MONITORING WATER QUALITY OF THE PERIALPINE ITALIAN LAKE GARDA THROUGH MULTI-TEMPORAL MERIS DATA
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
MONITORING WATER QUALITY OF THE PERIALPINE ITALIAN LAKE GARDA
THROUGH MULTI-TEMPORAL MERIS DATA
Gabriele Candiani(1), Dana Floricioiu(2), Claudia Giardino(1), Helmut Rott(2)
(1) Optical Remote Sensing Group-IREA, National Research Council, via Bassini 15, I-20133 Milan, Italy
(2) Institute of Meteorology and Geophysics, University of Innsbruck, Innrain 52, A-6020 Innsbruck, Austria
ABSTRACT
This study represents a preliminary test to build up an operational tool for the evaluation of water quality of Italian
perialpine lakes from space using Full Resolution MERIS data. To the aim, 9 Level-1P (L1P) and 7 Level-2P (L2P)
image data acquired close to in situ measurements of chlorophyll-a (chl-a) concentration in Lake Garda were analysed.
L2P data already comprises the chl-a concentration map, while the top of atmosphere radiances of L1P were converted
to remote sensing reflectances Rrs(λ). The Rrs(λ) values were obtained with an algorithm based on the 6S radiative
transfer model. When no in situ measurements were available the aerosol properties were estimated from MERIS data
with the Dark Dense Vegetation approach. To estimate concentrations of chl-a from Rrs(λ) values a bio-optical model,
parameterised with the specific inherent optical properties of the lake, was inverted using fast inversion procedures,
such as band-ratios or optimisation techniques. All the image products were then compared to in situ data measured
close to the image acquisitions. The optimisation technique (bands 5 to 9) and the band-ratio (B5/B7) applied to 6S-
corrected L1P data provided chl-a values in agreement with in situ data (RMSE=0.87 mgm-3 for optimisation,
RMSE=1.20 mgm-3 for band-ratio). Also the Algal2 products gave promising results (RMSE=1.60 mgm-3) even the
presence of several invalid pixels. Nevertheless, more tests with additional field data and MERIS L1P and L2P imagery
are required to validate and improve the method we shown.
1 INTRODUCTION
Lake water is an essential renewable resource and its sustainable use requires the combination of surface waters
assessment and monitoring programs, coupled with decision making and management tools. The Water Framework
Directive (WFD) [1] is the major reference to guide efforts for attaining a sustainable aquatic environment in the years
to come. The WFD includes guidelines which define the categories of quality and the corresponding components and
parameters. Because some of these parameters can be determined by Remote Sensing (RS) with a reasonable accuracy,
RS-related technologies may be integrated in the monitoring programs defined by the WFD. To this aim the MERIS
sensor, onboard of Envisat, offers an excellent choice in terms of costs, revisiting time, spectral and radiometric
resolutions.
Most of the largest European lakes are located in the Nordic countries and in the Alpine regions [2]. The most important
Italian lake district is located in the northern part of Italy and represents more than 80% of the total Italian lacustrine
volume [3]. With respect to the lakes of the district, Lake Garda was chosen because the RS-related activity has a pretty
long tradition at this lake and hence optical properties are studied [4, 5, 6, 7, 8]. The lake dimension is also in agreement
with the pixel size of FR MERIS data, at least for the southern part. The lake is therefore also used as a benchmark to
implement RS in the monitoring programs of the whole southern perialpine lake district, when appropriate Earth
Observation (EO) systems will be available. The Full Resolution (FR) acquisition mode of MERIS seems in fact still
too coarse for the smallest and narrowest lakes of the region.
2 MATERIAL AND METHODS
2.1 Water quality data
With an area of 368 km2, Lake Garda is the largest Italian lake and one of the most important lake of the European
region. The lake needs accurate care not only for its natural relevance, but also for its economical importance due to
tourist-related activities. The lake is oligomitic and oligo-mesotrophic in accordance with the Organisation for
Economic Cooperation and Development (OECD) classification [9].
The water quality parameters that were considered relevant for the bio-optical model used in this study were the
concentration of chlorophyll-a (chl-a), the concentration of suspended particulate matter (SPM) and the absorption of
coloured dissolved organic matter (CDOM) at 440 nm. The most likely concentrations of these parameters in the lake is
described by Probability Density Functions (PDFs). The PDF of each water quality parameter was built from a quasi-homogenous database which includes in situ measurements acquired since 1997. The statistical analysis performed has
retrieved the log-normal distribution to be the most suitable probability distribution for each parameter (Fig. 1).
