All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 265 119 8
PDF Downloads 166 81 8

Monitoring of Observation and Analysis Quality by a Data Assimilation System

View More View Less
  • 1 European Centre for Medium Range Weather Forecasts, Reading, U.K.
Restricted access

Abstract

The purpose of this paper is to demonstrate the ability of a modern data assimilation system to provide long-term diagnostic facilities to monitor the performance of the observational network. Operational data assimilation systems use short-range forecasts to provide the background, or first-guess, field for the analysis. We make a detailed study of the apparent or perceived error of these forecasts when they are verified against radiosondes. On the assumption that the observational error of the radiosondes is horizontally uncorrelated, the perceived forecast error can be partitioned into prediction error, which is horizontally correlated, and observation error, which is not. The calculations show that in areas where there is adequate radiosonde coverage, the 6-hour prediction error is comparable with the observation error.

This statement is discussed from a number of viewpoints. We demonstrate in the Northern Hemisphere midlatitudes, for example, that the forecasts account for most of the evolution of the atmospheric state from one analysis to the next, so that the analysis algorithm needs to make only a small correction to an accurate first-guess field; the situation is rather different in the Southern Hemisphere. If the doubling time for small errors is two days, then analysis error will amplify by less than 10% in 6 hours.

This being the case, the statistics of the forecast/observation differences have a simple statistical structure. Large variations of the statistics from station to station, or large biases, are indicative of problems in the data or in the assimilation system. Case studies demonstrate the ability of simple statistical tools to identify systematically erroneous radiosonde wind data in data sparse, as well as in data rich areas, errors which would have been difficult to detect in any other way. The statistical tools are equally effective in diagnosing the performance of the assimilation system.

The results suggest that it is possible to provide regular feedback on the quality of observations of winds and heights to operators of radiosonde networks and other observational systems. This capability has become available over the last decade through improvements in the techniques of numerical weather analysis and prediction.

Abstract

The purpose of this paper is to demonstrate the ability of a modern data assimilation system to provide long-term diagnostic facilities to monitor the performance of the observational network. Operational data assimilation systems use short-range forecasts to provide the background, or first-guess, field for the analysis. We make a detailed study of the apparent or perceived error of these forecasts when they are verified against radiosondes. On the assumption that the observational error of the radiosondes is horizontally uncorrelated, the perceived forecast error can be partitioned into prediction error, which is horizontally correlated, and observation error, which is not. The calculations show that in areas where there is adequate radiosonde coverage, the 6-hour prediction error is comparable with the observation error.

This statement is discussed from a number of viewpoints. We demonstrate in the Northern Hemisphere midlatitudes, for example, that the forecasts account for most of the evolution of the atmospheric state from one analysis to the next, so that the analysis algorithm needs to make only a small correction to an accurate first-guess field; the situation is rather different in the Southern Hemisphere. If the doubling time for small errors is two days, then analysis error will amplify by less than 10% in 6 hours.

This being the case, the statistics of the forecast/observation differences have a simple statistical structure. Large variations of the statistics from station to station, or large biases, are indicative of problems in the data or in the assimilation system. Case studies demonstrate the ability of simple statistical tools to identify systematically erroneous radiosonde wind data in data sparse, as well as in data rich areas, errors which would have been difficult to detect in any other way. The statistical tools are equally effective in diagnosing the performance of the assimilation system.

The results suggest that it is possible to provide regular feedback on the quality of observations of winds and heights to operators of radiosonde networks and other observational systems. This capability has become available over the last decade through improvements in the techniques of numerical weather analysis and prediction.

Save