Abstracts

Maryam Abdi-Oskouei (JCSDA)

Christian Sampson (JCSDA)

 Jerome Barre (JCSDA)

Anna Shlyaeva (JCSDA)

Yannick Tremolet (JCSDA)

Victor Marchais (JCSDA)

Thomas Auligne (JCSDA)

Christoph Keller (GMAO)

Tom Hill (UKMO)

Tim Payne (UKMO)

Joint Effort for Data Assimilation Integration (JEDI) is a unified data assimilation (DA) framework for Earth system prediction. JEDI can be used for both research and operational purposes to reduce redundant work within the community and increase the efficiency and flexibility of transition from development teams to operations. 

One major effort that JCSDA is undertaking is the implementation of a robust 4D-Var system for numerical weather prediction but also atmospheric composition applications (such as air quality and human emission monitoring). A crucial element in the 4D-Var methodology is the linear forecast model. Usually, the linear model is based on differentiating the non-linear forecast model. This can be easily done for the advection but can be more complicated when more non-linear processes are involved such as atmospheric physics and chemistry. In this presentation, we will show the recent advances that JCSDA has made in implementing a hybrid-tangent linear model following the methodology in Payne, 2020, in which a simplified conventional tangent-linear model (e.g., just the dynamical core) is used together with an ensemble-based adjustment every time step to account for the remaining processes, i.e., chemistry and physics. We will demonstrate this approach using output from the NASA GEOS Composition Forecast System (GEOS-CF), where we will show how the chemistry and physics can be represented in the linear model using ensemble information. We will expand and conclude on how methods in machine learning could be used to improve the computational efficiency and utility of hybrid tangent linear models. 

References: Payne, T. J., 2020: A Hybrid Differential-Ensemble Linear Forecast Model for 4D-Var. Mon. Wea. Rev., 149, 3–19, https://doi.org/10.1175/MWR-D-20-0088.1.

Clark Amerault, US Naval Research Laboratory

The assimilation of boundary layer height (BLH) measurements using a variational scheme will be discussed. The adjoint of a bulk Richardson number BLH operator is employed to project BLH measurement misfit information onto model variables above and below the BL. Successful BLH assimilation depends on the proximity between the BLH derived from the forward bulk Richardson number method and the visually inspected BLH from vertical profiles of model potential temperature and humidity data. If the difference is too large, adjoint sensitivities are insufficient to bring about meaningful changes in the analysis field. Additionally, we will present efforts to implement a BLH operator and its adjoint that align more closely with visually determined BLH from model profile data.

Nuo Chen, University of Oklahoma

Brett Hoover, University of Wisconsin-Madison, Space Science and Engineering Center

Due to its unique characteristics of conservation and invertibility, potential vorticity (PV) has proven invaluable as both a forecasting and pedagogical tool for comprehending large-scale dynamical processes in meteorology. The integration of PV concepts with adjoint sensitivity analysis introduces a novel approach for diagnosing synoptic weather events. Specifically, adjoint sensitivity to PV enables a comprehensive interpretation of sensitivities to both winds and temperature, enhancing our understanding of dynamical processes. The adjoint sensitivities to Quasi-geostrophic PV (QGPV) and to Ertel PV are compared in the case of the rapid intensification period of Hurricane Ian. The results reveal that sensitivity to Ertel PV captures features smaller than the Rossby radius of deformation, which sensitivity to QGPV overlooks. Geostrophically and nonlinearly balanced sensitivity to wind, recovered from sensitivity to QGPV and Ertel PV, respectively, closely resemble the original wind sensitivity fields. In addition, these balanced adjoint sensitivity fields can serve as the adjoint forcing to eliminate the high-frequency wave generated due to geostrophic adjustment during the adjoint model spin-up. Finally, we scrutinize the validity of approximating sensitivity to PV using the PV perturbation—a common practice in previous adjoint sensitivity studies. A comparison between PV perturbation and actual sensitivity to PV is presented, underscoring the need for caution when employing such approximations.

Yu-An Chen, Department of Atmospheric Science, Colorado State University, Fort Collins, CO

Peter Jan van Leeuwen, Department of Atmospheric Science, Colorado State University, Fort Collins, CO

The tropical cyclone (TC) boundary layer is the interface where energy is transferred from the ocean surface to the vortex aloft and plays a critical role in the TC intensification. Both previous numerical model and observational studies have shown that coherent turbulent structures (CTSs) in the TC boundary layer, which is driven largely by shear instabilities, can be involved in inverse energy cascade to larger-scale vortex via wave-mean flow non-linear interactions (Guimond et al., 2018; Li and Pu, 2023; Sroka and Guimond, 2021). However, the role of CTSs in shaping the TC structure, and thus predictability of TC, remains unclear. We will use the nonlinear data assimilation (DA) to examine the nonlinear interaction of the observed CTSs with the vortex thermodynamics. Specifically, we use the Particle Flow Filter, a fully nonlinear DA technique for high-dimensional geophysical systems (Van Leeuwen et al. 2019; Hu and Van Leeuwen, 2021) and is now available in the data-assimilation framework JEDI. This filter is fully nonlinear, appropriate for the highly nonlinear Hurricane physics in the boundary layer. The working of the filter is similar to that of ensemble 3DVars, with the difference that the ensemble members communicate with one another during the minimizations, and the prior does not have to be Gaussian (Hu and Van Leeuwen, 2021). 

In this research, we will perform data denial experiments by the state-of-the-art Model for Prediction Across Scales (MPAS), leveraging high-resolution (125 m horizontal and 30 m vertical grid spacing) Doppler radar observations in the TC boundary layer with the particle flow filter. We aim to investigate how to use fully nonlinear DA methods to integrate turbulence resolving radar observations into the numerical model; examine the influence of CTSs on the predictability of a hurricane; and to analyze the dynamical pathways from the CTSs to the large-scale vortex during intensification. In the presentation we will report on our findings in this ambitious project.

Sarah L Dance (University of Reading & National Centre for Earth Observation, UK)

Guannan Hu (University of Reading & National Centre for Earth Observation, UK)

Alison Fowler (University of Reading & National Centre for Earth Observation, UK)

David Simonin and Jo Waller (Met Office, UK)

An assessment of the value of these observations can guide us in the design of future observation networks, help us to identify problems with the assimilation system, and allow us to assess changes to the assimilation system. However, the assessment can be challenging in convection-permitting numerical weather prediction (NWP) due to (1) strong nonlinearities in the forecast model, (2) the use of limited area models and short forecasts, giving problems with verification and our ability to gather sufficient statistics. (3) the use of novel observations, which can be difficult to simulate in an observing system simulation experiment (OSSE), Furthermore, convection-permitting data assimilation often uses ensemble approaches, where there are fewer well-established observation influence metrics. For example, the degrees of freedom for signal (DFS) has long been used to assess the influence of observations on the analysis in variational data assimilation (DA) and EDA (ensemble of data assimilations) systems. While various methods exist for calculating the DFS in variational DA systems, calculating the DFS in ensemble-based DA systems, such as the ensemble transform Kalman filter (ETKF), is a largely unexplored area. Unlike in variational DA systems, the background error covariance matrix and the Kalman gain matrix are not static in the ensemble-based DA methods. Consequently, the DFS calculated at each assimilation step measures the observation influence for a certain background error covariance matrix. This means that the DFS estimates are flow dependent. In addition, domain localisation of observations is often used in ensemble-based DA systems (e.g., LETKF). This implies that the DFS should be calculated locally. In this work, we propose novel approaches for calculating the DFS in ensemble-based DA systems and investigate existing approaches applicable to such systems. We establish their consistency under idealised conditions and discuss their differences in practical applications. To validate our theoretical findings, we conduct simple numerical experiments. We hope that this work will provide useful information for assessing the influence of observations in ensemble-based DA systems.

