Download presentation
Presentation is loading. Please wait.
Published byHarry Mervyn Rose Modified over 8 years ago
1
NOAA-08: An Optimal Atmospheric Dataset for Algorithm Training and Covariance Matrix Generation Kevin Garrett, Sid-Ahmed Boukabara, and Fuzhong Weng 10 th Specialist Meeting on Microwave Radiometry and Remote Sensing of the Environment
2
Outline Goal/Motivation The Optimal Dataset Algorithm Training Impact on a 1D-Variational Retrieval Summary
3
Goal/Motivation Goal –To create a refined dataset which encompasses a multitude of individual datasets for use with: Regression-based retrieval algorithm training applied to satellite measurements (sounding) Covariance-matrix generation as a constraint to variational retrievals Motivation –No one dataset can represent ground truth to 100% accuracy; each have inherent issues –Need for maximizing performances of satellite-based retrievals when comparing to multiple independent datasets –Requires an optimal dataset which represents a broad range of observational systems and cancels out their limitations
4
An Optimal Dataset Extensive array of datasets available –Observation –Forecast –Re-analysis –Satellite EDRs –Climate (archived) Support wide range of application –Validation –NWP initialization –Satellite-based retrievals Algorithm training Retrieval constraint
5
An Optimal Dataset Caveats inherent to each dataset impact application, analysis, use, etc. Microcosm: Radiosonde Training on just one RAOB instrument type would effect retrieval performances on full RAOB set. Training on all RAOB instrument types would increase retrieval performances. AVK- BAR Vaisala RS80
6
An Optimal Dataset Macrocosm: Training on one dataset (e.g. Radiosonde) is not optimal for comparison to other datasets (e.g. ECMWF, GDAS etc.), and perhaps even to that dataset itself. SHOW: MIRS retr. Perf wrt GDAS trained on GDAS and ICDB SHOW: MIRS retr. Perf wtr ICDB trained on GDAS and ICDB
7
An Optimal Dataset Components of NOAA-08 –Integrated Global Radiosonde Archive (IGRA) 2006-2008 Highly quality-controlled (QC) Further QC by profile depth, profile gaps –NOAA-88 Highly QC’d radiosonde dataset –Global Data Assimilation System (GDAS) Sampled days from each season Y2007 QC for super-saturated profiles –European Centre for Medium-Range Weather Forecasts 40- year re-analysis (ERA-40) 60L-SD QC dataset of ERA-40 dataset 13495 sampled profiles encompassing natural variability, global/seasonal coverage
8
An Optimal Dataset NOAA-08 Components (continued) –DropSonde Representative of mid-upper troposphere in cloudy/precipitating atmospheres –Mesoscale Model 5 (MM5) Forecast model representative of tropospheric cloudy/precipitating atmospheres
9
An Optimal Dataset Representation/Variability (Statistics) –Number of profiles from each data type –Representativeness of time of year, time of day –Distributions of T, Q, CLW of oc,si,ld,sn
10
Applicability – Algorithm Training Create regression algorithm for retrieval of atmospheric and surface parameters –Use profiles to simulate brightness temperatures (TBs) at NOAA-18 AMSU/MHS frequencies –Calculate regression coefficients/correlation based on atm/sfc parameter, TBs at each channel, scan angle, and latitude of profile –Use algorithm to perform regression-based retrieval on independent N18 AMSU/MHS measurements and compare to collocated GDAS and radiosonde. –Compare to retrievals using algorithms trained on just GDAS, or just radiosonde, or NOAA-88.
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.