Upload
anon157324138
View
215
Download
0
Embed Size (px)
Citation preview
7/28/2019 Water Qual Hyperion 07
1/13
Assessment of water quality in Lake Garda (Italy) using Hyperion
Claudia Giardino a,, Vittorio E. Brando b, Arnold G. Dekkerb,Niklas Strmbeckc, Gabriele Candiani a
a Optical Remote Sensing Group, CNRIREA, Milano, Italyb Environmental Remote Sensing Group, CSIRO-Land and Water, Canberra, Australia
c Department of Limnology, EBC, Uppsala University, Uppsala, Sweden
Received 22 August 2006; received in revised form 20 December 2006; accepted 23 December 2006
Abstract
For testing the integration of the remote sensing related technologies into the water quality monitoring programs of Lake Garda (the largest
Italian lake), the spatial and spectral resolutions of Hyperion and the capability of physics-based approaches were considered highly suitable.
Hyperion data were acquired on 22nd July 2003 and water quality was assessed (i) defining a bio-optical model, (ii) converting the Hyperion at-
sensor radiances into subsurface irradiance reflectances, and (iii) adopting a bio-optical model inversion technique. The bio-optical model was
parameterised using specific inherent optical properties of the lake and light field variables derived from a radiative transfer numerical model. A
MODTRAN-based atmospheric correction code, complemented with an air/water interface correction was used to convert Hyperion at-sensor
radiances into subsurface irradiance reflectance values. These reflectance values were comparable to in situ reflectance spectra measured during
the Hyperion overpass, except at longer wavelengths (beyond 700 nm), where reflectance values were contaminated by severe atmospheric
adjacency effects. Chlorophyll-a and tripton concentrations were retrieved by inverting two Hyperion bands selected using a sensitivity analysis
applied to the bio-optical model. The sensitivity analysis indicated that the assessment of coloured dissolved organic matter was not achievable in
this study due to the limited coloured dissolved organic matter concentration range of the lake, resulting in reflectance differences below the
environmental measurement noise of Hyperion. The chlorophyll-a and tripton image-products were compared to in situ data collected during
the Hyperion overpass, both by traditional sampling techniques (8 points) and by continuous flow-through systems (32 km). For chlorophyll-a the
correlation coefficient between in situ point stations and Hyperion-inferred concentrations was 0.77 (data range from 1.30 to 2.16 mg m 3). The
Hyperion-derived chlorophyll-a concentrations also match most of the flow-through transect data. For tripton, the validation was constrained by
variable re-suspension phenomena. The correlation coefficient between in situ point stations and Hyperion-derived concentrations increased from
0.48 to 0.75 (data range from 0.95 to 2.13 g m3) if the sampling data from the re-suspension zone was avoided. The comparison of Hyperion-
derived tripton concentrations and flow-through transect data exhibited a similar mismatch. The results of this research suggest further studies to
address compatibilities of validation methods for water body features with a high rate of change, and to reduce the contamination by atmospheric
adjacency effects on Hyperion data at longer wavelengths in Alpine environment. The transferability of the presented method to other sensors and
the ability to assess water quality independent from in situ water quality data, suggest that management relevant applications for Lake Garda (and
other subalpine lakes) could be supported by remote sensing.
2007 Elsevier Inc. All rights reserved.
Keywords: Hyperspectral satellite data; Lake waters; Bio-optical modelling; In situ data
1. Introduction
Lake water is an essential renewable resource for mankind
and the environment and it is important for civil (drinking water
supply, irrigation, transportation), industrial (processing and
cooling, energy production, fishery) and recreational purposes.
Sustainable use of water resources requires the coupling of
surface waters assessment monitoring programs and decision
making and management tools. The Water Framework Di-
rective (WFD) of the European Commission (Directive 2000/
60/EC, 2000) is the major reference in Europe to guide efforts
for attaining a sustainable aquatic environment in the years to
Remote Sensing of Environment 109 (2007) 183195
www.elsevier.com/locate/rse
Corresponding author. Tel.:+39 0223699298; fax: +39 0223699300.
E-mail address: [email protected] (C. Giardino).
0034-4257/$ - see front matter 2007 Elsevier Inc. All rights reserved.doi:10.1016/j.rse.2006.12.017
mailto:[email protected]://dx.doi.org/10.1016/j.rse.2006.12.017http://dx.doi.org/10.1016/j.rse.2006.12.017mailto:[email protected]7/28/2019 Water Qual Hyperion 07
2/13
come. The WFD includes guidelines which define the cat-
egories of quality and the required components and parameters.
As some of these parameters can be determined by Remote
Sensing (RS) with a reasonable accuracy, RS-related technol-
ogies may be integrated in the monitoring programs defined by
the WFD, provided they can be demonstrated to independently
assess Water Quality Parameters (WQPs).Since the 1980s satellite RS represents an opportunity for
synoptic and multitemporal viewing of water quality. To estimate
WQPs from satellite data three different approaches can be used
(Cracknell et al., 2001; Dekker et al., 1995). The (1) empirical
approachis based on the development of bi-variate or multivariate
regressions between RS data and measured WQPs. Digital
numbers or radiance values at the sensor, as well as their band
combinations, are correlated with in situ measurements of WQPs,
usually collected in coincidence of the sensor overpass. A
summary of empirical approaches for lakes can be found in
Lindell et al. (1999). The (2) semi-empirical approach may be
used when spectral characteristics of the parameters of interestare known. This knowledge is included in the statistical analysis
by focusing on well-chosen spectral areas and appropriate
wavebands used as correlates. An example of a semi-empirical
approach with different sensors is reported by Hrm et al. (2001)
over Finnish lakes. In the (3) analytical approach, WQPs are
related to the bulk Inherent Optical Properties (IOPs) via the
Specific Inherent Optical Properties (SIOPs). The IOPs of the
water column are then related to the Apparent Optical Properties
(AOPs) and hence to the Top of Atmosphere (TOA) radiance,
such as described by the radiative transfer theory (Mobley, 1994;
Vermote et al.,1997). The analytical method involves inverting all
above relations (WQPsIOPsAOPsTOA radiances) to
determine the WQPs from RS data. An example of such approach,using Landsat over lakes, can be found in Dekker et al. (2001) for
the total suspended matter retrieval.
Quantitatively, the relationships developed to assess water
quality in lakes within empirical and semi-empirical approaches
are often scene dependent and only apply to the data from which
they are derived. Well-calibrated and validated physics-based
approaches are instead applicable to every scene acquired over
the selected lake (presuming constant SIOPs), giving the
opportunity to assess water quality independently from ground
measurements of WQPs. The monitoring of spatially heteroge-
neous parameters, as re-suspension phenomena due to local
variability in wind and circulation, or algal blooms at the surfacealso necessitates these (in situ independent) methods. Dekker
et al. (2002) investigated the capabilities of Landsat-TM and
SPOT data for retrospective analysis in Dutch lakes. Both
sensors were capable of describing larger concentration
gradients characterised by temporal changes that were not
represented by point in situ data. Kutser (2004) used Hyperion
data to map accumulation of aggregations of cyanobacteria in
the Gulf of Finland, an assessment unachievable by traditional
in situ sampling due to spatial and temporal issues. Kutser
(2004) also showed that flow-through systems were only
suitable to map chlorophyll from a fixed depth, and therefore
inappropriate for assessing cyanobacteria blooms closer to or at
the water surface.