Amal El Akkraoui, NASA

The three-cornered hat method (3CH) estimates error variances among three independent datasets, assuming negligible error covariances. Though promising in applications to characterize observation, analysis, and forecast errors, its practical use is challenging, particularly in selecting appropriate corners for real-world cases. This work addresses the 3CH method’s application with Earth system reanalysis and NWP datasets, where error correlations can’t be ignored. A geometric analog of the 3CH method is presented, which illustrates the method and the challenges arising from shifts in the reference point used to define the errors to be estimated. Using the GMAO Observing System Simulation Experiment (OSSE), the study explores analysis error correlations with forecast and analysis errors for different forecast lead times and analysis lags. Using ERA5, GFS and GMAO datasets, the method’s concepts, strengths, and limitations are demonstrated, and two practical frameworks are proposed around relaxed assumptions about error correlations with guidelines to ensure reliability of the method.

Takeshi Enomoto, Kyoto University

Saori Nakashita, Kyoto University

Sensitivity analysis plays a crucial role in targeted observations and identifying the source of forecast errors. Ensemble sensitivity analysis, based on pre-run ensemble forecasts, offers a low-cost alternative to the adjoint-based analysis–forecast system. The ensemble singular vector (Bishop and Toth 1999) and adjoint (Ancell and Hakim 2007) sensitivities can both be derived using the Lagrangemultiplier method (Enomoto et al. 2015, Hacker and Lei 2015). Remarkably, ensemble sensitivity analysis remains applicable to various phenomena even under the assumption of linearity. It enables the identification of sensitivities not only for extratropical cyclones but also for tropical cyclones. This success is partly attributed to the ensemble forecasts generated using a nonlinear model. Additionally, the ensemble singular vector works as an excellent additive inflation in local ensemble transform Kalman Filter (LETKF) by selecting fast-growing modes (Yang et al. 2015).

Keenan Eure, Pennsylvania State University

David Stensrud, Department of Meteorology and Atmospheric Science, Pennsylvania State University 

Yunji Zhang, Department of Meteorology and Atmospheric Science, Pennsylvania State University 

Matthew Kumjian, Department of Meteorology and Atmospheric Science, Pennsylvania State University 

Darrel Kingfield, Global Systems Laboratory, NOAA

Accurate forecasts of the development and evolution of deep, moist convection in convection-allowing models (CAMs) are both a priority and a challenge for the National Oceanic and Atmospheric Administration (NOAA). Additionally, modeling of the microphysical and internal structures of convection is difficult, as this can affect the storm mode, intensity, and longevity. Novel observations from the WSR-88Ds and GOES-16 have the potential to improve the forecasts of deep convection in CAM ensembles. Since the upgrade to the national network of WSR-88Ds was completed in 2013, polarimetric radar data offer a wealth of information about the shape, size, and type of hydrometeors present in clouds. Several distinct polarimetric signatures in early stages of deep convection have been identified, such as the differential reflectivity (ZDR) column. These columns are vertical protrusions of positive ZDR values above the environmental melting level and can aid significantly in characterizing storm updrafts. Information on the updraft location and intensity have potential to improve CAM representation of convection. In addition, GOES-16 infrared all-sky brightness temperatures provide complimentary information on cloud structures and cover that Doppler radars cannot directly measure. To explore the benefits of both types of data, an ensemble data assimilation approach is used with their simultaneous assimilation. The CAM selected for this study is the Advanced Research version of the Weather Research and Forecasting (WRF-ARW) model with the High-Resolution Rapid Refresh (HRRR) configuration. Observations are assimilated using the Ensemble Kalman Filter (EnKF). Different observations in the experiments are conducted jointly and separately, and all experiments include conventional observations. Analysis is conducted using a real case to realize the influence of these observations on different aspects of the convection, and results are presented and discussed.

Steven Fletcher, Cooperative Institute for Research in the Atmosphere, Colorado State University

Senne Van Loon

In almost all applications of data assimilation, a substantial assumption is made: all variables are well-described by Gaussian error statistics. This assumption has the advantage of making calculations considerably simpler, but it is often not valid, leading to biases in forecasts or, even worse, unphysical predictions. We propose a simple, but effective, way of replacing this assumption, by making use of transforming functions, while remaining consistent with Bayes' theorem. This method allows the errors to have any value of the skewness and kurtosis. We apply this framework to a 3D variational data assimilation method, and find improved performance in a simple atmospheric toy model, compared to an all-Gaussian technique.

Alison Fowler (NCEO, University of Reading)

Devon Francis, Amos Lawless (University of Reading)

John Eyre (Met Office)

Stefano Migliorini (Met Office)

Bias correction, via VarBC, is essential for satellite observations to have the dramatic impact on the skill of numerical weather prediction forecasts that they exhibit. To ensure VarBC identifies observation bias and not model bias, unbiased (anchor) observations are necessary. The importance of the precision, positioning and timing of these anchor observations in 4DVar-VarBC is explored theoretically and illustrated with the Lorenz 96 model. Anchor observations are shown to be especially helpful in distinguishing model and observation bias when they are collocated with biased observations and occur later in the assimilation window. However, the anchor observations themselves can be a vessel for model bias to contaminate VarBC if the anchor observations observe regions with model bias characteristics different from those of the observations being corrected. Lastly, we look at how metrics of observation impact, such as FSOI, can be used to diagnose the current role of anchor observations and guide network development.

Clémentine Hardy Gas, JCSDA

Yannick Trémolet, 

Tsz-Yan Leung, 

Kate Huxtable, 

Andrew Lorenc, 

Maryam Abdi-Oskouei, 

Fabio Diniz, 

Dom Heinzeller, 

Eric Lingerfelt, 

Benjamin Ruston, 

Christian Sampson

The Ensemble Data Assimilation (EDA) approach involves using multiple model simulations, each initialized with slightly different conditions, to account for uncertainty in the model and the observations. This method has been shown to be effective in improving the accuracy and uncertainty estimates of forecasts and has been widely used in weather forecasting applications.

The Joint Center for Satellite Data Assimilation (JCSDA) has developed a comprehensive workflow (SkyLab) to facilitate the execution of diverse EDA scenarios. These scenarios can involve different instruments, algorithm choices such as background error covariance or cost function, and even different weather models. The SkyLab workflow serves as the orchestrator, driving the underlying JEDI code (Joint Effort for Data assimilation Integration) to execute the different EDA experiments. 

