Table of contents

Rationale Input data Parameters Zones Quality control items Visual inspection Time series Histograms Scatter plots Anomalies

Rationale

Following the MERIS Science Advisory Group recommendation, ESA started the generation of some MERIS level 3 demonstration products, using the MKL3 tool developed by ACRI-ST. The MKL3 tool is implemented on the GRID On Demand Processing chain at ESA and the MERIS level 3 products are generated routinely and distributed to the user on the web.

Although the validation of level 3 data cannot replace the validation of level 2 data, it may contribute to the estimation of the measurements quality by providing information:

• where in-situ data does not exist (and will maybe never exist)
• at a global scale almost every day
• at daily, monthly, seasonal and annual temporal scales almost everywhere
• on the long-term instrument stability
• on instrument ageing

The statistical analysis of the spatial variations of the level 3 data combined with oceanographic knowledge can lead us to a certain level of validation or invalidation of the products. Of course, the huge number of available points in the level 3 products increase the quality of the statistics.

Comparisons of level 3 products between sensors can be also of great interest to estimate the coherency between the couples instrument + processing as well as to characterise some possible inter-calibration between the products, for example before any merging attempt.

Input data

The analysis has been applied to the MERIS monthly level 3 products available on the ESA web site (1/12° sinusoidal grid) and the MODIS and SeaWiFS level 3 products available on the NASA Ocean Colour web site (both at 9 km on an Plate-Carré grid). The MERIS products have been reprojected on the same 9km Plate-Carré grid before any computation (using the Sutherland-Hodgeman area clipping and flux-conserving algorithms).

• MERIS level 3 products
• MODIS level 3 products
• SeaWiFS level 3 products

Parameters

The following parameters are covered by the quality control tasks:

• chlorophyll-a, case 1 water
• normalised water leaving radiances (nLw) at 412, 443, 490, 510 and 560 nm
• aerosol optical thickness over water at 865 nm
• Angstrom coefficient over water at 865 nm

Note that there is not a full coherency between the MERIS, MODIS and SeaWiFS parameters:

• CHL1 is not obtained from the water leaving radiances using the same sets of equations for the three sensors. MERIS processing stage level is "Q". MODIS processing level is 1.1. SeaWifs processing level is 5.2.
• the spectral bands are not identical: MERIS sensor measures at 412, 443, 490, 510 and 560 nm while MODIS sensor measures at 412, 443, 488 and 551 nm (no 510 nm band) and SeaWiFS observes at 412, 443, 490, 510 and 555 nm,
• the aerosol optical thickness is computed at 865 nm for MERIS and SeaWiFS and 869 nm for MODIS
• the angstrom coefficient is provided at 865 nm for MERIS, 531 nm for MODIS and 510 nm for SeaWiFS.

An inter-calibration correction is applied to the MODIS and SeaWiFS CHL1 daily L3 products in order to get compatible concentrations with respect to the MERIS sensor.
Validity limits for this conversion: 0.01 < CHL1_{MODIS/SeaWiFS} < 13
A reverse formulation could be applied to the merged CHL1 products in order to get SeaWiFS-like CHL1 concentrations:
Validity limits for this conversion: 0.017 < CHL1_MERIS< 50

Zones

The statistical analysis has been performed at global level and on a set of dedicated zones. These zones are usually visible at all time of the year, except AtlS55 below 50°S and not fully visible during North Hemisphere summer and AtlN55 above 50°N and not fully visible during North Hemisphere winter.

• Global scale: Glob (all latitudes, any water depth), Glob50 (latitudes between 50°S and 50°N, any water depth), GlobDW (all latitudes, only deep water - z>1000m), Glob50DW (latitudes between 50°S and 50°N, only deep water), GlobEW (all latitudes, eutrophic water case), GlobMW (all latitudes, mesotrophic water case), GlobOW (all latitudes, oligotrophic water case), GlobSW (all latitudes, shallow water case),
• Zonal areas: slices of 10° from 50°S to 50°N between 170°W and 150°W.
• Regional diagnostic sites are defined from existing lists (cf. ClimZoo final report and Brian Franz methodology, 2005), including the Hawaii region, where the vicarious calibration of MODIS and SeaWiFS sensors is performed.

Quality control items

Visual inspection

All MERIS level 3 products are downloaded (netCDF+XML files) and converted in TIF format. The TIF files are opened with the openev software.

