Spaceborne detection of XCO2 enhancement induced by Australian mega-bushfires

The 2019–20 Australian mega-bushfires, which raged particularly over New South Wales and Victoria, released large amounts of toxic haze and CO2. Here, we investigate whether the resulting CO2 enhancement can be directly detected by satellite observations, based on National Aeronautics and Space Administration’s Orbiting Carbon Observatory-2 (OCO-2) column-averaged CO2 (XCO2) product. We find that smoke from wildfires can greatly obscure satellite observations, making the available XCO2 mainly locate over outer fringes of plumes downwind of the major mega-bushfires in eastern Australia in three orbit observations during November–December 2019, with their enhancements of approximately 1.5 ppm. This fire-induced CO2 enhancement is further confirmed using an atmospheric transport model, Goddard Earth Observing System-Chem, forced by satellite observation-derived fire product Global Fire Emissions Database, version 4.1 and wind observations, with comparable simulated XCO2 enhancements. Model simulation also suggests that the sensitivity of the downwind maximum XCO2 enhancement is 0.41 ± 0.04 ppm for 1 TgC d−1 fire emissions. In sum, though detectable to some extent, it remains a challenge to get the accurate maximum XCO2 enhancements due to the gaps in XCO2 detections obscured by smoke. Understanding the capability of OCO-2 XCO2 detection is prerequisite for monitoring and constraining wildfire CO2 emissions by inversions.


Introduction
The unprecedented 2019-20 Australian bushfires, which raged particularly over eastern Australia (the worst hit states were New South Wales and Victoria), drew worldwide attention. The worst fire incidents started at the beginning of November 2019. During November and December 2019, eastern Australia experienced recordbreaking temperatures and widespread severe and extreme drought with decreased precipitation (supplementary figure 1 (available online at https://stacks.iop.org/ERL/15/124069/mmedia)), as indicated by previous studies , Nolan et al 2020. This anomalous hot and dry climate may have been driven by natural atmospheric dynamics and anthropogenic global warming (Phillips and Nogrady 2020). The unusual positive Indian Ocean Dipole (IOD) in 2019 was one of the strongest such events in history. During the positive phase of IOD, decreased precipitation and warmer temperatures often occur over parts of Australia (Saji et al 1999, Saji and Yamagata 2003, Cai et al 2009. A sudden stratospheric warming above Antarctica can also cause the hot and dry conditions over Australia (Phillips and Nogrady 2020). These hot and dry conditions greatly contribute to catastrophic bushfires.
The mega-bushfires of 2019-20 burned vast areas of temperate broadleaf and mixed forests , Nolan et al 2020. Approximately 5.8 million hectares were burned between September 2019 and early January 2020, accounting for around 21% of Australia's temperate broadleaf and mixed forests . The fires greatly threatened the Gondwana Rainforests, a World Heritage site (Kooyman et al 2020). The fires caused losses of USD billions from the economy, as well as deaths of more than 30 humans and countless animals. Concurrently, the mega-bushfires released large amounts of toxic haze and CO 2 , further threatening human health and contributing to the increase of global CO 2 concentrations.
Orbiting Carbon Observatory-2 (OCO-2) is the Earth remote sensing satellite used by National Aeronautics and Space Administration (NASA) to measure the atmospheric CO 2 concentration (Crisp et al 2008, Crisp andOCO-2 Team 2015). The OCO-2 column-averaged CO 2 (XCO 2 ) products have been used to detect the XCO 2 enhancement over both megacity and volcanoes (Schwandner et al 2017), analyze the variations of XCO 2 over the Niño 3.4 regions during the 2015-16 El Niño event (Chatterjee et al 2017), estimate the land-atmosphere carbon flux and fire emissions during the El Niño event (Heymann et al 2017, Liu et al 2017, and estimate the CO 2 emissions by megacities (Zheng et al 2020) and single power plants (Nassar et al 2017). In the previous study (Heymann et al 2017), the XCO 2 enhancement induced by fires was calculated as the difference between OCO-2 XCO 2 and background XCO 2 values on 0.5 • × 0.5 • grids, where background XCO 2 was derived from a CO 2 model 'CarbonTracker' and observations unaffected by fires determined by the Stochastic Time-Inverted Lagrangian Transport model. In contrast, we in this study will investigate whether OCO-2 can directly and clearly detect the XCO 2 enhancement induced by Australian mega-bushfires, analogous to the study of Schwandner et al (2017), based on the combination of OCO-2 L2 XCO 2 product and Goddard Earth Observing System (GEOS)-Chem model simulations. Understanding the capability of OCO-2 to detect XCO 2 enhancement induced by eastern Australia's fires is important for monitoring and constraining wildfire CO 2 emissions over the whole of Australia.

