Merger of Ocean Color Information from Multiple Satellite Missions under the NASA SIMBIOS Project Office Ewa J. Kwiatkowska Giulietta S. Fargion Science Applications International Corporation SIMBIOS Project NASA Goddard Space Flight Center, Code 970.2 Greenbelt, MD 20771-5000, USA.
[email protected] Abstract - The purpose of data merger activities undertaken by the Sensor Intercomparison and Merger for Biological and Interdisciplinary Studies (SIMBIOS) Project is to create scientific quality ocean color data sets encompassing measurements from multiple satellite missions. To meet this goal, a number of image processing and data fusion methodologies have been developed within the Project Office. A backpropagation neural network has been employed to map ocean color products from one sensor along with extracted ancillary parameters into products from another sensor. This enabled seamless fusion of data from both sensors to improve ocean color daily global coverage. Concurrently, statistical objective analysis has been implemented to validate the neural network approach. Wavelet-based image multiresolution analysis has been used to merge measurements from sensors of different spatial resolutions and also to examine the prospect of enhancing oceanic features in lower resolution imagery through the use of higher resolution data. Finally, a merger of satellite and in situ measurements has been developed.
as an index of phytoplankton biomass [2]. Other factors which influence the backscatter signal are scattering by inorganic suspended material, scattering from water molecules, absorption by yellow substances, and reflection off the sea bottom. These spectral-radiance signatures of the ocean surface can be detected by remote sensing satellites. Through challenging sensor calibration and validation efforts [3,4], atmospheric and other corrections [5], and normalization for satellite and sun zenith angles, water-leaving radiances are obtained and converted to chlorophyll concentration using empirical algorithms [6]. One objective of the NASA SIMBIOS Project at Goddard Space Flight Center (GSFC) is to integrate information from past, present and any future satellite ocean color sensors and to create scientific quality data sets for routine distribution to the user community. The most obvious benefit of the data merger is improvement in spatial and temporal ocean color coverage because single sensor coverage is severely limited by data gaps between the orbits and data gaps caused by clouds, sun glint and other phenomena [7]. The other critical benefit is an increase in statistical confidence in extracted bio-optical parameters [8]. Ocean color satellite sensors are characterized by different calibration/validation accuracies and different spectral, spatial, temporal, and ground coverage attributes. A large variety of useful multi-sensor applications can be implemented which will take advantage of these sensorvarying characteristics. Several merged ocean color products are expected to be produced in the near future. These will include daily global chlorophyll concentration maps at the highest feasible spatial resolution using data from NASA’s Moderate Resolution Imaging Spectroradiometer (MODIS) on the Terra [9] and Aqua satellites combined with observations from ORBIMAGE and NASA’s Sea-viewing Wide Field-of-view Sensor
Keywords: remote sensing, ocean color, image data fusion, image processing, wavelets, multiresolution analysis, neural networks, SIMBIOS.
1. Introduction Phytoplankton are the principal source of organic matter in the oceans which sustain the marine food chain. They also act as a biological pump which sequesters carbon dioxide from the atmosphere to the deep ocean [1]. Some characteristics of the upper ocean, including phytoplankton concentrations, are differentiated in terms of solar radiance scattered upward in the visible part of the electromagnetic spectrum. The concentration of the main phytoplankton photosynthetic pigment, chlorophyll-a, is often considered
ISIF © 2002
291
(SeaWiFS). Regional and local products are planned for a variety of local applications along with climatological data sets and long-term time series using Coastal Zone Color Scanner (CZCS), Japanese Ocean Color and Temperature Scanner (OCTS), French Polarization and Directionality of the Earth’s Reflectances (POLDER), German Modular Optoelectronic Scanner (MOS), SeaWiFS, MODIS, European Medium Resolution Imaging Spectrometer (MERIS), and Japanese Global Imager (GLI).
