The Modified Near-IR Correction Algorithm

Wayne Robinson, SAIC
NASA / SeaWiFS project, Code 970.2

Introduction

The Siegel near infrared (NIR) approximation for high chlorophyll water (Siegel et. al., 2000) has helped to retrieve significantly more reasonable chlorophyll estimates in coastal waters from data gathered by the Sea Wide Field of view Sensor (SeaWiFS). This algorithm was implemented for the processing of SeaWiFS data in the third SeaWiFS reprocessing (Robinson et. al., 2000). With the fourth reprocessing, a near-IR correction algorithm based on the reflectance in the 670 nm band (Arnone, 1998) was adopted to perform the same
correction. Although the near-IR algorithm has performed well on the whole, some minor problems have arisen in the way the implementation of the algorithm handles low chlorophyll waters. This chapter discusses these problems in the algorithm implementation in low-chlorophyll water and the modifications made to address those problems.

Near-IR Algorithm Problems

The NIR algorithm uses an iteration during the areosol determination processing to account for the non-zero water-leaving radiance in the NIR bands. The iteration is required because the Lw(670) (perviously chlorophyll) - a product of the aerosol-corrected radiances - is used to determine the NIR water-leaving radiance.

In order to be relatively time-efficient, the current near-IR method assumes an initial chlorophyll of 0.3 when it derives the NIR water-leaving radiances. This is done so that for the majority of low chlorophyll water, no iterations need to be done of the computationally expensive aerosol determination. The initial chlorophyll of 0.3 was used so that the chlorophyll would have a smooth distribution when iterations began above the chlorophyll of 0.3. This approximation was considered safe because the effects of the NIR correction are very small at low chlorophyll concentrations. Although the vicarious calibration to MOBY removed any significant bias caused by this assumption, it meant that the Lwn and chlorophyll would come out slightly different from the computation without the near-IR algorithm.

The changes in the chlorophyll and water-leaving radiances at low chlorophyll concentrations raise some minor problems.

The first problem is that the Lwn and chlorophyll change even in low or no chlorophyll waters when the near-IR algorithm is switched on. This seemed to contradict the thinking that the NIR Lwn is zero at low chlorophyll values. Some SeaWiFS users pointed this out as an unsettling problem.

The second problem is that the vicarious calibration process in use until the third reprocessing performed an iterative approach to arrive at the calibration gain values. A set of gains would be assumed and the SeaWiFS data would be processed to Lwn, at which time they would be compared to the MOBY Lwn. The error in Lwn would be used to produce an adjustment to the gain and the process would be iterated until the error in Lwn was sufficiently small. With an ever-growing set of MOBY-SeaWiFS matchup pairs, the process became time consuming. A simpler method was devised that starts with the MOBY Lwn and adds back all the atmospheric radiance components to derive the top-of-atmosphere radiance. The ratio of this radiance to the SeaWiFS total radiance becomes the gain factor, without the need for iteration. This method cannot easily accomodate the near-IR algorithm iterations.

For these two reasons, it became important to make a modification to the NIR algorithm. An initial solution would be to start the aerosol determination without the NIR effects, ie., using a NIR of 0 and not the NIR for a chlorophyll of 0.3. Unfortunately, this causes a discontinuity to appear in the chlorophyll distribution when transitioning from no iterations at chlorophyll less than 0.3 to one or more iterations at chlorophyll greater than 0.3.

The modified algorithm must satisfy 3 conditions:

There should be no effect on Lwn and chlorophyll for low chlorophyll (< 0.3) and also no effect on the majority of chlorophyll values used at the vicarious calibration site (for an average chlorophyll of 0.08 at the MOBY calibration site, the 0.3 mg m^-3 limit should be sufficient)
No significant artifacts should appear in the chlorophyll distribution as an effect of the algorithm.
The algorithm should not significantly slow down the processing.

The modified Near-IR method

To achieve the goal of having no change of the algorithm at low chlorophyll, the nera-IR algorithm should have no effect - ie have no NIR component at low chlorophyll. At a higher chlorophyll, sufficiently below the 2 mg m^-3, where the chlorophyll has a significant effect, the near-IR algorithm should be used. An abrupt switching to a non zero NIR at a chlorophyll value would result in an abrupt change in the chlorophyll or nLw distribution. Instead of turning on the NIR at a single chlorophyll value, the modified algorithm will use a NIR value that transitions from zero at a low chlorophyll bound to the NIR at a high chlorophyll bound.

In order to get a transition that is as smooth as possible, it would be best to have the transition chlorophyll range fairly large and in a region where the near-IR algorithm's effect on chlorophyll is small. Generally, Siegel found that the chlorophyll error compared to no NIR correction was negative for low chlorophyll values and positive for higher values, with some dependance on the viewing geometry and the chlorophyll algorithm used. In a test of the current OC4v4 algorithm on a typical HRPT scene, it was found that the chlorophyll change from no Siegel to Siegel was the smallest near 1 mg m^-3 (see fig 1 - chl_chg_doc.gif). Thus, a transition from zero NIR to the Siegel NIR was set for the chlor range of .7 to 1.3 mg m^-3. This relatively broad transition, centered in a range where the effect on chlorophyll is less than 2%, should minimize any artifacts in the chlorophyll distribution. A start of the transition at .7 mg m^-3 will also permit a larger range of water conditions in which vicarious calibration can be performed.

