[Paper Review] Towards a Prototype Global CO2 Emissions Monitoring System for Copernicus

Nicolas Bousserez|arXiv (Cornell University)|Oct 25, 2019

Atmospheric and Environmental Gas Dynamics15 references4 citations

TL;DR

This paper proposes a hybrid ensemble-variational data assimilation system for ECMWF's IFS to enable joint state/parameter estimation in a prototype Copernicus CO2 emissions monitoring service. By integrating ensemble-based error covariances with adjoint-based increment propagation and extending the 4D-Var window using ensemble approximations of transport Jacobians, the method enhances spatial resolution, incorporates missing physical processes, and enables long-window source inversion for CO2 and co-emitted tracers like CO and NO2.

ABSTRACT

This document describes in detail a new hybrid ensemble-variational algorithm that generalizes existing ensemble or variational data assimilation approaches and would enable joint state/parameter estimations in ECMWF's Integrated Forecasting System (IFS). The proposed methodology is intended to serve as the basis for a prototype Copernicus CO2 emission monitoring service. The main characteristics of the system are: 1) 4D hybridization of ensemble information with full-rank statistical modeling by combining an ensemble-based increment with an adjoint-based increment propagation, allowing one to increase current spatial resolution and/or include forward model processes missing from the adjoint integration; 2) combination of tangent-linear and adjoint solvers with ensemble-based approximations of transport Jacobians to construct a long-window 4D-Var with timescales relevant to greenhouse gas source inversion. The proposed methodology is non-intrusive in the sense that the main structure of the current incremental 4D-Var algorithm remains unchanged, while the additional computational cost associated with the source inversion component is minimized.

Motivation & Objective

• Develop a robust, scalable methodology to integrate heterogeneous CO2 emission products (global to local) into a unified global inversion system.
• Address limitations of the current 12-hour 4D-Var window and transport-only adjoint models in the IFS for greenhouse gas source inversion.
• Enable joint inversion of CO2 and co-emitted tracers (e.g., CO, NO2) to improve source attribution and disentangle anthropogenic from biogenic emissions.
• Minimize computational overhead by designing a non-intrusive algorithm that preserves the existing incremental 4D-Var structure.
• Support policy-relevant monitoring of national and regional emissions reductions, including Nationally Determined Contributions (NDCs).

Proposed method

• Proposes a 4D hybrid-4D-Var algorithm that combines ensemble-based increments with adjoint-based increment propagation to improve background error covariance estimation.
• Introduces a 4D generalization of the hybrid B matrix using stochastic perturbations of physics tendencies (SPPT) to account for model errors in variational optimization.
• Utilizes ensemble approximations of transport Jacobians to extend the assimilation window beyond 12 hours, enabling long-window 4D-Var for greenhouse gas source inversion.
• Employs tangent-linear and adjoint models alongside ensemble-based approximations to construct a long-window 4D-Var system with timescales relevant to atmospheric CO2 transport.
• Enables assimilation of external posterior emission products by approximating error covariances between observations and posterior fluxes using sample deviations from prior and posterior ensembles.
• Provides a non-intrusive framework that maintains the core structure of the existing incremental 4D-Var while adding source inversion capabilities with minimal computational cost.

Experimental results

Research questions

• RQ1 How can a hybrid ensemble-variational system be designed to enable joint state and parameter estimation in a global atmospheric chemistry model?
• RQ2 What modifications are required to extend the 12-hour 4D-Var window in the IFS to support long-term greenhouse gas source inversion?
• RQ3 How can chemical reaction processes missing in the adjoint model be effectively incorporated into the increment propagation without increasing computational cost?
• RQ4 Can co-emitted tracers such as CO and NO2 improve the accuracy and specificity of anthropogenic CO2 emission attribution in a global inversion system?
• RQ5 How can posterior emission products from diverse models and spatial scales be consistently assimilated into a global IFS-based inversion system?

Key findings

• The proposed hybrid 4D-Var system enables long-window 4D-Var for greenhouse gas source inversion by using ensemble approximations of transport Jacobians, overcoming the 12-hour window limitation of the current IFS.
• The method achieves partial 4D localization of ensemble-based error covariances through the combination of ensemble and adjoint-based increments, improving spatial resolution and error representation.
• Chemical processes such as reactions and higher-resolution transport features can be included in the ensemble-based increment while keeping the adjoint model integration computationally efficient.
• Joint inversion of CO2 and co-emitted tracers (e.g., CO, NO2) provides additional constraints on source attribution, improving the separation of anthropogenic and biogenic CO2 fluxes.
• External posterior emission products from regional or local inversion systems can be assimilated into the global IFS using sample-based approximations of prior and posterior deviations, enabling multi-scale fusion.
• The framework is non-intrusive, preserving the existing incremental 4D-Var structure while adding source inversion capabilities with minimal computational overhead.