This study is part of ongoing research efforts aimed at de-
veloping RS strategies towards the implementation of the WFD,
ensuring systematic monitoring of water quality in Lake Garda,
the largest Italian lake. Empirical and semi-empirical ap-
proaches were previously investigated to retrieve chlorophyll
concentrations in the lake (Brivio et al., 2001; Candiani et al.,
2003; Giardino et al., 2005) but their results were scenedependent. The aim of this study is to provide a RS-based
measurement tool, transferable to different RS-instruments. It
would be useful for water management authorities of Lake
Garda for coarse scale regular monitoring (with high revisiting
time spaceborne sensors), for intermediate/fine scale studies
(with high spatial resolution satellite and airborne sensors) and
for retrospective analysis with time-series imagery (as in
Dekker et al., 2005). As a test, hyperspectral Hyperion data
(with a 30 m pixel size and a choice of more than 200 spectral
channels), analytical modelling, and in situ measurements
coincident with the satellite overpass for a validation or com-
parison of the image-derived products, were considered ap-propriate. The approach used in this study builds on the method
developed for Hyperion imagery of a sub-basin of a subtropical
bay in Australia (Brando & Dekker, 2003). Based on a bio-
optical model sensitivity analysis, Hyperion bands were
selected and concentrations of chlorophyll-a and tripton (the
non-algal particles of the suspended particulate matter) were
retrieved. Point in situ data for an initial validation of the
products, followed by a comparison of concentrations retrieved
from Hyperion using high spatial resolution flow-through
estimates of chlorophyll-a and tripton were used.
2. Materials and methods
2.1. Study area and fieldwork activities
Approximately 500,000 lakes over 1 ha surface area exist in
Europe. Most of the largest European lakes are located in the
Nordic countries and in the Alpine regions (EEA, 1999). The
most important Italian lake district is located in the subalpine
region and represents more than 80% of the total Italian
lacustrine volume (Premazzi et al., 2003). Lake Garda, located
65 m a.s.l. around 4540 N and 1041 E at the eastern border
of the subalpine region, is the largest Italian lake. It has a surface
area of 368 km2, a volume of 49 million m3 and a maximum and
a mean depth of 350 m and 133 m, respectively. The averagevalue of the Secchi disk depth is 4.5 m in summer and 16 m in
winter. Chlorophyll-a (CHL-a) and suspended particulate matter
(SPM) concentrations range from 0.5 to 12 mg m3 and from
0.1 to 5.5 g m3, respectively. The coloured dissolved organic
matter concentration (absorption coefficient at 440 nm, aCDOM(440)) ranges from 0.017 to 0.36 m1. Average concentrations
of CHL-a, SPM and aCDOM(440) are around 2.7 mg m3, 2.5 g
m3 and 0.09 m1, respectively (Premazzi et al., 2003; Zilioli,
2002). According to the OECD guidelines (Vollenweider &
Kerekes, 1982), Lake Garda can be classified as an oligo-
mesotrophic basin.
In collaboration with the local agencies in charge of
limnological monitoring of Lake Garda, intensive fieldwork
184 C. Giardino et al. / Remote Sensing of Environment 109 (2007) 183195
7/28/2019 Water Qual Hyperion 07
3/13
activities were run in the whole basin, within several national
and international projects (Lindell et al., 1999; Zilioli, 2002,
2004). More than 30 days over 5 years of in situ measurements
were performed to achieve a comprehensive dataset of
concentrations of WQPs, of IOPs and AOPs, leading to the
parameterisation of a three-component bio-optical model
according to Strmbeck et al. (2003). In this study in situ datacollected on 22nd July 2003 to validate the atmospheric
correction of Hyperion data, to validate the image-derived
WQPs products, as well as to improve the bio-optical
parameterisation are presented.
Water samples of the first integrated meter of water column in
9 point stations (Fig. 1) were analysed for CHL-a and tripton
(TR) concentrations. The in situ samples were collected
within a 3-hour interval around the image acquisition. Water
was filtered through Watman GF/F glass fiber filters and the
material retained was analysed for CHL-a concentrations
according to the analytical method ISO 10260-E (1992).Phytoplankton was composed by 70% of Chlorophyta
species and by 30% of almost equal parts of Cryptophyceae,
Diatomeae and Cyanophyta species. Because the suspended
particulate matter can be divided into phytoplankton and into
the non-algal component (i.e., tripton), TR was indirectly
estimated from SPM concentrations. They were measured
on pre-combusted and pre-weighted Watman GF/F filters,
dried at 95 C overnight. The biomass of phytoplankton
was considered correlated to CHL-a and the formula
TR=SPM0.07CHL-a (with TR and SPM in g m3 andCHL-a in mg m3) was used to separate TR from SPM. Gons
et al. (1992) observed that for fresh water algae the part of
SPM determined by the biomass of phytoplankton can vary
between 0.02 and 0.1. The average value of 0.07 had been
successfully adopted in Dutch lakes (Hoogenboom et al.,
1998), Finnish lakes (Kutser et al., 2001) as well as in coastal
waters (Brando & Dekker, 2003). Thus, itwas supposed to be
valid for Lake Garda too.
Five water samples collected during the day were used to
measure the absorption spectra of phytoplankton aph()
and tripton aTR(), according to the method showed in
Strmbeck and Pierson (2001). The absorption spectra ofparticles ap() retained onto the GF/F filters, were measured
using a laboratory spectrophotometer and the filter-pad
technique (Tassan & Ferrari, 1995). The filters were then
treated with cold Methanol to extract pigments and the
absorption spectra of tripton aTR() of these bleached filters
were measured. The absorption spectrum of phytoplankton
aph() was derived by subtracting aTR() from ap() spectra.
A 32-km-long transect (Fig. 1) of fluorescence and turbidity
data were collected using a flow-through system. The system
is composed of a hydraulic device, (essentially an intake
pipe) continuously pumping water from 0.5 m depth into a
Turner Design Scufa-II fluorometer/turbidimeter, and a GPS,
both logged by a Campbell data-logger. Logged values offluorescence (in mV) and turbidity, in Nephelometric
Turbidity Units (NTU), were corrected for delays caused
by the flow-through system. Flow-through data were con-
verted into chlorophyll-a and tripton using the concentrations
derived from laboratory analysis on water samples. Eight
laboratory-concentrations were regressed against the average
of logged values over the nearest 100 m to the GPS location
where the water samples were collected. By means of linear
regression analysis, the measured in vivo fluorescence was
transformed into chlorophyll-a concentrations (R2=0.55),
and turbidity into tripton concentrations (R2=0.68) (Fig. 2).
It was hence assumed that flow-through data were able todescribe both CHL-a and TR concentrations along the 32-
km-long transect although turbidity, because it includes
phytoplankton scattering, is more closely a measure of SPM.
Spectroradiometric measurements of water radiance using
the PR-650 spectroradiometer were performed to calculate
the subsurface irradiance reflectance R(0, ) in three
pelagic stations (4, 6 and 7 in Fig. 1). R(0, ) values were
computed from remote sensing reflectances Rrs(0+, ),
measured above-water according to the SeaWifs protocol
(Fargion & Mueller, 2000). Effects of the lake surface
roughness on above-water Rrs(0+, ) determinations were
corrected by a sky-radiance reflectance factor and by an
offset term, that does not impose a constrained normalisation
Fig. 1. Study area and location of fieldwork activities performed within a 3-hour
interval around the image acquisition. 1 to 9 are the stations where water was
sampled for laboratory analysis for chlorophyll-a and tripton concentrations
(stations 4, 6 and 7 have also PR-650 radiometric measurements). The flow-
through system was cruised throughout all the stations, for a length of about32 km. The 7.5-km-wide portion of lake imaged by Hyperion is outlined.