All resuts shown presently were performed using the Siegel implementation of the near-IR but should apply equally well to the new algorithm. Tests will be run to verify this for reprocessing 4.

Figure 1. Average (boxes, solid curve) and standard deviation (error bars) of the percent change of the chlorophyll between the Siegel with iterations and no Siegel in use. The ranges are 0.1 mg m^-3 chlorophyll ranges from 0.2 mg m^-3 to 2.0 mg m^-3. The Siegel closely matches the no Siegel around 1 mg m^-3.

Results

A version of the level-1 to 2 processing program was made with the modified Siegel algorithm and the resulting chlorophyll and Lwn were examined and compared to the same data produced from the operational program containing the standard Siegel method. Test data was used from 1 to 3 days of GAC data for the period of 12 - 14 July, 1998 to assure global coverage for the test.

Figure 2 (anal_a_day_doc.gif) shows histograms of the chlorophyll distribution for 3 days of GAC data made using the operational Siegel algorithm and the modified version. The distribution made using the modified Siegel method is smooth and contains no artifacts at the transition chlorophyll values of .7 and 1.3 confirming that this method leaves no artifacts. The distribution also closely follows the operational one.

Figure 2. Chlorophyll distributions for the modified Siegel (solid line) and the current Siegel (dotted line) algorithms. Chlorophyll is taken from 46 level-2 files with global coverage for 12 - 14 July, 1998.

The switch to the modified Siegel method has some small effects on the chlorophyll values. Figure 3 (plt_pdif_chl_doc1.gif) shows the average percentage change in chlorophyll for different chlorophyll value ranges for 16 GAC passes on 12 July 1998. For low to moderate chlorophyll (less than 1 mg m^-3), the chlorophyll from the modified Siegel algorithm is larger than the operational chlorophyll: about 2% higher for the range from 0.02 - 0.5 mg m^-3 and up to 20% for chlorophyll less than 0.02 mg m^-3. For high chlorophyll values, the modified Siegel decreases chlorophyll by up to 18.5%. The average overall increase in chlorophyll is about 1.6%. These are relatively small changes overall in the chlorophyll.

Figure 3. Average of the chlorophyll change from the standard Siegel method to the modified Siegel method as a function of the initial chlorophyll. The difference is compiled from 16 GAC files for the 12th of July, 1998.

The effect on water-leaving reflectances is shown in figure 4 (plt_pdif_chl_doc2.gif). The maximum change in average reflectance occured in the 2-5 mg m^-3 chlorophyll range with a 2x10-4 change in the 412 nm reflectance. This change is roughly 3% of the average reflectance of 6.8 x10^-3. The highest change in reflectance relative to the average reflectance is at 555 nm with an 8% change. This is mainly due to the 555 nm average reflectance being lower. The average change in the reflectance for all the data is shown in the heavy line in figure 4. The maximum change in reflectance is less than 5x10^-5 and the maximum change relative to the average reflectance is 1.8%. It should also be noted that for the higher chlorophyll ranges, the 412 and 443 nm water-leaving radiances were frequently negative. The switch to the modified Siegel algorithm had little effect in the amount of negative water-leaving radiances.

Figure 4. Change in the normalized water-leaving reflectance as a function of wavelength for various ranges of SeaWiFS chlorophyll. Ranges are: .5 - 1 mg m^-3 (astrisk), 1 - 2 mg m^-3 (diamond), 2 - 5 mg m^-3 (triangle), 5 - 10 mg m^-3 (square), 10 - 20 mg m^-3 (plus sign), and 20 - 50 mg m^-3 (x sign). The heavy line is the average reflectance change for all chlorophyll data. Average reflectances for the 5 wavelengths are (using this scale): 68., 63., 50., 34., and 17.

Although some testing is required with 8-day data, the modified Siegel algorithm meets the required goals and has minimal effect on the chlorophyll and water-leaving reflectances, with respect to the current Siegel algorithm. The processing speed of the new algorithm is also acceptable. It is expected that the modified Siegel algorithm will be used for the upcoming fourth SeaWiFS reprocessing.

This modification for the near-IR correction works equally well with the Arnone implementation of the near-IR correction. For a sample of 1000 level-2 files created with the new NIR method, a histogram of the chlorophyll was collected and plotted below. There are no artifacts of the technique in the chlorophyll distribution.

References

Arnone, R. A., P. Martinolich, R.W. Gould,Lr., R. Stumpf, and S. Lander, 1998, "Coastal Optical Properties using SeaWiFS", (a personal communication)

Robinson, W.D., G.M.Schmidt, C.R. McClain, P.J. Werdell, 2000: "Changes Made in the Operational SeaWiFS Processing." In: McClain, C.R., R.A. Barnes, R.E. Eplee, Jr., B.A. Franz, N.C. Hsu, F.S. Patt, C.M. Pietras, W.D. Robinson, B.D. Schieber, G.M. Schmidt, M. Wang, S.W. Bailey, and P.J. Werdell, 2000: SeaWiFS Postlaunch Calibration and Validation Analyses, Part 2. NASA Tech. Memo. 2000- 206892, Vol. 10, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 12-28.

Siegel, D.A., M. Wang, S. Maritorena, and W. Robinson, 2000: "Atmospheric correction of satellite ocean color imagery: the black pixel assumption", Appl. Opt., 39., 20 Jul 2000, pp 3582-3591.