SeaWiFS data are available for processing and display via SeaWiFS Data Analysis System (SeaDAS) [15]. MODIS imagery can now be displayed in SeaDAS [16]. The Project has been operating a thorough ocean color validation program to quantify the accuracies of the missions’ products in comparison to in situ measurements [17]. Finally, the Project Office has initiated research and development of methodologies for generating merged multi-sensor ocean color products. The main emphasis has been to define the algorithms which can uniformly overcome mission-specific parameters and be applied to different products and sensors.
There are many difficulties associated with ocean color data merger. Sensors have varying designs and characteristics. They rely on different calibration approaches [3], processing algorithms, and means of vicarious calibration [4]. With new sensors, like MERIS and soon-to-be-launched MODIS-Aqua, GLI, and POLDER-II, merger activities may have to depend on the calibration and validation quality of data products generated by the respective science teams. Nevertheless, before integrating data, the differences in standard products among sensors, such as chlorophyll concentration, normalized water-leaving radiances (nLw) at different spectral bands, aerosol optical thickness (AOT) and others, need to be assessed [10,11]. The same products can be derived using different bands which may or may not cause incompatibilities. Data acquisition and retrieval between the missions are not straightforward because of the large volume of data, differences in file formats and geometric projections, and limited availability of subsampled data sets for developing algorithms and sensor validation.
2. SIMBIOS Project Office data merger achievements In the year 2001, the SIMBIOS Project Office has developed a number of image processing and data fusion methodologies and algorithms to gain expertise and meet the goals of data merger. Four major issues which concern the generation of multi-sensor ocean color products have been addressed: 1. 2. 3.
4.
Improvement of ocean color global coverage (MODIS and SeaWiFS). Merger of ocean color data of different spatial resolutions (MOS and SeaWiFS). Merger of satellite and in situ measurements (SeaWiFS and California Cooperative Oceanic Fisheries Investigation (CalCOFI)). Creation of diagnostic data sets.
Diagnostic data sets are defined as scientist-sponsored sites around the globe for which data are collected from various satellite platforms to facilitate future data merger activities [18]. In the following sections, the first three of the prior mentioned data merger issues will be discussed.
The SIMBIOS Project Office has been addressing the data merger difficulties. It has acquired expertise with reading and analyzing data formats from different missions, including recent MODIS-Terra data sets. Software for uniform calibration and processing of selected ocean color missions was created to improve the level of compatibility among products. SeaWiFS nLw and AOT were used to calibrate MOS radiances [12]. OCTS and POLDER ocean color data were compared and calibrated vicariously using contemporary in situ measurements [11]. A comparative study and inter-calibration were performed between the Korean Ocean Scanning Multispectral Imager (OSMI) and SeaWiFS [13]. Cross-sensor comparisons were made to evaluate product differences between MODIS and SeaWiFS [10]. The entire data set of OCTS Global Area Coverage (GAC) 4km-resolution files was reprocessed by the SeaWiFS and SIMBIOS Projects in collaboration with the National Space Development Agency of Japan (NASDA) and Japanese scientists and made available to the user community through the GSFC Distributed Active Archive Center (DAAC) [14]. CZCS, OCTS, MOS, and
2.1 Improvement of ocean color global coverage MODIS and SeaWiFS are global ocean color sensors currently on orbit. They are both in descending sunsynchronous, near-polar, circular orbits with a 10:30 local Equator crossing time for MODIS and a 12:20-noon crossing time for SeaWiFS. The MODIS swath is 2330 km cross track and SeaWiFS GAC coverage is 1502 km cross track. Gaps between the orbits are filled on the next day. MODIS is equipped with 36 spectral bands of which 9 are used for ocean color studies and SeaWiFS has 8 spectral bands – all of which are defined for ocean color research. MODIS and SeaWiFS have been chosen to investigate merger algorithms leading to the improvement of daily
292