185C. Giardino et al. / Remote Sensing of Environment 109 (2007) 183195
7/28/2019 Water Qual Hyperion 07
4/13
at 750 nm (Toole et al., 2000). Literature data (Toole et al.,
2000) for clear sky and high wind speeds were used (on 22nd
July, 2003 the average wind speed on the lake was 6 m s
1
).Assuming an air/water interface parameter of 0.533 (Lee
et al., 1994) and a Q-factor of 4.2 sr1 the roughness-cor-
rected Rrs(0+, ) were then transformed into R(0, )
values. As reported in Strmbeck et al. (2003) the Q-factor
was the average, between 400 and 750 nm, of a spectral Q-
factor computed using the HYDROLIGHT 4.2 model
(Mobley, 1994; Mobley & Sundman, 2001).
2.2. The bio-optical model
The bio-optical model used in this study was similar to
previously published three-components (i.e., chlorophyll-a,
tripton and coloured dissolved organic matter) Case-2 or lakewater models, e.g., Pierson and Strmbeck (2001). The sub-
surface irradiance reflectance R(0, ) was calculated as a
function of absorption and backscattering coefficients according
to Walker (1994):
R 0;k 1
1 Pld k Plu k
dbb k
a k bb k 1
where, a() is the spectral total absorption coefficient, bb() is
the spectral total backscattering coefficient, and d()/u() is
the ratio of the average cosine of the downwelling light to that of
the upwelling light (Mobley, 1994).The spectral total absorption coefficient a() was computed
as:
a k aw k CHL a aph k aCDOM 440 e
SCDOM k440
TR aTR 440 eSTR k440
2
where, aw() is the pure water absorption (Pope & Fry, 1997;
Smith & Baker, 1981), aph () is the chlorophyll-specific
phytoplankton absorption, SCDOM is the slope factor of the
absorption spectra of coloured dissolved organic matter,
aTR (440) is the absorption coefficient at 440 nm specific
for 1 g m3 of tripton, STR is the slope factor of the absorp-
tion spectra of coloured dissolved organic matter. In this study
SCDOM and STR were equal to 0.021 and 0.012, respectively.
The spectral total backscattering coefficient bb() wascomputed as:
bb k bbw k CHL a bbph
k
TR bbTR
550 k
550
ni
3
where, bbw() is the backscattering coefficient of pure water
(Morel, 1974; Dall'Olmo & Gitelson, 2006), bbph () is the
specific backscattering caused by phytoplankton, bbTR () is the
specific backscattering coefficient at 550 nm for 1 g m3 of
tripton, ni is an exponent describing the spectral dependency of
tripton backscattering (mainly due to its inorganic components).
The specific backscattering by phytoplankton was computedusing an expression based on Gordon et al. (1988), Morel
(1988), Ammenberg et al. (2002), and Roesler and Boss (2003):
bbph k bph
555 aph 555
k555
nph
kaph k
bbph
bph
4
where, bph (555) is the chlorophyll-a specific scattering, nph is
an exponent describing the spectral dependency of the phy-
toplankton beam attenuation, k is an empirical coefficient reg-
ulating the effect by phytoplankton absorption and bbph /bph is
the average spectral backscattering efficiency of phytoplankton.The parameterisation of the bio-optical model used in this
study is largely based on the data presented in Strmbeck et al.
(2003), which have been acquired in Lake Garda on 10th and
11th October 2002. The dataset contains discrete measurements
of WQPs, total absorption a and total scattering b coefficients
at 9 wavelengths obtained with a WET Labs ac-9, and total
backscattering bb coefficients at 6 wavelengths obtained by a
HOBILabs HydroScat-6. In particular, scattering b and back-
scattering bb data of the lake, were used to parameterise Eq. (4)
originally adopted for oceanic phytoplankton. Because the
average specific absorption coefficients of phytoplankton,
aph () and tripton, aTR () derived from data collected on 22nd
July 2003 were comparable to data collected on 10th and 11th
Fig. 2. Scatter plots of fluorescence vs. chlorophyll-a (using stations 1 to 8) and of turbidity vs. tripton (using stations 2 to 9) with the calibration lines (turbidity in
Station 1 and fluorescence in Station 9 were no available).
186 C. Giardino et al. / Remote Sensing of Environment 109 (2007) 183195
7/28/2019 Water Qual Hyperion 07
5/13
October 2002, they were averaged and integrated in the existing
dataset. Apparently, the natural modifications of algal and
tripton compositions occurred between the two periods had a
negligible effect on their absorption spectral properties.
Together with the d()/u() ratio (average value from
450 nm to 750 nm equal to 0.327), that was derived running
HYDROLIGHT 4.2 with inputs typical of Lake Garda (e.g.,
IOPs, averages values of wind speed and visibility ranges,
summertime Sun zenith angles at 111 h UTC) (Strmbeck
et al., 2003), these SIOPs uniquely characterise the parameter-isation of Eq. (1) for Lake Garda waters. Table 1 summarises the
bio-optical model parameters with the day of acquisition and the
data provider, Fig. 3 shows the SIOPs used in this study.
The performance of the above parameterisation was
evaluated using the 22nd July 2003 dataset. It consisted of the
PR-650-derived measurements of R(0, ), collected in the
stations 4, 6 and 7 (Fig. 1), in which concentrations of CHL-a
and TR were also known. These concentrations, together with
the long-term (i.e., 0.09 m1) average value of aCDOM(440) of
Lake Garda (as aCDOM(440) concentrations were not measured
in this campaign), were given as input to the bio-optical model
to simulate R(0
, ) spectra, assuming that SIOPs were thesame of October 2002. The optical closure between in situ
determinations of R(0, ) and the simulated values from
forward modelling was quantified with the Root Mean Square
Error (RMSE) and the relative RMSE (in %):
RMSE
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPNi1
P
Xi Xi
2N1
vuuut5
Relative RMSE RMSE
1
NPNi1
P
Xi
d 100 6
where, N is the number of bands, and Xi and Xi are the
subsurface reflectance values from in situ data and forward
modelling, respectively. The number of bands N was 28; 22 of
these in the visible (VIS) range, from 480 to 690 nm, and the
remaining in the near-infrared (NIR) range, from 700 to 750 nm.
The optical closure between in situ determinations of R(0,
) and the simulated values from forward modelling wasconsidered satisfactory in all stations (Fig. 4, Table 2). In
particular, the convergence was good in the VIS range (average
RMSE of the three stations 0.006, relative RMSE 12%) while
beyond 700 nm a larger divergence was observed (average
RMSE 0.012, relative RMSE 55%).
2.3. Hyperion data and pre-processing analyses
On 22nd June 2003, image data from an area of 7.5 by 42 km
was acquired by Hyperion with a near-nadir viewing. At the
time of the overpass Sun zenith and azimuth angles were 32
and 136, respectively. For this study 28 Hyperion spectralbands ranging from 480 nm to 750 nm were selected to be
relevant for WQPs estimation and reliable for the sensor
calibration (Green et al., 2003). Following the approach by
Brando and Dekker (2003), the image was convolved using a
Table 1
List of the bio-optical parameters and in situ water quality data presented in this
studywith information about the day of acquisition and data source (APPA is the
Environmental Protection Agency of Trento, ARPAV is the Environmental
Protection Agency of Veneto)
Parameter Day of acquisition Source
aw()
Smith and Baker (1981),Pope and Fry (1997)
bbw() Morel (1974), Dall'Olmo
and Gitelson (2006)
aph () 10th11th October 2002 HelsinkiUniversity and
Luode Consulting Oy
22nd July 2003 APPA
bbph () 10th11th October 2002 HelsinkiUniversity and
Luode Consulting Oy
aTR (440), STR 10th11th October 2002 HelsinkiUniversity and
Luode Consulting Oy
22nd July 2003 APPA
bbTR () 10th11th October 2002 HelsinkiUniversity and
Luode Consulting Oy
SCDOM 10th11th October 2002 APPA
u(), d() HYDROLIGHT 4.2CHL-a, TR 22nd July 2003 ARPAV
Fluorescence, turbidity 22nd July 2003 CNRIREA
Fig. 3. SIOPs of Lake Garda: spectra of absorption (upper graph) and
backscattering (lower graph) coefficients used in the bio-optical model. aw is the
absorption coefficient of pure water, aph is the chlorophyll-specific absorption
coefficient of phytoplankton, aTR is the specific absorption coefficient of tripton,
and aCDOM is the specific absorption coefficient of coloured dissolved organic
matter. bbw is the backscattering coefficient of pure water, bbph is the
chlorophyll-specific backscattering coefficient of phytoplankton, and bb
TR isthe specific backscattering coefficient of tripton.
