Spatial Statistical Estimation for Massive Sea Surface Temperature Data

Abstract:

We combine several large remotely sensed sea surface temperature (SST) datasets to create a single high-resolution SST dataset that has no missing data and provides an uncertainty associated with each value. This high resolution dataset will optimize estimates of SST in critical parts of the world’s oceans, such as coastal upwelling regions. We use Spatial Statistical Data Fusion (SSDF), a statistical methodology for predicting global spatial fields by exploiting spatial correlations in the data. The main advantages of SSDF over spatial smoothing methodologies include the provision of probabilistic uncertainties, the ability to incorporate multiple datasets with varying footprints, measurement errors and biases, and estimation at any desired resolution. In order to accommodate massive input and output datasets, we introduce two modifications of the existing SSDF algorithm. First, we compute statistical model parameters based on coarse resolution aggregated data. Second, we use an adaptive spatial grid that allows us to perform estimation in a specified region of interest, but incorporate spatial dependence between locations in that region and all locations globally. Finally, we demonstrate with a case study involving estimations on the full globe at coarse resolution grid (30 km) and a high resolution (1 km) inset for the Gulf Stream region.