global ocean color coverage. SIMBIOS Project started a collaboration with the MODIS Oceans Team and the MODIS Group at the GSFC DAAC. Common data formats and products from both sensors have been identified for the initial comparisons, data mining, and fusion. Equalarea binned global products created by a standardized binning algorithm were studied at daily 4.63-km resolution [19] to assure exactly the same ground coverage from both sensors. For chlorophyll comparisons, the MODIS chlor_a_2 product was applied because this represents the OC3M algorithm which is most similar to the OC4v4 algorithm used to obtain the SeaWiFS chlor_a product. Software was developed for the combined extraction and analysis of binned MODIS and SeaWiFS data files. As a benchmark for merger evaluation, the Project Office has implemented a daily binning at 9km resolution of combined MODIS and SeaWiFS chlorophyll products. This basic merged product will soon be made available from the DAAC and through the SeaWiFS and SIMBIOS web pages. MODIS, 8 April 2001
were made using spatially overlapping bins for each day with nLw at corresponding bands, chlorophyll concentration, nLw ratios, AOT, and diffuse attenuation coefficient (K490). Product differences were evaluated using density scatter plots (i.e. matchups) and statistics. The matchups were made using both total global coverage data and open-ocean/clear-atmosphere data. The openocean/clear-atmosphere observations were analyzed separately to eliminate ambiguous coastal water and high AOT conditions. Matchups on data from three dates – December 2000, April 2001, and June 2001 – show that there are no time-dependent trends in the comparisons [10]. SeaWiFS and MODIS products compare relatively well for nLw, nLw ratios, K490, and chlorophyll concentration, although some non-functional relationships among data are visible. When considered exclusively, open ocean and clear atmosphere conditions show the same statistical trends as the entire global data set. To better evaluate product correlations and differences between the two missions, more analyses are needed for global and
processing ver. March 2002
50% MODIS and 50% SeaWiFS, 8 April 2001
SeaWiFS, 8 April 2001
processing ver. 4
Figure 1. Original MODIS and SeaWiFS daily-binned chlorophyll-concentration files at 4.63-km resolution and the result of merger of both data sets based on the data mapping approach. As a part of the data mining process preceding the merger, MODIS and SeaWiFS product histograms and scatter plots were investigated. The histograms enabled the examination of data distributions and the product inter-comparisons – the transfer functions between the two sets of data. The presence of time-dependent trends and non-functional data correspondence caused by sensor calibration or data processing inaccuracies were studied because they can severely complicate data merger efforts. The comparisons
local coverages and cross-seasonal temporal scales. However, because the availability of MODIS data processed using the newest algorithms has been restricted, the development of the merger approach has progressed on the limited data set. Over the last year, both the SIMBIOS Science Team and the SIMBIOS Project Office have independently investigated a number of merger methodologies which
293
would increase ocean color coverage [20,21]. The wellestablished statistical objective analysis [22] requires the knowledge of statistical properties of geophysical measurements, such as signal and noise variances, which are not independently available for chlorophyll concentration data sets. Problems with establishing statistics are due to the three scales of magnitude change in the chlorophyll data (0.01-100 mg/m3) and because the only evaluation of satellite product accuracy comes from very sparse matchups with in situ measurements (there are just over 100 of such matchups up to date). Some other merger algorithms, like blended analysis [23], require the spatial geophysical-value field to be relatively smooth and assume the existence of a sufficiently extensive network of global in situ observations to treat them as a ‘benchmark” to propagate over shape-of-the-field defining satellite data. Such extensive ground measurement networks are present, for example, in the case of sea-surface temperature data but not for ocean color. To calculate data at grid points, these analyses either use spatial lag correlations among observation points within defined areas of influence [24] or solve a Poisson equation in regions of sufficient satellite data given the internal boundary condition defined by in situ observations [23]. Members of the SIMBIOS Science Team have investigated blended analysis for the merger of CZCS and in situ data [25] and have also tried it with MODIS and SeaWiFS imagery.