187C. Giardino et al. / Remote Sensing of Environment 109 (2007) 183195
7/28/2019 Water Qual Hyperion 07
6/13
5 5 low pass filter to reduce the environmental noise-
equivalent reflectance differences NER(0, )E. According
to Brando and Dekker (2003) and Wettle et al. (2004), the value
of NER(0, )E is about 0.001 for Hyperion data acquired
over water. The filtered image was atmospherically corrected to
R(0, ) using the MODTRAN-based c-WOMBAT-c procedure
(Brando & Dekker, 2003). The procedure consists of: (i) a three
step atmospheric inversion from at-sensor-radiance to apparent
reflectance and (ii) a two step inversion of the airwater
interface from apparent reflectance to subsurface irradiancereflectance. c-WOMBAT-c was run with actual measurements
of visibility range (15 km) derived from sun-photometer ob-
servations performed synchronously to the sensor overpass
(Vermote et al., 1997). A Q-factor of 4.2 sr1, an air/water
interface parameter of 0.533, a nadir-viewing geometry and a
maritime extinction for aerosols were also given as inputs to the
atmospheric correction code.
The atmospherically corrected image was geo-located and
image-derived R(0, ) values were compared to in situ data
measured during the Hyperion overpass in the pelagic stations
4, 6 and 7 (Fig. 1). The optical closure (Fig. 4, Table 2) between
in situ and Hyperion spectra was on average good in the VISrange (from 480 to 690 nm, the average reflectance RMSE of
the three stations was 0.007 and the relative RMSE 14%) and
inferior in the NIR bands (beyond 700 nm, the average
reflectance RMSE was 0.011 and the relative RMSE 77%).
More in detail, in Station 7 the near-infrared wavelength re-
flectance values from image data were over-estimated compared
to in situ values (beyond 700 nm the RMSE was 0.015 and the
relative RMSE 159%). The most likely cause of such over-
estimation in the atmospherically corrected image data was the
contamination of Hyperion radiances by adjacency effects, due
to multiple reflections of radiation coming from the neighbour-
ing environment, specifically in the northern part of Lake
Garda. Here adjacency effects were probably caused by thevegetation growing over the steep sides, laterally delimiting the
northern narrow part of the lake, where Station 7 is located
(Fig. 1). The contribution of reflections from the vegetated
environment on the water surface increases the signal measured
by the sensor at longer wavelengths (Floricioiu & Rott, 2005)
and c-WOMBAT-c can fail in removing these quantities.
2.4. Band selection and model inversion
A direct inversion of the bio-optical model was applied to
the Hyperion image using a linear Matrix Inversion Method
(MIM), as in Brando and Dekker (2003). They ran MIM on a
Table 2
List of RMSE and relative RMSE (in %) measuring the optical closure of in situ determinations of R(0, ) vs. forward-modelled and Hyperion-derived R(0, )
values
In situ vs. forward-modelled In situ vs. Hyperion-derived
St. 4 St. 6 St. 7 AvMod St. 4 St. 6 St. 7 AvHyp AvAll
VIS 0.006 0.007 0.005 0.006 0.005 0.006 0.010 0.007 0.006
12% 13% 10% 12% 10% 10% 21% 14% 13%
NIR 0.015 0.020 0.002 0.012 0.009 0.009 0.015 0.011 0.011
72% 73% 22% 55% 42% 31% 159% 77% 66%
RMSE and relative RMSE for each station are reported together with their average value (in bold); AvMod is the average for stations 4, 6 and 7 of RMSEs computed
from in situ and forward-modelled R(0, ) values, AvHyp is the average for stations 4, 6 and 7 of RMSEs computed from in situ and Hyperion-derived R(0, )
values. AvAll is the average of AvMod and AvHyp RMSEs and measures how R(0
, ) values from Hyperion, forward bio-optical modelling and in situ optical dataconverge. RMSEs and relative RMSEs are separately computed for VIS (22 bands from 480 to 690 nm) and NIR (6 bands from 700 to 750 nm) ranges.
Fig. 4. In situ, forward-modelled and Hyperion-derived subsurface irradiance
reflectance R(0, ) spectra in stations 4, 6 and 7 ( Fig. 1): In situ spectra are
derived from above-water measurements of Rrs(0+, ); Model spectra are
calculated from forward bio-optical modelling using CHL-a (in mg m3) and TR
(in g m3) concentrations measured in situ (shown in each graph) and assuming
aCDOM(440)=0.09 m1 in all stations; Hyp spectra are computed from
Hyperion imagery using the c-WOMBAT-c atmospheric correction code. All the
data were collected on 22nd July 2003. Average reflectance RMSEs (see also
Table 2) are 0.006 (relative RMSE 13%) between 480 and 690 nm, and 0.011
(relative RMSE 66%) beyond 700 nm.
188 C. Giardino et al. / Remote Sensing of Environment 109 (2007) 183195
7/28/2019 Water Qual Hyperion 07
7/13
Hyperion image of a coastal site to retrieve chlorophyll, tripton
and aCDOM(440) by inverting three bands (490 nm, 670 nm and
700740 nm). The bands were chosen through some iterations
starting from wavelengths closest to spectral feature typical for
each of the WQPs and avoiding the shortest wavelengths (blue)
where Hyperion was noisy. In this study, the large number
of Hyperion bands was exploited by performing a sensitivityanalysis of the bio-optical model on aCDOM(440), CHL-a
and TR independently, based on the first derivative approach
(Hoogenboom et al., 1998).
The first derivative spectra relative to variations of aCDOM(440) were obtained from 10 forward runs of the bio-optical
model changing aCDOM(440) from 0.03 to 0.30 m1 and fixing
CHL-a and TR to 1 mg m3 and to 1 g m3, respectively. The
first derivative was rescaled by a 0.01 factor to appreciate the
sensitivity of the model in discriminating aCDOM(440) at con-
centration ranges of the lake of around 0.09 m1. The first
derivative spectra relative to variations of CHL-a were computed
by 10 forward runs of the bio-optical model incrementing CHL-aby 1 mg m3 withina range between 1 and 1 mg m3, and fixing
TR to 1 g m3 and aCDOM(440) to 0.09 m1. Similarly, the first
derivative spectra relative to variations of TR were computed by
10 forwardruns of the bio-opticalmodel incrementing TR by 1 g
m3 within a range between 1 and 10 g m3, and fixing CHL-a
to 1 mg m3 and aCDOM(440) to 0.09 m1.
