Experimental SST Statistics using Adaptive QC (Release 2.0)

1. Introduction

   An experimental set of sea surface temperature (SST) statistics for 2° latitude x 2° longitude boxes was completed in 2002 using a new "adaptive" quality control (QC) procedure. An initial stage of the new procedure was to define the large-scale SST signal over a 1961-90 base period, when the data are relatively dense and homogeneous. The procedure improved over the existing ICOADS climatological "trimming" (Wolter, 1997) by incorporating large time and space climate signals (e.g., El Niño and La Niña ) into the QC. However, the product is still considered experimental, and since its completion ICOADS has been extensively updated (thus it does not represent the latest ICOADS data).

   This procedure (Smith and Reynolds, 2003) was adapted for ICOADS processing, and applied to the entire 1784-1997 period of Release 2.0 observations (LMRF format). A flag (SQZ) was output for each extant SST observation within LMRF defining the relationship of the SST to the QC limits in 0.5 standard-deviation (sigma) increments. Data outside 3.0 sigma limits were rejected; however, because the SQZ values were rounded to the nearest 0.5 sigma, this actually represents a slightly broader (3.25 sigma) cutoff. If SST data were missing, or outside broad physical limits (-5 to 40C), SQZ was output but left blank.

2. Monthly summary product description

   Only group 3 (SST, air temperature, and specific and relative humidity) in the Monthly Summary Groups (MSG) format was produced, and only the 10 SST statistics are extant (those for the other three variables are missing) (this special product is archived at NCAR and can be made available upon request). The data mixture input for this product was the same as that for the "enhanced" ICOADS statistics (ships plus other platform types). This contrasts with the "standard" statistics (ship data only). A further difference should be noted: data outside 4.5 sigma (3.5 sigma) trimming limits are rejected from the existing enhanced (standard) statistics.

3. Comparison results

   For each LMRF report, Table 1 shows the relationship between the old trimming flag (SF) and the new adaptive flag (SQZ).

   The following figures provide a variety of information about the new flags and comparisons with other existing products:

   Figures showing the distributions of the old and new QC flags:

   Figure 1. Distribution of 1784-1997 old trimming (SF) flags.

   Figure 2. Distribution of 1784-1997 new adaptive (SQZ) flags.

   Discussion: For the enhanced statistics, 1.56% of the SST data were trimming for being outside 4.5 sigma limits. For the new adaptive QC, 1.34% of the SST data were rejected for being outside 3 sigma limits. (The sigma values are not comparable across the two procedures--tighter limits can be used with the adaptive procedure because the large-scale climate signal is removed from the data.)

   January gif animations over selected 30-year periods (and a shorter 18-year period at the end), showing the global spatial coverage of the old enhanced trimming (top) versus the new adaptive QC (bottom). The color/greyscale bar shows the ratio of QC'd over untrimmed observations. Thus red indicates that all observations were removed by QC, blue indicates that few or none were, and greyscale values fall in between:

   Figure 3a. 1830-1859.

   Figure 3b. 1860-1889.

   Figure 3c. 1890-1919.

   Figure 3d. 1920-1949.

   Figure 3e. 1950-1979.

   Figure 3f. 1980-1997.

   Discussion: Systematic data problems (possible mislocated data), such as in January 1972 off Madagascar in some cases were resolved by both procedures. The new adaptive QC shows coverage improvements over areas without the old trimming limits, e.g., east of Cape Horn.

   El Niño examples. For a Pacific region, the three panels compare resultant SST December monthly mean values for the old enhanced trimming (top), and the new adaptive QC (middle). The bottom panel is a comparison with the Smith and Reynolds Extended Reconstructed SST (ERSST; Smith and Reynolds, 2003):

   Figure 4a. 1972.

   Figure 4b. 1982.

   Figure 4c. 1997.

   La Niña examples:

   Figure 5a. 1970.

   Figure 5b. 1975.

   Figure 5c. 1988.

   Discussion: For each example event, the new procedure generally fills in some boxes that were missing in the enhanced product, possibly due to rejection by the old trimming of genuine extreme values. The analyzed ERSST product is much smoother and preserves less detail.

4. Status

   Following completion of this product, the LMRF observations were converted into the newer International Maritime Meteorological Archive (IMMA) format. During this process the SQZ flag associated with each SST observation was stored in IMMA, together with an accompanying SQA flag providing a measure of the reliability of the QC (SQA has a roughly inverse relationship with the number of observations available nearby, such that smaller alpha values indicate more data). Although this makes the adaptive QC flags available to observational users, it is important to note that the adaptive processing was only applied to the earlier Release 2.0 data, and data subsequently updated or replaced will not have the adaptive QC flags (i.e., only a subset of the currently available data includes the adaptive flags).

References

Smith, T.M. and R.W. Reynolds, 2003: Extended reconstruction of global sea surface temperatures based on COADS data (1854-1997). J. Climate, 16, 1495-1510.

Wolter, K., 1997: Trimming problems and remedies in COADS. J. Climate, 10, 1,980-1,997.

[Documentation and Software][Links to additional]