In this study, we present an overview of our workflow's architecture and a comparative analysis of three distinct methods employed to conduct EDA experiments. The first method involves a distinct DA run for each member. The second approach leverages block-based methods: since many similar optimization problems are solved it is possible to use information from all the members to construct a better approximation of the eigen-structure of the matrix at the heart of the optimization problem and accelerate the convergence. The block Lanczos algorithm is such an example. Finally, we introduce a novel application developed at the UK MetOffice of an EDA solving for the full unperturbed control run and only the perturbations of the ensemble members.

 During our presentation, we will illustrate how the SkyLab workflow developed at JCSDA helps us set up and compare these different experiments. We will highlight the advantages and trade-offs associated with each ensemble method, with results from experiments using operational models. Furthermore, we will discuss practical considerations and implications for future operational implementation.

Shay Gilpin, Department of Mathematics, University of Arizona

Tomoko Matsuo: Smead Aerospace Engineering Sciences, and Department of Applied Mathematics, University of Colorado Boulder 

Stephen E Cohn: Global Modelling and Assimilation Office, NASA Goddard Space Flight Center

For continuum states governed by the advection equation, continuity equation, and related hyperbolic differential equations, we show that the discrete variance evolution implied by the usual discrete covariance propagation M(MP)^T, with M a typical discrete propagator for the state dynamics, is surprisingly inaccurate, more so than traditional discretization error analysis would suggest. The underlying cause of these errors is the continuum covariance dynamics, which admits two solutions along the covariance diagonal: the first solution being the continuum variance dynamics, and the second solution being the dynamics of the continuous spectrum of the covariance operator. The latter solution makes its presence known as correlation lengths approach grid scale, which occurs naturally in shear flow, for example. Through extensive error analysis, we show that the conventional discrete variance propagation derived from M(MP)^T gives rise to error terms that depend on the ratio of the grid scale to the correlation length. These error terms introduce both loss and gain of variance by altering the continuum dynamics being approximated along the covariance diagonal. We verify our analytical results and illustrate the corresponding degree of variance loss and/or gain through one-dimensional examples. The analytical results and insights gained from this work may help explain the surprisingly inaccurate variance propagation often observed in the chemical constituent data assimilation community.

Behzad Golparvar (Department of Civil and Environmental Engineering, Rutgers University)

Shaghayegh Vosoughitabar (WINLAB, Department of ECE, Rutgers University) 

David Bazzett (Department of Civil and Environmental Engineering, Rutgers University) 

Joseph F. Brodie (AKRF, Inc) 

Chung-Tse Michael Wu (WINLAB, Department of ECE, Rutgers University) 

Narayan B. Mandayam (WINLAB, Department of ECE, Rutgers University) 

Ruo-Qian Wang (Department of Civil and Environmental Engineering, Rutgers University)

The allocation of the 5G mmWave spectrum in the 26 GHz range, known as 3GPP band n258, has raised widespread concern among the remote sensing and weather forecast communities due to the adjacency of this band with a frequency band used by passive sensors in Earth Exploration-Satellite Service (EESS). The concern stems from the potential radio frequency interference (RFI) caused by transmissions within the n258 band that interfere with the 23.8 GHz frequency, one of the key frequencies employed by weather satellite passive sensing instruments, such as AMSU-A and ATMS, to measure atmospheric water vapor using its emission spectrum. Such RFI can bias satellite observation and compromise weather forecasting. In this project, we estimate the spatio-temporal distribution of 5G mmWave base stations at the county-level throughout the contiguous United States. Then, by performing a parametric study for a wide range of interference power received by the AMSU-A radiometer from a single 5G base station (tower), the aggregate interference power for each satellite observation footprint is calculated to model contaminated microwave observations. Using WRFDA and WRF, a sensitivity analysis is conducted to investigate which meteorological variables are affected by assimilating contaminated microwave observations in terms of brightness temperature, the number of observations that are rejected by quality control processes, and the impact of 5G-induced contamination on weather forecasting accuracy.

Parisa Heidary, Research affiliate of George Washington University

Leila Farhadi, George Washington University, Washington, DC, United States 

Muhammad Umer Altaf, Division of Earth Sciences, King Abdullah University of Science and Technology, Saudi Arabia

Quantifying the magnitude and the spatial patterns of groundwater recharge is important for understanding and managing groundwater systems. Groundwater recharge is a complex process, which depends on several factors, including the moisture and the hydraulic properties of soil in the vadose zone. Despite the importance of this flux there are no direct measurements that can allow any mapping and regional estimation of the rate of recharge. The research objective of this study is to propose a data assimilation framework to quantify and map the patterns and dynamics of diffuse recharge flux by accurately estimating its key state and parameters such as soil moisture profile and effective soil hydraulic parameters. In order to achieve this objective, a state-of-the-art data assimilation technique is used to estimate these parameters from implicit information contained in surface Soil Moisture (SM) observations which are widely available by remote sensing in a wide range of spatial and temporal scales. The proposed approach uses the reduced order variational data assimilation scheme called Reduced-Adjoint Variational Data assimilation that assimilates SMAP SM data into HYDRUS-1D model, a highly nonlinear soil water model, to produce accurate estimates of the effective soil hydraulic parameters and the soil moisture profile. Consequently, more accurate spatial patterns of diffusive recharge flux are obtained following the Darcy equation. The method is implemented over an area of 12,000 km2 in the U.S. Southern Great Plains for water year 2021 (Oct 2020-Sep 2021), and its performance is evaluated through comparison tests using available ground measurements datasets.

Tom Hill, Met Office

Tim Payne (Met Office) 

Christian Sampson (JCSDA)

Yannick Trémolet (JCSDA), many other contributors from Met Office and JCSDA

A tangent linear model (TLM) aims to predict the difference between reference and perturbed forecasts of a nonlinear model. Its adjoint gives the gradient of the nonlinear model. Both are key components of 4D-Var, the algorithm used for the Met Office’s global data assimilation system. Traditionally, the TLM and adjoint are produced using line-by-line differentiation and transposition of the nonlinear model code, which is time-consuming. Additionally, the derivative does not always exist (or has only very local validity) for physical parameterizations, which contain discontinuities and sharp changes in gradient. 

One option is to produce smoother versions of the parameterizations and differentiate those, but this degrades the TLM’s accuracy. Over the past decade, alternatives which use ensemble forecasts to construct the TLM have been developed. For these to be accurate, the number of ensemble members must equal at least the number of variables in the nonlinear model’s computational stencil. 

The Met Office will use one of these methods, the hybrid TLM, in its next-generation global 4D-Var system, following promising trials with the current system. Importantly, this method still uses the traditional approach for the dynamical core of the nonlinear model code. Since the physical parameterizations have a smaller computational stencil, only a small number of ensemble members is required. 

Hybrid TLM code has been developed within the model-generic Joint Effort for Data assimilation Integration system through collaboration with the Joint Center for Satellite Data Assimilation, where scientific trials have begun. Ways to augment the hybrid TLM to improve accuracy and reduce cost, including sub-timestepping and multiresolution methods, have also been developed. This presentation will focus on these. 

Finally, the presentation will look ahead to the Met Office’s own scientific trials which will begin once all the necessary components are in place. These include the LFRic atmospheric model and the TLM and adjoint to its dynamical core, named GungHo.

