CoAuthors

Jeffrey Early (NorthWest Research Associates, United States)

Mapped sea surface height (SSH) anomaly products from satellite altimetry are already an integral part of modern oceanography, widely used in both scientific and operational contexts. However, we note three significant limitations of the present mapped products and distribution system. 1) The correlation function imposed by the most common mapped products is physically unrealistic and inconsistent with the underlying along-track observations. This correlation function, originally proposed in 1985, has good statistical properties in general but artificially steepens spectral slopes in mapped SSH products, essentially throwing away information about small scales. 2) Effective resolution, errors and uncertainties in mapped products are often misunderstood by oceanographers, leading to mis-application of altimetry products. 3) Scientists often require the SSH field on a custom grid, and either apply additional interpolation or improvise their own mapping procedures.

We attempt to address these three issues. Our approach is based on standard Gaussian process regression methods with a carefully chosen parametric covariance function that is consistent with observations. Covariance parameters (space and time decorrelation scales, spectral slopes, and propagation speeds) are estimated by Maximum Likelihood Estimation, and will be presented as part of the mapped data product. This presentation will primarily focus on the third issue. We introduce an open-source, reproducible tool that allows statistical SSH mapping on custom grids. The final distributed products are fully open source and flexible to meet the needs of scientists. The data are not constrained to a fixed grid, but will be provided in functional form, reproducing the interpolation and its error estimate at any point in time.