Daily CO 2 flux estimates over Europe from continuous atmospheric measurements: 1, inverse methodology

Philippe Peylin, P. J. Rayner, Philippe Bousquet, C. Carouge, F. Hourdin, Philippe Heinrich, Philippe Ciais, Aerocarb contributorsAtmospheric 2005 · DOI

TL;DR

This paper introduces a novel inverse method to estimate daily CO2 fluxes over Europe using continuous atmospheric measurements. By leveraging high-frequency data and a sophisticated atmospheric transport model, the researchers aim to improve our understanding of regional carbon cycles, a crucial step for monitoring and managing atmospheric CO2 levels.

What Problem Does This Paper Address?

Accurately determining the sources and sinks of atmospheric CO2 is fundamental to understanding and managing the global carbon cycle. Traditional "bottom-up" approaches aggregate local flux estimates, which can be prone to errors from extrapolation and model reliance. "Top-down" or inverse methods infer fluxes from atmospheric concentration measurements, but these have historically been limited by sparse data and uncertainties in atmospheric transport models. This paper tackles the challenge of achieving high temporal (daily) and spatial resolution in flux estimates, particularly for regional applications, by developing an inverse methodology that can effectively utilize continuous, high-frequency atmospheric CO2 measurements.

Methods

The core of this work is a new inverse method designed for regional applications using continuous atmospheric CO2 concentration data. The methodology is built upon a "matrix approach" and Bayesian formalism for regularization. Key aspects include:

  • High-Resolution Inverse Modeling: The method aims to infer fluxes at the resolution of the atmospheric transport model's grid (approximately 50x50 km over Europe) and on a daily timescale.
  • "Retro-tracer" Approach: To efficiently compute the sensitivity of measurements to fluxes and initial conditions, the researchers employ a "retro-transport" method within the LMDZ global circulation model. This involves running the transport model backward in time from measurement sites.
  • Treatment of Initial Conditions: Recognizing that short study periods (like the one-month inversion here) require accounting for the initial atmospheric CO2 concentration field, the method explicitly includes and optimizes this field as part of the inversion.
  • Spatial Correlation of Fluxes: To handle the ill-posed nature of the inverse problem and limit error amplification, the method incorporates prior spatial correlations between fluxes of neighboring model grid cells. This is implemented using exponentially decaying correlation functions based on distance.
  • Two-Step Inversion: A preliminary global inversion using monthly data from a broader network of sites is performed first to establish a reasonable prior flux field. This prior is then refined in a second, higher-resolution inversion over Europe using the continuous daily data.

Datasets

  • Continuous CO2 Measurements: Data from six continuous monitoring sites in Europe (Schauinsland, Mace Head, Hegyhatsal tower, Westerland, Monte Cimone, and Plateau Rosa) collected during November 1998.
  • GLOBALVIEW-CO2: Used for the preliminary global inversion, this dataset comprises smoothed monthly CO2 concentrations from approximately 120 surface stations.
  • ECMWF Analyzed Fields: Used for "nudging" the LMDZ model's meteorological simulation to better represent atmospheric transport.
  • Prior Flux Information: Monthly flux patterns from the TURC model for land biosphere and Takahashi et al. (2002) for the ocean were used as prior information. Fossil fuel emissions were taken from Andres et al. (1996) and Marland et al. (2001).

Key Insights and Ideas

  • The Power of High-Frequency Data: The study demonstrates that continuous, daily measurements, even from a limited number of sites, can provide significant information for constraining regional CO2 fluxes, going beyond what is achievable with monthly averaged data.
  • Initial Conditions Matter: For short-term inversions (like a month), the initial state of the atmosphere is not negligible and can significantly influence flux estimates. Optimizing this initial condition is crucial for accurate results.
  • Spatial Correlation is Key: The choice of spatial correlation length for prior flux uncertainties critically impacts the spatial extent and magnitude of inferred flux changes. A longer correlation length leads to a more widespread influence of information from observing sites.
  • Global Context is Essential: A reasonable global flux field is a prerequisite for a regional inversion. The influence of far-field fluxes grows throughout the inversion period, highlighting the interconnectedness of regional and global carbon cycles.
  • Model-Data Fit as a Diagnostic: A good prior model-data fit is essential. Significant misfits can indicate issues with the transport model or the prior flux estimates, which need to be addressed before or during the inversion.

Key Results

  • Improved Model-Data Fit: The optimized model concentrations, incorporating the inverted fluxes, captured most of the daily variations in CO2 concentrations at the six European sites, resulting in a significantly reduced chi-square diagnostic (0.09).
  • Dominance of European Land Fluxes: For the European sites, the contribution of European land fluxes to the daily CO2 concentration variations was found to be dominant, with North Atlantic fluxes being smaller.
  • Time-Dependent Contributions: The contribution of distant regions to the modeled concentration increased over time, while the contribution of initial conditions decreased, approaching a non-zero mean.
  • Impact of Correlation Length: Increasing the correlation length for prior flux uncertainties (from 500 km to 2000 km for land pixels) led to a more uniform spatial impact of flux increments across Europe and a significant increase in the overall error reduction, suggesting that larger regions in previous inversions might have underestimated the power of continuous measurements.
  • Significant Flux Adjustments: The inversion resulted in flux changes, particularly around the Baltic area, exceeding 70 gC/m²/month, indicating substantial adjustments to the prior estimates.

Limitations and Future Work

  • Diurnal Cycle Neglect: The study explicitly neglects diurnal variations in both CO2 concentrations and fluxes, acknowledging that this is a simplification and that diurnal cycles can be important, especially for low-elevation continental towers. Future work will address this.
  • Transport Model Uncertainties: While the LMDZ model is deemed capable of representing major transport features, the paper acknowledges that transport errors, particularly in simulating boundary layer ventilation, can affect flux estimates.
  • Limited Data Density and Duration: The study uses data from only six sites for one month. The authors emphasize that this is a methodological exploration and that more stations and longer study periods are needed for more robust flux estimates.
  • Future Work: The authors plan to extend this work to include more stations, longer time series (e.g., the year 2001), and to address the diurnal cycle and diurnal rectification. Comparisons with independent bottom-up calculations are also planned.

Summary updated

Abstract

This paper presents an inverse method for inferring trace gas fluxes at high temporal (daily) and spatial (model grid) resolution from continuous atmospheric concentration measurements. The method is designed for regional applications and for use in intensive campaigns. We apply the method to a one month inversion of fluxes over Europe. We show that the information added by the measurements depends critically on the smoothness constraint assumed among the source components. We show that the initial condition affects the inversion for 20 days, provided one has enough observing sites to constrain regional fluxes. We show that the impact of the far-field fluxes grows throughout the inversion and hence a reasonable global flux field is a prerequisite for a regional inversion.
BibTeX
@inproceedings{peylin2005daily,
  author = {Philippe Peylin and P. J. Rayner and Philippe Bousquet and C. Carouge and F. Hourdin and Philippe Heinrich and Philippe Ciais and AEROCARB contributors},
  title = {{Daily CO <sub>2</sub> flux estimates over Europe from continuous atmospheric measurements: 1, inverse methodology}},
  booktitle = {Atmospheric},
  year = {2005},
  doi = {10.5194/acp-5-3173-2005},
}
Figure