the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Quantification of regional terrestrial biosphere CO2 flux errors in v10 OCO-2 MIP models using airborne measurements
Brendan Byrne
Brad Weir
Lesley E. Ott
Kathryn McKain
Bianca Baier
Luciana V. Gatti
Abstract. Multi-inverse modeling inter-comparison projects (MIPs) provide a chance to assess the uncertainties in inversion estimates arising from various sources such as atmospheric CO2 observations, transport models, and prior fluxes. However, accurately quantifying ensemble CO2 flux errors remains challenging, often relying on the ensemble spread as a surrogate. This study proposes a method to quantify the errors of regional terrestrial biosphere CO2 flux estimates from 10 inverse models within the Orbiting Carbon Observatory-2 (OCO-2) MIP by using independent airborne CO2 measurements for the period 2015–2017. We first calculate the root-mean-square error (RMSE) between the ensemble mean of posterior CO2 concentration estimates and airborne observations and then isolate the CO2 concentration error caused solely by the ensemble mean of posterior terrestrial biosphere CO2 flux estimates by subtracting the errors of observation and transport in seven regions. Our analysis reveals significant regional variations in the average monthly RMSE over three years, ranging from 0.90 to 2.04 ppm. The ensemble flux error projected into CO2 space is a major component that accounts for 58–84 % of the mean RMSE. We further show that in five regions, the observation-based error estimates exceed the atmospheric CO2 errors computed from the ensemble spread of posterior CO2 flux estimates by 1.37–1.89 times, implying an underestimation of the actual ensemble flux error, while their magnitudes are comparable in two regions. By identifying the most sensitive areas to airborne measurements through adjoint sensitivity analysis, we find that the underestimation of flux errors is prominent in eastern parts of Australia and East Asia, western parts of Europe and Southeast Asia, and midlatitude North America, suggesting the presence of systematic biases related to anthropogenic CO2 emissions in inversion estimates. The regions with no underestimation were southeastern Alaska and northeastern South America. Our study emphasizes the value of independent airborne measurements not only for the overall evaluation of inversion performance but also for quantifying regional errors in ensemble terrestrial biosphere flux estimates.
- Preprint
(1494 KB) - Metadata XML
-
Supplement
(760 KB) - BibTeX
- EndNote
Jeongmin Yun et al.
Status: open (until 27 Dec 2023)
-
CC1: 'Comment on egusphere-2023-2258', Andrew Jacobson, 15 Nov 2023
reply
1. The OCO-2 v10 MIP sampled a much wider set of aircraft data than
those used in this study. In particular NOAA operates a light aircraft
program that produces regular profiles of CO2 measurements over North
America and Raratonga. These data should be well suited to the
analysis conducted here due to the regular sampling frequency, nearly
continuous coverage, and altitudes sampled. For some reason, of these
timeseries stations, only the data from Dahlen, North Dakota (DND) and
Marcellus, Pennsylvania (MRC) were included in Table 1 of the
manuscript. In addition to these two sites, there are evaluation data
in the OCO-2 MIP samples from timeseries over:Briggsdale, Colorado - (CAR)
Offshore Cape May, New Jersey - (CMA)
Carbon in Arctic Reservoirs Vulnerability Experiment (CARVE) - (CRV)
Estevan Point, British Columbia - (ESP)
East Trout Lake, Saskatchewan - (ETL)
Homer, Illinois - (HIL)
INFLUX (Indianapolis Flux Experiment) - (INX)
Park Falls, Wisconsin - (LEF)
Offshore Portsmouth, New Hampshire (Isles of Shoals) - (NHA)
Poker Flat, Alaska - (PFA)
Rarotonga - (RTA)
Offshore Charleston, South Carolina - (SCA)
Southern Great Plains, Oklahoma - (SGP)
Offshore Corpus Christi, Texas - (TGC)
Trinidad Head, California - (THD)
West Branch, Iowa - (WBI)2. This reviewer's experience with simulation of aircraft measurements
is that model residuals are strongly affected by altitude and by
season. The analysis here does not discriminate by either of these
factors, except to choose an altitude range apparently chosen to
minimize the effect of residuals closer to the surface. Should the
model residuals have significant variability by these factors, the
evaluation criteria would be affected and possibly dominated by those
factors, which would confound the statistical conclusions of this
work. I suggest that a factor analysis, possibly an analysis of
variance, is needed to determine whether model residuals are driven by
these factors.
3. Lines 124-125: "measurements made between 1 and 5 km altitude" does
not specify whether this means above ground level or above sea
level. This needs to be specified. Furthermore, if this altitude range
is above sea level then it is entirely possible that highly-variable
PBL measurement data are included in the evaluation data, since many
aircraft data were collected over topography with surface elevations
of hundreds of meters ASL. This would cloud the analysis with noisy
measurements having strong signals of local exchange.4. It is not clear whether the analysis excludes measurements that
were assimilated in the LNLGIS experiment. This is a fundamental piece
of information needed to understand the analysis and should absolutely
be explicitly stated. If assimilation data are included, then the
entire analysis needs to be considered differently.5. The INPE PFP used in this study data have not been screened for water
vapor contamination. This is a known problem with PFPs in humid
environments and can lead to both a low bias and spurious variability
in CO2 measurements. This is a particular concern with tropical
aircraft samples due to expected high humidity of sampled air. There
are indications that water vapor contamination can persist in PFP
flasks so that even dry high-altitude samples may be affected. This
water vapor issue in aircraft PFPs has been documented in Baier et
al. (2019,
https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JD031339) and
reported at various meetings
(e.g. https://gml.noaa.gov/publications/annual_meetings/2019/abstracts/74-190401-B.pdf).As reported to the authors in OCO-2 meetings, about one-third of
historical NOAA PFP measurements have been flagged due to suspected
water vapor contamination. In the same meetings the authors were
cautioned about this issue affecting INPE PFP data. In ObsPack
products, INPE PFP data are all flagged as "do not assimilate",
indicating that they are neither suitable for assimilation nor for
evaluation purposes. Finally, these data are distributed in a special
ObsPack product labeled "restricted" in part to warn users about the
problem.6. The CO2 measurement data used in this study have not been correctly
cited. It also is not clear whether ObsPack data providers have been
properly acknowledged. The OCO-2 ObsPack product is a "composite"
product created from seven source ObsPacks. The source products need
to be cited following the instructions at
https://gml.noaa.gov/ccgg/obspack/citation.php (available also in the
distributed metadata). Use of an ObsPack product also includes usage
terms which suggest that it may be appropriate to offer coauthorship
to the data providers. The seven source ObsPacks are listed in the
metadata directory of the downloaded product. In the current draft,
only the obspack_co2_1_GLOBALVIEWplus_v6.1_2021-03-01 product is
cited, whereas apparently there are data used from five other
ObsPacks: the NRT product, the Manaus product, the INPE product, the
CONTRAIL product, and the AirCore product.Citation: https://doi.org/10.5194/egusphere-2023-2258-CC1
Jeongmin Yun et al.
Jeongmin Yun et al.
Viewed
HTML | XML | Total | Supplement | BibTeX | EndNote | |
---|---|---|---|---|---|---|
150 | 62 | 14 | 226 | 18 | 6 | 7 |
- HTML: 150
- PDF: 62
- XML: 14
- Total: 226
- Supplement: 18
- BibTeX: 6
- EndNote: 7
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1