Fig. 5 presents the first derivative spectra of R(0, ) vs.
each of the WQPs. The maximum variation of the first
derivative of R(0, ) for aCDOM(440) occurred at shortest
wavelengths (Fig. 5a), in a region where Hyperion data are too
noisy and ill calibrated (Green et al., 2003). Moving towards the
region where Hyperion provides calibrated data (i.e.,N
480 nm),the variation of the first derivative dR(0, ) / daCDOM(440)
falls within the Hyperion NER(0, )E of 0.001. This implies
that estimates of aCDOM(440) at the concentration range of the
lake was not achievable and therefore fixed to the long-term
average value for Lake Garda (i.e., 0.09 m1). This, a priori
determination of an unmeasurable signal due to small aCDOM(440) variations illustrates the usefulness of the first derivative
approach. Based on sensitivity analysis for CHL-a and TR, the
Hyperion bands at 490 nm and 550 nm were selected for the
inversion. The 490 nm band was chosen because of the location
of the maximum variation of the first derivative of R(0, ) for
CHL-a concentration (Fig. 5b). The 550 nm band was chosen
because it is the hinge point of the first derivative of R(0, )for CHL-a concentration (Fig. 5b), as well as being in the region
where the maximum variation of the first derivative of R(0, )
for TR concentration occurred (Fig. 5c). However, since Hy-
perion spectra show band to band spikes or dips (Cairns et al.,
2003), a selection based on single bands could match some
spikes (Fig. 4). The binning of bands 480500 nm and of bands
550560 nm were thus used instead of the two single channels,
centred at 490 nm and 550 nm, respectively.
To implement the MIM algorithm to retrieve CHL-a and TR
concentrations, Eq. (1) (in which Eqs. (2) and (3) were
substituted) was rewritten to a set of 2 equations (for 2 wave-
lengths), where each equation has the form:
CHL a aph ki bbph
ki 11 Pld ki =
Plu ki
R 0; ki
TR aTR ki bbTR
ki 1
1 Pld ki =Plu ki
R 0; ki
aw ki bbw ki 11 Pld ki =
Plu ki
R 0; ki
aCDOM 440 eSCDOM ki440 7
Fig. 5. Sensitivity analysis for the band selection of the MIM image inversion.
(a): first derivative spectra ofR(0, ) vs. aCDOM(440), rescaled by a 0.01 factor
(aCDOM(440) concentration was changed from 0.03 to 0.30 m1, the other two
WQPs were kept constant to CHL-a=1 mg m3 and TR=1 g m3); (b): first
derivative spectra of R(0, ) vs. CHL-a (CHL-a concentration was changed
from 1 to 10 mg m3, the other two WQPs were kept constant to aCDOM(440)
=0.09 m1 and TR=1 g m3); (c): first derivate spectra of R(0, ) vs. TR (TR
concentration was changed from 1 to 10 g m3, the other two WQPs were kept
constant to aCDOM(440)=0.09 m1 and CHL-a=1 mg m3). The NER(0,
)E of Hyperion is overlaid on each of the graphs as dot lines. The box indicates
the wavelength range of Hyperion bands that may be used. Note the different yaxis ranges of each plot.
189C. Giardino et al. / Remote Sensing of Environment 109 (2007) 183195
7/28/2019 Water Qual Hyperion 07
8/13
where, R(0, i) is the Hyperion-derived subsurface irradiance
reflectance, i is the band used in the inversion and aCDOM(440)
is fixed to 0.09 m1.
3. Results and discussion
Fig. 4 and Table 2 show how subsurface irradiance re-flectance values from atmospherically corrected image data, in
situ optical data and bio-optical modelling converge. A
reasonable optical closure was achieved in the range from 480
to 690 nm (average reflectance RMSE 0.006 and relative RMSE
13%, Table 2). Below 480 nm Hyperion data were not reliable
for calibration (Green et al., 2003) while beyond 700 nm
Hyperion reflectances did not seem very well corrected for
adjacency effects and the optical closure was inferior (average
reflectance RMSE 0.011 and relative RMSE 66%, Table 2). At
the two wavelengths (490 and 550 nm) where bands used for
MIM were located and a good convergence to the same
subsurface irradiance reflectance was obtained (the reflectancedifferences were on average less than 0.005, or 7% as relative
value).
Fig. 6 presents the pseudo true colour Hyperion image and
the two WQP maps retrieved applying the MIM to the image.
The pseudo true colour Hyperion image qualitatively describes
the diversity of waters within the lake. The turquoisecyan
colours in the southern part of the lake are due to bottom effects
of bright substrates. Most of the deep waters are dark-blue but
lighter blue waters come up about the middle of the scene, as
well as in the south, on the western side of the peninsula. On
22nd July 2003, re-suspension of sediments was caused by the
strong wind action resulting in variable patterns of more highly
scattering waters (i.e., light-blue colours in the pseudo true
colour Hyperion image). The two MIM-retrieved Hyperion
WQP maps describe the chlorophyll-a and tripton concentra-
tions. In these product maps, pixels where bottom depth was
less than 10 m, were masked because the bio-optical model we
used (Eq. (1)) is applicable in optically deep waters only. Both
maps show ranges of CHL-a and TR between 0 and 5 mg m3
and 0 and 5 g m3, respectively. The patterns in TR map seem
correlated to the patterns in pseudo true colour Hyperion image:
TR concentrations are higher in the light-blue waters, whilst
they are lower in the dark-blue (less-scattering) waters. The fact
that the CHL-a map is uncorrelated to the TR map indicates
successful decomposing of the R(0, ) signal and independent
assessment of CHL-a and TR concentrations following this
method.
Validation of CHL-a and TR Hyperion-derived maps was
performed using in situ point stations. Fig. 7 shows the two
scatter plots depicting the Hyperion-derived CHL-a and TR
estimations vs. in situ concentrations measured in 8 pelagicstations (all the stations in Fig. 1, except Station 1 which is
located in shallow waters where Hyperion data were masked to
avoid bathymetric effects). Hyperion data was averaged on a 3
by 3 pixel region of interest centred on the location of in situ
sampling stations. The Hyperion-derived CHL-a was in good
agreement with in situ point data, showing a correlation
coefficient (r) of 0.77, a determination coefficient (R2) of
0.59, a RMSE of 0.36 mg m3 (relative RMSE 20%), a bias of
0.12 mg m3 (relative bias 6%), and being close to the 1:1 line
(Fig. 7). The RMSE and the relative RMSE were computed with
Eqs. (5) and (6), where Nis now equal to 8 (i.e., the number of
stations), and Xi and X i are the in situ observed and the
Fig. 6. The pseudo true colour Hyperion image (with locations of point in situ stations) and the two MIM-retrieved products obtained from Hyperion data acquired on22nd July 2003. In the product maps shallow waters are masked.
190 C. Giardino et al. / Remote Sensing of Environment 109 (2007) 183195
7/28/2019 Water Qual Hyperion 07
9/13
Hyperion-derived concentrations of WQPs, respectively. The
Hyperion-derived TR did not match in situ point data (r=0.48,
R
2
=0.23 and RMSE 1.1 g m3
), due to divergence betweenHyperion and in situ point data observed in Station 5. Removing
this measurement from the dataset the regression analysis
performed better (r=0.75, R2=0.57, RMSE 0.55 g m3,
relative RMSE 31%, bias 0.27 g m3, and relative bias 15%)
and data become closer to the 1:1 line (Fig. 7). Station 5 is
located in the region of the light-blue, more scattering waters.
Due to the strong wind blowing on the day of the image
acquisition the light-blue water pattern was characterised by a
high rate of change (both in time and space). The water at
Station 5 was probably sampled when these more scattering
waters (imaged by Hyperion at 9:50 UTC) had already moved
elsewhere (assuming that the 6 m s1 northsouth direction
wind resulted in a 0.06 m s1 water current, within 10 min a oneHyperion pixel displacement may occur).