120 0.25 50 0.30 30 0.60
100 0.20 40 0.25 25 0.50
80 0.20 20 0.40
0.15 30
60 0.15 15 0.30
0.10 20
40 0.10 10 0.20
20 0.05 10 0.05 5 0.10
0 0.00
0 0.00 0 0.00
0.10
1.90
3.71
5.51
7.32
9.12
10.92
12.73
14.53
16.34
18.14
0.10
0.59
1.08
1.57
2.06
2.55
3.05
3.54
4.03
4.52
5.01
5.50
0.02
0.12
0.21
0.31
0.40
0.50
0.60
0.69
chl-a [mgm -3] SPM [gm -3] a(440)CDOM [m-1 ]
Fig. 1. The PDFs of the investigated water quality parameters of Lake Garda. For each panel: left y-axis is frequency,
the right y-axis is the probability and the x-axis is the parameter concentration.
2.2 The bio-optical model
An important parameter in the interpretation of water quality from remote sensing is the remote sensing reflectance
Rrs(λ). Rrs(λ), in fact, can be related to the water quality parameters by means of a function of the spectral absorption
and backscattering coefficients.
The spectral absorption coefficient a(λ) is often divided into four components which account for the contribution of
pure water (w), phytoplankton (F ), nonalgal particles (NAP) and CDOM [10]:
a(λ ) = a w (λ ) + a Φ (λ , chl − a ) ⋅ chl − a + a NAP (λ , SPM ) ⋅ SPM + a CDOM (λ , a CDOM (440 ))⋅ a CDOM (440 ) .
* * *
(1)
While for the pure water absorption was chosen [11], the absorption coefficients of the other three components were
modelled using simple models able to describe different water conditions. These models were parameterized using
absorption spectra measured in laboratory. In particular, aF *( λ, chl-a) was calculated according to [12], aNAP*( λ, SPM)
and aCDOM*( λ, aCDOM(440)) terms were calculated according to [10]. Similar to [13], the analysed NAP data shown a
relationship between aNAP*(440) and SPM, and an average SNAP value equal to 0.0079. For CDOM absorption an
inverse relationship was found between SCDOM and aCDOM(440), with an average SCDOM value close to 0.14.
The spectral backscattering bb(λ) is expressed as the sum of three absorption coefficient for water, phytoplankton and
Suspended Inorganic Particulate Matter (SPIM):
bb (λ ) = bbw (λ ) + 0.091⋅ λ−0 .515 ⋅ chl − a + 0.040⋅ λ−0 .138 ⋅ SPIM . (2)
In this case the backscattering value of pure water was taken from [14]. Backscattering coefficients of phytoplankton
and SPIM were modelled using measurements achieved with HydroScat-6 during several field campaigns and they were
considered to be constant. For the purposes of the model also a relationship between SPM and SPIM was found using
114 discrete measurements of both parameters.
Defining
bb (λ )
u (λ ) ≡ , (3)
a(λ ) + bb (λ )
Rrs(λ) can be expressed as
Rrs(λ ) = g ⋅ u (λ ) , (4)
where g itself is a function of u [15]:
g = g 0 + g1 ⋅ u(λ )g2 . (5)To define g0, g1 and g2, Hydrolight 4.2 was employed to generate Rrs(λ) values. These Hydrolight-simulated values
were compared with values given by approximations from (4) and (5) varying the set of g. The g0, g1 and g2 coefficients
were determined by minimizing the following quantity as in [15]
Rrs B (λ )
δ = average ln , (6)
Rrs H (λ )
where RrsB(λ) is a bio-optical model simulated value and RrsH(λ) is an Hydrolight-generated value.
The resulting best-fit set of g values gave us g0 ≈ 0.045, g1 ≈ 0.088 and g2 ≈ 1.192, and hence the final bio-optical model
parameterization was:
( )
Rrs (λ ) ≈ 0.045 + 0.088 ⋅ u(λ )1.192 ⋅ u(λ ) . (7)
2.2 MERIS data
In this study 9 Full Resolution (FR) Level-1P data were analysed and, for all the dates expect two, also Level-2P data
were available (Tab. 1). The MERIS imagery was selected according to data already acquired within AO533 and
AO164 ESA PI projects and to available in situ chl-a data, routinely measured by local agencies (Tab. 1).