sensor given data from the other sensor. Consequently, the missing sensor data can be emulated in regions where only a single sensor coverage exists. A weighted average of data from both sources can then be performed using respective sensor accuracy levels as in the joint coverage case. An example is shown in Figure 1. This approach consists of the mapping of one sensor’s data so as to imitate data from the other sensor. Although the mapping can be performed using linear or non-linear regression, the use of an artificial neural network is preferred because any complex mapping functions can be approximated using a neural network methodology [26]. The mapping is obtained using data from overlapping bins from both sensors, i.e. the MODIS and SeaWiFS bins used in the matchups. In preliminary studies, the backpropagation neural network mapped chlorophyll values from one sensor into chlorophyll values from the other sensor given additional information on nLw at different spectral bands as well as quality-control ancillary parameters. The use of nLw and ancillary information has been justified by the discovery of non-functional relationships between MODIS and SeaWiFS products shown in the scatter plots. These relationships were found from matchups to be dependent on the geographical locations of the data pixels. Further dependencies of sensor chlorophyll on nLw and ancillary data were discovered using a genetic algorithm. The algorithm evaluated and propagated the fitness of different combinations of nLw and ancillary data inputs through generations of neural networks trained on scaled down data sets combining chlorophyll and these nLw and ancillary inputs.
Because MODIS and SeaWiFS are just two measurement sources with vague statistical properties, non-smooth chlorophyll fields, and patchy global distributions that cannot easily form internal data boundaries, an alternate merger method is also being studied by the SIMBIOS Project Office. The algorithm is aimed at producing global merged products of consistent accuracy for all data pixels independent of their spatial location relative to other data. The approach is independent of sensor-to-sensor differences in instrument design and characteristics, calibration peculiarities, and data processing and it should scale well when more than two global sensors become available. In the process, multisensor chlorophyll values contributing to each pixel are scaled according to instrument accuracy levels defined by matchups with in situ measurements.
The most time consuming stages of this method are the determination of an optimal set of input products and parameters for the neural network as well as training the network on representative global and temporal data sets so as to cover the widest range of chlorophyll conditions. To improve the mapping, other input features such as spatial and temporal chlorophyll distributions can also be included. Once trained, the neural network mapping is very efficient. The scheme can also be used to map sensorto-sensor nLw at different spectral bands and other ocean color products, like AOT.
In practice, the goal is to eliminate discontinuities in merged product data in areas where ocean coverage by a single sensor abuts joint satellite coverages or a single coverage from the other sensor. Since algorithms which smooth out the coverage conversions (e.g. statistical objective and blended analyses) produce uneven accuracies for distant pixels, a different approach is proposed. The approach reproduces the response from one
To evaluate the neural network approach, the Project Office is currently implementing the objective analysis with simplistic statistical assumptions on chlorophyll data and chlorophyll error variances. Additional merged product quality gauge will be provided by the combined MODIS and SeaWiFS binned products, matchups with in situ measurements, and analytic and visual scrutiny.
294
the resolution and adds high frequency variation to the lower resolution scene. This process enables spatial resolution enhancement without altering the average magnitudes of the ocean color values. Because of this, the wavelet method is particularly useful when the sensor qualities are different and the measurement accuracy of the lower resolution sensor should be preserved.
2.2 Merger of ocean color data of different spatial resolutions The SIMBIOS Project Office has been studying ocean color merger opportunities at local spatial scales to provide useful tools for scientists interested in smaller-size geophysical phenomena. The Project has examined the feasibility of merging chlorophyll concentration products from ocean color sensors of different spatial resolutions for cases where there is overlapping ground coverage for individual scenes [27]. The prospect of enhancing oceanic features in lower resolution imagery through the use of higher resolution data has also been studied. The algorithm
The wavelet algorithm has been tested on SeaWiFS and MOS imagery a rare opportunity as these two missions have been cross-calibrated, uniformly processed, and analyzed for overlapping concurrent ground-coverage within the SIMBIOS Project. Original SeaWiFS scenes are
30% MOS and 70% SeaWiFS
MOS