Chih-Chi Hu, Princeton University

Mingjing Tong, NOAA/GFDL

The probability density function (pdf) of the observation error plays a crucial role in data assimilation (DA). In practice, we need to build the observation error pdf before actually doing DA, typically by estimating its mean and variance and assume the pdf is Gaussian. However, due to the presence of the representation error, the Gaussian assumption on the observation error pdf can be inappropriate, e.g., for all-sky satellite radiances. The strategy for assimilating the all-sky radiances is to build a state-dependent and conditionally Gaussian observation error pdf. Specifically, one proposes a predictor and assumes that the observation error pdf is Gaussian with its mean and variance depending on the predictor. Nevertheless, one can show that a state-dependent and conditionally Gaussian observation error pdf still leads to a non-Gaussian likelihood, which does not satisfy the Gaussian assumption in most DA methods commonly used in the operational centers. 

We have recently proposed a new method, called the evolving-Gaussian method, to deal with the non-Gaussian likelihood in the incremental variational methods. The evolving-Gaussian method opens up new possibilities to examine the observation with non-parametric observation error pdf using the variational DA system. In this presentation, we will first revisit the state-dependent observation error model and examine its non-Gaussian impact on the analysis in an idealized experiment. Next, we will use the evolving-Gaussian method to explore different strategies to address the non-Gaussianities in the all-sky microwave radiances in the Geophysical Fluid Dynamics Laboratory (GFDL) System for High‐resolution prediction on Earth‐to‐Local Domains (SHiELD) model.

Kian (Qien) Huang, University of Utah

Zhaoxia Pu, University of Utah

In this study, we developed a coupled land-atmosphere data assimilation capability with the Joint Effort for Data Assimilation Integration (JEDI) system using a limited area version of the NOAA Unified Forecast System (UFS) and Noah-MP land surface model. We examined the cross-covariances between land and atmosphere. The effectiveness of various horizontal and vertical localization schemes was evaluated and tested. The Soil Moisture Active Passive (SMAP) satellite-derived soil moisture observations and near-surface atmospheric data (2-m temperature and humidity) are assimilated into a coupled land-atmosphere data assimilation at different configurations. The impacts of the coupled data assimilation on the short -range weather forecasts, especially the prediction of near-surface atmospheric conditions with the UFS model, are examined.

Kate Huxtable (Met Office)

Tsz Yan Leung (Met Office)

Andrew Lorenc (Met Office)

Neill Bowler (Met Office)

Clémentine Gas (JCSDA)

The Met Office is currently building its next generation DA system as part of the Joint Effort for Data Assimilation Integration (JEDI). For the global EDA, we are exploring two options: an ensemble of independent data assimilations (EIDA) or a more novel approach, Control-Pert. 

The Control-Pert method is designed as a cheaper alternative to an EIDA, and involves running the best available variational DA method on a control member to give a best estimate of an analysis increment, which is added to all ensemble members. For the ensemble members, a similar but cheaper variational approach is used to give an additional increment for each member's perturbation to the control member, which has the function of adjusting the ensemble spread. These ‘pert’ member assimilations run a simpler DA scheme than the control member and introduce linear assumptions..

These methods have both been implemented into the Object-Oriented Prediction System (OOPS) within JEDI, a collection of model-agnostic DA libraries. We are running scientific experiments to compare Control-Pert and EIDA. We note that Control-Pert without additional inflation is not expected to be as accurate as an EIDA, due to the linear assumptions. We are hoping to see how important the approximated nonlinearities are, and if Control-Pert will be a suitable, affordable implementation. 

This presentation will explain the theory behind Control-Pert, some of the challenges in implementing this in JEDI and a look at early results from the comparative experiments.

Kayo Ide (University of Maryland College Park)

Radiance observations from earth-observing satellites have a significant positive impact on numerical weather prediction (NWP) forecasts, but some spectral regions are not fully exploited. Observations from hyperspectral IR sounders in the longwave region, for instance, are routinely assimilated in many NWP models, but observations in the shortwave region are not. Each of these regions provides information on the temperature structure of the atmosphere, but the shortwave IR region is considered challenging to assimilate due to higher noise and phenomena that are challenging to model, like non-Local Thermodynamic Equilibrium (NLTE) and solar reflectance. With recent advances in small-satellite technology, shortwave IR temperature sounders may provide an agile and cost-effective complement to the current constellation of IR sounders, and a better understanding of the use of these observations in NWP is therefore necessary. In this study, the value of shortwave IR observations in global NWP is assessed by assimilating shortwave IR observations from the Cross-track Infrared Sounder (CrIS) in NOAA’s Global Data Assimilation System (GDAS). The methodologies used to enable the assimilation of these observations are discussed, as are the results of a series of Observing System Experiments (OSEs) conducted to test the assimilation of shortwave IR observations. The results show that shortwave IR assimilation produces similar forecast impacts to longwave IR assimilation. The ability to demonstrate that the assimilation or shortwave IR observations in NWP is a realistic prospect may help to shape future constellations of small-satellites. 

Bryan M. Karpowicz, UMBC/GESTARII/NASA GMAO

Erica McGrath-Spangler Morgan State University/GESTAR II/NASA GMAO

Hyperspectral Infrared sounders such as IASI, AIRS, and CrIS have long been an integral part of radiance assimilation in numerical weather prediction (NWP), providing vertical profiles of water vapor and temperature information. Principal Component Scores (PCS) are a lossy form of compression that retains most information, such as temperature and moisture, by using a large training set of atmospheric profiles. However, PCS may not well represent profiles which are rare events, such as volcanic eruptions, and drops some sources of random noise. There has been an increased interest in the use of PCS as EUMETSAT plans to distribute future geostationary sounder radiances from MTG-IRS via PCS only. NWP centers use two approaches to deal with PCS: direct assimilation of the PCS by modifying the radiative transfer model to produce PCS and the associated Jacobians, or a simpler approach of decompressing the PCS and reconstructing the radiances back into channel space to allow assimilating radiances without modifications to the data assimilation system. EUMETSAT has developed a PCS product for IASI that has been operational since 2011. We utilize this product opting for the simpler approach, decompressing IASI PCS into channel space, and assimilating those radiances using the GEOS-ADAS. We then compare this with a control using the standard IASI radiance product. Resulting differences in global forecast statistics, differences in Forecast Sensitivity to Observation Impact, along with implications for implementation and quality control are discussed. 

Hyerim Kim, The University of Iowa

Gregory R. Carmichael, The University of Iowa

South Korea is experiencing severe air pollution events, which often exceeds the National Ambient Air Quality Standard. Air pollution in Korea is due to local emissions—from industrialized sources and high population density—and transboundary emissions due to prevailing westerlies from adjacent countries. In order to control air quality, knowing the influences from different locations of sources is crucial. Sensitivity methods from chemical transport models can provide these source contributions to guide us to a better understanding of the sources and pathways. In particular, backward sensitivity method (adjoint) can be computationally efficient when focusing on specific receptors and seeing the sensitivity to their sources and parameters. In this study, we use the Community Multiscale Air Quality (CMAQ) and its adjoint to analyze source contribution focused on bad air quality episodes in South Korea during recent years. We investigate local impacts as well as transboundary impacts from neighboring countries.

