Extension of MODIS Ocean Processing Capabilities to Include
the 250 & 500-meter Land/Cloud Bands

Bryan Franz
NASA Ocean Biology Processing Group
21 June 2006


The 36 spectral channels of the MODIS instrument were selected to support observation of clouds, land, and oceans. The traditional channels used for ocean color observation are the 9 bands in the 412-869 nm spectral regime, with a spatial resolution of 1000-meters at nadir. These ocean bands were designed with high sensitivity over the dynamic range of reflectances typical over open oceans, including contributions from the surface and the atmosphere. Over highly turbid coastal and inland waters, it is possible for this dynamic range to be exceeded, such that the bands saturate and the true signal is unkown. Other bands on MODIS were specifically designed for land and cloud observations, with both increased spatial resolution and reduced sensitivity over a broader dynamic range. These land/cloud bands overlap the spectral range of the ocean bands and extend into the short-wave infrared (SWIR), from 469 nm to 2130 nm. The ocean processing code developed by the OBPG, the Multi-Sensor Level-1 to Level-2 code (l2gen), has been extended to support these additional bands. The primary purpose of this effort is to provide a mechanism for exploring the potential value of the increased spectral information, as well as the higher spatial resolution and saturation limits of the land/cloud bands, for application to coastal and inland waters. The extended band suite is shown in the Table 1 below. For completeness, the additional thermal bands used for 4um and 11-12um SST retrievals are shown in Table 2.

It must be emphasized that our goal in this effort is to provide the capability to the research community for investigating potential applications of the additional spectral bands for ocean remote sensing. Within the OBPG, we will not be producing any standard geophysical products that require the 250 and 500-meter bands. Our primary effort has been to characterize the instrument response (e.g., polarization sensitivities, relative spectral response functions) in the additional channels, in a manner consistent with what we have done for the standard ocean bands, and to develop the appropriate software and tables to facilitate retrieval of oceanic optical properties at the additional spectral channels. We have also developed some mechanisms for utilizing the increased spatial resolution, and we have added additional options to our atmospheric correction algorithm for utilizing the SWIR information. We are providing these additional capabilities to the research community through SeaDAS, with the hope that it will encourage evaluation and application development.

Table 1: Extended MODIS Band Suite for Oceans

Band Number Wave Length (nm) Band Width (nm) Spatial Resolution (m) Signal-to-Noise SNR Radiance
Saturation Radiance
8 412 15 1000 880 4.49
9 443 10 1000 838 4.19
3 469 20 500 243 3.53
10 488 10 1000 802 3.21
11 531 10 1000 754 2.79
12 551 10 1000 750 2.10
4 555 20 500 228 2.90
1 645 50 250 128 2.18
13 667 10 1000 910 0.95
14 678 10 1000 1087 0.87
15 748 10 1000 586 1.02
2 859 35 250 201 2.47
16 869 15 1000 516 0.62
5 1240 20 500 74 0.54
6 1640 * 35 500 275 0.73
7 2130 50 500 110 0.10
* the 1640 band on MODIS-Aqua is non-functional

Table 2: Thermal Bands Used for SST

Band Number Wave Length (nm) Spatial Resolution (m) NEdT (K) Radiance
22 3.9 1000 0.07 0.067
23 4.0 1000 0.07 0.079
31 11 1000 0.07 0.955
32 12 1000 0.05 0.894

Input Level-1B Products

The standard MODIS Level-1B format divides the calibrated radiances fields into three separate files corresponding to the three distinct spatial resolutions of 250, 500, and 1000-meters, with filename identifiers QKM, HKM, and 1KM used to distinguish the quarter, half, and 1-kilometer variants, respectively. The 1KM file is also sometimes called LAC, based on historical conventions. The spectral bands associated with each file are provide in Table 3. Note that the HKM file also includes the two 250-meter bands (marked with * in Table 3). The radiances from the 250-meter bands are averaged to 500-meter spatial resolution, and written to special "aggregated" fields in the HKM file. Similary, the radiances from the 250 and 500-meter bands are averaged to 1000-meter resolution and stored in the 1KM file.

Table 3: Spectral bands as distributed in each Level-1B file
* aggregated to 500-meters, ** aggregated to 1000-meters

645 nm 645 nm * 645 nm **
859 nm 859 nm * 859 nm **
469 nm 469 nm **
555 nm 555 nm **
1240 nm 1240 nm **
1640 nm 1640 nm **
2130 nm 2130 nm **
412 nm
443 nm
488 nm
531 nm
551 nm
667 nm
678 nm
748 nm
869 nm
3.9 um
4.0 um
11 um
12 um

Level-2 Processing

As discussed in the User's Guide, l2gen supports the processing of observed radiances from a variety of sensors, with the option to ouput a host of geophysical products. Two of the sensors supported by l2gen are MODIS on Aqua and MODIS on Terra, identified as MODISA and MODIST, respectively. To these we now add HiRes MODIS variants HMODISA and HMODIST. It is important to understand that these are treated as separate sensors within l2gen, with separate atmospheric tables and default parameter sets. This was done to simplify maintenance of standard MODIS ocean color processing, during the development and evaluation of HiRes MODIS capablities. l2gen determines which sensor it is processing by examining the input file.