SeaWiFS Figure 2. Original MOS and SeaWiFS chlorophyll-concentration scenes and the result of wavelet-based merger of both data sets binned at 0.5-km resolution and mapped onto a rectilinear latitude/longitude grid. is based on a signal processing approach — wavelet multiresolution analysis — which enables an image to be examined at different frequency/scale intervals [28]. The resolution of an image, which is a measure of detail information in the scene, can be defined and changed by a combination of high pass and low pass filtering operations [29]. The scale of an image is changed by downsampling and upsampling operations. The high frequency, low-scale spatial detail in higher resolution scenes is extracted using the high pass filters of the wavelet transform and is combined with the complete pre-processed lower resolution image [30]. Reversal of the transform increases
binned at 1 km and MOS scenes at 0.5-km resolution to assure correct data coregistration. Bins are then projected onto a rectilinear latitude/longitude grid map to facilitate image processing and to preserve the spatial resolution of the bins in the mapped image. Any missing grid points caused by the mapping of spherical coordinates onto a flat grid are approximated. This preprocessing provides data with the desired size and resolution for the wavelet analysis. One pass of the wavelet transform is applied to the MOS image to extract pixel-to-pixel spatial detail from MOS data and to subsample the MOS scene by 2 which gives it the 1km spatial resolution – same as the native
295
SeaWiFS resolution [31]. The SeaWiFS scene is preprocessed to bring the magnitude of chlorophyll values to the level corresponding to a single application of the low pass filter. Then, the entire SeaWiFS scene replaces the output of the MOS low pass filter and the inverse wavelet transform is calculated. Consequently, the inverse transform produces an enhanced 0.5-km resolution SeaWiFS image with the added MOS high-resolution detail. To merge the data sets, a weighted addition of the wavelet-enhanced SeaWiFS image and the original MOS scene is calculated where weights depend on the established relative accuracies of the products from each instrument, Figure 2.
frequency spatial information contained in MODIS highresolution bands (i.e. 500m or 250m).
2.3 Merger of satellite and in situ measurements The SIMBIOS Project Office has been investigating another application concerned with the integration of ocean color information at local spatial scales – merger of satellite and in situ measurements. The major purpose is to provide a utility to expose changes in remotely sensed chlorophyll range and distribution when collected in situ measurements are overlaid onto the satellite scenes. As the project routinely validates ocean color products using matchups with in situ observations, it is known that there is a significant scarcity of contemporaneous satellite and in situ data, mainly because of the presence of clouds, sun glint, coverage gaps between satellite orbits, and other satellite viewing and meteorological conditions. Consequently, the emphasis has been placed on the development of the ability to spread single in situ observation points onto satellite imagery. The method is based on the application of the wavelet transform which spatially extends in situ data point values onto corresponding areas in satellite scenes. These areas are defined by a radius of influence and depend on the geographical location of in situ measurements [35]. The Hann window function is applied to scale the effects of the in situ data points away from the area centers. Low frequency coefficients of the original ocean color subscenes are replaced with the low frequency coefficients of these subscenes updated with the in situ data points. The wavelet forces the resulting satellite pixels to be interpolations of in situ data points which, at the low resolution, are distributed smoothly around their areas of influence. Leaving the higher frequency coefficients associated with the image unchanged preserves the original high-resolution spatial variations within the areas of influence and protects pixel-to-pixel data variabilities.
To validate the wavelet algorithm, the original MOS scenes were compared against wavelet-enhanced SeaWiFS scenes and SeaWiFS scenes which were bi-linearly interpolated to the MOS resolution. Bi-linear interpolation does not provide the benefits of the higher resolution feature extraction which enabled SeaWiFS imagery to acquire spatial detail inherent in MOS data. Quantitatively, the correlation of original MOS imagery with bi-linearly interpolated SeaWiFS data is considerably smaller (~10%) than the correlation for the wavelet-enhanced SeaWiFS scenes. Qualitatively, the gain in spatial detail obtained by the wavelet approach is consequential and unique. There have been some difficulties associated with the application of the wavelet transform. Although the wavelet-enhanced SeaWiFS scenes appear sharper, there is a degree of high-frequency noise introduced from MOS which is peculiar to this sensor's data. As it happens, wavelets also provide a means for denoising speckled imagery and this has been implemented as an option in the algorithm [32]. This option is based on the softthresholding of wavelet coefficients and is equivalent to removing Gaussian noise from an arbitrary image [33]. Manipulation of wavelet coefficients causes occasional undesirable ringing effects in images because of the presence of high frequency features [34]. To limit this ringing, a selected number of transformed solutions based on different wavelet functions are averaged. Daubechies_20, Coiflet, Haar, and spline functions are examples of the wavelet functions used. Finally, the flags and masks from both sensors’ ocean color products are also merged in the final product.
