Online Observation of Non-Gaussian and Lognormal Probability Distributions within GFS Forecasts
Abstract
Many operational data assimilation and retrieval systems assume that the variables and errors come from a Gaussian distribution. Previous research has indicated this is not necessarily true, specifically for water vapor mixing ratio and temperature. This was done by employing a suite of hypothesis tests to time series of forecast observations. These tests included the Jarque-Bera test, the Shapiro-Wilk test, the Chi-squared goodness-of-fit test, and a composite test which incorporated the results of the former tests. These testing procedures have been updated from a so-called "static" version that requires a complete time series to an "online" system that employs a sliding window to allow variables to be analyzed in near real-time for non-Gaussian characteristics. For this study, results will be presented using 0.5 degree six hour forecasts from the 0, 6, 12, and 18Z analyses from the National Oceanic and Atmospheric Administration (NOAA) Global Forecast System (GFS) from 2016 and 2017. These results indicate that mixing ratio and temperature are often non-Gaussian, and, moreover exhibit a lognormal probability distribution. This indicates the purpose and value of a mixed lognormal-Gaussian distribution retrieval system (C1DOE) that is presented in the accompanying poster.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFMNG21B0923F
- Keywords:
-
- 3315 Data assimilation;
- ATMOSPHERIC PROCESSES;
- 0555 Neural networks;
- fuzzy logic;
- machine learning;
- COMPUTATIONAL GEOPHYSICS;
- 1640 Remote sensing;
- GLOBAL CHANGE;
- 3275 Uncertainty quantification;
- MATHEMATICAL GEOPHYSICS