OCO-2 data
The space-based XCO 2 measurements used to detect XCO 2 enhancement in this study are retrieved from the OCO-2 mission (Crisp et al 2008, Crisp andOCO-2 Team 2015), which was launched into a nearpolar orbit on 2 July 2014. The OCO-2 instrument has a narrow swath with the footprint <1.3 km × 2.25 km. Its retrieval algorithm was described by Crisp et al (2012) andO'Dell et al (2012). In this study, the XCO 2 retrievals are the version 9r level 2 product, which can be retrieved from the Goddard Earth Sciences Data and Information Services Center. Version 9r is the latest version of the XCO 2 dataset, containing biascorrected XCO 2 and other selected fields. In this analysis, we used the good quality XCO 2 data, which were determined by the variable 'xco2_quality_flag' in the file.

Atmospheric transport model
The GEOS-Chem model version 12.5.0 (The International GEOS-Chem User Community 2019) was adopted in this study to simulate the XCO 2 enhancement induced by Australian bushfires. It is a global 3D atmospheric chemistry model driven by meteorological fields which are freely available at the GEOS of the NASA Global Modeling Assimilation Office.
Specifically, the driving meteorological fields and boundary carbon fluxes are as follows: (a) The meteorological fields are the Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) (Gelaro et al 2017). MERRA-2 can utilize the newer microwave and hyperspectral infrared radiance observations, and is the first long-term global reanalysis to assimilate space-based aerosol measurements. In this study, we ran the GEOS-Chem model at the horizontal resolution of 4 • × 5 • and 47 hybrid sigma-pressure levels up to 0.01 hPa. In addition, in the analysis, we further used its variables of surface air temperature, U and V winds at 940 hPa, and aerosol optical depth (AOD). (b) The fossil fuel emissions are from the Open-Data Inventory for Anthropogenic Carbon Dioxide (ODIAC) which is a global CO 2 emission product from fossil fuel combustion Maksyutov 2011, Oda et al 2018). In this study, we used the latest version of the ODIAC fossil fuel emission product 'ODIAC2019' with the original resolution of  (Wang et al 2018). The VEGAS NRT was operationally run to present at 0.5 • × 0.5 • with the automatically updated meteorological fields such as precipitation, surface air temperature, and radiations. The details for the meteorological fields were described in Wang et al (2018). In this study, we regarded the difference between F TA and simulated fire emissions (F fire ) as 'land-atmosphere carbon flux without fires' .

Sensitivity experiments
Two sets of experiments were conducted for the main text, with their experimental designs as follows: (a) Control experiment (denoted as 'CTL'): full surface carbon emissions considered, including extrapolated ODIAC fossil fuel emissions, extrapolated oceanic carbon exchange, land-atmosphere carbon flux without fires simulated by VEGAS NRT, and GFED4.1 wildfire emissions. (b) Sensitivity experiment (denoted as 'S1'): same as CTL, but wildfire carbon emissions are set to zero over Australia from November to December 2019.
Thus, we can derive the XCO 2 and 3D CO 2 enhancement caused by the Australian bushfires from the difference between CTL and S1. We ran these two experiments from 1 January 2018 to 31 December 2019 with the CO 2 restart fields from previous longterm simulations.

