spacer
spacer
spacer
top of banner nasa logo and GSFC title, go to GSFC home    + NASA Homepage  
spacer

Evaluation Summary of Mandiva Chain Test

The L1B and the downstream Land and Atmosphere products from the MODAPS system may be created on computers running any one of three different operating systems: Irix, Mandrake Linux, and Mandriva Linux (2008). The Irix and the Mandriva (2008) products are nearly identical, but the Mandrake products differ in a small percentage of the pixels. The reason for this is that the compilers for these three operating systems each have slightly different implementations for floating point operations. As a result, the floating point numbers come out slightly different on each platform.

However most of the MODIS L1B and downstream L2/L3/L4 land and atmosphere products are not stored as floating point values, but as scaled integer (SI) values with a floating point scale and offset. When converted to an integer, a very small difference in a floating point value can, in a small number of cases, cause the resulting integer to be rounded either to a higher or a lower integer, causing a larger difference in the integer values. In addition to the difference caused by the floating point operation, minor differences introduced in the L1B output, often as small as 1 count, could be slightly amplified by the downstream land and atmosphere PGEs by certain non-linear operations involving fixed threshold. However overall the differences are negligible and happen to very small percentage of global data and hence should not be of any concern.

MODAPS recently ran a 32-day chain test of all the land PGEs on two processing systems with different operating systems: mandrake and the mandriva-2008. Following table contains percentage of global retrievals with a relative absolute difference of 1% (relative to the values retrieved by mandrake) or less in values between the products produced from these two platforms. This analysis confirms that the number of pixels affected is very small and the difference is within the calibration error. Please note that since a difference of 1 or 2 counts in data values is generally present at most pixels, and so an absolute difference of 2 or less at any pixel is truncated to zero. The global browse images of the land products from the two processing systems are available at L630 test browse page.


Mandriva test results - Daily products

All daily product results are from 1 day worth of global data, from day 2005042.

ESDTSDS NameResult Summary
MOD09 250m Band 1 reflectance 99.9999% within 1% (less than 50,000 out of approximately 6 billion pixels has difference >1%.)
250m Band 2 reflectance 99.9999% within 1% (less than 80,000 out of approximately 6 billion pixels has difference >1%.)
500m Band 3 reflectance 99.9999% within 1% (about 25,000 out of approximately 1.7 billion pixels has difference >1%.)
500m Band 4 reflectance 99.9999% within 1% (about 25,000 out of approximately 1.7 billion pixels has difference >1%.)
500m Band 5 reflectance 99.9999% within 1% (about 17,000 out of approximately 1.7 billion pixels has difference >1%.)
500m Band 6 reflectance 99.9999% within 1% (less than 4,000 out of approximately 1.7 billion pixels has difference >1%.)
500m Band 7 reflectance 99.9999% within 1% (less than 4,000 out of approximately 1.7 billion pixels has difference >1%.)
MOD09CMG Band 1 CMG reflectance 99.9999% within 1% (about 1,000 out of approximately 26 million pixels has difference >1%.)
Band 2 CMG reflectance 99.9999% within 1% (about 1,000 out of approximately 26 million pixels has difference >1%.)
Band 3 CMG reflectance 99.9999% within 1% (about 1,100 out of approximately 26 million pixels has difference >1%.)
Band 4 CMG reflectance 99.9999% within 1% (about 1,100 out of approximately 26 million pixels has difference >1%.)
Band 5 CMG reflectance 99.9999% within 1% (about 850 out of approximately 26 million pixels has difference >1%.)
Band 6 CMG reflectance 99.9999% within 1% (about 175 out of approximately 26 million pixels has difference >1%.)
Band 7 CMG reflectance 99.9999% within 1% (about 185 out of approximately 26 million pixels has difference >1%.)
MOD10A1 Snow_Cover_Daily_Tile 100% of data were within 1%
Snow_Albedo_Daily_Tile 100% of data were within 1%
Fractional_Snow_Cover 100% of data were within 1%
MOD11 LST 99.9999% within 1% (less than 30 out of 780 million + pixels has difference >1%.)
Emis_31 99.9999% within 1% (less than 15,000 out of 780 million + pixels has difference >1%.)
Emis_32 99.9999% within 1% (less than 15,000 out of 780 million + pixels has difference >1%.)
MOD15A1 1km FPAR 99.9999% within 1% (about 2500 out of about 400 million pixels has difference >1%.)
1km LAI 99.9999% within 1% (about 375 out of approximately 26 million pixels has difference >1%.)
MOD29 Sea Ice by reflectance 99.9999% match (less than 300 out of approximately 275 million pixels do not match)
Ice surface temperature 100% of data were within 1%
MOD29E1D Sea Ice by reflectance - NP 99.9999% match (less than 100 out of approximately 20 million pixels do not match)
Sea Ice by reflectance - SP 99.9999% match (less than 10 out of approximately 20 million pixels do not match)
Ice surface temperature - NP 100% of data were within 1%
Ice surface temperature - SP 100% of data were within 1%