The radius of the area of influence can be defined using texture extraction. The more irregular the texture around the in situ measurement point, the smaller the radius; the smoother the texture, the bigger the radius. The merger is also dependent on the established relative accuracies assigned to in situ and satellite data. Another condition for merger is the maximum time difference between the satellite and in situ observations.
Future tasks will include the application of the wavelet algorithm to the merger of MODIS and SeaWiFS overlapping scenes so that SeaWiFS imagery could be enhanced by the spatial detail contained in MODIS data. Also, a useful experiment would be to combine MODIS ocean color products at 1km resolution with high
Results of the satellite and in situ measurement merger algorithm were analyzed for the years 1997 and 1998 using SeaWiFS and CalCOFI data sets. Although the largest time difference between the SeaWiFS overflight and in situ data
296
collection was set for a high value of 12 hours, there were just 13 SeaWiFS files for which the merger could have been performed, with a single maximum of three points per scene. To limit the cases where small clouds (a few pixels long) and similar conditions cause ocean color pixels to be masked out from the imagery, a gap-filling algorithm has been designed and implemented. The algorithm is based on the multiresolution image analysis supported by the wavelet transform. Its goal is to preserve spatial patterns of chlorophyll distributions in ocean color imagery without smoothing. The gap-filling algorithm is used to interpolate missing pixels within the areas of influence of in situ data points. This contributed an increase in the number of matchups of around 10%.
Field, R. Frouin, W.E. Esaias, D.D. Kolber, and N.H. Pollack, Biospheric Primary Production During an ENSO Transition, Science, vol. 291, pp. 2594-2597, 2001. [2] S. Sathyendranath, editor, Remote Sensing of Ocean Colour in Coastal, and Other Optically-Complex, Waters, Reports of the International Ocean-Colour Coordinating Group, no. 3, pp. 5-21, 2000. [3] R.A. Barnes, R.E. Eplee Jr., G.M. Schmidt, F.S. Patt, and C.R. McClain, Calibration of SeaWiFS. I. Direct Techniques, Applied Optics, vol. 40, no. 36, pp. 66826700, 2001. [4] R.E. Eplee Jr., W.D. Robinson, S.W. Bailey, D.K. Clark, P.J. Werdell, M. Wang, R.A Barnes, and C.R. McClain, Calibration of SeaWiFS. II. Vicarious Techniques, Applied Optics, vol. 40, no. 36, pp. 67016718, 2001.
3. Conclusions Confronting the increasing number of ocean color missions launched and planned by the international community, the data merger activities have become one of the priorities of the SIMBIOS Project. During the year 2001, the Project’s Office gained experience in a variety of different approaches to ocean color data merger. The most pressing assignment has been to produce daily global chlorophyll data sets encompassing MODIS and SeaWiFS measurements. Both the Project Science Team members and the Project Office have been working on the implementation and testing of their respective algorithms. The Project Office approach is based on artificial intelligence and statistical and signal processing data mining – to imitate the response from one sensor given data from the other sensor. A neural network has been employed to perform the mapping between MODIS and SeaWiFS data. This work will continue to define the most optimal mapping and to create the most accurate daily merged global MODIS and SeaWiFS chlorophyll products. The SIMBIOS Project Office has also been investigating ocean color merger opportunities at local spatial scales to provide useful tools for the science community. A waveletbased algorithm to enhance oceanic features in lower resolution imagery through the use of higher resolution data has been implemented as well as the merger of satellite and in situ measurements. The Project Office will continue its cooperation with the SIMBIOS Science Team, MODIS Team, and DAAC on sensor intercomparisons, data merger, and algorithm implementation. The developed scientific tools and expertise will be used when subsequent ocean color data, such as from MODIS-Aqua, MERIS, GLI and POLDER II, become available.