Australian bushfire carbon emissions and OCO-2 detections
Out-of-control wildfires hit the temperate forests, dominated by fire-prone Eucalyptus, in New South Wales and Victoria from the beginning of November 2019, causing an enhanced amount of carbon (or CO 2 ) release ( figure 1(a)). On average, the carbon release in 0.25 • × 0.25 • grid from the mega-fires between November and December 2019 is in general higher than 15 gC m −2 d −1 , in which emissions can reach an extreme of up to 126.9 gC m −2 d −1 (figure 1(a)). Australia's fossil fuel carbon emissions are also centered over its eastern coastal areas, around the major cities as well as the surrounding power plants ( figure 1(b)). It is therefore a concern that fossil fuel emissions may interfere with the spaceborne detection of XCO 2 enhancement induced by wildfires. Fortunately, the strength of fossil fuel carbon emissions is found to be much lower than that of the wildfire emissions in these two months. The total fossil fuel carbon emissions in the region of 144-154 • E, 40-26 • S is approximately 0.2 TgC d −1 , leading to a total of 12.2 TgC (or 44.7 TgCO 2 ) for the two months. In contrast, the strength of total wildfire carbon emissions in this region is close to zero before November, and then it becomes much stronger and more variable leading to emissions of approximately 119.0 TgC (or 436.3 TgCO 2 ) during November and December 2019 ( figure 1(d)). The wildfire carbon emissions are therefore approximately ten times stronger than fossil fuel emissions in these two months. In addition, the simulated terrestrial ecosystems in this region show a weak absorption (excluding the simulated fire emissions) with the amplitude of 0.08 TgC d −1 (figure 1(d)) because November and December are the austral late spring and early summer. Hence, the surface carbon emissions in this region are dominated by the mega-bushfire carbon emissions.
In this region, prevailing winds in the planetary boundary layer are west, west-southwest, and southwest, which blow the emitted CO 2 to the ocean ( figure 1(c)). The average wind speed in November and December at 940 hPa is 6.3 m s −1 , resulting in about 5.6 • longitude advection for 1 d.
We checked the OCO-2 XCO 2 orbit observations around eastern Australia's mega-bushfires between November and December 2019, and selected three orbits that could be used to detect XCO 2 enhancement induced by wildfires. These three orbits were observed at around 03:30 Coordinated Unviersal Time (UTC) on 14 November 2019, 21 November 2019, and 28 December 2019. Figure 2 shows their respective distributions, associated with corresponding wildfire carbon emissions and atmospheric circulation. The XCO 2 values are gradually enhanced from the regions far away from the wildfires to the wildfire downwind areas (figures 2(a)-(c)). These variations are significantly different from the pulse of XCO 2 enhancements induced by the single power plant fossil fuel CO 2 emissions (supplementary figure 2), which is similar to the previous results (Nassar et al 2017, Zheng et al 2020. Taking the XCO 2 values furthest from the wildfires as the background levels, we used the available XCO 2 values over the fringe areas of plumes to calculate maximum XCO 2 enhancement in these three orbits with their respective   figure) patterns, we choose the XCO2 data far away from the fires to calculate background XCO2 levels (blue dots here) and data over the fringes of plumes (coral dots) to calculate XCO2 enhancement against background XCO2. (d) Box plots for background XCO2 levels and enhanced XCO2 for the three orbits. (e)-(g) as (a)-(c), but with the spatial patterns of AOD.
It is worth mentioning that the gaps of XCO 2 in these three orbits over the wildfire downwind areas are mainly obscured by the amount of smoke emitted by the wildfires. This smoke can enhance the AOD at the local and downwind regions (figures 2(e)-(g)). The local AOD can be higher than 2.7, greatly contaminating the OCO-2 XCO 2 observations. The missed observations were often located in the core wildfire-induced XCO 2 enhancement regions, probably leading to an underestimate of the maximum XCO 2 enhancement.

GEOS-Chem model simulations
The observed XCO 2 variations can be influenced by surface carbon emissions and atmospheric circulation. To further verify that the XCO 2 enhancement observed from the three orbits was caused by wildfire CO 2 emissions, we use the GEOS-Chem model to disentangle the wildfire-induced XCO 2 enhancement with sensitivity experiments (figure 3).
The simulated XCO 2 during November and December in the control experiment has similar spatial patterns and amplitudes compared with OCO-2 observations, although the simulated XCO 2 is approximately 1 ppm higher than results from OCO-2 (supplementary figure 3). In combination with the sensitivity experiment that sets the wildfire carbon flux over Australia as zero emissions during November and December 2019, we can derive the wildfire-induced atmospheric CO 2 variations. Figure 3 shows the spatial patterns for simulated XCO 2 enhancement at around 03:30 UTC on 14 November 2019, 21 November 2019, and 28 December 2019. The higher XCO 2 values from these three orbits observed by OCO-2 coincide with distributions of the simulated XCO 2 enhancement. Importantly, the simulated XCO 2 enhancements are basically comparable to the XCO 2 enhancements detected by OCO-2 (figure 2(d)), confirming that these directly detected XCO 2 enhancements are caused by wildfires. The wildfire CO 2 emissions can also greatly enhance surface CO 2 concentration (supplementary figure 4). From the height-longitude cross-section, we can see that significantly enhanced CO 2 concentrated under 1 km height, and that the enhanced CO 2 in the free atmosphere will be quickly advected (supplementary figure 5).
The sensitivity of XCO 2 maximum enhancement to the wildfire carbon emissions is presented in figure 4. The wildfire carbon emissions were calculated as the total emissions in the region of 144-154 • E, 40-26 • S (box in figure 1(a)) in the 24 h before 03:30 UTC each day; the XCO 2 maximum enhancement is the average of the simulated XCO 2 values at 03:30 UTC higher than 90th percentile in the region of 145-160 • E, 40-26 • S. Regression analysis suggests that 1 TgC d −1 of wildfire carbon emissions can lead to 0.41 ± 0.04 ppm XCO 2 maximum enhancement with the R 2 of 0.63 (P < 0.01). The relationship between the directly calculated XCO 2 maximum enhancements from OCO-2 orbits and wildfire carbon emissions is also consistent with the simulated relationship.