 
PNG files are also generated, which can be opened with Google Earth. The original MERIS level 3 products at 1/12° have been resampled on a 0.25° Plate-Carré grid using a flux-conserving algorithm and plotted using a parameter-specific colour scale. All products available for any month have been packaged in a single kmz file accessible from a dedicated web page.

Time series

Time series of monthly mean values are plotted for MERIS sensor alone and for the three sensors. Mean values are computed inside the various zones for the common valid bins of the three sensors (even for MERIS alone analysis).

Following the recommendations of the IOCCG group (see Chapter 8 of the IOCCG report 4, 2004), a weighted arithmetic mean formulation is used for all the parameters, taking also into account the fact that the bins are defined in a Plate-Carré grid (i.e. including a weighting factor equal to cos(Φ) where Φ is the average bin latitude.

with An exemple of time serie plot is shown below, where we can see the temporal evolution of the monthly mean value of the normalised water leaving radiance at 490 nm provided by the three sensors inside the PacN45 zone.

Histograms

Chlorophyll histograms are build from the three sensors monthly averaged data. They display the distribution frequency of the chlorophyll data set in each region. The histogram classes are defined in logarithmic scale and the actual size of each bin of the Plate-Carré grid is taken into account (histogram classes are increased by cos(Φ) instead of 1). Histogram frequencies are computed using only the common valid bins of the three sensors.

Some statistical information (mean value, coefficient of variation) is provided in MERIS-only histogram plots. Note that the mean value is computed here as the weighted geometrical mean (i.e. average of the logarithmic value of the chlorophyll concentration.

The standard deviation measures the dispersion of the parameters around the geometrical mean. As for the mean value, we use the weighted standard deviation definition based on log(CHL):
with The coefficient of variation is another measure of the dispersion of a data set. It is defined as the ratio of the standard deviation to the mean. It is often expressed in %. Note that the coefficient of variation is meaningless when the mean value is close to 0.
The histograms are normalised on each plot (i.e. all histogram frequencies are divided by the highest one). The normalisation is applied simultaneously to the three histograms on the MERIS/MODIS/SeaWiFS histogram plot (i.e. all histogram frequencies are divided by the highest one of the three sensors).

Histogram difference plots display the absolute difference between the histograms of MERIS or MODIS sensor and SeaWiFS.

Scatter plots

2D scatter plots are used to compare two variables representing the same set of data: CHL1 concentrations measured by various sensors. By construction, only the common valid bins are used to build the plot.

The following scatter plots are available:

• MERIS CHL1 versus MODIS CHL1,
• MERIS CHL1 versus SeaWiFS CHL1,
• MODIS CHL1 versus SeaWiFS CHL1,

A density colour scale is applied to the plot points to highlight the regions of high frequencies.

Statistical information is provided on each plot.

• Slope of the linear regression line,
• Intercept of the linear regression line,
• Correlation coefficient,
• Root Mean Square Deviation,
• Bias.

To perform a linear regression on a scatter plot of two measurement data sets, we have selected the reduced major axis regression (RMA) rather than the ordinary least square method (OLS) as both data sets involved are measurements affected by errors. The RMA method minimises the sum of the horizontal and vertical distances between the data set points and the regression line. Note that the regression is performed in a logarithmic scale for chlorophyll.

The slope b of the linear regression line is given by:

with The intercept a is:
The correlation coefficient is:
The root mean square deviation (or RMSD) and bias give a statistical measure of the magnitude of the absolute difference between two data sets coming from different origins (sensors, processors...). Note that the RMSD and bias quantities are computed in logarithmic scale for the chlorophyll concentration and in linear scale for all other parameters.

Linear scale:

Logarithmic scale:

Anomalies

MERIS chlorophyll (CHL1) monthly anomaly maps are available. They are based on the 6 years of available MERIS data. In a first step, climatological monthly products are computed from the MERIS monthly level 3 products. Then, the anomaly maps are produced by subtracting the climatological monthly averages from the monthly products.

The difference is expressed in a dual-log color scale between -10 and +10 milligrams per cubic meter.

The grey color represents land areas. The black color represents areas where no data is available (e.g. latitude limitation due to observation condition, cloud cover, etc). Blue/purple colors indicate that the monthly average is lower than the climatology while yellow/red colors indicate a higher value. White color is used when the difference is not significant (1.e-3 mg/m3).