Conference PaperPDF Available

True color analysis of natural waters with SeaWiFS, MODIS, MERIS and OLCI by SNAP

Authors:
  • Royal Netherlands Institute for Sea Research - Royal NIOZ

Abstract and Figures

The colors from natural waters differ markedly over the globe, depending on the water composition and illumination conditions. Space-borne " ocean color " sensors are operational instruments designed to retrieve important water-quality indicators, based on the measurement of water leaving radiance in a limited number (5 to 10) of narrow (≈10 nm) bands. Surprisingly, the analysis of the satellite data has not yet paid attention to color as an integral optical property that can also be retrieved from such multispectral satellite data. Based on a paper in Sensors [Van der Woerd and Wernand, 2015] we demonstrate that color, expressed mainly by the hue angle (α), can be derived accurate and consistently from SeaWiFS, MODIS, MERIS and OLCI data. The algorithm consists of a weighted linear sum of the remote sensing reflectance in all visual bands plus a correction term for the specific band-setting of each instrument. The algorithm is validated by a set of hyperspectral measurements from inland-, coastal-and near-ocean waters and makes a natural connection to hyperspectral measurements, smart phone measurements and the historic color observation of the 20 th century. The algorithm is now available for these four instruments in the Sentinel-3 Toolbox, which is built on the Sentinel Application Platform (SNAP). We will present a number of SNAP-based applications that demonstrate that (α) is a parameter that adds new insight to ocean-color images.
Content may be subject to copyright.
1
True color analysis of natural waters with SeaWiFS,
MODIS, MERIS and OLCI by SNAP
Hendrik Jan van der Woerd1, Marcel R. Wernand1, Marco Peters2, Muhammad Bala2,
Carsten Brockmann2.
1 NIOZ Royal Netherlands Institute for Sea Research, NL-1790 AB Den Burg, the Netherlands.
2 Brockmann Consult, Max-Planck-Str. 2, D-21502 Geesthacht, Germany.
Abstract
The colors from natural waters differ markedly over the globe, depending on the water
composition and illumination conditions. Space-borne “ocean color” sensors are operational
instruments designed to retrieve important water-quality indicators, based on the
measurement of water leaving radiance in a limited number (5 to 10) of narrow (≈10 nm)
bands. Surprisingly, the analysis of the satellite data has not yet paid attention to color as an
integral optical property that can also be retrieved from such multispectral satellite data.
Based on a paper in Sensors [Van der Woerd and Wernand, 2015] we demonstrate that
color, expressed mainly by the hue angle (α), can be derived accurate and consistently from
SeaWiFS, MODIS, MERIS and OLCI data. The algorithm consists of a weighted linear sum of
the remote sensing reflectance in all visual bands plus a correction term for the specific
band-setting of each instrument. The algorithm is validated by a set of hyperspectral
measurements from inland-, coastal- and near-ocean waters and makes a natural connection
to hyperspectral measurements, smart phone measurements and the historic color
observation of the 20th century. The algorithm is now available for these four instruments in
the Sentinel-3 Toolbox, which is built on the Sentinel Application Platform (SNAP). We will
present a number of SNAP-based applications that demonstrate that (α) is a parameter that
adds new insight to ocean-color images.
Introduction
Color is a concept that originates in the human perception of radiation between the
wavelengths of 380 to 720 nm. The human eye has three cone receptors that are very
sensitive in the red, green or blue. Since the start of the 20th century the sensitivity of
human color perception was well documented [CIE, 1932]. Also scientists like Forel and Ule
found a way for a consistent measure of ‘the water color by using human perception to
compare colors of natural waters. This Forel-Ule (FU) scale is a historical standard that has
recently been very well calibrated [Novoa et al., 2013]. The scale was developed because of
technological limitations that existed at the end of the 19th century. However, new
initiatives in participatory science like within the EU-Citclops project, indicate that the color-
comparison methodology can be transferred to nowadays measuring techniques using smart
2
phones and other new kind of devices [Wernand et al., 2016]. Past (SeaWiFS, MERIS) and
present (MODIS, OLCI) satellite ocean color imaging instruments are spectrometers that
select only a limited number of narrow (≈10 nm) spectral bands in the visual domain and
therefore a straight-forward calculation of the true color is not possible, like in the case of
hyperspectral data. Nevertheless, in [Van der Woerd and Wernand, 2015] we demonstrate
that the true color can be approximated very well. In this presentation we show some
examples of how color can help in the image analysis.