Spatial Resolution and Geolocation

If the HKM and QKM files are provided, l2gen now has the ability to process at 500 or 250-meter resolution. This is controlled through a new input parameter called resolution, which recognizes values of 250, 500, 1000. When processing at 250-meters, l2gen bilinearly interpolates the 500-meter and 1000-meter band radiances to 250-meter resolution as well, so that the full band set is co-registered. It should be recognized, however, that only the 645 and 859-nm channels are truly at 250-meter resolution. Processing at 250-meter resolution requires the existance of all three Level-1B files: QKM, HKM, and 1KM (or LAC). When processing at 500-meter resolution, the QKM file is not required, as the code will use the aggregated fields within the HKM file for the 645 and 859-nm radiances, as well as the standard 500-meter channels, and it will interpolate the 1000-meter radiances from the 1KM file to 500-meter resolution. When processing at 1000-meter resoution, only the 1KM (or LAC) file is required, as this contains all spectral channels in native or aggregated form.

The processing of MODIS from Level-1B calibrated radiances to Level-2 geophysical products also requires a separate geolocation file. The geolocation file can be obtained from the Goddard DAAC or generated from the Level-1A files using SeaDAS. This file defines the centers of the 1km pixels, so l2gen will perform appropriate interpolations when processing at higher resolutions. There is no HKM or QKM equivalent geolocation file. The interpolation follows the methodology outlined by Liam Gumley, University of Wisconsin, as presented here (section 5). All path geometry (solar and sensor view zenith and azimuth) are similarly interpolated, as are the radiances when going from lower to higher spatial resolutions.

Input File Rules

The parameter "ifile" is used to specify the input Level-1 file to l2gen. However, for MODIS processing at resolutions above 1000-meters, 2 or more input files are required. l2gen locates these additional files by assuming some standard naming conventions are being maintained. It is assumed that the three files are identical in filename, varying only in the existence the letters QKM, HKM, and 1KM (or LAC).

New Atmospheric Correction Capabilities

The additional spectral channels in the NIR and SWIR provide an opportunity to explore a variety of new atmospheric correction options that may be of particular value in highly reflective, turbid waters. The standard atmospheric correction approach used of global ocean processing makes use of the NIR bands at 748 and 869nm to determine aerosol type and concentration. The method requires a priori knowledge of the water-leaving radiance in these longer wavelengths. In Morel case 1 waters, where the spectral distribution of water-leaving radiances can be assumed to vary with chlorophyll concentration alone, a simple iteration scheme (i.e., Stumpf 2002) can be used to model and predict the water-leaving radiances in the NIR. However, in turbid waters this case 1 assumption does not hold, so water-leaving radiances in the NIR are difficult or impossible to estimate, and the resulting aerosol determination based on the NIR will be corrupted and erroneous. In contrast, water is so strongly absorbing in the SWIR spectral regime that even highly reflective turbid waters appear black at these longer wavelengths. Following the recent work by Wang & Shi (2005), the SWIR bands on MODIS can be used to determine aerosol type and concentration. This information can then be used either to predict the water-leaving radiance in the NIR (thereby allowing the standard algorithm to procede), or the aerosol determination from the SWIR can be extrapolated all the way to the visible. These capabilities have been incorporated into l2gen to facilitate evaluation. It should be noted, however, that the signal-to-noise in the SWIR bands (Table 1) is quite low, and this may be a limiting factor in any advantage gained by using the SWIR. The aerosol correction options for l2gen are described in Table 6.

Table 6: Relevant Aerosol Correction Options in l2gen

aer_wave_short lower wavelength used for aerosol model selection 748
aer_wave_long upper wavelength used for aerosol model selection and aerosol concentration 869
aer_swir_short lower SWIR wavelength used for aerosol model selection 1240
aer_swir_long upper SWIR wavelength used for aerosol model selection and aerosol concentration 2130
aer_opt Option for aerosol calculation mode.
1 - 12 extrapolation to visible with fixed model (aer_opt) with concentration from aer_wave_long
-1 Model selection using aer_wave_short and concentration from aer_wave_long
-3 Model selection using aer_wave_short and concentration from aer_wave_long,
with iteration for non-zero water-leaving radiance
-9 Model selection using aer_wave_short and concentration from aer_wave_long,
with NIR water-leaving radiance determined using aer_swir_short and aer_swir_long

