Oahe Basin SWE Comparison

Oahe Basin SWE Comparison

The following sets of images and plots contain 2 sets of comparisons between Corps of Engineers data and SSM/I-derived SWE. For each date, the upper left subimage contains Army SWE data (converted from inches to mm) for the Oahe basin. Each subsequent subimage is one of the passive microwave algorithm outputs for the same date, masked for the basin area only. SSM/I algorithms include:

  1. NSIDC1: Chang SMMR algorithm (combination of horizontally polarized channels) modified for use with SSM/I channels
  2. NSIDC3: Chang's AMSR algorithm, a decision tree built on top of the basic horizontal channel difference algorithm
  3. GW: Goodison & Walker algorithm (combination of verticlaly polarized channels)
  4. Nagler: Thomas Nagler algorithm, H-pol channels, uses 85 GHz
  5. GB: Grody & Basist, snow extent only
The red outline in the passive microwave images is the outline of the Army data SWE extent. The plots contain scatter plots of the Army data vs. each algorithm output. Only SWE values are considered (i.e. "wet" snow flags and pixels with missing TBs are eliminated).

  1. Army Survey SWE vs. SSM/I
  2. Army Survey SWExSCA (survey SWE multiplied by AVHRR-derived snow-covered area mask) vs. SSM/I
General notes:

  1. The comparison data for Survey x SCA for SSM/I data from March 1 was dated March 4_3, indicating that survey data for March 1 were crossed with SCA data from the closest dates possible, March 4-6 (ref. e-mail from Emily to Mary Jo).
  2. Scatter plots were created by regridding the Army data to a 25-km EASE-Grid, using drop-in-the-bucket average interpolation. Resulting Army SWE pixels were compared with non-missing SSM/I-derived SWE for that pixel.
  3. Nagler algorithm outputs depth. I converted this to SWE (mm) using a factor of 3, so that images are compared quickly visually.
  4. Statistical correlation values for SWE improved from ~0.25 for the comparison to the survey data to ~0.75 for the comparison to the survey data multiplied by the AVHRR-derived snow-covered area fractions. Note that the prominent vertical artifact at about 50 mm Army SWE in the Survey comparisons is removed in the Survey X SCA comparisons.
  5. For the two dates compared (March 1 and 15), the SWE algorithms performed consistently, relative to one another. In both cases, the decreasing order of correlation (best to worst) was
    1. GW (Goodison & Walker)
    2. Nagler
    3. NSIDC1
    4. NSIDC3
  6. Due to the differences and theoretical improvements in NSIDC3 versus NSIDC1, Richard and I expected NSIDC3 to improve the correlation when compared with NSIDC1, but this was not the case. I don't yet know why. Then again, difference in r values of 0.755 vs. 0.743 probably isn't significant?
  7. GW was better than all of the rest, not particularly surprising, since this area is very similar to the Canadian prairies, so that algorithm really ought to do well, here.
  8. Hard to say much about GB (Grody-Basist) since it's only a snow cover algorithm. It puts down too much snow when compared to Survey SWExSCA.

M. J. Brodzik <brodzik@nsidc.org>
Last modified: Thu Mar 29 12:13:26 2001