Optimal Network Design Based on Spatial Sampling Error Study
Jun She 1,2 and S. Nakamoto 1,3
1. Japan Marine Science and Technology Center, 2-15, Natsushima Yokosuka 237 Japan
2. First Institute of Oceanography, 3a, Hongdaozhi Road, Qingdao 266003 P.R.C.
3. Earth Science and Technology Organization, 1-9-9 Roppongi Minato-Ku, Tokyo 106 Japan
Abstract
In this paper, the optimal network design is investigated based on sampling error study with high resolution sea surface temperature (SST) measurement. The logic procedures and the practical examples are demonstrated to determine the optimal sampling distances which correspond to a minimum sampling points of the whole network for an accepted mean sampling error of anomaly SST both from observed dataset and sampling error formula.
Firstly the relationship between mean sampling error and the sampling parameters is analyzed. It is found that the sampling error does not increase monotonously with sampling parameters. When meridional sampling distance varies from 2 degrees to 3 degrees, the sampling error decreased for a prescribed zonal sampling distance.
For a mean sampling error of 4% of the area-average variance, the optimal merid- ional sampling distance derived from observed anomaly SST is 5 degrees. However, the optimal zonal sampling distance is rather inhomogeneous. Warm water areas, equatorial front area and north tropical Pacific need short zonal sampling distance of 5-7 degrees while central near-equatorial Pacific (NEP), south-east NEP and south tropical Pacific have large zonal sampling separations of 11-25 degrees. It is estimated that the whole tropical Pacific areas need 140 in situ sampling points (70 in NEP and 70 in other tropical areas) to obtain a mean sampling error of 4%.
To solve the optimal network design from the formula, a new sampling error formula is developed to simulate sampling error of anomaly SST based on the existing sampling error formula by Nakamoto et al. The previous formula is validated for high-passed SST. It is found that the formula underestimates the sampling error for high-passed SST owing to its simplification in the integration to obtain the variance of the areal average. A new sampling error formula for high-passed SST is derived by improving the integration and then validated. The results agree with the observation better than the previous formula. Another formula is developed to estimate the sampling error of anomaly SST. For the prescribed correlation length scales of high-passed SST