With the above options, it is possible to test a multitude of processing permutations. For example, the aer_wave_short and aer_wave_long can be set to 1240 and 1640, respectively, with aer_opt set to -1, and the aersol correction will be determined using the two SWIR bands rather than the standard NIR bands. Setting aer_wave_long to 2130 and aer_opt to 4 will yield aerosol concentration determined at 2130nm (where water is very dark), with the resulting aerosol reflectance extrapolated to the visible via the fixed model (maritime 90%, see User's Guide). Setting aer_wave_long to 859 and aer_opt to 4 will yield aerosol concentration determined with the 250-meter channel at 859nm, with the result extrapolated to the visible via the same fixed model (yielding true 250-meter aerosol retrievals). The aer_wave_short and aer_wave_long can be set to any sensor wavelength (though only wavelengths longward of 600nm make any sense). The additional parameters aer_swir_short and aer_swir_long are only relevant to aer_opt=-9, where the aerosol determination in the SWIR is used to estimate water-leaving radiance in the NIR. This is basically the Wang & Shi (2005) approach.

Gaseous Absorption

One issue with the SWIR bands is high sensitivity to water-vapor and CO2. New corrections were developed to account for these effects. In l2gen, these corrections can be enable via the gas_opt parameter.

gas_opt=gaseous transmittance bitmask selector (default: 1)
1: Ozone
2: CO2
4: NO2
8: H2O
0: no correction

Note that gas_opt is a bit mask. For example, to enable Ozone and H2O (water-vapor) correction, set gas_opt=9 (i.e., 1+8). The water-vapor and ozone concentrations are determined from ancillary files. A method to determine water-vapor directly from the MODIS bands is in development. The correction for NO2 is also still in development (currently not active).

Artifacts and Caveats

Image Striping

User's of standard Level-2 MODIS ocean data are familiar with the cross-track striping artifacts that are sometimes visible in the imagery. This effect is due to imperfect relative corrections between the 10 along-track detectors associated with each 1KM band. For the HKM and QKM bands, 20 and 40 detectors are distributed along-track to provide the higher along-track resolution. As such, cross-track striping artifacts at 20 and 40 line intervals can occur for the HKM and QKM bands, respectively. This problem is potentially exacerbated by the reduced sensitivity (higher digitization error) of each detector over the dynamic range of ocean observations. In addition, within each physical scan of MODIS, the HKM detector sets are sampled at double the rate of the 1KM, and the QKM detectors are sampled at four times the rate. This temporal sub-sampling is what provides the cross-track pixel resolution. Slight variation in this temporal subsampling rate, and imperfect reset of the sampling registers, can give rise to vertical (along-track) striping in the higher resolution bands. Corrections have been applied to correct for this effect, but the corrections are never perfect. In conclusion, both along-track and cross-track striping can be seen in the HKM and QKM bands, when viewed at native resolution.

File Size & Processing Time

When processing at higher resolutions, or when including a large number of products in the Level-2 output, file size can exceed 2GB. It is strongly recommended that users limit the processing to the minimum region of interest. The best way to do this is to use the MODIS subscene tool, available in SeaDAS, to extract a subscene from a Level-1A file. The Level-1A extract can than be processed to Level-1B and Level-2. Alternatively, the full Level-1B can be processed to Level-2, but the output can be limited to a subregion of the granule (via l2gen parameters sline, eline, spixl, epixl). The former method is more efficient, as the latter requires some level of processing on the full file before it can be reduced to the subscene. In either case, for the same region, processing at 250-meter resolution will generally increase processing time by a factor of 16 relative to processing at 1000-meter resolution.


Gao, B.-C., M. J. Montes, Z. Ahmad, and C. O. Davis (2000). Atmospheric correction algorithm for hyperspectral remote sensing of ocean color from space, Appl. Opt., 39, 887-896. Gordon, H.R., and Wang, M. (1994). Retrieval of water-leaving radiance and aerosol optical-thickness over the oceans with SeaWiFS - a preliminary algorithm. Appl. Opt., 33, 443-452.

Li, Rong-Rong, Yoram J. Kaufman, Bo-Cai Gao, and Curtiss O. Davis (2003). Remote Sensing of Suspended Sediments and Shallow Coastal Waters, IEEE Transaction on Geoscience and Remote sensing, Vol. 41, No. 3 PP. 559.

O'Reilly, J.E., Maritorena, S., Mitchell, B.G., Siegel, D.A., Carder, K.L., Garver, S.A., Kahru, M., and McClain, C. (1998). Ocean color chlorophyll algorithms for SeaWiFS. J. Geophys. Res., 103, 24937-24953.

Wang, M. and W. Shi (2005). Estimation of ocean contribution at the MODIS near-infrared wavelengths along the east coast of the U.S.: Two case studies, Geophys. Res. Lett., 32, L13606, doi:10.1029/2005GL022917.