An efficient method of selecting ocean observing locations for capturing the leading modes and reconstructing the full field

Abstract

We present a method of selecting a small number of observing locations for capturing the leading spatial modes of an ocean field and reconstructing the field, assuming sufficient historical data for extraction of the field's empirical orthogonal functions (EOFs). The selection metric contains two main requirements: (1) The EOFs should have large magnitudes at the selected locations. (2) The cross products between the EOFs should be small at the selected locations. We select observing points sequentially: adding points one by one, without modifying the already picked ones. Selection of each location is by a simple sorting based on the presented metric. We apply the method to the Pacific sea surface temperature (SST) field, using the Smith-Reynolds SST data set and the Tropical Atmosphere Ocean Project (TAO) buoys' SST measurements. We demonstrate that using observations at a small number of selected locations, one can quite accurately estimate the leading modes' amplitudes. Consequently, the full field can be reconstructed by observations at the selected locations. To evaluate the performance gain realized by employing the presented algorithm, we also provide two comparisons. In the first, we confine selections to regions considered logistically accessible. In the second, we use locations of existing buoys. The reconstruction error using observing locations selected by our algorithm is significantly lower than that using the buoys' locations. The presented algorithm takes into account mismatch between the base data set and the observations. Copyright 2008 by the American Geophysical Union.