[5] H.R. Gordon and M. Wang, Retrieval of WaterLeaving Radiance and Aerosol Optical Thickness over the Oceans with SeaWiFS: a Preliminary Algorithm, Appl. Opt., vol. 33, pp. 443-452, 1994. [6] J.E. O’Reilly, S. Maritorena, B.G. Mitchell, D.A. Siegel, K.L. Carder, S.A. Garver, M. Kahru, and C.R. McClain, Ocean Color Chlorophyll Algorithms for SeaWiFS, J. Geoph. Research, vol. 103, no. C11, pp. 24,937-24,953, 1998. [7] W.W. Gregg, W.E. Esaias, G.C. Feldman, R. Frouin, S.B. Hooker, C.R. McClain, R.H. Woodward, Coverage Opportunities for Global Ocean Color in a Multimission Era, IEEE Trans. Geosci. and Remote Sens., vol. 36, no. 5, pp. 1620-1627, 1998. [8] H.J. Thiebaux and M.A. Pedder, Spatial Objective Analysis with Applications in Atmospheric Sciences, Academic Press, pp. 129-204, 1987. [9] W.E. Esaias, M.R. Abbot, I. Barton, O.B. Brown, J.W. Campbell, K.L. Carder, D.K. Clark, R.L. Evans, F.E. Hoge, H.R. Gordon, W.P. Balch, R. Letelier, and P.J. Minnett, An Overview of MODIS Capabilities for Ocean Science Observations, IEEE Trans. Geosci. and Remote Sens., vol. 36, no. 4, pp. 1250-1265, 1998. [10] K. Kilpatrick, E. Kearns, E.J. KwiatkowskaAinsworth, and R.L. Evans, Time Series of Calibrated Ocean Products from NASA’s Moderate Resolution Scanning Spectrometer (MODIS), Proceedings of the Ocean Sciences Meeting, Honolulu, Hawaii, USA, Feb. 2002. [11] M. Wang, A. Isaacman, B.A. Franz, and C.R. McClain, Ocean-Color Optical Property Data Derivation from the Japanese Ocean color and Temperature Scanner and the French Polarization and Directionality of the
References [1] M.J. Behrenfeld, J.T. Randerson, C.R. McClain, G.C. Feldman, S.O. Los, C.J. Tucker, P.G. Falkowski, C. B.
297
Earth’s Reflectances: a Comparison Study, Applied Optics, vol. 41, no. 6, pp. 974-990, 2002.
[22] H.J. Thiebaux, Maximally Stable Estimation of Meteorological Parameters at Grid Points, J. Atmospheric Science, vol. 30, pp. 1710-1714, 1973.
[12] M. Wang and B.A. Franz, Comparing the Ocean Color Measurements between MOS and SeaWiFS: A Vicarious Intercalibration Approach for MOS, IEEE Trans. Geosci. and Remote Sens., vol. 38, no. 1, pp. 184197, 2000.
[23] R.W. Reynolds, A Real-Time Global Sea Surface Temperature Analysis, Climate, vol. 1, pp. 75-86, 1988. [24] P.R. Julian and H.J. Thiebaux, On some Properties of Correlation Functions Used in Optimum Interpolation Schemes, Monthly Weather Review, vol. 103, pp. 605-616, 1975.
[13] B.A. Franz and Y. Kim, A Comparative Study and Intercalibration Between OSMI and SeaWiFS, Trans. American Geophysical Union, vol. 82, no. 47, Fall Meeting, San Francisco, USA, pp. F661, Dec. 2001.
[25] W.W. Gregg and M.E. Conkright, Global Seasonal Climatologies of Ocean Chlorophyll: Blending In Situ and Satellite Data for the CZCS Era, J. Geoph. Research – Oceans, vol. 106(C2), pp. 2499-2515, 2001.
