modelled
ce spectrum.
AM. All the
t the number
for the angle
y are for this
3. RESULTS AND DISCUSSION
Results of the physic based approach can be considered as
preliminary because the modelled spectral library is very small.
The number of different CDOM concentrations used in the
model was three and the number of mineral suspended matter
concentrations was four. This was too crude to describe the
whole range of concentrations we had in the studied lakes.
Therefore, the correlation between measured and estimated
concentrations was poor. The number of different chlorophyll-a
concentrations was six. The correlation between measured and
estimated concentrations of chlorophyll-a was very good as seen
in Figure 1, although one of the reasons behind the good
correlation is the Lake Harku where chlorophyll was above 200
mg/m’.
250 -
t
15200 - *
E y 2 0.9862x - 4.9113 ë
= R?= 0.9574
= 150 -
=
g
2
= 100
uo ,
9 doo
E s0- et.
ré mate +
0 Ses ! ; :
0 50 100 150 200 250
Measured chlorophyll-a mg/m?
Figure 1. Correlation between the chlorophyll concentration
measured from water samples and the chlorophyll concentration
estimated using a modelled spectral library and Spectral Angle
Mapper procedure.
We are currently in the process of creating a comprehensive
spectral library that contains thousands of spectra modelled for
different concentrations of optically active substances. Our
previous results (Kutser et al. 2001, where we used a simple
bio-optical model and slightly different procedure for
comparing measured and modelled spectra) showed that using
of variable concentration steps is a reasonable approach to
reduce the number of model simulations needed i.e.
concentrations used in the model are increased with small steps
when the concentrations are small and the step are increased
with increasing concentrations. Otherwise the number of
different combinations of the tree optically active constituents
becomes exhaustive.
There is also some need to improve parameterisation of the
Hydrolight model for our lakes. For example the model was not
able to replicate as strongly as needed the features characteristic
to cyanobacterial blooms — a peak at 650 nm and phycocyanin
absorption feature at 620 nm. One of the reasons may be that
the Anabaena circinalis specific optical properties do not mach
exactly these of the species present in the studied lakes.
However, more likely cause is too low and too flat scattering to
backscattering ratio used in Hydrolight as it is known that
cyanobacteria are very efficient backscatterers and their specific
backscattering coefficient spectra may have more sophisticated
spectral shape than just monotonous decrease with increasing
wavelength. Optical properties of cyanobacteria differ
significantly from other phytoplankton. It may mean that
creating a comprehensive spectral library that will work well
during the whole ice free season may require modelling with
two sets of optical properties of phytoplankton — cyanobacteria
and all others.
The “classical” statistical approach was used in parallel with the
physic based approach i.e. we were looking for band ratio type
or more sophisticated algorithms that are in correlation with any
of the water properties. Different algorithms were used for
retrieval of chlorophyll-a, CDOM, suspended matter. A few
results are presented below.
It has been shown (Kutser et al. 2005a,b, 20092) that lake
CDOM can be mapped with multispectral satellites using ratio
of green (B2) and red (B3) bands. We calculated average
reflectance values for 525-605 nm and 630-690 nm spectral
ranges from Ramses reflectance data in order to test suitability
of the B2/B3 ratio for CDOM retrieval in lakes under
investigation. Results from two measurements in Lake Harku
did not fit the general picture. These two measurements were
rather exceptional as chlorophyll-a was above 200 mg/m’ in
both cases. Figure 2 illustrates the correlation between B2/B3
ratio and lake CDOM when the two extreme samples were
removed from the analysis.
9
8 De Hy
ud X
E 6 AN
S Ne +
gs #
am SN
a es
ë 5 + >
j | 779691189 Tutt
R? = 0.7999 E
1
0 : ;
0 0.5 1 15 2 25 3
B2/83
Figure 2. Correlation between in situ measured absorption of
CDOM at 420 nm and B2/B3 ratio calculated from Ramses
reflectance spectra.
Glint-free reflectance spectra were used in the analysis as was
mentioned above. For the comparison we used also reflectance
spectra measured above the water surface. The correlation
between CDOM and the band ratio shown in Figure 2 dropped
significantly (R°=0.54) when the above water reflectance
spectra were used. This confirms the need to remove sun and
sky glint from reflectance spectra. On the other hand we have
developed a procedure to remove glint from field radiometry
data (Kutser et al. 2012). Thus, the field reflectance spectra can
be corrected to the same level than the data used by us in this
study. Comparing the CDOM retrieval algorithm obtained here
for field radiometers with the algorithm we obtained for
Advanced Land Imager (Kutser et al. 2005b) shows that there is
slight shift in the power function (i.e. the coefficients in the
retrieval algorithm are not the same). One of the reasons may be
the glint as it was not removed from the satellite data in the
earlier study.
The peak in reflectance spectra near 700 nm is often used as an
indicator of chlorophyll-a concentration in eutrophic waters
(Gitelson 1992, Kutser 1997, Kallio et al. 2001, Gower et al.
2005). In the present study we used a simple difference between
the reflectance in the read peak, Max(red), and minima in the