Methods
The derivation of the true color of natural waters is based on the calculation of the
Tristimulus values that are the three primaries (X,Y,Z) that specify the color stimulus of the
human eye [CIE, 1932]. Suppose the radiation spectrum that comes from the water is given
by I that is a function of wavelength (λ), then the tristimulus values are given by (example for
red):
( )
λλλ
=dxIX )(
(1)
The standard colorimetric 2 degree Color Matching Functions (CMFs) are presented byx
(red), y (green) andz (blue). These serve as weighting functions for the determination of
the tristimulus values. Because ocean color satellites do not provide full-spectral coverage,
the spectrum must be first reconstructed by linear interpolation, based on the remote
sensing reflection (Rrs) measured at the spectral bands. It is important to realize that
satellite Rrs derived from standard image processing is already corrected for atmospheric
influences and illumination conditions and is therefore a quasi-inherent optical property
[Ocean Optics Book, 2016]. We showed in [Van der Woerd and Wernand, 2015] that for
ocean color instruments, for example MERIS with nine bands, equation (1) can be
approximated by equation (2). This implies that each tristimulus value can be calculated as a
linear sum of the band information. M(i) are the linear coefficients that are pre-calculated
for each color and each instrument and can be found in the article.
=
=
=
9
1RS
)()(
i
i
iRiMX
(2)
Once the tristimulus values (X,Y,Z) have been calculated, the three values are normalized
and the color is expressed in the chromaticity coordinates (x,y):
ZYX X
x++
=
ZYX Y
y++
=
(3)
In the (x,y) chromaticity plane, the coordinates are transformed to polar coordinates with
respect to the white point W (x=y=1/3) and the hue angle (α) is derived. This angle (α) lies
between the vector to a point with coordinates (x-xw, y-yw) and the positive x-axis (at y-yw =
0), giving higher angles in an anti-clockwise direction (see Figure 1).
3
Figure 1. A chromaticity diagram showing the hue color angle (α) match relative to the
white point (xw, yw). The dominant wavelength of the specific segment is indicated in
nm, together with the position of the 21 FU scale colors.
Finally each (α), derived from equations (1-3), is corrected by a small number(Δ) to get the
best linear relation between the calculated hyperspectral hue angle and the ocean color
instrument hue angle. Also these corrections can be found in the article. To couple to
historical data, the classification of (α) in FU-numbers is added to the end result.
In order to make this algorithm rapidly available for a wider audience, the hue-angle
algorithm is now available for four instruments (SeaWiFS, MODIS, MERIS, OLCI) in the
Sentinel-3 Toolbox, which is built on the Sentinel Application Platform (SNAP)
[http://step.esa.int]. SNAP is a platform for Earth Observation data processing and analysis
characterized by the following technological highlights: Extensibility, Portability, Modular
Rich Client Platform, Generic EO Data Abstraction, Tiled Memory Management, and a
powerful Graph Processing Framework. SNAP and the ESA toolboxes, namely S1, S2 and S3
Toolbox, support the scientific exploitation of the Sentinel missions and also a range of
national and third party missions. The color analysis is included in the SNAP 4.0 release and
can be downloaded from ESA [SNAP 2016].
After installation of SNAP and importing a satellite image, the processor will show up in the
menu at "Optical/Thematic Water Processing/FU Classification". The menu is very simple,
see Figure 2, and a help file is included. The main menu is related to input and output files
and automated display of results in the SNAP environment. In the second menu the
instrument can be indicated manually or left to the auto-detection. Copying of all source
bands is advised, since that will provide all the original input and processing information
(flags) in the SNAP environment. As a start we also recommend to include the intermediate
products of the hue angle calculation.
