The Orbiting Carbon Observatory (OCO-2) tracks 2–3 peta-gram increase in carbon release to the atmosphere during the 2014–2016 El Niño

The powerful El Niño event of 2015–2016 – the third most intense since the 1950s – has exerted a large impact on the Earth’s natural climate system. The column-averaged CO2 dry-air mole fraction (XCO2) observations from satellites and ground-based networks are analyzed together with in situ observations for the period of September 2014 to October 2016. From the differences between satellite (OCO-2) observations and simulations using an atmospheric chemistry-transport model, we estimate that, relative to the mean annual fluxes for 2014, the most recent El Niño has contributed to an excess CO2 emission from the Earth’s surface (land + ocean) to the atmosphere in the range of 2.4 ± 0.2 PgC (1 Pg = 1015 g) over the period of July 2015 to June 2016. The excess CO2 flux is resulted primarily from reduction in vegetation uptake due to drought, and to a lesser degree from increased biomass burning. It is about the half of the CO2 flux anomaly (range: 4.4–6.7 PgC) estimated for the 1997/1998 El Niño. The annual total sink is estimated to be 3.9 ± 0.2 PgC for the assumed fossil fuel emission of 10.1 PgC. The major uncertainty in attribution arise from error in anthropogenic emission trends, satellite data and atmospheric transport.

The powerful El Niño event of 2015-2016 -the third most intense since the 1950s -has exerted a large impact on the Earth's natural climate system. The column-averaged CO 2 dry-air mole fraction (XCO 2 ) observations from satellites and ground-based networks are analyzed together with in situ observations for the period of September 2014 to October 2016. From the differences between satellite (OCO-2) observations and simulations using an atmospheric chemistry-transport model, we estimate that, relative to the mean annual fluxes for 2014, the most recent El Niño has contributed to an excess CO 2 emission from the Earth's surface (land + ocean) to the atmosphere in the range of 2.4 ± 0.2 PgC (1 Pg = 10 15 g) over the period of July 2015 to June 2016. The excess CO 2 flux is resulted primarily from reduction in vegetation uptake due to drought, and to a lesser degree from increased biomass burning. It is about the half of the CO 2 flux anomaly (range: 4.4-6.7 PgC) estimated for the 1997/1998 El Niño. The annual total sink is estimated to be 3.9 ± 0.2 PgC for the assumed fossil fuel emission of 10.1 PgC. The major uncertainty in attribution arise from error in anthropogenic emission trends, satellite data and atmospheric transport.
Uncertainties in estimates of regional sources (+ve flux) and sinks (−ve flux) of CO 2 and other greenhouse gases, derived from direct inventory methods or inferred from atmospheric observations, have hindered the development of effective policy for reduction of emissions from anthropogenic activity 1 . The large uncertainties obscure the relative roles of management approaches for terrestrial biospheric fluxes and the energy intensity of the industrial activities. For example, the sources and sinks of CO 2 by the tropical land biosphere has remained uncertain 2 and the CO 2 emissions from industries in China are frequently revised by the state and international research communities 3 . While the inventory method suffers from a lack in completeness and transparency, the atmospheric constraint has hitherto been compromised by both the sparseness of observational network, and uncertainties in models employed for regional CO 2 flux calculations 4 .
To improve the time and spatial coverage of the atmospheric CO 2 measurements, NASA launched the OCO-2 satellite in July 2014 5 . Since early September of 2014, OCO-2 has been routinely returning almost one million soundings each day over the sunlit hemisphere. While clouds and large aerosols abundances preclude full-column measurements of CO 2 from most of these soundings, more than 10% (~100,000 soundings/day) yield estimates of the column-averaged dry air mole fraction, XCO 2 . The OCO-2 XCO 2 retrievals, after bias correction, agree well globally with the TCCON for nadir, glint, and target observations, with median differences less than 0.5 parts per million (ppm) and root-mean-square differences typically below 1.5 ppm 6 . If regional scale biases are controlled to similar levels, these data can provide the precision and accuracy needed to characterize CO 2 sources and sinks 7 .
The other factor that affects estimates of CO 2 fluxes from XCO 2 measurements is the biases in the inverse methods using chemistry-transport models (CTMs). The role of such bias has been illustrated using the XCO 2 observations from the first dedicated Greenhouse Gases Observing Satellite "IBUKI" (GOSAT), which was launched on 23 January 2009 by the Japan Aerospace Exploration Agency (JAXA) 8 . Using multiple flux inversions of in situ and satellite CO 2 data, Howeling et al. find that the model-model flux differences quickly increase to >100% of the annual flux on the scale of the subcontinental regions 9 . It is generally understood that the differences in inversion-derived CO 2 fluxes are caused by a variety of the underlying modeling components in the inversion systems, not the CTMs alone 4,9 . The modeling components include a priori flux and uncertainty assumptions, screening and treatment of observational data, and uncertainties in transport models 4 .
The efficiency of the terrestrial ecosystem at absorbing atmospheric carbon dioxide (CO 2 ) depends on the availability of sunlight, soil moisture (fed by precipitation), and air temperature 10,11 . Thus droughts and high temperatures associated with El Niño reduce the ability of the terrestrial ecosystem to assimilate carbon while additional release by frequent occurrence of fires further reduces the uptake of carbon by the terrestrial biosphere [12][13][14][15][16]  The study regions are all dominated by the Indonesian fires despite varying in their exact definitions ("Tropical Asia", "Maritime Southeast Asia" etc.). The range of estimates provides some measure of the considerable uncertainty in our knowledge of the pyrogenic carbon flux. However, each of these anomalies is smaller than those estimated for the 1997/1998 El Niño event for Southeast Asia (~1 PgC) 14,16 .
In addition to the relatively large uncertainties, the above-mentioned carbon flux estimates are limited only to the emission mechanism of biomass burning. CO 2 observations, on the other hand, have the advantage of being more directly linked to the net carbon flux to the atmosphere, i.e., they are not limited to a specific emission mechanism like biomass burning.
Although the equatorial east Pacific Ocean experiences weaker ventilation of deep-water CO 2 during an El Niño, thus a negative CO 2 flux anomaly 22 , but the effect of the ocean component on global total CO 2 flux anomaly is not clear 15,23 . For simplicity of this work, no attempt is made to partition land and ocean fluxes.
Here, we analyze early OCO-2 observations of XCO 2 to quantify the impact of the powerful El Niño event 24 on large scale CO 2 flux anomalies. A record CO 2 rise is predicted for 2016, sufficient to keep the atmospheric level above 400 ppm at Mauna Loa, Hawaii 25 for the foreseeable future. The OCO-2 observations along with CTM simulations are used to make an impact assessment of the ongoing El Niño event on the terrestrial carbon cycle. We estimated monthly CO 2 flux corrections from the differences in OCO-2 measurements and transport model simulations. Comparisons with in situ, ground-based remote sensing and GOSAT observations provide a test of the robustness of the estimated carbon exchange based on the OCO-2 observations.