The ability to monitor water quality in highly dynamic
systems could be hindered by the spatial or temporal density of
point sampling offered by traditional sampling techniques
rendering them inappropriate to validate RS-derived products.
Lindfors et al. (2005) suggested that validation of remotely
sensed data products and locations of point measurements
needed for monitoring work should be based on continuously
measured flow-through values. They discussed IOPs, salinity
and temperature in Lake Vnaren (Sweden) and in the Gulf of
Finland. In this study, to qualitatively evaluate the spatial
variation of the WQP retrievals, the flow-through calibratedtransects of CHL-a and TR were resampled according to the
30 m size of Hyperion pixels. Flow-through data were first
cleaned of anomalies, e.g., spikes due to bubbles, saturation and
lack of data (caused by wave-related difficulties in pumping
water from subsurface into the onboard instrumentation). Fig. 8
illustrates the comparison between two indirect estimates of
WQPs: the Hyperion-derived concentrations and the flow-
through-derived transect in situ data. In Fig. 8, the location of
transect in situ data with respect to the time of the Hyperion
overpass is indicated to show the temporal mismatch between
the acquisition of transect in situ data and image data.
The spatial trend of CHL-a concentrations, derived from
Hyperion data and from flow-through data is plotted in Fig. 8
(a). Overall, Hyperion-based estimations are in agreement with
transect measurements. The first part of the transect (04 km of
the length) is not shown since it is not included in the imagefootprint. In Section I of the transect (46 km of the length),
only calibrated transect data are plotted because Hyperion data
were masked to avoid bathymetric effects. Section II (614 km
of the length) shows a good agreement in range and spatial
behaviour between the Hyperion-derived CHL-a and the
transect fluorescence-derived CHL-a data, even if the peak
occurrences in concentration are sometimes shifted in phase. In
Section III (1420 km of the length), when the transect data
were acquired almost at the same time of Hyperion data,
imagery-derived CHL-a match flow-through-derived CHL-a
data. Unfortunately this is the section where many flow-through
data were missed through filtering of wave-related anomalies.
In Section IV of the transect (2022 km of the length),Hyperion-derived CHL-a values were lower than flow-through-
derived concentrations but presented a similar ascending
gradient. Section V (2232 km of the length), shows a rea-
sonable agreement in range and spatial behaviour between
imagery-derived CHL-a and the flow-through-derived CHL-a
transect data with a descending gradient towards the transect
end. The last kilometre depicts Hyperion data only because the
fluorescence measurements were not available.
Fig. 8(b) describes the spatial variation for tripton. As for the
chlorophyll-a, the first 4 km of the transect were outside the
image footprint. In Section I of the transect, only calibrated
transect data are plotted because Hyperion data were masked toavoid bathymetric effects. In Section II, estimations derived
from Hyperion and flow-through transect data are comparable
in values, except for the 3 g m3 peak assessed by Hyperion at
the 7th kilometre. This peak also exists in the transect flow-
through data but it is located at the 5th kilometre, in Section I.
Hyperion-derived concentrations appeared shifted in phase with
respect to flow-through data. As forFig. 8(a), Section III is the
region where collections of flow-through data were closer to the
Hyperion overpass. Within this section, both Hyperion-derived
TR concentrations and flow-through-derived data show a close
agreement exhibiting a steep ascending gradient and compara-
ble concentration ranges. Unfortunately between the 18th and
the 20th kilometres many flow-through data were missed due to
Fig. 7. Scatter plots of Hyperion-derived products and in situ concentrations measured in point stations: on left for chlorophyll-a, on right for tripton. Dot lines indicate
the 1:1 relation. Both graphs do not include data from Stations 1 because it is located in the shallow waters. The statistic in the tripton graph does not include the black
symbol (i.e., Station 5).
191C. Giardino et al. / Remote Sensing of Environment 109 (2007) 183195
7/28/2019 Water Qual Hyperion 07
10/13
wave-related anomalies. In Section IV of the transect, flow-
through data exhibit a flat trending, contrary to the image data
that presents a peak with two times higher concentrations. This
is the region where the transect track crosses the light-blue
waters. As observed before, when point in situ data at Station 5were compared to image-derived TR estimations (Fig. 7), it
seems that the light-blue water front, quickly changing its
position in time and space, was not described by in situ
observations. A re-suspension of tripton due to the strong wind
action, which was synoptically imaged at 9:50 UTC by
Hyperion, had been easily missed by in situ observations
collected at about 0.30.5 h later. In Section V, both the flow-
through data and the Hyperion retrieved TR concentrations
present a descending gradient but with different concentration
ranges and slopes. Moreover, the peak of 4 g m3 of TR
observed by Hyperion (at the 25th kilometre), did not occur in
the transect in situ data. Hyperion synoptically assessed the
tripton distribution at 9.50 UTC: higher concentrations in the
first part of Section V, lower concentrations in the last part of
Section V. The flow-through system gave the TR concentrations
(actually geo-coded to Hyperion pixels) 11.5 h later with
respect to Hyperion, when the front had probably moved
elsewhere and the distribution of tripton was changed. Ingeneral, Fig. 8(b) shows that Hyperion-derived tripton con-
centrations were not comparable to flow-through-derived
values, expect for few kilometres in Section III (from 15th to
19th kilometre) where Hyperion and transect acquisitions match
in time. These results suggest that Hyperion-derived tripton
concentrations, in occasion of events subjected to local vari-
ability in wind, re-suspension and circulation, are difficult to
compare to in situ data due to the incompatibilities of methods
used for tripton assessments. Even fast monitoring methods like
flow-through measurements are time consuming (3 h for a 32-
km-long transect) and they could become inappropriate to
describe natural events with a high rate of change as may occur
in wind driven currents in lakes.
Fig. 8. Comparison of Hyperion-derived products and in situ concentrations estimated from the flow-through data along the horizontal transect: (a) chlorophyll-a, (b)
tripton. Hyperion and flow-through transect data are extracted from a 30 m per 30 m pixel grid (see text for labels I to IV). The approximate location of transect in situ
data with respect to the time of Hyperion overpass is also indicated.
192 C. Giardino et al. / Remote Sensing of Environment 109 (2007) 183195
7/28/2019 Water Qual Hyperion 07
11/13
4. Conclusions
This work presents a procedure to map CHL-a and TR
concentrations in Lake Garda from hyperspectral satellite data
based on forward and inverse bio-optical modelling. The per-
formance of the analytical inversion approach was measured by
the optical closure between forward-modelled reflectance, insitu reflectance and atmospherically corrected Hyperion
reflectance. The closure was sufficient for the purposes of this
study but further investigations on atmospheric adjacency ef-
fects, focused on surface reflected vegetation spectra from steep
slopes, are recommended to obtain a better closure at longer
wavelengths. The bio-optical model sensitivity analysis indi-
cated the optimal bands to run the inversion as well as the
inability to detect aCDOM(440) in this study. The matrix in-
version method was applied to run the inversion on the spatially
and spectrally convolved Hyperion image. The MIM algorithm
provided ranges of CHL-a concentrations comparable to in situ
data collected the day of the satellite overpass. Results fortripton were less satisfactory but an improvement was found if
data from a re-suspension zone were avoided. A further eval-
uation of image-products was based on high spatial resolution
transect in situ data: about 32 km (some transect in situ data
were missed because of the wave-related anomalies) of flow-
through-derived measurements of CHL-a and TR were
qualitatively compared to concentrations retrieved from Hyper-
ion. For chlorophyll-a the Hyperion-derived concentrations
were on average comparable to transect in situ data. The com-
parison was more difficult for tripton since some incompatibil-
ities of methods happened. On the day of the Hyperion overpass
a strong wind occurred over the lake resulting in re-suspension
of sediment (tripton). Further investigations are therefore nec-essary, mainly addressing the compatibilities of methods for
monitoring water body features with high rate of wind or wave
driven change. Matthews et al. (2001) already observed how
continuous fluorometers towed behind boats may offer an
increased capability to monitor chlorophyll-a with respect to
traditional sampling technique in highly dynamic coastal zone
but, the linear track estimates, may be themselves inadequate to
describe the wide-scale heterogeneous phenomenon as synop-
tically retrieved by RS. The results also indicate that fast
processing of hyperspectral images is feasible: once the pre-
processing was done the Hyperion image processing took only
180 s on a standard desktop PC. Another advantage of themethod is that each good set of in situ AOPs, IOPs and SIOPs
measurements added to the spectral library of the lake will
improve the algorithm performance (thus at a certain moment
no further in situ measurements will be required as all source
materials, e.g., inflowing waters, re-suspended material and
algal populations are characterised properly). In such a context,
next studies could also benefit from more information about
variability in the SIOPs over time and space. The presented
procedure is also transferable to other lakes, for which the
optical characterisation of the water body is known and in-
formation about atmospheric properties during the satellite
overpass is accessible. In particular, accurate visibility ranges
are required since Keller (2001) observed how incorrect values
may produce large errors in obtaining R(0, ) and conse-
quently in retrieving concentrations of water quality parameters.
