Both atmospheric meteorology and chemistry models provide valuable societal and research services, from forecasting of weather to development of strategies for addressing the challenges of ambient air pollution and climate change. A specific atmospheric constituent of great concern is black carbonaceous aerosol (BC), a byproduct of both natural (e.g., fires) and anthropogenic (e.g., diesel emissions) combustion. Exposure to concentration of BC is associated with a range of deleterious human health outcomes. Further, owing to its absorptive radiative properties, and its direct and indirect (cloud) effects, BC aerosol is estimated to induce a global positive radiative forcing (relative to preindustrial times) second only to carbon dioxide. However, model estimates of BC concentrations and sources are often quite uncertain, with estimates from difference inventories often differing by up to an order of magnitude. Historically treated separately, with NWP models for weather forecasting and “offline” chemical transport models for air quality, dynamic online models of weather and atmospheric chemistry are becoming increasingly prevalent. Coupling chemistry and atmospheric processes as components of an earth system modeling framework offers advantages for both fields, potentially improving extreme event forecasts and reducing the computational overhead for chemical weather forecasting. For BC, such models can account for the interaction of biomass burning plumes on dynamic meteorology, which have been shown to influence convection and extreme events such as tornadoes. Detailed simulations of BC aerosol microphysics and chemistry can incur large computational burdens. The increase in computational cost for coupled systems becomes especially acute in the context of coupled data assimilation systems such as 4D-Var, for which many integrations of the model are required. To address these challenges, we implement and evaluate a new formulation of 4D-Var for coupled chemistry-weather systems that has potential to be operationally scalable. Our work focuses on exploiting a recently developed assimilation algorithm – the Randomized Incremental Optimal Technique (RIOT). RIOT takes a state-of-the-art data assimilation approach (incremental 4D-Var) and allows for better code parallelization by replacing the key sequential step (conjugate gradient minimization) with a highly scalable, parallelizable algorithm (randomized singular value decomposition). Previous study and preliminary results with the WRFDA system (for inverse modeling of BC fluxes) suggest that RIOT could reduce the wall-time computation of incremental 4D-Var by at least an order of magnitude. In short, this algorithmic improvement presents a new opportunity for enhancing the sophistication and scalability of coupled atmospheric-chemical systems, including extension to larger volumes and varieties of observations. We will present our evaluation of the computational performance of RIOT for chemical data assimilation, as applied to the WRFDA-Chem data assimilation system. In addition, we will present our progress toward enabling operational air quality forecasting with flux and state estimation constraints provided by remotely sensed aerosol optical depth products from, e.g., TERRA, AQUA, JPSS, GOES-R. These components are the initial steps toward constructing a coupled atmospheric meteorological-chemical 4D-Var system for use on operational time-scales. Methodological improvements of this scale are best realized in a modular software environment which facilitates rapid transition from research to operations, and where the new approaches can be efficiently evaluated and benchmarked against previous algorithms. The Joint Center for Satellite Data Assimilation (JCSDA) initiated such an effort in the framework of the Joint Effort for Data Assimilation Integration (JEDI) project, which proposes to apply object-oriented concepts to the community model. As early adopters of the JEDI software development platform, our work on a coupled-data assimilation system (i.e., operators, state vectors, models, DA routines) will contribute to JEDI’s growth and acceptance in the community.
This publication was presented at the following: