Satellite-derived time series of Particulate Organic Carbon (POC) the California Current area

Mati Kahru, mkahru@ucsd.edu                                   Updated on:    11/13/2017

 

Abstract

 

Global mapped datasets of various satellite products, including Particulate Organic Carbon (POC) concentration, are available for downloading at the NASA Ocean Color website https://oceancolor.gsfc.nasa.gov/ and other data centers but not at full resolution. For example, mapped SeaWiFS Level-3 products are available at 9 km resolution and MODIS-Aqua products at 4 and 9 km resolutions. Full-resolution data, i.e. approximately 1 km for SeaWiFS and MODISA ocean products, are available at Level-2, i.e. not mapped and not composited over time. This document is an addition to the project that produces full-resolution, mapped and composited satellite time series of the California Current area (here). The standard NASA POC products in Level-2 datasets are mapped to a common 1-km map and then merged from all available sensors: OCTS (1996-1997), SeaWiFS (1997-2010), MODIS-Terra (MODIST, 2000-present), MODIS-Aqua (MODISA, 2002-present), MERIS (RR processed by NASA for 2002 to 7-April-2012), VIIRS-NPP (2-January-2012 to present). Overlapping data from multiple sensors are simply averaged pixelwise. The temporal periods of compositing are 1 day, 5 days, 15 days, 1 month. Software from Wimsoft (http://wimsoft.com), WIM Automation Module (WAM), is used to process Level-2 data files, map, composite and merge the various data products.

Processed data files are available at http://www.wimsoft.com/CAL/. Each dataset has 3 files: the 1-km dataset as float32 in HDF4 (*.hdf), the annotated and log-scaled dataset at reduced 4-km resolution in both HDF4 (*annotated.hdf) and PNG (*annotated.png) formats. Please note that the files are not available through FTP but through HTTP. For downloading multiple files at once you can use the widely available utility wget (e.g. http://www.gnu.org/software/wget/).

 

 

Methods

The standard NASA POC algorithm (Stramski et al 2008) is sensitive to errors in Rrs that are common along cloud edges (see below). These errors cause outliers that are often an order of magnitude higher than neighboring pixels and make it difficult to create longer-term composites as those will be strongly affected by the outliers.

Fig. 1. Example of cloud-edge produced errors in the standard NASA POC product (MODIS, 27-Jan-2017). Obvious errors are marked with red ovals.

 

In order to eliminate errors caused in Rrs at cloud-edges we dilate the cloud mask. As the “cloud” mask actually includes also land, pixels along land are also masked. Below is an example that in  the left has dilate=1, i.e. the cloud mask dilated once that is default for Chl-a processing in our California ocean color time series and  (on  the right) dilate=11, i.e. 11 times that eliminates (masks) the obvious errors. Of course, that also expands missing data are along coast and other clouds that do not show obvious eutliers.as land mask is expanded together with cloud mask. In our POC processing we have adopted dilate=9 as a compromise value for removing most errors.

 

Fig. 2. Effect of dilating cloud mask once (left panel) to dilating 11 times (right panel). Obvious cloud edge errors (inside red circle) were eliminated. Data from   standard NASA POC product (MODIS, 2017, day 304).

 

The datasets are mapped to an Albers Conic equal area projection with 3840 columns and 3405 rows of pixels with an area of 1.0 km2. The latitude and longitude values of all pixels can be obtained from a HDF file named cal_aco_3840_Latitude_Longitude.hdf in http://wimsoft.com/CAL/files/.

A mapped composite is generated for each calendar day using all available passes with a WAM utility wam_l2_map. The following L2 flags (if set) make a pixel invalid when using the default option:  ATMFAIL, LAND, PRODWARN, HIGLINT, HILT, HISATZEN, CLDICE, HISOLZEN, LOWLW, CHLFAIL, CHLWARN, SEAICE, NAVFAIL. In addition to eliminating the flagged pixels, the cloud image determined with the flag CLDICE is dilated (expanded) 9 times to eliminate affected pixels near cloud edges.

If POC data are available from multiple sensors, they are averaged pixelwise. While systematic differences between sensors are expected, they are assumed to be relatively small and their spatial and temporal dependence is not well understand; therefore taking an average is a good compromise. An example of a pixelwise comparison of daily POC from VIIRS to those from MODISA is shown below. Both datasets were averaged for 4x4 pixels to remove small spatial and temporal shifts due to different overpass times during the same day. At POC values below approximately 100 mg m-3 POC data from both sensors agree well but at higher levels POC from VIIRS are mostly higher than those from MODISA.

Fig. 3. Pixelwise comparison of daily standard NASA POC from VIIRS against those from MODISA, days 1-99 of 2016 along a strip from coast to offshore. This plot and the associated match-up datasets were produced with the following command:

wam_pixelwise_match E:\CAL\2016\A2016_poc_day_red4\A20160*Average.hdf E:\CAL\2016\V2016_poc_day_red4\V20160*Average.hdf  E:\Satmasks\California\Mask_strip4_aco960_10x550km.hdf xMin=1 xMax=3 yMin=1 yMax=3

 

References

Stramski, D. et al. (2008). Relationships between the surface concentration of particulate organic carbon and optical properties in the eastern South Pacific and eastern Atlantic Oceans. Biogeosciences, 5(1), 171-01. doi:10.5194/bg-5-171-2008