4
Figure 2. Main menus of the hue-angle FU processor to determine the input/output parameters (left)
and processing parameters (right)
Results
What is the value of this hue angle and how can it be used in the future interpretation of sensor data
for scientists who study the biology and hydrology of oceans, lakes and estuaries? Here we present 3
examples of the potential of this parameter. From the NASA website [oceancolor.gsfc.nasa.gov] the
level-2 products MER_RR_2PQBCM20060504_101124_xx (MERIS) and a2006124115500.l2_LAC_OC
(MODIS) were retrieved. These images cover the North Sea area, notable the central North Sea and
the coastal water of The Netherlands, Germany and Denmark. The images were collected at 4 May
2006 (UTC 10:14 and 11:55 hours respectively). First the satellite images were re-projected into
Lambert Azimuthal Equal Area by the SNAP geometry processor.
Example 1: Image quality control
In Figure 3 the results of the processor is displayed for part of the MERIS image near the
Dutch coast. The discrete FU-color legend is fully consistent with [Novoa et al., 2013]. Note
that the processor does not process land- or cloud-pixels. The blue at the center of the North
Sea corresponds to FU= 3 and 4, while at the Dutch coast, within the barrier islands and in
Lake IJssel values range from 8 to 14. The total impression is a fine-scale image of complex
changes in composition. In more detail, some aspects in this color image are of interest: In
Fig. 4 (left) the estuaries and parts of the Wadden Sea have a very blue color instead of the
typical CDOM and silt dominated brown water color, which points to error in the
atmospheric correction due to sand banks and nearby land; part of the blue atmospheric
radiation is not taken out sufficiently.
5
Figure 3. Map of MERIS FU values at the North Sea with the discrete FU color scale below.
Fig. 4. Details of MERIS FU values at the Northern German coast (left) and cloud formations near the
west Norwegian coast (right).
The clouds detected algorithm provides the cloud mask of the MERIS image in SNAP.
However, there are always pixels close to recognized clouds that are impacted by residual
water or ice droplets in the atmosphere or limited success of the atmospheric correction
close to clouds. These pixels can be quickly identified, as can be seen in the right panel in
Figure 4. Predominantly at the East to South-East pixels with significant lower hue angle are
found that seem to be correlated with the position of nearby clouds and are not related to
the underlying blue waters. This does not imply that the derived standard products (for
example chlorophyll-a) are wrong, but it should be interpreted as an quick and easy indicator
that products were derived from level-2 spectra that are markedly different from the nearby
pixels.
Example 2: Vicarious calibration MERIS and MODIS
In SNAP it is possible to connect the information from two sensors (Figure 5). To get the
different images in 1 file use: RASTER-GEOMETRIC OPERATIONS-COLLOCATION and define
the master and slave parameters. Scatter plots can be defined in dedicated masks (see
example areas A, B and C) with the mask manager: VIEW-TOOL WINDOWS-MASK MANAGER.
As an example we show some results taken from the hue images of MERIS and MODIS.
6
Figure 5. Collocated images of the hue angle derived for MERIS (left) and MODIS (right).
The results for box A are encouraging, see Figure 6. Although the image of MODIS was taken
101 minutes after the MERIS image, the color of each pixel of the master (in this case MERIS)
correlates well. The data are not perfect on the 1:1 line, which can be due to (small)
differences in the water composition or differences in the calibration of the spectrometers
and atmospheric correction software.
Figure 6. Scatter plot of the hue angle derived from the pixels in box A of Figure 5.
7
Figure 7. Scatter plot of the hue angle derived from the pixels in box B (left) and C (right) of Figure 5.
As expected the correlation deteriorates closer to the coast, see the right panel of Figure 7
that shows the scatter plot in the German Bight. Larger inconsistencies between the two
instruments become apparent. Part could be due to tidal actions (re-suspension, transport)
but also the algorithms of both instruments might show larger deviations in case-2 waters.
Example 3: The color of case-2 waters and their composition
In a third example of the use of color as a research tool, we demonstrate how well color can
be a proxy for the water composition, notably the Chlorophyll-a content. In the PLOS one
article on the FU historical data base[Wernand et al., 2013] we demonstrated, based on
Hydrolight calculations, that for the open ocean type waters (FU= 1 to 9] the color was
related to the Chlorophyll-a content.
Figure 8. MERIS algal-1 product (left) in case-1 waters close to Norway and MERIS algal-2 product
(right) in case-2 waters of the German Bight.