In order to obtain the spectral reflectance from the L1P MERIS data a correction based on the 6S radiative transfer
model [16] was applied. The atmospheric profiles of air temperature, pressure and absolute humidity were derived from
radiosonde data at Milano acquired at 12:00 UTC on the days with MERIS overflights. The average Aerosol Optical
Thickness value at 550 nm (AOT550) and the Ångström α coefficient were derived from sun photometric measurements
carried out on the shore of Lake Garda or by means of the Dense Dark Vegetation (DDV) approach [17] applied to the
area around the lake. The Junge size distribution was chosen with a parameter, ν, related to α by ν=α+3. A detailed
description of the atmospheric correction procedure can be found in [18].
A reflectance spectrum derived from the MERIS spectral radiance over an area in the south basin of Lake Garda on the
day with field measurements is shown in Fig. 2. The agreement between MERIS and SpectraScan PR-650 spectra is
generally good. Some differences appear in the blue/green bands, possibly due to path radiance, which is large in this
spectral region. The maximum at-surface reflectance values ρ of Lake Garda are around 4% in all processed MERIS
scenes, which reveals a weak signal coming from the lake.
Tab. 1. List of MERIS data investigated in this study and in situ chl-a data close to satellite overpasses (only on 22-Jul-
03 there is coincidence between in situ and MERIS data)
In situ chl-a data 09-Jun-03 17-Jun-03 09-Jul-03 22-Jul-03 23-Jul-03 14-May-04 14-Jul-04 03-Aug-04 09-Aug-04
MERIS overpasses 04-Jun-03 19-Jun-03 06-Jul-03 22-Jul-03 07-Aug-03 19-May-04 15-Jul-04 31-Jul-04 13-Aug-04
Level-1P ü ü ü ü ü ü ü ü ü
Level-2P ü ü ü ü - - ü ü ü
To directly compare imagery data to the forward bio-optical modeling the atmospherically corrected ρ data were
converted into Rrs (since now the λ-dependency of Rrs is omitted for simplicity). This conversion was simply
performed dividing L1P-derived ρ data by π. The assumption implies that Rrs of our MERIS data may include
components due to specular reflection effects occurring at the surface.
A geo-located region of interest (ROI) of 195 pixels, defining a common lake area for all the images, was defined for
the data analysis. All the lake-water pixels of the Lake Garda ROI (LG-ROI) were located in the southern part of the
basin, were image data are less affected by residuals of adjacency effects and were Algal2 products all shown valid
pixels.10 MERIS FR LGarda South
AOT550=0.28 ν=4.8
8 8 SpectraScan
ρ [%]
6 6
4 4
2 2
0 0
700 800 900 1000 400 500 600 700 800 900 1000
λ [nm] λ [nm]
Fig. 2. MERIS reflectance in the south basin of Lake Garda after atmospheric correction compared to reflectance
spectrum above the water surface measured in situ on 22-Jul-03. The aerosol properties measured at the time of the
satellite overpasses and used in the radiative transfer model, AOT550 and the Junge parameter ν, are indicated.
The average Rrs spectra extracted from the LG-ROI for each date were then plotted together with the average values of
bio-optical derived Rrs spectra (Fig. 3). For the generation of these simulated spectra, the bio-optical model was run 200
times using random combinations of chl-a, SPM and CDOM distributed according to their PDFs (as in Fig. 1). In this
way, the 200 simulated spectra of Rrs could be considered to be realistic for Lake Garda. The order of magnitude of 6S-
corrected L1P data was generally in agreement with the average value of the 200 forward runs of the bio-optical model.
At longer wavelengths, 6S-corrected L1P data appeared in less agreement with modelled spectra showing Rrs values
larger than simulated ones. Also at shorter wavelengths, 6S-coderived Rrs were generally larger than simulated spectra.
