Events / Media

Data Assimilation Seminar Series

Mr. Sam Hatfield (Aug. 16, 2017, 15:30-)

Affiliation Oxford University
Title How low can you go? Reducing the precision of data assimilation to improve forecast skill
Abstract I will discuss the potential benefits of reducing precision within data assimilation. Data assimilation is inherently uncertain due to, for example, the use of noisy observations and imperfect models. Thus, rounding errors incurred from reducing precision may be within the tolerance of the system. Lower precision arithmetic is cheaper, and so by reducing precision in ensemble data assimilation, computational resources can be redistributed towards, for example, a larger ensemble size.

I will present results using the SPEEDY intermediate complexity GCM with a local ensemble transform Kalman filter (LETKF). I will discuss challenges and opportunities for the use of half precision arithmetic and compare the impact of rounding and model errors. Then, I will compare results from a double precision SPEEDY/LETKF system with an equivalent-cost reduced precision system, using a larger ensemble. Trading precision for more ensemble members could provide a boost to weather forecasting skill. Additionally, I will give some preliminary findings on the benefits of running the LETKF at single precision.
PDF Link to PDF file