8
Now we can investigate this claim directly in the observations and select more waters and
do an empirical correlation between the hue angle and the chlorophyll-a content, as derived
by the standard processors. In this case we use the MERIS image and the hue angle band
that was calculated by the FU-classification processor. Note that the composition
information comes from two standard processors (algal-1) and (algal-2) that give the
concentration in equivalent Chlorophyll-a concentration (mg m-3).
The left panel of Figure 8 shows the scatter of data in a box near Norway with very blue
waters. Although the extreme oligotrophic waters (FU=1,2) are missing, the relation
between Chlorophyll-a and hue angle is well described, although the spread in the vertical
increases towards lower hue angle. This spread should be investigated and maybe we can
learn something more about intrinsic changes in the composition or the error budget of the
algal-1 processor that is presently estimated at 30% (MERIS ATBD).
The right panel expands the relation for more eutrophic waters in the German Bight. The
Chlorophyll-a concentration is derived with the algal-2 processor. Most pixels have a hue
angle lower than 170 degrees. Again a relation exists, although the uncertainties increase,
likely due to the fact that these waters are variable in composition, notably the Chla and
CDOM concentrations that have a main effect on the color.
References
CIE, 1932. Commission Internationale de l’Éclairage proceedings, 1931. Cambridge University
Press, 1932.
Novoa, S., Wernand, M.R., Van der Woerd, H.J., 2013. The Forel-Ule scale revisited
spectrally: preparation protocol, transmission measurements and chromaticity. J. Eur. Opt.
Soc. Rapid Public. 8 (13057).
Ocean Optics Book, 2016. www.oceanopticsbook.info/view/overview-of-optical-
oceanography/ reflectances
SNAP, 2016. Downloaded from ESA [http://step.esa.int/main/download/]
Van der Woerd, H.J. and Wernand, M.R., 2015. True colour classification of natural waters
with medium-spectral resolution satellites: SeaWiFS, MODIS, MERIS and OLCI. 2015. Sensors
15, 25663-25680. doi: 10.3339/s151025663
Wernand, M.R, Van der Woerd, H.J., Gieskes. W.W.C, 2013. Trends in ocean colour and
chlorophyll concentration from 1889 to present. Plos One 0063766.
Wernand, M.R. et al., 2016. App and SNAP; the color of grace between surface and space.
Ocean Optics conference, Victoria (BC), Canada, October 23-27, 2016.
... However, this exercise is not meant to be a complete analysis of the differences, since this would involve a perfect error budget of each band for each instrument (σ i in Equation (6)). Even with complete knowledge of instrument measurement errors, the accuracy of atmospheric correction is instrument-dependent, introducing errors and bias in the derived R rs [42]. ...
Article
Full-text available
Biomass estimation of multiple phytoplankton groups from remote sensing reflectance spectra requires inversion models that go beyond the traditional band-ratio techniques. To achieve this objective retrieval models are needed that are rooted in radiative transfer (RT) theory and exploit the full spectral information for the inversion. HydroLight numerical solutions of the radiative transfer equation are well suited to support this inversion. We present a fast and flexible Python framework for forward and inverse modelling of multi- and hyperspectral observations, by further extending the formerly developed HydroLight Optimization (HYDROPT) algorithm. Computation time of the inversion is greatly reduced using polynomial interpolation of the radiative transfer solutions, while at the same time maintaining high accuracy. Additional features of HYDROPT are specification of sensor viewing geometries, solar zenith angle and multiple optical components with distinct inherent optical properties (IOP). Uncertainty estimates and goodness-of-fit metrics are simultaneously derived for the inversion routines. The pursuit to retrieve multiple phytoplankton groups from remotely sensed observations illustrates the need for such flexible retrieval algorithms that allow for the configuration of IOP models characteristic for the region of interest. The updated HYDROPT framework allows for more than three components to be fitted, such as multiple phytoplankton types with distinct absorption and backscatter characteristics. We showcase our model by evaluating the performance of retrievals from simulated Rrs spectra to obtain estimates of 3 phytoplankton size classes in addition to CDOM and detrital matter. Moreover, we demonstrate HYDROPTs capability for the inter-comparison of retrievals using different sensor band settings including coupling to full spectral coverage, as would be needed for NASA’s PACE mission. The HYDROPT framework is now made available as an open-source Python package. SEE Link: https://github.com/tadz-io/hydropt
Article
Full-text available
The colours from natural waters differ markedly over the globe, depending on the water composition and illumination conditions. The space-borne "ocean colour" instruments are operational instruments designed to retrieve important water-quality indicators, based on the measurement of water leaving radiance in a limited number (5 to 10) of narrow (≈10 nm) bands. Surprisingly, the analysis of the satellite data has not yet paid attention to colour as an integral optical property that can also be retrieved from multispectral satellite data. In this paper we re-introduce colour as a valuable parameter that can be expressed mainly by the hue angle (α). Based on a set of 500 synthetic spectra covering a broad range of natural waters a simple algorithm is developed to derive the hue angle from SeaWiFS, MODIS, MERIS and OLCI data. The algorithm consists of a weighted linear sum of the remote sensing reflectance in all visual bands plus a correction term for the specific band-setting of each instrument. The algorithm is validated by a set of 603 hyperspectral measurements from inland-, coastal- and near-ocean waters. We conclude that the hue angle is a simple objective parameter of natural waters that can be retrieved uniformly for all space-borne ocean colour instruments.
