Spatial Sampling Study for the Tropical Pacific with Observed Sea Surface Temperature Fields

View More View Less
  • 1 Ocean Research Department, Japan Marine Science and Technology Center, Yokosuka. Japan
© Get Permissions
Full access

Abstract

Using weekly averaged sea surface temperature (SST) data with a horizontal resolution of 1° latitude and 1° longitude during 1984-93, basic statistics [such as mean-square error (MSE), correlation length scales, etc.] for unfiltered, low-passed (including interannual change and monthly mean climatology) and high-passed (with a timescale shorter than 1 yr) fields are analyzed. Sampling error formalism proposed by Nakamoto et al. was examined for its practical application.

It is found that the MSE caused by the low-frequency-high-frequency interference term is so small that MSE of unfiltered data can be expressed as the sum of MSEs from low-frequency parts and high-frequency parts. The root-mean-square error (rmse) from unflitered data is smaller than 0.3°C except in the near-coastal regions. A maximum rmse belt is found in the equatorial front area in eastern near-equatorial Pacific. The spatial patterns of rmse for unfiltered, high- and low-passed datasets are similar and also similar to those of high-frequency correlation length scales, especially zonal correlation scales. For high-frequency data, the zonal length scales vary from 5° to 30°, and meridional scales from 4° to 16°. Small zonal scales less than 15° are mainly found in the equatorial front and warm water area.

By subsampling from the SST dataset with TOGA-TAO (Tropical Ocean and Global Atmosphere-Tropical Atmosphere Ocean) buoy array sampling parameters, sampling error fields are estimated from SST data and the formalism, respectively. In the latter, sampling error depends on the design parameters and observation derived length scales only. The authors found that the formalism can simulate the observed sampling error fields when stationary and homogeneous conditions are satisfied. Length scales and averaging time am the other two main factors that decide the spatial sampling error besides design parameters. The correlations between length scales and sampling error and between averaging time and sampling error are studied for a high-frequency dataset, which also match the formalism quantitatively.

For the TOGA-TAO buoy array, the spatial sampling error for high-frequency monthly averaged data is about 10%-20% of intraannual area-averaged variance in the 10°S-10°N area, and the high value is in the northeast near-equatorial Pacific. However, if we focus on the ratio of high-frequency MSE to the area-averaged variance of interannual variation, a high value greater than 5% appears mainly in the warm water area and the coastal area, while the other interior ocean is of the ratio of less than 5%.

Abstract

Using weekly averaged sea surface temperature (SST) data with a horizontal resolution of 1° latitude and 1° longitude during 1984-93, basic statistics [such as mean-square error (MSE), correlation length scales, etc.] for unfiltered, low-passed (including interannual change and monthly mean climatology) and high-passed (with a timescale shorter than 1 yr) fields are analyzed. Sampling error formalism proposed by Nakamoto et al. was examined for its practical application.

It is found that the MSE caused by the low-frequency-high-frequency interference term is so small that MSE of unfiltered data can be expressed as the sum of MSEs from low-frequency parts and high-frequency parts. The root-mean-square error (rmse) from unflitered data is smaller than 0.3°C except in the near-coastal regions. A maximum rmse belt is found in the equatorial front area in eastern near-equatorial Pacific. The spatial patterns of rmse for unfiltered, high- and low-passed datasets are similar and also similar to those of high-frequency correlation length scales, especially zonal correlation scales. For high-frequency data, the zonal length scales vary from 5° to 30°, and meridional scales from 4° to 16°. Small zonal scales less than 15° are mainly found in the equatorial front and warm water area.

By subsampling from the SST dataset with TOGA-TAO (Tropical Ocean and Global Atmosphere-Tropical Atmosphere Ocean) buoy array sampling parameters, sampling error fields are estimated from SST data and the formalism, respectively. In the latter, sampling error depends on the design parameters and observation derived length scales only. The authors found that the formalism can simulate the observed sampling error fields when stationary and homogeneous conditions are satisfied. Length scales and averaging time am the other two main factors that decide the spatial sampling error besides design parameters. The correlations between length scales and sampling error and between averaging time and sampling error are studied for a high-frequency dataset, which also match the formalism quantitatively.

For the TOGA-TAO buoy array, the spatial sampling error for high-frequency monthly averaged data is about 10%-20% of intraannual area-averaged variance in the 10°S-10°N area, and the high value is in the northeast near-equatorial Pacific. However, if we focus on the ratio of high-frequency MSE to the area-averaged variance of interannual variation, a high value greater than 5% appears mainly in the warm water area and the coastal area, while the other interior ocean is of the ratio of less than 5%.

Save