The presentation will review Professor Phillips' many contributions to data assimilation. His work on initialisation in the late 1950s identified key issues that needed to be resolved for primitive equation models. The linear formulation he developed at this time presaged the approach of Machenhauer and of Baer and Tribbia in the mid 1970s. Indeed, with insight, it appears that Phillips came within an ace of formulating a non-linear Normal Mode Initialisation, a suggestion firmly denied by Professor Phillips.

Professor Phillips' interest in the Russian literature on numerical weather prediction led him to a deep study of the work of Gandin and others on the Optimum Interpolation analysis technique. Professor Phillips used the technique to estimate the analysis errors arising from the operational networks of the 1960s and to estimate the downstream consequences for forecasters. One conclusion from his study was that the problem of medium-range forecasting is more difficult for Europe than for any other part of the Northern Hemisphere.

Professor Phillips was entrained into the use of satellite data for numerical weather prediction soon after the launch of TIROS-N in 1979. The first problem he addressed was the identification of the effects of heavy rain in the MSU data. This led to a collaboration with NESDIS which helped advance our understanding and our capabilities in using satellite sounding data in numerical weather prediction.

In the 1980s Professor Phillips provided much stimulus to the study of the problems of sequential estimation. The determination of the statistics of forecast errors is a key issue in this area. Professor Phillips' theoretical study of the statistics of an ensemble of Rossby waves showed remarkable agreement with empirical results from elsewhere, thus providing a secure theoretical under-pinning for a great deal of work on assimilation since the mid 1980s.

The presentation will conclude with a summary of recent developments flowing from Professor Phillips' pioneering work in several areas of data assimilation.