Kenta Kurosawa, Chiba University

Atsushi Okazaki, Chiba University

As the effects of global warming intensify, extreme weather events such as typhoons and heavy rains are increasingly causing severe hydrological disasters. Addressing these challenges requires not only the development of infrastructure and the utilization of forecast information for disaster prevention but also exploring ways to adjust the intensity, timing, and scale of such weather events to potentially prevent or reduce their impact. Control theory represents a new research area that bridges meteorological models with control technologies, promising diverse contributions and novel developments from various scientific fields. The current study focuses on model predictive control (MPC), a technique established in engineering for nonlinear dynamic control, and advances its application in the realm of meteorology. This method has the potential to enable small interventions to modify these systems, especially under extreme conditions such as heavy precipitation or typhoons. It determines the optimal amount of correction and input values by numerically considering the time evolution of the model meteorological systems. Significantly, in MPC, determining optimal control inputs necessitates the minimization of a cost function, a process remarkably similar to the variational methods in data assimilation. This similarity highlights the potential for synergies between these two approaches, bridging the strengths of each to enhance weather prediction and management. In this presentation, we will delve into both the concepts of data assimilation and MPC, explaining their similarities and how they can be effectively combined. We will also discuss the application of these concepts in control theory, including the results obtained from implementing these theories in toy models and numerical weather prediction models.

Chengzhe Li, Department of Chemical and Biochemical Engineering, University of Iowa

Jun Wang, Department of Chemical and Biochemical Engineering, University of Iowa

Atmospheric aerosols play an important role in Earth's environment, climate change, and public health. Atmospheric chemical transport models (CTMs), such as Unified Inputs (Initial and Boundary conditions) Weather Research and Forecasting model coupled with Chemistry (UI-WRF-Chem), can provide forecast of aerosol distribution and surface air quality, and fill in data gaps where and when satellite data is not available. However, UI-WRF-Chem, like other air quality models, has uncertainties inherently associated with the deficiency in parametrization schemes, emissions, and description of different atmospheric processes. This proposed work seeks to improve UI-WRF-Chem simulation of aerosol mass concentration and aerosol properties (including vertical and size distribution) in United States through the data assimilation (DA) of ground-based observations (AERONET, EPA and PurpleAir PM2.5) and satellite aerosol products (TROPOMI and TEMPO). DA and inverse modeling methods are developed here, from the statistical optimization of emissions based on source-receptor relationship to the Ensemble Transfer Kalman Filter (ETKF) method for optimizing both emission and the atmospheric composition. At locations such as Ethiopia with limited surface observations and lack of satellite-based data of atmospheric composition due to high cloud cover, statistical optimization can be a viable method for using ground-based measurement of PM2.5 concentration to correct the emission inventory, thereby improving the simulation results of PM2.5 concentration by UI-WRF-Chem. ETKF method can be used for DA of concentration and vertical distribution of multiple species of aerosol and trace gases and is proposed here for using TEMPO-based aerosol and trace gas data products. Finally, the machine-learning based approach will be used to improve the model predictions. Preliminary findings indicate that the enhanced UI-WRF-Chem model, incorporating adjusted emission data, exhibits improved agreement with observation diurnal variation curves. Furthermore, preliminary findings demonstrated that the ETKF method can reduce the relative error by 20-50% in model simulations of trace gases.

Zhendong Lu (The University of Iowa)

Jun Wang (The University of Iowa)

Yi Wang (China University of Geosciences) 

Daven K. Henze (University of Colorado Boulder)

Xi Chen (The University of Iowa)

Tong Sha (Shaanxi University of Science and Technology) 

Kang Sun (University at Buffalo)

Due to the lockdown during the COVID-19 pandemic in China from late January to early April in 2020, a significant reduction of primary air pollutants has been identified by satellite and ground observations. However, this reduction is in contrast with the increase of surface O3 concentration in many parts of China during the same period. The reasons for this contrast are studied here from two perspectives: emission changes and inter-annual meteorological variations. Based on top-down constraints of NOx emissions from TROPOMI measurements and GEOS-Chem model simulations, our analysis reveals that NOx and volatile organic compound (VOC) emission reductions as well as meteorological variations lead to 8%, -3%, and 1% changes in O3 over North China, respectively. In South China, however, we find that meteorological variations cause ~30% increases in O3, which is much larger than -1% and 2% changes due to VOC and NOx emission reductions, respectively, and the overall O3 increase is consistent with the surface observations. The higher temperature is the main reason that leads to the surface O3 increase in South China. Overall, inter-annual meteorological variations have a larger impact than emission reductions on the aggravated surface O3 pollution in China during the early lockdown period of COVID-19 pandemic.

A. Biswas (University of Maryland Baltimore County)

Rasika Mahawattege (University of Maryland Baltimore County)

J.T. Webster

This talk bridges the fields of data assimilation, boundary control, and Luenberger compensator theory for partial differential equations to enhance system estimation and control in the presence of localized/partial observations. While data assimilation techniques have traditionally been employed to estimate the state variables of a system using a diverse range of interior observations, their integration with boundary control methods and Luenberger compensators introduces a powerful framework for real-time system monitoring and control. The proposed methodology combines the principles of data assimilation, which update system state estimates by assimilating boundary (or localized interior) measurements, with boundary control theory, which focuses on manipulating system behavior through boundary or spatially localized feedback. Such integration can be achieved through the design of a Luenberger compensator, a widely used tool in control theory, to simultaneously estimate the state and control input in the localized observation region.

Sharanya J. Majumdar (University of Miami)

Lisa R. Bucci (NOAA/NWS/National Hurricane Center)

Robert Atlas (NOAA/Atlantic Oceanographic and Meteorological Laboratory),

G. David Emmitt (Simpson Weather Associates)

Steve Greco (Simpson Weather Associates)

This study uses a regional Observation System Simulation Experiment (OSSE) framework to investigate the influence of assimilating different configurations of vertical wind profiles on analyses and forecasts of a tropical cyclone. Three OSSEs are examined. The first establishes a “best-case” benchmark by assimilating idealized profiles throughout the parent domain. The second uses swaths of idealized profiles that would be sampled by polar-orbiting satellites. The final OSSE assesses the role of tropical cyclone inner-core observations. All observations are simulated from a high-resolution regional Nature Run, and are assimilated using a cycled ensemble Kalman filter in NOAA’s regional Hurricane Weather Research and Forecasting framework. The observation impact, error statistics, and tropical cyclone structural differences are compared across the three OSSEs. The most accurate representation of the tropical cyclone is achieved via the simultaneous assimilation of collocated and uniform thermodynamic and kinematic observations. Intensity forecasts are improved with increased inner-core wind observations, even if the observations are only available once daily. Domainwide errors are reduced when the cyclone is observed during a period of structural change, such as rapid intensification. The OSSEs emphasize the importance of wind observations and the role of inner-core surveillance when analyzing and forecasting tropical cyclone structure. In this presentation, particular attention will be given to the OSSE framework development and lessons learned.