This study aimed to use Hyperion imagery as a bench-mark
for moving towards operational use of RS-related technologies
that, integrated with traditional survey programmes, could
provide useful information to implement the European WFD.
Within the WFD it is possible, for each water body, to monitoronly the water quality elements most sensitive to a certain risk or
pressure. For Lake Garda this could be the deviation from a
trophic level assessed with two causal elements (i.e., phospho-
rous and nitrogen) and with one response parameter, the
chlorophyll-a concentration. The Hyperion data processing
presented in this study will be transferred to the assessment of
lake water quality (mainly chlorophyll-a) using more operational
instruments (being a part of a technology validation/demonstra-
tion mission, Hyperion cannot be considered suitable for a long-
term monitoring). Large swath MODIS and MERIS sensors
(both having the spectral bands used by MIM) offer almost-daily
imagery of northern Italy and the method presented could beextended to the other large (relative to the spatial resolution of
the remote scanner that is) lakes of the subalpine region, where
visibilities ranges are provided by airports or Aeronet stations.
An onerous activity needs however to be completed mainly to
asses the lakes SIOPs or to evaluate how they differ from the
Lake Garda ones. To start, Premazzi et al. (2003) discussed that
the in the subalpine region composition of the phytoplankton
communities would register marked similarities from one lake to
another, as regards density, biomass and species.
Acknowledgements
This work is in memory of Eugenio Zilioli who passed awayin 2004 and who made a considerable effort in Europe to
establish remote sensing of lakes as a tool for environmental
monitoring. Hyperion data were acquired by the Helge Axson
Johnsons Foundation, Sweden. IOP data were collected by A.
Lindfors and K. Rasmus at the Dep. of Geophysics, Helsinki
University and Luode Consulting Oy, Helsinki. This study was
funded by the Italian Space Agency (Ninfa Project), and by
ESA and Regione Lombardia with financial support grants to N.
Strmbeck and to G. Candiani, respectively. The CNR/CSIRO
Agreement (200406 Program) and the Scientific Office at the
Embassy of Italy in Canberra supported the collaboration
among our institutes. This work would not be possible withoutthe assistance and contributions provided during these years by
L. Alberotanza from CNR-ISMAR, G. Zibordi from JRC, and
by C. Defrancesco and G. Franzini from the Environmental
Protection Agencies of Trento and Veneto, respectively. We are
grateful to T. Kutser from Estonian Marine Institute for his
valuable suggestions. Constructive comments from the anon-
ymous reviewers, including ones on an earlier version of the
manuscript, were greatly appreciated.
References
Ammenberg, P., Flink, P., Lindell, T., Pierson, D., & Strombeck, N. (2002). Bio-
optical modelling combined with remote sensing to assess water quality.International Journal of Remote Sensing, 23, 16211638.
193C. Giardino et al. / Remote Sensing of Environment 109 (2007) 183195
7/28/2019 Water Qual Hyperion 07
12/13
Brando, V. E., & Dekker, A. G. (2003). Satellite hyperspectral remote sensing
for estimating estuarine and coastal water quality. IEEE Transaction on
Geoscience and Remote Sensing, 41, 13781387.
Brivio, P. A., Giardino, C., & Zilioli, E. (2001). Determination of chlorophyll
concentration changes in Lake Garda using an image-based radiative transfer
code for Landsat TM images. International Journal of Remote Sensing, 22,
487502.
Candiani, G., Giardino, C., Strmbeck, N., & Zilioli, E. (2003). Imaging spec-trometry to estimate chlorophyll-a in Lake Garda. 3rd EARSeL Workshop on
Imaging Spectroscopy, H errsching, Germany, 1316 May (pp. 389394).
Cairns, B., Carlson, B. E., Ruoxian, Ying, Lacis, A. A., & Oinas, V. (2003).
Atmospheric correction and its application to an analysis of Hyperion data.
IEEE Transaction on Geoscience and Remote Sensing, 41, 12321245.
Cracknell, A. P., Newcombe, S. K., Black, A. F., & Kirby, N. E. (2001). The
ADMAP (algal bloom detection, monitoring and prediction) concerted
action. International Journal of Remote Sensing, 22, 205247.
Dall'Olmo, G., & Gitelson, A. A. (2006). Effect of bio-optical parameter
variability and uncertainties in reflectance measurements on the remote
estimation of chlorophyll-a concentration in turbid productive waters:
Modeling results. Applied Optics, 45, 35773592.
Dekker, A. G., Malthus, T. J., & Hoogenboom, H. J. (1995). The remote sensing
of inland water quality. In F. M. Danson & S.E. Plummer (Eds.), Advances
in environmental remote sensing (pp. 123142). John Wiley & Sons Ltd.Dekker, A. G., Vos, R. J., & Peters, S. W. M. (2001). Comparison of remote
sensing data, model results and in situ data for total suspended matter (TSM)
in the southern Frisian lakes. Science of the Total Environment, 268,
197214.
Dekker, A. G., Vos, R. J., & Peters, S. W. M. (2002). Analytical algorithms for
lake water TSM estimation for retrospective analyses of TM and SPOT
sensor data. International Journal of Remote Sensing, 23, 1535.
Dekker, A. G., Brando, V. E., & Anstee, J. M. (2005). Retrospective seagrass
change detection in a shallow coastal tidal Australian lake. Remote Sensing
of Environment, 97, 415433.
Directive 2000/60/EC (2000, December 22). Water Framework Directive of the
European Parliament and of the Council of 23 October 2000 establishing a
framework for community action in the field of water policy. Official
Journal L, 327.
EEA (1999). Lakes and reservoirs in the EEA area. Topic report No 1/1999Copenhagen: European Environment Agency (EEA).
Fargion, G. S., & Mueller, J. L. (2000). Ocean optics protocols for satellite ocean
color sensor validation, revision 2. NASA/TM-2000209966, Greenbelt,
Maryland, USA.
Floricioiu, D., & Rott, H., (2005). Atmospheric correction of MERIS data over
perialpine regions, MERIS-(A)ATSR Workshop, Frascati, Italy, 2630
September 2005, CD-ROM ISBN 9290929081 ESA.