0.018 04-Jun-03
19-Jun-03
0.016 06-Jul-03
22-Jul-03
07-Aug-03
0.014
19-May-04
15-Jul-04
0.012 31-Jul-04
13-Aug-04
Rrs [sr ]
-1
0.010
-1
Av(+1.96StDev)
Av bio-optical modelled
0.008 Av(-1.96*StDev)
0.006
0.004
0.002
0.000
400 500 600 700 800
λ [nm]
Fig. 3. Comparison between MERIS-derived and forward bio-optical modelled Rrs values. MERIS spectra are the
average values extracted from the geo-located LG-ROI, modelled Rrs spectra are plotted as average values (of 200
spectra) ± 1.96 the standard deviation.
3 RESULTS
There are several approaches to derive water quality parameters from image-derived Rrs data (or other quantities such
as the subsurface irradiance reflectance), when an analytical bio-optical model is available for a certain lake. In this
study an optimisation technique and a band-ratio method, the second already tested on Lake Garda [7], were considered.In the optimisation technique (opt) the quantity in equation (6) was minimised. With respect to equation (6) the
denominator was now the MERIS-derived Rrs. In order to avoid uncertainties related to the atmospheric correction, the
inversion was applied to MERIS spectra from band 5 (560 nm) to 9 (708 nm). For the generation of the band-ratio
algorithm (br), the 200 Rrs values simulated using the bio-optical model and the PDFs were used. The Rrs value of each
band was divided by each other and regressed against each of the chl-a concentration. In this way, the regression
function based on the br with highest correlation with chl-a was derived. The B5/B7 (r=-0.7) was selected to describe
the chl-a variability even if the chl-a changes were better described by band-ratios located in the red-NIR wavelengths.
Nevertheless, these bands were not considered to avoid wavelengths where the atmospherically corrected MERIS data
shown some anomalies (see Fig. 3).
15
In situ br opt Algal2
12
chl-a [mgm-3]
9
6
3
0
19-May-04
04-Jun-03
19-Jun-03
06-Jul-03
22-Jul-03
15-Jul-04
31-Jul-04
07-Aug-03
13-Aug-04
Fig. 4. Comparison between in situ measurements (plotted with their standard deviations) and the average values
(extracted form the common LG-ROI of 195 pixels) of MERIS-derived chl-a concentrations, using br and opt
techniques and, when available, Algal2 products.
Fig. 4 shows the comparison between in situ data, measured close to image acquisitions, and the MERIS-derived chl-a
concentrations, estimated according to the opt and br techniques: on 22-Jul-03, when in situ data were collected the
same day of MERIS acquisition, all the MERIS-derived chl-a agree to in situ observations; for all dates except for 19-
Jun-03 and 13-Aug-04, the chl-a derived with opt technique falls within average ± the standard deviation of in situ
observations (even if on 07-Aug-03 the opt technique value is slightly larger than in situ data, ± its standard deviation).
Erroneous estimation on 13-Aug-04 may be due to the location of in situ stations (all in the northern part and no
matching any pixel of the LG-ROI); on 19-Jun-03 all the MERIS-derived chl-a concentrations are larger that in situ data
(this behaviour seems in agreement with a surface Anabaena bloom observed in situ on 20-Jun-03 that maybe was not
revealed by laboratory analysis of the 1-integrated meter of subsurface sampled water).
Excluding data acquired on 19-Jun-03 and on 13-Aug-04 (due to problems mentioned above) the Root Mean Square
Error (RMSE) computed using in situ data was 0.87 mgm-3 for the opt technique and 1.20 mgm-3 for the br algorithm.
Still excluding the 19-Jun-03 and the 13-Aug-04 dataset, the RMSE computed using 5 Algal2 products was 1.60 mgm-3.
4 CONCLUSIONS
Preliminary results obtained from atmospherically corrected L1P FR MERIS data are promising to implement a scene-
independent method to assess chl-a concentration in Lake Garda. Up to now the estimation using the opt technique
applied to the 6S-corrected Rrs MERIS data performed better than the br algorithm.
Algal2 products also provided encouraging results, even if the spatial information seems minor due to the presence of
several invalid pixels (more in the 2003 than in the 2004 dataset). More images, as close as possible to field data, are
necessary to verify the method concerning the inversion of Rrs spectra using br algorithms and opt techniques (as well
as other inversion methods).
Moreover, the responsibility of IOPs on chl-a assessment had to be better understood and new data on optical properties
of Lake Garda waters are going to be collected.5 ACKNOWLEDGEMENTS
MERIS data were supplied by ESA within AO553 and AO164 PI projects. Limnological in situ data were provided by
APPA Trento, ARPAV Veneto and ASL-Brescia. IOP in situ data were measured by G. Zibordi (JRC-EI), L.