Article
Full-text available
The Forel-Ule colour comparator scale has been applied globally and intensively by oceanographers and limnologists since the 19th century, providing one of the oldest oceanographic data sets. Present and future Forel-Ule classifications of global oceanic, coastal and continental waters can facilitate the interpretation of these long-term ocean colour data series and provide a connection between the present and the past that will be valuable for climate-related studies. Within the EC-funded project CITLOPS (Citizens’ Observatory for Coast and Ocean Optical Monitoring), with its main goal to empower end users, willing to employ community-based environmental monitoring, our aim is to digitalize the colours of the Forel-Ule scale to establish the colour of natural waters through smartphone imaging. The objective of this study was to reproduce the Forel-Ule scale following the original recipes, measure the transmission of the solutions and calculate the chromaticity coordinates of the scale as Wernand and Van der Woerd did in 2010, for the future development of a smartphone application. Some difficulties were encountered when producing the scale, so a protocol for its consistent reproduction was developed and is described in this study. Recalculated chromaticity coordinates are presented and compared to measurements conducted by former scientists. An error analysis of the spectral and colourimetric information shows negligible experimental errors. [DOI: http://dx.doi.org/10.2971/jeos.2013.1303]
Article
Full-text available
Marine primary productivity is an important agent in the global cycling of carbon dioxide, a major 'greenhouse gas', and variations in the concentration of the ocean's phytoplankton biomass can therefore explain trends in the global carbon budget. Since the launch of satellite-mounted sensors globe-wide monitoring of chlorophyll, a phytoplankton biomass proxy, became feasible. Just as satellites, the Forel-Ule (FU) scale record (a hardly explored database of ocean colour) has covered all seas and oceans - but already since 1889. We provide evidence that changes of ocean surface chlorophyll can be reconstructed with confidence from this record. The EcoLight radiative transfer numerical model indicates that the FU index is closely related to chlorophyll concentrations in open ocean regions. The most complete FU record is that of the North Atlantic in terms of coverage over space and in time; this dataset has been used to test the validity of colour changes that can be translated to chlorophyll. The FU and FU-derived chlorophyll data were analysed for monotonously increasing or decreasing trends with the non-parametric Mann-Kendall test, a method to establish the presence of a consistent trend. Our analysis has not revealed a globe-wide trend of increase or decrease in chlorophyll concentration during the past century; ocean regions have apparently responded differentially to changes in meteorological, hydrological and biological conditions at the surface, including potential long-term trends related to global warming. Since 1889, chlorophyll concentrations have decreased in the Indian Ocean and in the Pacific; increased in the Atlantic Ocean, the Mediterranean, the Chinese Sea, and in the seas west and north-west of Japan. This suggests that explanations of chlorophyll changes over long periods should focus on hydrographical and biological characteristics typical of single ocean regions, not on those of 'the' ocean.
App and SNAP; the color of grace between surface and space
  • M R Wernand
Wernand, M.R. et al., 2016. App and SNAP; the color of grace between surface and space. Ocean Optics conference, Victoria (BC), Canada, October 23-27, 2016.
Commission Internationale de l'Éclairage proceedings
CIE, 1932. Commission Internationale de l'Éclairage proceedings, 1931. Cambridge University Press, 1932.