Mandriva test results - n-day products

Day 2005033

ESDTSDS NameResult Summary
MOD09A1 Band 1 reflectance 99.9999% within 1% (about 110,000 out of approximately 1.7 billion pixels has difference >1%.)
Band 2 reflectance 99.9999% within 1% (less than 93,000 out of approximately 1.7 billion pixels has difference >1%.)
Band 3 reflectance 99.9999% within 1% (about 135,000 out of approximately 1.7 billion pixels has difference >1%.)
Band 4 reflectance 99.9999% within 1% (about 150,000 out of approximately 1.7 billion pixels has difference >1%.)
Band 5 reflectance 99.9999% within 1% (about 86,000 out of approximately 1.7 billion pixels has difference >1%.)
Band 6 reflectance 99.9999% within 1% (less than 44,000 out of approximately 1.7 billion pixels has difference >1%.)
Band 7 reflectance 99.9999% within 1% (less than 44,000 out of approximately 1.7 billion pixels has difference >1%.)
MOD10A2 Maximum snow extent 99.9999% of pixels match (about 27,000 out of approximately 1.7 billion pixels do not match.)
MOD11A2 LST 1km Day 99.9999% within 1% (about 2,400 out of approximately 456 million pixels has difference >1%.)
LST 1km night 99.9999% within 1% (about 5,400 out of approximately 456 million pixels has difference >1%.)
MOD13A1 NDVI 99.9999% within 1% (about 52,000 out of approximately 1.7 billion pixels has difference >1%.)
EVI 99.9999% within 1% (less than 100,000 out of approximately 1.7 billion pixels has difference >1%.)
MOD13A2 NDVI 99.9997% within 1% (less than 100,000 out of approximately 420 million pixels has difference >1%.)
EVI 99.9998% within 1% (about 70,000 out of approximately 420 million pixels has difference >1%.)
MOD14A2 Fire Mask 99.9999% of pixels match (about 30 out of approximately 410 million pixels do not match.)
MOD15A2 FPAR 99.9999% within 1% (less than 6,700 out of approximately 410 million pixels has difference >1%.)
LAI 99.9999% within 1% (less than 2,200 out of approximately 410 million pixels has difference >1%.)
MOD17A2 GPP 99.9998% within 1% (less than 60,000 out of approximately 410 million pixels has difference >1%.)
PSNNet 99.9998% within 1% (less than 60,000 out of approximately 410 million pixels has difference >1%.)

For details on test and evaluation please contact Sadashiva Devadiga .


footer image with FirstGov logo image, go to FirstGov web page

+ NASA Privacy Statement, Disclaimer,
and Accessibility Certification

  footer image with NASA logo image, go to NASA home page

Curator: Demi Feng
NASA Official: Edward Masuoka
Last Updated: Mar. 21, 2013

spacer