Juliana Matranga - UCSC

Andrew Moore - UCSC

Data assimilation (DA) is computationally demanding, and the increasing availability of meteorological and oceanic observations, as well as the use of high resolution grids, continue to add to the computational costs. One possible way to reduce the computational burden is by performing some numerical operations with reduced arithmetic precision. Of particular concern though is whether the performance of an analysis-forecast system is significantly altered using such an approach. A simple case study in the form of a zonally periodic ocean channel with an initial meridional temperature front was used to explore the impact of reduced arithmetic precision on the efficacy of DA and subsequent forecasts. Specifically, the resulting baroclinically unstable circulation was analyzed using Observing System Simulation Experiments (OSSEs) comprising different combinations of arithmetic precision and model resolutions. The OSSEs include 40 day forecasts initialized every 2 days, and the forecast error variance was analyzed by using the Empirical Orthogonal Functions (EOFs) of the resulting ensemble of overlapping forecasts. It was found that the use of single precision does not significantly degrade the results of the DA experiments, and differences in the forecast error structure can be linked to non-linearities in the flow. 

Erica McGrath-Spangler, NASA GMAO/GESTAR II/Morgan State University

Nikki Privé (NASA GMAO/GESTAR II/Morgan State University 

Bryan Karpowicz (NASA GMAO/GESTAR II/University of Maryland Baltimore County) 

Andrew K. Heidinger (GeoXO Program Office NOAA/NESDIS/STAR)

The United States’ satellite programs are currently in the process of evaluating the future of satellite observations of the Earth’s atmosphere, both as an individual nation and within the context of international partnerships. In an effort to better understand the impacts of various alternatives, the Global Modeling and Assimilation Office (GMAO) observing system simulation experiment (OSSE) framework has been utilized to evaluate several configurations of the international hyperspectral infrared sounder constellation from the perspective of global numerical weather prediction (NWP). This was done with an emphasis on the proposed NOAA/NASA Geostationary eXtended Observations (GeoXO) Sounder (GXS), planned to launch in the mid-2030s, and similar planned missions from international partners. GXS, in addition to contributions from EUMETSAT and JMA, will form a global ring of geostationary sounders consistent with the WMO’s 2040 vision. In addition to a novel geostationary IR sounder program, considerations extend to the future of the Low Earth Orbit (LEO) sounder program with the upcoming demise of several existing instruments and plans for future missions ongoing. The GMAO OSSE framework has examined the potential roles of GEO and LEO sounders in weather prediction improvement. Overall, the inclusion of both GEO sounders in a global ring and LEO sounders produces the most beneficial observation impact and the most accurate global weather forecasts as evaluated using several metrics, including hurricanes and the FSOI. 

Saori Nakashita (Graduate school of science, Kyoto University)

Takeshi Enomoto (Disaster Prevention Research Institute, Kyoto University)

Regional atmospheric models require lateral boundary conditions obtained from global circulation models. The regional analysis sometimes suffers from deterioration of the large-scale structure than that of the global analysis due to limitations in the domain size and observations. The large-scale error may cause the displacement error for disturbances such as typhoons or synoptic-scale fronts and degrade the performance of convective-scale DA. Although several scale-dependent blending methods of global and regional analyses have been proposed to alleviate those large-scale errors, these blending methods may hinder the optimality of individual DA. Guidard and Fischer (2008) and Dahlgren and Gustafsson (2012) introduced the augmented information vector with the global analysis into the regional variational assimilation and reported promising results. However, their formulations require several assumptions for the error correlations and ignore the covariance between the global and the regional forecast errors. In this study, we extend their augmented variational formulation to an ensemble variational method to relax those assumptions and take the flow-dependency of the forecast error into account. We test the proposed method in the one-way nesting system using ideal one-dimensional models proposed by Lorenz (2005) and compare the results with those of the separate assimilations and of the previous studies. The effect of the cross-covariance between the two models will also be discussed.

Olivier PANNEKOUCKE (INPT-ENM/CNRM/CERFACS)

Martin SABATHIER (ONERA) 

Vincent MAGET (ONERA)

The parametric Kalman filter (PKF) is an implementation of the Kalman filter (KF) which approximates the covariance dynamics by the parametric evolution of a covariance model all along the analysis and the forecast steps. In this talk we briefly review the ideas behind this approach (univariate and multivariate formulation) and show some applications when the covariances are parameterized from the error variance and local anisotropy tensor. Then we focus on the specification of boundary conditions (BCs) for the PKF and how this helps to better specify BCs in an ensemble KF, or to take into account BCs in a KF at a theoretical level. We conclude by addressing the potential of the forecast and analysis PKF equation in observation targeting.

Nikki Privé, GESTARII/Morgan State University

In a world where the true state of the atmosphere is known, what would be commonly understood characteristics of observation impacts and data assimilation behavior? What does background error look like? What are the actual consequences when observations are improperly weighted? How do observations affect the short term-forecast quality? What would actually happen if nearly-perfect initial conditions were used in numerical weather prediction? How much observation information is retained between cycle times for different types of data? These and other questions will be addressed based on results from observing system simulation experiments. This presentation will be in the form of an interactive game "poster".

Visweshwaran Ramesh (National Centre for Earth Observation & University of Reading, UK)

Elizabeth Cooper (UK Centre for Ecology and Hydrology, Wallingford, UK) 

Sarah L Dance (National Centre for Earth Observation & University of Reading, UK)

Accurate soil moisture estimates play a crucial role in the effective management and operational planning of various applications including response to floods and droughts. Gridded soil moisture values, derived from land surface models, often exhibit significant deviations from in-situ observations. Employing Four-Dimensional ensemble variational (4D-EnVar) hybrid data assimilation, we ingested in situ field-scale soil moisture observations (Cosmic-ray soil moisture monitoring network: COSMOS-UK) into the Joint UK Land Environment System (JULES) land surface model. In this study, we applied 4D-EnVar for joint state-parameter estimation considering a single soil column from several COSMOS-UK sites across the UK. This joint optimization has the potential to enhance not only the model's representation of current hydrological conditions but also to improve the underlying parameters governing physical processes, leading to an improved forecast. Preliminary findings indicate that optimizing the parameters leads to a notable enhancement in the accuracy of soil moisture predictions by the JULES model. Nonetheless, further investigations are necessary to understand this behaviour over larger areas and across different climatic zones.

Laura Risley, University of Reading 

Amos Lawless (University of Reading, NCEO)

Matthew Martin (UK Met Office)

Anthony Weaver (CERFACS)

When specifying the background error covariance matrix in variational data assimilation (DA), model variables are transformed into control variables that can be assumed to be approximately uncorrelated. This process is known as the control variable transform. With a view of assimilating ocean surface currents, we seek to develop a transformation that decorrelates horizontal velocity. Helmholtz theorem can be used to separate horizontal velocity into nondivergent and irrotational components. The transformed variables can then be defined as relative vorticity and horizontal divergence, as is typically done in atmospheric DA. In the NEMOVAR ocean data assimilation system, the horizontal velocity control variables are taken to be the ageostrophic components of the vector. Here, we investigate using alternative velocity control variables based on the ideas above.

