Enhancement of the ASI algorithm

Lars Kaleschke, 9/2001

A major drawback of the 85 GHz SSM/I channels compared to the 19 and 37 GHz channels is the far more opaque atmosphere at 85 GHz. This can cause severe biases in ice concentrations obtained from 85 GHz SSM/I data, particularly in cloud-covered areas.
In order to overcome the problem of false high ice concentrations caused by clouds or water vapor a correction term is introduced to the (ARTIST SEA ICE algorithm (ASI)).

ARTIST SEA ICE Algorithm Version 3.2 (ASI-v3.2)

The ASI-v3.2 algorithm uses fixed tiepoints for sea ice (P1=7.5 K) and open water (P0=47 K).

ARTIST SEA ICE Algorithm Version 4.1 (ASI-v4.1)

ASI-v4.1 utilizes adaptive tie points for each pixel. The differences between ASI-v3.2 and reference (NTA) ice concentrations are used to calculate new tie points for each pixel.

The proposed new method includes the following steps:

1. Calculate ice concentrations C(ASI-v3.2) with fixed tie points.
2. Calculate ice concentrations C(NTA).
3. Degrade (e.g. gaussian smoothing) the resolution of C(ASI-v3.2) to those of C(NTA).
4. Calculate differences DC=C(NTA)-C(ASI-v3.2)
5. Calculate new tiepoints P1new=P1-DC*1/(1+b/a)*d.
6. Calculate ice concentrations C(ASI-v4.1) using P0 and P1new.

The empirical factor d (d=0.25; ASI-v4.1, d=0.19; ASI-v4.2) was invented in order to obtain a positive bias because the NTA underestimates the ice concentration in the interior ice regions. It is not a bug it is a feature ;-)

First Results

ASI-v3.2 (old)

Please take a closer look at the corrupt clusters at high ice concentrations and compare them to the new version.

ASI-v4.1 (new)

Summary

The deviations of NASA-TEAM and ASI-v4.1 sea ice concentration data has been significantly reduced as compared with those calculated with ASI-v3.2. The correlation coefficients have been significantly increased.

Correlation coefficients (NASA-TEAM and different ASI versions).

Region CC (v3.2) CC (v4.1)
Greenland Sea0.982 0.991
Central Arctic0.917 0.975
Laptev Sea0.963 0.987
Weddell Sea0.977 0.987
Ross Sea0.985 0.993
Bellinghausen Sea0.983 0.987
Davis Sea0.990 0.993

TODO

There are still some nasty errors due to the gridding process. These errors (osciallations from surface gridding) could be removed easily by using a finer grid cell size. Unfortunately, this would increase the computational effort. (Please wait until we get a more powerful computer than our old 400 MHz PC ;-)

The bootstrap or the new NASA-TEAM-2 algorithm should also be used instead of NASA-TEAM for comparisons.