Discussion
Although we have used GEOS-Chem model to verify that the OCO-2 XCO 2 enhancements observed at these three orbits are very likely induced by wildfire CO 2 emissions, one concern is the effect of the uncertainties of the OCO-2 data. Worden et al (2017) suggested that effects of aerosols or surface albedo can influence the accuracy of XCO 2 data. Therefore, we present the respective XCO 2 uncertainties at these three orbits in supplementary figure 6. The XCO 2 at 03:30 UTC on 21 November 2019 shows higher uncertainties with values approximately 0.6-0.8 ppm (supplementary figure 6(b)), while the uncertainties at the other two orbits show smaller values approximately 0.4-0.6 ppm (supplementary figures 6(a) and (c)). Indeed, aerosols emitted by wildfires can increase the uncertainties at some pixels, but majorities of the uncertainties are smaller than 0.8 ppm. Therefore, it is convincing that the approximately 1.5 ppm XCO 2 enhancements downwind of the major mega-bushfires in eastern Australia at the three orbits cannot be caused by the variations of retrieval uncertainties.
Additionally, we ran two extra sensitivity experiments (supplementary text 2) to explore the influence of carbon flux anomalies induced by the ecosystem and wildfires on the XCO 2 variations. We found that the XCO 2 can be significantly enhanced over the southeast regions of Australia and surrounding oceans by wildfire anomalies in November and December 2019, whereas the enhanced XCO 2 induced by ecosystem carbon anomalies without fires occurs over the northwest regions of Australia (supplementary figure 7). The amplitudes of XCO 2 anomalies induced by wildfire anomalies peak at approximately 0.45 ppm. We also calculated the OCO-2 XCO 2 anomalies based on the method of Chatterjee et al (2017). Compared with the simulated amplitudes of XCO 2 anomalies over the southeast regions of Australia and surrounding oceans (supplementary figure 7(d)), the OCO-2 XCO 2 anomalies significantly underestimate the amplitudes over the corresponding regions (supplementary figure 8(a)). This phenomenon is largely caused by smoke contamination from wildfires, which can lead to higher AOD (supplementary figure 8(b)). In addition, the pattern of OCO-2 XCO 2 anomalies contains other information that makes it hard to clarify the influence of eastern Australia's mega-bushfires. Therefore, it is hard to explore the anomalous XCO 2 variations as in the previous study that investigated the anomalous XCO 2 variations caused by the extreme El Niño (Chatterjee et al 2017).

Conclusions
Base on the NASA OCO-2 XCO 2 product, we in this study investigated whether the recent satellite observations can detect the XCO 2 enhancement induced by the Australian mega-bushfires in November-December 2019. We picked out three orbits downwind of the mega-bushfires in eastern Australia. Though the XCO 2 values at these three orbits exhibit the gradual increase from the regions furthest from the wildfires to the wildfire downwind areas, the XCO 2 values in the core wildfire-induced XCO 2 enhancement regions are obscured by the amount of smoke, making the spaceborne detection challenging. We suggest the approximately 1.5 ppm XCO 2 enhancements at these three orbits, calculating from the difference of available XCO 2 values over outer fringes of plumes and XCO 2 values furthest from the wildfires. This fire-induced XCO 2 enhancement was further confirmed by GEOS-Chem simulations. Based on the simulated results, we point out that the sensitivity of the downwind maximum XCO 2 enhancement is 0.41 ± 0.04 ppm with 1 TgC d −1 fire emissions in the region of 144-154 • E, 40-26 • S. study was supported by the National Natural Science Foundation of China (Grant No. 41807434), the National Key R&D Program of China (Grant Nos. 2016YFA0600204 and 2017YFB0504000). NZ's participation was supported by Grant Nos. NOAA-NA18OAR4310266, NASA-80NSSC18K0908, and NIST-70NANB14H333. Jun Wang (Grant No. 201906195014) thank the China Scholarship Council for funding. The work was conducted while J W was a visiting scientist at the University of Maryland, whose hosting is gratefully acknowledged. We are grateful to the High Performance Computing Center (HPCC) of Nanjing University for doing the numerical calculations on its blade cluster system.
All data that support the findings of this study are included within the article (and any supplementary information files).