Christian Sampson

Tom Hill (UKMO)

Tom Fearon (UKMO)

Maryam Abdi-Oskouei (JCSDA)

Jerome Barre (JCSDA)

Yannick Tremolet (JCSDA)

Anna Shlyaeva (JCSDA)

4d-Var has been shown to provide some of the most reliable weather forecasts to date, but is not without its pitfalls. In particular, 4d-Var depends heavily on a tangent linear model (TLM) and an adjoint to the tangent linear model. While conceptually simple, coding these two elements is extremely time intensive and difficult. A small change in the larger weather model can induce months of work on its TLM and adjoint delaying the benefits of improvements on the model side. In this talk I will introduce the Hybrid Tangent Linear Model (HTLM), developed in [Payne 2021], which is aimed at avoiding these pitfalls as well as present a generic implementation of it in the JEDI system. The HTLM is similar to other ensemble linear models such as the Localized Ensemble Tangent Linear Model (LETLM) [Frolov 2018] which use nonlinear ensembles to directly estimate the tangent linear model, but require large ensembles and large samples with in them for accuracy. In contrast, the HTLM leverages any available incomplete TLM (perhaps dynamics only) by first forwarding an ensemble of perturbations with it and then employing an LETLM with sampling only on a column to find coefficients for a corrective update of the TLM forwarded perturbation. This is done for each time step in the 4dvar window. These coefficients can then be used in a 4d-Var assimilation updating the incomplete TLM and adjoints at each time step. The HTLM formulation reduces the number of ensemble members, and size of samples within them, that are needed for accuracy when compared to a full LETLM. It also provides update coefficients that can be re-used later in other 4d-Var assimilations. The generic implementation of the HTLM in JEDI provides the opportunity for any model with a JEDI interface to do 4d-Var assimilation with it. I will also present some current results with the JEDI-HTLM, plans for further development, and challenges of the method. 

Elizabeth Satterfield, U.S. Naval Research Laboratory, Monterey, CA

Sarah A. King, NRL 

Nancy L. Baker, NRL 

Bill Campbell, NRL 

Justin S. Tsu, NRL 

Bailey Stevens, SAIC

Fei Lui, SAIC

Francois Vandenberghe, JCSDA

Alex Reinecke, NRL 

James Doyle, NRL

The upcoming unified FALCON data assimilation (DA) system is the Navy’s implementation of the Joint Effort for Data assimilation Integration (JEDI) developed by the Joint Center for Satellite Data Assimilation (JCSDA). The FALCON DA system contains Navy unique observation preprocessing, background error specification, and cylc workflow. FALCON DA utilizes JEDI solvers with a combination of Navy, JEDI Skylab, and METplus diagnostic and verification tools. The planned initial FALCON transition will be in support of the global NEPTUNE atmospheric model and will consist of a 3DVar solver with plans in place to eventually transition to a hybrid 4DVar solver. We will provide a broad overview of improvements made to the FALCON DA system in recent months. For the solver we have undertaken to implement dual-resolution and the JEDI 3DVar FGAT. We have increased the observation types we are assimilating as well as implemented variational bias correction. Additionally, we have improved the tuning of our static background error covariance. We will discuss the impacts of these improvements in our cycling experiments and provide context for future development. 

FALCON: Flexible Assimilation Linking Collaborations to Operations for NEPTUNE 

NEPTUNE: Navy Environmental Prediction sysTem Utilizing a Nonhydrostatic Engine

Narges Shahroudi, NASA-GSFC/UMD

Antonia Gambacorta (NASA-GSFC)

Isaac Moradi(NASA-GSFC/UMD)

John Blaisdel (NASA-GSFC/SAIC)

Robert Rosenberg(NASA-GSFC/SAIC)

Hyperspectral microwave sensors, operating within the frequency range of 6-250 GHz, have gained substantial endorsement from space and meteorological agencies worldwide. These sensors exhibit the potential to enhance atmospheric temperature and water vapor profiling, thereby advancing operational numerical weather prediction (NWP) capabilities through spaceborne observations. To evaluate the potential impact of assimilating hyperspectral microwave observations from a Microwave Photonic Instrument (HyMPI) on NWP, this study conducted an observing system simulation experiment (OSSE). The OSSE framework utilized is the Global Modeling and Assimilation Office (GMAO) OSSE system. The generation of simulated hyperspectral microwave data was done using CRTM forward model and the GEOS-5 Nature Run as its inputs. This presentation provides an overview of the simulation and assimilation of all-sky hyperspectral microwave observations, along with some preliminary results highlighting their impacts on analyses quality and forecasts.

Johnathan Stoddard, University of Utah

Dr. Zhaoxia Pu, University of Utah

This study examines the effects of assimilating NASA CYGNSS (Cyclone Global Navigation Satellite System) satellite-derived ocean surface winds and surface heat fluxes into the NCEP's new operational Hurricane Analysis and Forecast System (HAFS) using the GSI-based 3DEnVAR data assimilation system. Past studies have found that assimilating CYGNSS ocean surface winds into the hurricane weather research and forecasting (HWRF) model resulted in an improved hurricane inner-core structure and wind field. Positive impacts were also observed in hurricane track and intensity forecasts. This study will transition CYGNSS ocean surface wind data assimilation from HWRF to HAFS. In addition, developments will be conducted to assimilate the CYGNSS satellite-derived surface heat fluxes into the HAFS with GSI 3DEnVar. Experiments will be conducted to examine the effects of assimilating ocean surface winds and surface heat fluxes together and separately. Best track data will be used to verify the impacts of these experiments. The diagnoses will also be conducted to understand the role of surface heat fluxes on tropical cyclone evolution and intensity changes. Results will be presented.

Filip Vana, European Centre for Medium-Range Weather Forecasts (ECMWF)

New method for semi-implicit time scheme will be presented. It defines the linear implicit model by time variable tangent-linear approximation around the actual true atmosphere. This formulation of implicit model allowing also for diabatic processes is believed to be the most accurate linear model representation of the full nonlinear model. The methods to invert this implicit model by no matrix methods will be presented including first results from its IFS model implementation. Being it exclusively grid-point and local, the new scheme represents interesting alternative to deal with computational efficiency on massively parallel platforms.

Zackary VANCHE (ONERA and CNES)

Vincent MAGET (ONERA) 

Olivier PANNEKOUCKE (CNRM)

Salammbô is a model of the radiation belts dynamics that has been developped at the ONERA since the 90’. In order to make it operationnal, an Ensemble Kalman Filter (EnKF) is used since 2007 to assimilate satellite observations. For the EnKF to be optimal, it is important to know and master the uncertainties on these observations. One of the error sources is the representativity error on their position. The reason is that the position of the satellite is known in geographic coordinates, while Salammbô works with magnetic coordinates. It is therefore necessary to convert these positions from one system of coordinates to the other. Many observation operators based on different magnetic models exist for this conversion, but none of them is perfect : they all create a position error that has a consequence on the observation error. This contribution present the estimation of this error statistics calculated using several observation operators. The representaty error appears multiplicative, and a parameterization of the variance is proposed as a function of the distance and the solar activity. Finally, a twin experiment is proposed that shows an improvement on the forecast accuracy when accounting for the parameterized representativity error variance.

