Satellite Estimation of Chlorophylla Concentration and ... - CALMIT

0 downloads 0 Views 166KB Size Report
non, including introduction of Mnemiopsis leidyi and a reduced anthropogenic impact. To fully understand this effect, it should be carefully studied using various ...
ISSN 00124966, Doklady Biological Sciences, 2010, Vol. 432, pp. 216–219. © Pleiades Publishing, Ltd., 2010. Original Russian Text © G.G. Matishov, V.V. Povazhnyi, S.V. Berdnikov, W.J. Moses, A.A. Gitelson, 2010, published in Doklady Akademii Nauk, 2010, Vol. 432, No. 4, pp. 563– 566.

GENERAL BIOLOGY

Satellite Estimation of Chlorophylla Concentration and Phytoplankton Primary Production in the Sea of Azov Academician of the RAS G. G. Matishova, V. V. Povazhnyia, S. V. Berdnikova, W. J. Mosesb, and A. A. Gitelsonb Received July 14, 2009

DOI: 10.1134/S0012496610030142

In recent years, satellite technologies have been widely used for marine studies. On the basis of chloro phylla (chla) measurements performed with the aid of the satellite scanners Sea WiFS (Seaviewing Wide Field of view Sensor) and MODIS (MODerate Imag ing Specroradiometer), the data on the annual and seasonal dynamics of the primary production in the northern seas of Russia have been summarized [1]. Satellite images of the chla distribution in the south ern seas are currently published (NASA, NPC Priroda). However, there is no calibrated and validated algorithm for estimating the chla concentration from the satellite data in socalled secondary (Case II) waters, which are the coastal and estuarine productive and turbid waters [2]. The Sea of Azov, northwestern Black Sea, and northern Caspian Sea belong to these waters. In the report [1], the authors assume that some areas of the Barents, Kara, and Chukchi seas, which are influenced by the river flows, should be classified as the Case II waters; the algorithms commonly used here can give misleading results. The algorithms using the blue–green spectral band for estimating the chla concentration in the open ocean fail when applied to Case II waters, where phy toplankton is not the only factor that determines the optical parameters [2, 3]. The threeband model and its special case, the twoband model, utilizing the red and near infrared (NIR) spectral regions [4, 5] were proposed for estimating the chla concentration in the turbid productive waters. The results published in [4, 6, 7] demonstrate fairly accurate chla assessment in the Case II waters with a wide range of biological and physical parameters. This was confirmed by the data obtained with field spectrometers, and during simulating the spectral bands characteristic of the sen

a

Southern Scientific Center of the Russian Academy of Sciences, RostovonDon, 344006 Russia b Center for Advanced Land Management Information Technologies (CALMIT), School of Natural Resources, University of NebraskaLincoln, Lincoln, NE 68583 United States

sors MODIS and MERIS (MEdium Resolution Imaging Spectrometer). Since 1988, an increase in the primary production in the Sea of Azov was about twice as high as in 1977– 1987 [8]. There are many reasons for this phenome non, including introduction of Mnemiopsis leidyi and a reduced anthropogenic impact. To fully understand this effect, it should be carefully studied using various available approaches, such as mea surement of the primary production in situ, use of satellite technologies, modeling the primary pro duction cycle, etc. In this study, we discuss the calibration and valida tion of the spectral algorithms used for measurement of the chla concentrations in the Sea of Azov on the basis of field observations; the correlations between the chla concentration and primary production determined in situ were also studied. Expedition observations. The water samples were obtained in the period from April 2008 to March 2009 in the Taganrog Bay of the Sea of Azov to determine the chla concentration and to perform experimental analysis of the primary production. The water samples were filtered onto the Sartorius MGF microfiber glass filters (an analogue of Whatman GF/F); the filters were dried in an exsiccator and kept at –4°C until fur ther treatment in the laboratory. To detect chloro phylls, the spectrophotometric method was used as described in [9]. Both primary production and destruction of plank ton were assessed in water column by an oxygen mod ification of the bottle method described in [10]. In four horizons, four bottles were exposed (two light and two dark) at a depth corresponding to transparencies of 0.5, 1, 2, 3 according to the Secchi disk. Simulta neously, a light and a dark bottles filled with water from the surface horizon were exposed in a tank with run ning water on board the ship. Models, satellite images, and methods of their treatment. The models used in this study were the fol lowing.

