Four Dimensional Variational Inversion of Atmospheric Chemical Sources in WRFDA
Atmospheric aerosols are known to affect health, weather, and climate, but their impacts on regional scales are uncertain due to heterogeneous source, transport, and transformation mechanisms. The Weather Research and Forecasting model with chemistry (WRF-Chem) can account for aerosol-meteorology fe...
Main Author: | |
---|---|
Format: | Text |
Language: | unknown |
Published: |
CU Scholar
2016
|
Subjects: | |
Online Access: | https://scholar.colorado.edu/mcen_gradetds/132 https://scholar.colorado.edu/cgi/viewcontent.cgi?article=1132&context=mcen_gradetds |
id |
ftunicolboulder:oai:scholar.colorado.edu:mcen_gradetds-1132 |
---|---|
record_format |
openpolar |
spelling |
ftunicolboulder:oai:scholar.colorado.edu:mcen_gradetds-1132 2023-05-15T15:17:51+02:00 Four Dimensional Variational Inversion of Atmospheric Chemical Sources in WRFDA Guerrette, Jonathan J. 2016-01-01T08:00:00Z application/pdf https://scholar.colorado.edu/mcen_gradetds/132 https://scholar.colorado.edu/cgi/viewcontent.cgi?article=1132&context=mcen_gradetds unknown CU Scholar https://scholar.colorado.edu/mcen_gradetds/132 https://scholar.colorado.edu/cgi/viewcontent.cgi?article=1132&context=mcen_gradetds Mechanical Engineering Graduate Theses & Dissertations data assimilation inverse modeling least squares optimization randomization source attribution Atmospheric Sciences Computer Sciences Environmental Chemistry text 2016 ftunicolboulder 2018-10-07T09:02:07Z Atmospheric aerosols are known to affect health, weather, and climate, but their impacts on regional scales are uncertain due to heterogeneous source, transport, and transformation mechanisms. The Weather Research and Forecasting model with chemistry (WRF-Chem) can account for aerosol-meteorology feedbacks as it simultaneously integrates equations of dynamical and chemical processes. Here we develop and apply incremental four dimensional variational (4D-Var) data assimilation (DA) capabilities in WRF-Chem to constrain chemical emissions (WRFDA-Chem). We develop adjoint (ADM) and tangent linear (TLM) model descriptions of boundary layer mixing, emission, aging, dry deposition, and advection of black carbon (BC) aerosol. ADM and TLM model performance is verified against finite difference derivative approximations. A second order checkpointing scheme is used to reduce memory costs and enable simulations longer than six hours. We apply WRFDA-Chem to constraining anthropogenic and biomass burning sources of BC throughout California during the 2008 Arctic Research of the Composition of the Troposphere from Aircraft and Satellites (ARCTAS) field campaign. Manual corrections to the prior emissions and subsequent inverse modeling reduce the spread in total emitted BC mass between two biomass burning inventories from a factor of x10 to only x2 across three days of measurements. We quantify posterior emission variance using an eigendecomposition of the cost function Hessian matrix. We also address the limited scalability of 4D-Var, which traditionally uses a sequential optimization algorithm (e.g., conjugate gradient) to approximate these Hessian eigenmodes. The Randomized Incremental Optimal Technique (RIOT) uses an ensemble of TLM and ADM instances to perform a Hessian singular value decomposition. While RIOT requires more ensemble members than Lanczos requires iterations to converge to a comparable posterior control vector, the wall-time of RIOT is x10 shorter since the ensemble is executed in parallel. This work demonstrates that RIOT improves the scalability of 4D-Var for high-dimensional nonlinear problems. Overall, WRFDA-Chem and RIOT provide a framework for air quality forecasting, campaign planning, and emissions constraint that can be used to refine our understanding of the interplay between atmospheric chemistry, meteorology, climate, and human health. Text Arctic black carbon Human health University of Colorado, Boulder: CU Scholar Arctic |
institution |
Open Polar |
collection |
University of Colorado, Boulder: CU Scholar |
op_collection_id |
ftunicolboulder |
language |
unknown |
topic |