Annika Vogel (Environment and Climate Change Canada)

Olivia E. Clifton (NASA and Columbia University, NY)

Richard Ménard (Environment and Climate Change Canada)

An accurate error representation is crucial for atmospheric data assimilation as it determines the weighting of the available information. Inspired by the increasing amount of atmospheric model simulations and observations, this contribution exploits the application of statistical error estimation to a multi-model ensemble and observations. In contrast to the standard approach assuming observations to accurately reproduce the truth, this approach allows for an estimation of error (co)variances from all datasets, including the observations. Instead, a certain number of assumptions about the cross-correlation between datasets is required to close the estimation problem, and the remaining cross-correlations can be estimated. While the problem is well-described in theory (as developed in Vogel and Menard, 2023) and easy to compute, it is shown that the actual preparation of data and selection of assumptions become the most critical parts for real applications. Different possibilities on how to determine an optimal set of assumptions based on quantitative measures are discussed. Additionally, a-posteriori measures are introduced that show fundamental violations of assumption and thus give insights on the reliability of the results. The effects on the results are demonstrated with an application to a multi-model ensemble for ozone dry deposition. 

Reference (error estimation theory): Vogel, A. and Ménard, R.: How far can the statistical error estimation problem be closed by collocated data?, Nonlin. Processes Geophys., 30, 375–398, https://doi.org/10.5194/npg-30-375-2023, 2023.

Annika Vogel (Environment and Climate Change Canada)

Richard Ménard (Environment and Climate Change Canada)

Jack Chen (Environment and Climate Change Canada)

2023 was record-breaking for wildfires in Canada and around the globe, with unprecedented impacts on local ecosystems as well as large scale smoke hazards. These exceptional fire impacts rose the public demand for accurate forecasts of smoke plumes as well as analysis of air quality impacts. However, smoke dispersion forecasts rely heavily on highly uncertain estimates of fire activity and emissions. Fire smoke plumes are extreme air quality events with exceptionally high local concentrations and related uncertainties fall outside statistical ranges. Thus, error estimates for data assimilation require specialized approaches which are able to capture their high uncertainties and spatial gradients. Here, a novel assimilation approach, called parametric Kalman filter (PKF), is applied to fire aerosols. By explicitly evolving the main error parameters, the PKF has been proven to provide accurate uncertainty estimates at very low computational costs. 

In this work, error variances are dynamically propagated in the online-coupled air quality forecast model GEM-MACH. Preliminary results indicate that the forecast error distributions for extreme air quality events can be sufficiently approximated by a passive variance-tracer. It is demonstrated that variance source terms which represent the uncertainties in emissions are a critical component for dynamical variance simulation of extreme air quality events. The results are compared to the operational objective analysis at ECCC (Environment and Climate Change Canada) and validated against independent observations.

Longtao Wu (NASA JPL)

Derek J. Posselt (JPL)

Mathias Schreier (JPL)

Jacola Roman (JPL)

Masashi Minamide (University of Tokyo)

Bjorn Lambrigtsen (JPL)

Forecast observing system simulation experiments (OSSEs) are conducted to assess the potential impact of geostationary microwave (GeoMW) sounder observations on numerical weather prediction forecasts. A regional OSSE is conducted using a tropical cyclone (TC) case that is very similar to hurricane Harvey (2017), as hurricanes are among the most devastating of weather-related natural disasters, and hurricane intensity continues to pose a significant challenge for numerical weather prediction. A global OSSE is conducted to assess the potential impact of a single GeoMW sounder centered over the continental United States vs two sounders positioned at the current locations of the National Oceanic and Atmospheric Administration Geostationary Operational Environmental Satellites (GOES) East and West.

It is found that assimilation of GeoMW soundings result in better characterization of the TC environment, especially before and during intensification, which leads to significant improvements in forecasts of TC track and intensity. TC vertical structure (warm core thermal perturbation and horizontal wind distribution) is also substantially improved, as are the surface wind and precipitation extremes. In the global OSSE, assimilation of GeoMW soundings leads to slight improvement globally and significant improvement regionally, with regional impact equal to or greater than nearly all other observation types.

Hao-Lun Yeh, Department of Atmospheric Science, Colorado State University, Fort Collins, Colorado, USA

Peter Jan van Leeuwen, Department of Atmospheric Science, Colorado State University, Fort Collins, Colorado, USA

The increasing significance of nonlinearities in numerical models for the geosciences and in observation operators, which map model states to observation space, demands attention. The recently developed particle flow filter (PFF) is a fully nonlinear and efficient sequential Monte Carlo filter that removes the weight degeneracy problem in particle filters. This is achieved by iteratively transporting the equal-weighted particles from the prior to the posterior distribution. The deterministic version of the PPF has been successfully applied to high-dimensional systems and is unbiased in the limit of an infinite number of particles. However, with a small number of particles, there may be a low-bias ensemble spread, particularly in the observed part of the state space. 

To address this challenge, we have extended this nonlinear filter in two ways. First, we developed a stochastic version that is unbiased for any finite ensemble size. Second, we extended the filter to a smoother, resulting in the Stochastic Particle Flow Smoother (SPFS). This methodology has similarities with an ensemble of 4DVars, with the difference that the particles communicate during the minimization, such that the full nonlinear smoother problem is solved. We demonstrate the performance of the SPFS in high-dimensional systems such as the 1000-dimensional Lorenz-96 model. Our results demonstrate that the SPFS successfully avoids particle collapse and accurately captures the evolutions of particles. We also discuss the difference between using the adjoint sensitivities versus the ensemble sensitivities in the smoother.

Zoe Zibton (National Research Council) 

James Doyle (Naval Research Laboratory)

The onset of tropical cyclone (TC) rapid intensification (RI) remains a challenge as there is still uncertainty about key processes important for development. Uncertainty in the environment surrounding the TC can add additional challenges to predicting the onset of RI. Through targeted changes to a variety of the TC’s structure and/or its surrounding environment (e.g., sea level pressure, wind speed, moisture, upper-level divergence), an ensemble can be made where members seek to change a particular aspect of the forecast. Adjoint-derived sensitivities and perturbations can be used to evaluate the importance of these processes and environmental conditions, as perturbations to the forecast are dependent on the definition of the response function (e.g., forecast aspect being evaluated). This concept has been successfully developed and applied in the mid-latitudes for a mesoscale convective system (Xu et al., 2001). Since adjoint sensitivities typically contain both structured mesoscale, and synoptic-scale patterns along with smaller-scale, highly noisy patterns it is possible that by using the ensemble approach, the dominant physical processes can be revealed. Evaluating the evolution of perturbations and variance of the ensemble members can elucidate the likely scenarios for potential impacts to development throughout the forecast. Furthermore, through analyzing initial perturbation distribution, regions where there is enhanced uncertainty for how the TC will develop can be elucidated.

The presented work uses the U. S. Navy's Coupled-Ocean Atmosphere Mesoscale Prediction System (COAMPS) and its adjoint to assess the application of an adjoint-derived ensemble for TCRI onset. Through this approach, the ensemble is designed to capture the potential spread in the forecast to assess the predictability of the TC. This includes what regions and processes in the forecast are important for multiple metrics related to development. Results can be compared to typical ensemble members to understand the ability to capture the forecasted spread in intensity and track.