Giardino, C., Candiani, G., & Zilioli, E. (2005). Detecting chlorophyll-a in Lake
Garda (Italy) using TOA MERIS radiances. Photogrammetric Engineering
& Remote Sensing, 71, 10451052.
Gons, H. J ., Burger-Wiersma, T., Otten, J. H., & Rijkeboer, M. (1992). Coupling
of phytoplankton and detritus in a shallow, eutrophic lake (Lake Loosdrecht,
The Netherlands). Hydrobiologia, 233, 5159.
Gordon, H., Brown, O., Evans, R., Brown, J., Smith, R., Baker, K., et al. (1988).
A semianalytical radiance model of ocean color. Journal of GeophysicalResearch, 93, 1090910924.
Green, R. O., Pavri, B. E., & Chrien, T. G. (2003). On-orbit radiometric and
spectral characteristics of EO-1 Hyperion derived with an underflight of
AVIRIS and in situ measurements at Salar de Arizaro, Argentina. IEEE
Transaction on Geoscience and Remote Sensing, 41, 11941203.
Hrm,P.,Vepslinen,J., Hannonen,T., Pyhlahti, T., Kmri, J., Kallio, K., et al.
(2001). Detection of water quality using simulated satellite data and semi-
empirical algorithms in Finland. Science of the Total Environment, 268,
107121.
Hoogenboom, H. J., Dekker, A. G., & Althuis, I. J. A. (1998). Simulation of
AVIRIS sensitivity for detecting chlorophyll-a over coastal and inland
waters. Remote Sensing of Environment, 65, 333340.
ISO 10260-E (1992). Water quality measurement of biochemical parameters
spectrophotometric determination of chlorophyll-a concentration Geneva,
Switzerland: ISO (E).
Keller, P. A. (2001). Comparison of two inversion techniques of a semi-
analytical model for the determination of lake water constituents using
imaging spectrometry data. Science of the Total Environment, 268,
189196.
Kutser, T. (2004). Quantitative detection of chlorophyll in cyanobacterial
blooms by satellite remote sensing. Limnology and Oceanography, 49,
21792189.
Kutser, T., Herlevi, A., Kallio, K., & Arst, H. (2001). A hyperspectral model forinterpretation of passive optical remote sensing data from turbid lakes.
Science of the Total Environment, 268, 4758.
Lee, Z., Carder, K. L., Hawes, S. K., Steward, R. G., Peacock, T. G., & Davis,
C. O. (1994). Model for the interpretation of hyperspectral remote-sensing
reflectance. Applied Optics, 33, 57215731.
Lindell, T., Pierson, D., Premazzi, G., & Zilioli, E. (Eds.). (1999). Manual for
monitoring European lakes using remote sensing techniquesEUR Report,
vol. 18665. Luxembourg: Office for Official Publications of the European
Communities (EN).
Lindfors, A., Rasmus, K., & Strmbeck, N. (2005). Point or pointless-quality of
ground data. International Journal of Remote Sensing, 26, 415423.
Matthews, A. M., Duncan, A. G., & Davison, R. G. (2001). An assessment of
validation techniques fore estimating chlorophyll-a concentration from
airborne multispectral imagery. International Journal of Remote Sensing,
22, 429447.Mobley, C. D. (1994). Light and water radiative transfer in natural waters San
Diego: Academic Press.
Mobley, C. D., & Sundman, L. K. (2001). HYDROLIGHT 4.2, users guide
Redmond: Sequoia Scientific.
Morel, A. (1974). Optical properties of pure water and pure seawater. In E.
Jerlov & Steeman Nielsen (Eds.), Optical aspects of oceanography:
Academic.
Morel, A. (1988). Optical modeling of the upper ocean in relation to its
biogenous matter content (Case I waters). Journal of Geophysical Research,
93, 1074910768.
Pierson, D. C., & Strmbeck, N. (2001). Estimation of radiance reflectance and
the concentrations of optically active substances in Lake Malaren, Sweden,
based on direct and inverse solutions of a simple model. Science of the Total
Environment, 268, 171188.
Pope, R. M., & Fry, E. S. (1997). Absorption spectrum (380700 nm) of purewater. II. Integrating cavity measurements. Applied Optics, 36, 87108723.
Premazzi, G., Dal Miglio, A., Cardoso, A. C., & Chiaudani, G. (2003). Lake
management in Italy: The implications of the water framework directive.
Lakes and Reservoirs, Research and Management, 8, 4159.
Roesler, C. S., & Boss, E. (2003). Spectral beam attenuation coefficient
retrieved from ocean color inversion. Geophysical Research Letters, 30.
doi:10.1029/2002GL016366
Smith, R. C., & Baker, K. S. (1981). Optical properties of the clearest natural
waters (200800 nm). Applied Optics, 20, 177184.
Strmbeck, N., & Pierson, E. (2001). The effects of variability in the inherent
optical properties on estimations of chlorophyll a by remote sensing in
Swedish freshwater. Science of the Total Environment, 268, 123137.
Strmbeck, N., Candiani, G., Giardino, C., & Zilioli, E., (2003). Water
quality monitoring of Lake Garda using multi-temporal MERIS data.
MERIS Users Workshop, Frascati, Italy, 1013 November 2003, CD-ROMISBN 9290928603 ESA.
Tassan, S., & Ferrari, G. (1995). An alternative approach to absorption
measurements of aquatic particles retained on filters. Limnology and
Oceanography, 40, 13581368.
Toole, D. A., Siegel, D. A., Menzies, D. W., Neumann, M. J., & Smith, R. C.
(2000). Remote-sensing reflectance determinations in the coastal ocean
environment: impact of instrumental characteristics and environmental
variability. Applied Optics, 39, 456469.
Vermote, E., Tanr, D., Deuz, J. L., Herman, M., & Morcrette, J. J. (1997).
Second simulation of satellite signal in the solar spectrum (6S). 6S user
guide (v. 2) and 6S code (v. 4.1), July 1997.
Vollenweider, R. A., & Kerekes, J. J. (1982). Eutrophication of waters:
monitoring assessment and control Paris: Organisation for Economic Co-
operation and Development (OECD).
Walker, R. E. (1994). Marine light field statistics New York: Wiley.
194 C. Giardino et al. / Remote Sensing of Environment 109 (2007) 183195
http://dx.doi.org/10.1029/2002GL016366http://dx.doi.org/10.1029/2002GL0163667/28/2019 Water Qual Hyperion 07
13/13
Wettle, M., Brando, V. E., & Dekker, A. G. (2004). A methodology for retrieval
of environmental noise equivalent spectra applied to four Hyperion scenes of
the same tropical coral reef. Remote Sensing of Environment, 93, 188197.
Zilioli, E. (2002). Experimentation of a remote sensing integrated system for
lake water monitoring. Ninfa technical report 1, 29 November 200128
November 2002, ASI-CNR n. I/R/073/01 (http://milano.irea.cnr.it/ninfa/
ninfa.htm)
Zilioli, E. (2004). Experimentation of a remote sensing integrated system for lake
water monitoring. Ninfa technical report 2, 29 November 200201 April
2004, ASI-CNR n. I/R/073/01 (http://milano.irea.cnr.it/ninfa/ninfa.htm)
195C. Giardino et al. / Remote Sensing of Environment 109 (2007) 183195
http://milano.irea.cnr.it/ninfa/ninfa.htmhttp://milano.irea.cnr.it/ninfa/ninfa.htmhttp://milano.irea.cnr.it/ninfa/ninfa.htmhttp://milano.irea.cnr.it/ninfa/ninfa.htmhttp://milano.irea.cnr.it/ninfa/ninfa.htmhttp://milano.irea.cnr.it/ninfa/ninfa.htm