GOCE Gravity Field Recovery Using Massive Parallel Computing

Abstract

The Gravity Field and Steady-State Ocean Circulation Explorer (GOCE) mission has been selected by the Earth Sciences Advisory Committee (ESAC) of the European Space Agency (ESA) as the first of two core missions of ESA’s Earth Explorer Programme to be launched in 2004. The main objective of the GOCE mission is to provide a high-accuracy high-resolution global model of the Earth’s static gravity field and the geoid from a combination of spaceborne gravity gradiometry (SGG) and high-low satellite-to-satellite tracking (hl-SST). Of the order of a hundred thousand gravity field parameters have to be estimated from tens of millions measured gravity gradients, which makes a brute force approach in terms of a least-squares solution of the observation equations an enormous numerical task. This huge number of observations and gravity field parameters does not allow to set-up the design matrix and the normal matrix explicitly. In turn this implies that existing software libraries for the platform of choice cannot be used directly. Therefore, we discuss a (parallellized) data processing approach for SGG data on the CRAY T3E supercomputer at Delft University of Technology, which only requires the application of the design matrix or its transposed to some vectors. Moreover, we discuss the performance of the algorithm in terms of degree of parallelism and scalability properties based on simulations.