data assimilation inverse modeling least squares optimization randomization source attribution Atmospheric Sciences Computer Sciences Environmental Chemistry |
spellingShingle |
data assimilation inverse modeling least squares optimization randomization source attribution Atmospheric Sciences Computer Sciences Environmental Chemistry Guerrette, Jonathan J. Four Dimensional Variational Inversion of Atmospheric Chemical Sources in WRFDA |
topic_facet |
data assimilation inverse modeling least squares optimization randomization source attribution Atmospheric Sciences Computer Sciences Environmental Chemistry |
description |
Atmospheric aerosols are known to affect health, weather, and climate, but their impacts on regional scales are uncertain due to heterogeneous source, transport, and transformation mechanisms. The Weather Research and Forecasting model with chemistry (WRF-Chem) can account for aerosol-meteorology feedbacks as it simultaneously integrates equations of dynamical and chemical processes. Here we develop and apply incremental four dimensional variational (4D-Var) data assimilation (DA) capabilities in WRF-Chem to constrain chemical emissions (WRFDA-Chem). We develop adjoint (ADM) and tangent linear (TLM) model descriptions of boundary layer mixing, emission, aging, dry deposition, and advection of black carbon (BC) aerosol. ADM and TLM model performance is verified against finite difference derivative approximations. A second order checkpointing scheme is used to reduce memory costs and enable simulations longer than six hours. We apply WRFDA-Chem to constraining anthropogenic and biomass burning sources of BC throughout California during the 2008 Arctic Research of the Composition of the Troposphere from Aircraft and Satellites (ARCTAS) field campaign. Manual corrections to the prior emissions and subsequent inverse modeling reduce the spread in total emitted BC mass between two biomass burning inventories from a factor of x10 to only x2 across three days of measurements. We quantify posterior emission variance using an eigendecomposition of the cost function Hessian matrix. We also address the limited scalability of 4D-Var, which traditionally uses a sequential optimization algorithm (e.g., conjugate gradient) to approximate these Hessian eigenmodes. The Randomized Incremental Optimal Technique (RIOT) uses an ensemble of TLM and ADM instances to perform a Hessian singular value decomposition. While RIOT requires more ensemble members than Lanczos requires iterations to converge to a comparable posterior control vector, the wall-time of RIOT is x10 shorter since the ensemble is executed in parallel. This work demonstrates that RIOT improves the scalability of 4D-Var for high-dimensional nonlinear problems. Overall, WRFDA-Chem and RIOT provide a framework for air quality forecasting, campaign planning, and emissions constraint that can be used to refine our understanding of the interplay between atmospheric chemistry, meteorology, climate, and human health. |
format |
Text |
author |
Guerrette, Jonathan J. |
author_facet |
Guerrette, Jonathan J. |
author_sort |
Guerrette, Jonathan J. |
title |
Four Dimensional Variational Inversion of Atmospheric Chemical Sources in WRFDA |
title_short |
Four Dimensional Variational Inversion of Atmospheric Chemical Sources in WRFDA |
title_full |
Four Dimensional Variational Inversion of Atmospheric Chemical Sources in WRFDA |
title_fullStr |
Four Dimensional Variational Inversion of Atmospheric Chemical Sources in WRFDA |
title_full_unstemmed |
Four Dimensional Variational Inversion of Atmospheric Chemical Sources in WRFDA |
title_sort |
four dimensional variational inversion of atmospheric chemical sources in wrfda |
publisher |
CU Scholar |
publishDate |
2016 |
url |
https://scholar.colorado.edu/mcen_gradetds/132 https://scholar.colorado.edu/cgi/viewcontent.cgi?article=1132&context=mcen_gradetds |
geographic |
Arctic |
geographic_facet |
Arctic |
genre |
Arctic black carbon Human health |
genre_facet |
Arctic black carbon Human health |
op_source |
Mechanical Engineering Graduate Theses & Dissertations |
op_relation |
https://scholar.colorado.edu/mcen_gradetds/132 https://scholar.colorado.edu/cgi/viewcontent.cgi?article=1132&context=mcen_gradetds |
_version_ |
1766348107355783168 |