216

SATELLITE ESTIMATION OF CHLOROPHYLLa CONCENTRATION

The threeband model: Chl a ∝

–1 ( R λ1



–1 R λ2 )R λ3

–1

Threeband model 0.25

or

(a)

217 Twoband model 2.0

(1)

–1

Chl a ∝ ( R 665 – R 708 )R 753 . The twoband model: –1

–1

Chl a ∝ ( R λ1 )R λ3 or Chl a ∝ ( R 665 )R 708 ,

–1

Chl a = 232.29 [ ( R 665 – R 708 )R 753 ] + 23.17, and with the twoband algorithm:

(3)

–1

Chl a = 61.324 ( R 665 )R 708 – 37.94. (4) For the above stations, the values of both three band and twoband models closely and linearly DOKLADY BIOLOGICAL SCIENCES

Vol. 432

r2 = 0.97

(2)

where Rλ is the coefficient of spectral radiance (CSR) within a 10nm band centered at the wavelength λ (nm). Calibration and validation of the models were based on the data of the following MERIS sensor spectral channels: 7 (665 nm), 9 (708 nm), and 10 (753 nm). The use of MERIS is more preferable than that of MODIS because of a higher spatial resolution of the former (260 × 290 m versus 1 × 1 km in MODIS) and availability of the spectral channel at 708 nm. When the satellite images performed at the day of sampling in situ were unavailable, the MERIS images obtained no more than two days before or immediately after sampling were used. The data on the water sur face were restored on a satellite images using the Bright pixel atmosphere correction as described in [11, 12]. The values of chla concentration obtained in situ were compared with the values obtained with the aid of the treeband (1) or twoband (2) model. The prepa ration and selection of the satellite data are described in [7] in detail. The results of model calibration and validation. For all five in situ data collection series, the satellite images used to determine the chla concentration were obtained at 26 stations. At these stations, the chla concentration ranged from 0.6 to 65.5 mg/m–3, with an average value of 27.5 mg/m–3. At 20 stations, the gross primary production (GPP) was assessed simulta neously; in this case, the chla concentration ranged from 3.8 to 58.2 mg/m–3 (25.7 mg/m–3 on average), whereas the GPP of the surface horizon, which was reduced to the daily interval according to [10], ranged from 0.1 to 3.93 gCorgm–3 day–1 (the average value was 1.17 gCorgm–3 day–1). At 8 stations where GPP was estimated in water column, the average GPP values in the photic layer have been calculated. The data obtained in 2008 (18 stations) were used for calibration of model (1) and model (2), i.e., for determining the quantitative correlations between the coefficients of spectral radiance and chla concentra tion in accordance with the threeband algorithm: –1

1.6

0.15

2010

1.2 r2 = 0.95

0.05

0.8 –0.05 1 2

0.4

0 20 30 40 50 60 60 10 Chla concentration, mg m–3 Estimate of chlorophyll concentration, mg m–3 60 –0.15

0

(b) 45 1:1 Line 30 1 2 15

45 60 30 Chla concentration in situ, mg m–3

Fig. 1. (a) Calibration and (b) validation of the threeband (1) and twoband (2) models.

approximated the chla concentrations in situ with the coefficient of determination (r2) exceeding 0.95 (Fig. 1a). The data obtained in 2009 (8 stations) served for validation of the calibrated algorithms (a) using the CSR of the water surface and with the aid of algo rithms (3) and (4), the chla concentrations have been predicted, (b) the predicted chla values were com pared with the chla concentration determined in situ (Fig. 1b). Both algorithms predicted chla concentration at a high accuracy. The treeband algorithm yielded a root mean square error of 5.02 mg m–3, whereas the same parameter characteristic of the twoband algorithm was 3.65 mg m–3. Comparison with the data of field spectrometers [6, 8] suggested that the results yielded by the threeband model were more accurate than those of the twoband model, because the influence on CSR of the components other than chla was effec tively eliminated in the threeband model by means of subtracting of the reverse reflection λ2 (1). The influ

218

MATISHOV et al.

GPP surf = 0.0479Chl a,

GPP, gCorg/m–3day 4.5

(a)

4.0 3.5 3.0 r2 = 0.72

2.5 2.0 1.5 1.0 0.5 0

10

20

30 40 50 60 Chla concentration, mg m–3

GPP average in the photic layer, gCorg/m–3day 3.0

(b)

2.5 2.0 1.5 r2 = 0.96

1.0 0.5 0

1

2

(5)

GPP aver = 0.67GPP surf ,

3 4 5 GPP, surface, gCorg/m–3day

Fig. 2. (a) Daily GPP in the surface horizon versus chla concentration; (b) correlation between the surface GPP and the average GPP along the photic layer.

ence of theses components increased at low chla con centration. In our study, the λ3 of a longer wavelength has been established for the threeband model (3) as compared to the same parameter of the twoband model (4). Thus, unlike the twoband model, the threeband one was more sensitive to the uncertainties of the method of atmospheric correction because of a low signaltonoise ratio, especially at the stations with low CSR values. This may account for a larger scatter of the points corresponding to the chla con centration below 10 mg m–3 and for a slightly higher root mean square error characteristic of the three band model. In constructing the fields of primary production on the basis of chla concentrations determined from the satellite data, the experimental in situ estimates aver aged for the euphotic layer can be used as a first approximation.