[14] T. Tanaka, B.A. Franz, J.G. Acker, I. Asanuma, S. Bailey, R.E. Eplee Jr., H. Fukushima, J.M. Gales, S. Maritorena, Y. Mitomi, H. Murakami, J.E. O’Reilly, S. Shen, P. Smith, M. Wang, J. Wilding, and B. Woodford, Reprocessing of the OCTS Global Datataset, a Collaborative Effort Between NASDA and the NASA SIMBIOS Project, Trans. American Geophysical Union, vol. 82, no. 47, Fall Meeting, San Francisco, USA, pp. F660-F661, Dec. 2001.
[26] P.M. Atkinson and A.R.L. Tatnall, Neural Networks in Remote Sensing, Int. J. Remote Sens., vol. 18, no. 4, pp. 699-709, 1997. [27] E.J. Kwiatkowska-Ainsworth, Merger of Ocean Color Information of Different Spatial Resolution: SeaWiFS and MOS, Eos. Trans. American Geophysical Union, vol. 82, no. 47, Fall Meeting, San Francisco, USA, pp. F765, Dec. 2001.
[15] G. Fu, K.S. Baith, and C.R. McClain, SeaDAS: The SeaWiFS Data Analysis System, Proceedings of the 4th Pacific Ocean Remote Sensing Conference, pp. 73-79, Qingdao, China, Jul. 1998.
[28] S.G. Mallat, A Theory for Multiresolution Signal Decomposition: The Wavelet Representation, IEEE Trans. on Pattern Analysis and Machine Intelligence, vol. 11, no. 7, pp. 674-693, 1989.
[16] K.S. Baith, R. Lindsay, G. Fu, and C.R. McClain, Data Analysis System Developed for Ocean Color Satellite Sensors, Eos. Trans. American Geophysical Union, vol. 82, no. 18, 2001.
[29] O. Rioul and M. Vetterli, Wavelets and signal Processing, IEEE SP Magazine, pp. 14-38, Oct. 1991.
[17] J.L. Mueller and G.S. Fargion, Ocean Optics Protocols for Satellite Ocean Color Sensor Validation, Revision 3, NASA Technical Memorandum, no. 200221004, vol. 1 & 2, pp. 1-184, 2002.
[30] J. Núñez, X. Otazu, O. Fors, A. Prades, V. Palà, R. Arbiol, Multiresolution-Based Image Fusion with Additive Wavelet Decomposition, IEEE Trans. Geosci. and Remote Sens., vol. 37, no.3, pp. 1204-1211, 1999.
[18] G. Fargion and C.R. McClain, SIMBIOS Project 2001 Annual Report, NASA Technical Memorandum, no. 200221005, pp. 1-308, 2002.
[31] P. Blanc, T. Blu, T. Ranchin, L. Wald, and R. Aloisi, Using Iterated Rational Filter Banks Within the ARSIS Concept for Producing 10m Landsat Multispectral Images, Int. J. Remote Sens., vol. 19, no. 12, pp. 2331-2343, 1998.
[19] J.W. Campbell, J.M. Blaisdell, and M. Darzi, Level-3 SeaWiFS Data Products: Spatial and Temporal Binning Algorithms, NASA Technical Memorandum, no. 104566, vol. 32, pp. 1-73, 1995.
[32] J.-L. Starck and F. Murtagh, Image Restoration with Noise Supression Using The Wavelet Transform, Astron. Astrophys., vol. 288, pp. 342-348, 1994. [33] D.L. Donoho, De-Noising by Soft-Thresholding, IEEE Trans. on Information Theory, vol. 41, no. 3, pp. 613-627. 1995.
[20] S. Maritorena, D.A. Siegel, A.R. Peterson and M. Lorenzi-Kayser, Tuning of a Pseudo-analytical Ocean Color Algorithm for Studies at Global Scales, AGU, OS12M-05, San Antonio, Texas, USA, Jan. 2000.
[34] S.L Barnes, A Technique for Minimizing Details in Numerical Weather Map Analysis, J. Applied Meteorology, vol. 3, pp. 396-410, 1964.
[21] W.W. Gregg and R.H. Woodward, Improvements in coverage Frequency of Ocean Color: combining Data from SeaWiFS and MODIS, IEEE Trans. Geosci. and Remote Sens., vol. 36, no. 4, pp. 1350-1353, 1998.
298
