The quality of and characteristics of errors in surface temperature analyses in global and regional data assimilation systems

Tom Hamill

NOAA/ESRL Physical Sciences Laboratory

Thursday, Aug 01, 2019, 1:30 pm
DSRC Room 1D403

Abstract

This hour-long seminar will be comprised of two parts. In the first part, we consider differences between operational analyses of global surface temperature from NCEP, ECMWF, Canada, and Great Britain. There are large differences in the surface temperature analyses, especially during the cold season over snow pack. These differences are large enough to make one question the validity of using them for forecast verification. A method of decomposition of the error is proposed to try to determine how much of the analysis differences are systematic and how much they are random. The results indicate that there are substantial systematic errors in the analyses and thus substantial room for improvement through improvement of the background forecast and assimilation procedures.

The second part applies a relatively simple method of benchmarking the background, 1-h forecasts that are used in rapidly updated surface temperature analyses such as NOAA's High-Resolution Rapid Refresh (HRRR) system. The benchmark is based on a simple persistence of the anomaly from a diurnally and seasonally dependent climatology The benchmarking procedure is used to indicate whether HRRR 1-h surface-temperature forecasts in warm-season are competitive with the benchmark. While HRRR forecasts in 2015 were not competitive, its 2018 forecasts are much improved. It is hypothesized that this is because of the introduction of a simplified coupled data assimilation procedure, whereby surface temperature and humidity observation increments are used to make adjustments to the soil state.