where chla is the chla concentration in the surface horizon, mg/m–3; GPPsurf is the daily gross primary production (GPP) in the surface layer, gCorg/m–3day; GPPaver is the daily GPP averaged for the euphotic layer, gCorg/m–3day. Comparison of the GPP in the surface layer with the chla concentration suggests that this factor may account for 70% of the variance (Fig. 2a). The corre lation between the GPP in the surface layer and the average GPP in the euphotic layer is highly significant (with the coefficient of determination (r2) exceeding 0.96 (Fig. 2b). Our results demonstrate a high accuracy of the esti mates of the chla concentration in the productive tur bid waters on the basis of the satellite data, when the developed tree and twoband models are used. To our knowledge, we have been the first to perform success ful calibration and validation of these models on the basis of the satellite data. Nevertheless, several prob lems of the model calibration should be resolved to ensure the universal application of these models to the satellite images. The models need to be calibrated and validated on the basis of a larger dataset; hence, the technique of in situ data collection should be adjusted in order to maximize the number of stations that can be compared with a single satellite image. The spatial heterogeneity of the water mass around each station within a satellite pixel area should be also taken into account, as well as the changes in the biological, phys ical, and biooptical properties of the water mass, which occur during the time interval between the sat ellite survey and in situ data collection. Accurate and reliable atmospheric correction of the satellite data is still a major problem of the turbid productive waters. Provided that these factors are accounted for, it is possible to develop the calibrated algorithms using the spectral channels in the red and near infrared spectral bands for the realtime quantita tive measurements of the chla concentration on the basis of satellite data. The information retrieved would, in turn, be helpful in rational management of the inland, coastal, and estuarine ecosystems. ACKNOWLEDGMENTS This study was supported by the NASA Land Cover Land Use Change program (project no. NNG06GG17G), Program of the Presidium of the Russian Academy of Sciences no. 17 “Fundamental Problems of Oceanology: Physics, Geology, Biology, Ecology,” and Russian Foundation for Basic Research (project nos. 090410146k, 090792500IR a, and 070401409a). DOKLADY BIOLOGICAL SCIENCES

Vol. 432

2010

SATELLITE ESTIMATION OF CHLOROPHYLLa CONCENTRATION

REFERENCES 1. Vetrov, A.A. and Romankevich, E.A., Okeanologiya, 2008, vol. 48, no. 3, pp. 371–380. 2. Morel, A. and Prieur, L., Limnol. Oceanogr., 1977, vol. 22, no. 4, pp. 709–722. 3. Dall’Olmo, G., Gitelson, A., Rundquist, D., et al., Remote Sens. Environ., 2005, vol. 96, no. 2, pp. 176– 187. 4. Dall’Olmo, G. and Gitelson, A., Appl. Opt., 2005, vol. 44, no. 3, pp. 412–422. 5. Gitelson, A., Int. J. Remote Sens., 1992, vol. 13, no. 17, pp. 3367–3373. 6. Gitelson, A., Dall’Olmo, G., Moses, W., et al., Remote Sens. Environ., 2008, vol. 112, no. 9, pp. 3582–3593. 7. Gitelson, A.A., Gurlin, D., Moses, W.J., and Barrow, T., Envoron. Res. Lett., 2009, vol. 4, p. 045003. 8. Grebnevik Mnemiorsis leidui (A. Agassiz) v Azovskom i Chernom moryakh: biologiya i posledstviya vseleniya

DOKLADY BIOLOGICAL SCIENCES

Vol. 432

2010

219

(The Ctenophoran Mnemiorsis leidui (A. Agassiz) in the Sea of Azov and Black Sea: The Biology and Conse quences of Invasion), Volovik, S.P., Ed., Rostovon Don: Azovsk. Nauch.Issl. Inst. Rybn. Khoz., 2000. 9. Gosudarstvennyi kontrol’ kachestva vody (Governmental Control over Water Quality), Moscow: Izd. Standartov, 2001. 10. Abakumov, V.A., Rukovodstvo po metodam gidrobiolo gicheskogo analiza poverkhnostnykh vod i donnykh otlo zhenii (A Manual on the Methods of Hydrobiological Analysis Surface Waters and Bottom Sediments), Leningrad: Gidrometeoizdat, 1983. 11. Moore, G., Aiken, J., and Lavender, J., Int. J. Remote Sens., 1999, vol. 20, no. 9, pp. 1713–1733. 12. Aiken, J. and Moore, G., ATBD Case 2s Bright Pixel Atmospheric Correction, Rept. POTNMELGS0005, Plymouth: Center for Coastal and Marine Sciences, 2000, vol. 4.