Alberotanza (CNR-ISMAR), and A. Lindfors and K. Rasmus (Dep. of Geophysics, Helsinki University). We are very
grateful to N. Strömbeck (Uppsala University) and to A. G. Dekker and V. E. Brando (CSIRO-Land and Water) for the
continuous feedback on our researches on Lake Garda. This work was partially supported by Agenzia Spaziale Italiana
(ASI) in the framework of the project “NINFA” under the contracts ARS.
6 REFERENCES
[1] Directive 2000/60/EC, 2000, 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,
1-73, 22 December 2000.
[2] EEA, Lakes and reservoirs in the EEA area, Topic report No 1/1999, European Environment Agency (EEA),
Copenhagen, 1999.
[3] Premazzi G., Dalmiglio A., Cardoso A. C. and Chiaudani G., Lake management in Italy: the implications of the
Water Framework Directive, Lakes & Reservoirs: Research and Management, 8, 41-59, 2003.
[4] Zilioli E., Brivio P. A., and Gomarasca M. A., A correlation between optical properties from satellite data and
some indicators of eutrophication in Lake Garda (Italy), the Science of Total Environment J., 158, 127-133, 1994.
[5] Zilioli E. and Brivio P. A, The satellite derived optical information for the comparative assessment of lacustrine
water quality, the Science of Total Environment J., 196, pp. 229-245, 1997.
[6] Brivio P. A., Giardino C. and Zilioli E., 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, 487-
50, 2001.
[7] Strömbeck N., Candiani G., Giardino C. and Zilioli E., Water quality monitoring of Lake Garda using multi-
temporal MERIS data, MERIS Users Workshop, Frascati, Italy, CD-ROM, 10-13 November 2003.
[8] Giardino C., Candiani G. and Zilioli E., Detecting chlorophyll-a in Lake Garda (Italy) using TOA MERIS
radiances, Photogrammetric Engineering & Remote Sensing, 71, 1045-1052, 2005.
[9] Vollenweider R. A. and Kerekes J. J., Eutrophication of waters: monitoring assessment and control, Paris:
Organisation for Economic Co-operation and Development (OECD), 150 p, 1982.
[10] Babin M., Stramski D., Ferrari G. M., Claustre H., Bricaud A., Obolensky G. and Hoepffner N., Variations in the
light absorption coefficients of phytoplankton, nonalgal particles, and dissolved organic matter in coastal waters
around Europe, Journal of Geophysical Research, 108, 3211, 4,1-4,20, 2003.
[11] Pope M. and Fry E. S., Absorption spectrum (380-700 nm) of pure water. II: Integrating cavity measurements,
Applied Optics, 36, 8710-8723, 1997.
[12] Bricaud A., Babin, M., Morel, A. and Claustre, H., Variability in the chlorophyll-specific absorption coefficients
of natural phytoplankton: analysis and parameterization, Journal of Geophysical Research, 100, 13,321-13,332,
1995.
[13] Bowers D. G., Harker G. E. L. and Stephan B., Absorption spectra of inorganic particles in the Irish Sea and their
relevance to remote sensing of chlorophyll, International Journal of Remote Sensing, 17, 2449-2460, 1996.
[14] Morel A., Optical properties of pure water and pure seawater. In N. G. Jerlov & E. S. Nielson (Eds.): Optical
Aspects of Oceanography. Academic Press, 1974.
[15] Lee Z, Carder K. L., Mobley C. D., Steward R. G. and Patch J. S., Hyperspectral remote sensing for shallow
waters. I. A semianalytical model, Applied Optics, 37, 6329-6338, 1998.
[16] Vermote E., Tanré D., Deuzé J.L., Herman M. and Morcrette J. J., Second Simulation of the Satellite Signal in the
Solar Spectrum, 6S: An Overview, IEEE TGRS, 35 (3), 675 – 686, 1997.
[17] Santer R., Carrère V., Dubuisson P. and Roger J. C., Atmospheric corrections over land for MERIS, International
Journal of Remote Sensing, 20,1819-1840, 1999.
[18] Floricioiu D. and Rott H., 2005, Atmospheric correction of MERIS over alpine regions, this issue.You can also read