Results
Model-observation comparison. Figure 1 shows the latitude-time distributions of XCO 2 obtained from NASA's OCO-2 and JAXA's GOSAT instruments 26,27 and the differences with JAMSTEC's atmospheric chemistry-transport model (ACTM) simulations for the period from September 2014 through October 2016 (up to May for GOSAT). Details on observational data selection, ACTM simulations and their processing are given in the Methods section. The OCO-2 minus ACTM results are shown for three combinations of terrestrial and oceanic CO 2 fluxes, namely, CYC64 (Fig. 1b), IAV84 (Fig. 1c) and IAV84 + GFAS (Fig. 1d). The simulated XCO 2 growth rates by ACTM_CYC64 and ACTM_IAV84 overestimated (typically by ~0.5 ppm) and underestimated (by up to 2.0 ppm), respectively, the observed growth rate over this 25-month period. The underestimation of ACTM_IAV84 develops most strongly during Sep-Nov 2015. The ACTM_IAV84 + GFAS simulation most closely follows the OCO-2 observations, compensating in particular for the underestimation after Nov 2015 (referred to as 'best' a priori for flux corrections). All ACTM simulations use the same emissions from FFC at the rate of ~10 PgC yr −1 (Table 1). However, the annual total land and ocean fluxes vary, e.g., −2.86, −6.24, and −4.77 PgC yr −1 , respectively, for CYC64, IAV84 and IAV84 + GFAS cases for period July 2015 to June 2016. One striking difference for the April-July period is that GOSAT -ACTM differences (Fig. 1f,g,h) in the high northern latitudes (>30°N) are more negative than the OCO-2 -ACTM differences (Fig. 1b,c,d). This suggests a surface sofluxurce inversion would produce stronger sosinkurces in the northern high latitudes when GOSAT observations are used compared to using the OCO-2 observations. Figure 2a,b,c show comparisons of XCO 2 as measured by OCO-2 and simulated by ACTM as zonal means for three broad latitude ranges for the period from September 2014 through October 2016. The latitude bands of 10°S-10°N (hereinafter referred to as tropics) and 10°-90° cover 88.6 and 210.7 million km 2 , respectively. When combined into 2.5° × 2.5° grid boxes, the OCO-2 data coverage for the latitude bands poleward of 10° varies from 30% to 50% of the total area. The region south of 10°S has the largest model-observation mismatches, with values up to 2 ppm, with major contributions from the American and Asian sectors, during April to August 2015. The ACTM_IAV84 simulation, on the other hand, most closely follows the OCO-2 observations until July 2015 for the region north of 10°N (Fig. 2a), suggesting that the FFC emissions are reasonably prescribed at an increase of 0.  . Note the striking similarities between OCO-2 and GOSAT measurements and ACTM_IAV84 + GFAS simulation case, particularly over the tropics. Further detailed comparisons of GOSAT and ACTM, with separation for soundings over land and water surfaces suggests the positive model biases in the high latitude regions arise mainly over the ocean surface. Similar plots cannot be made using data from the TCCON or NOAA network sites without significant interpolation in space and time due to the geographically sparse sampling of the ground-based networks. to October 2016 could be underestimated by up to 0.2 PgC, which is assigned as FFC emission increase rate in our a priori model. The ACTM -OCO-2 differences show systematic decrease following the peak in February-March 2016, in particular for the southern latitudes, until October 2016, as the El Niño condition weakens (Fig. 2c).
Because the OCO-2 measurements started less than 6 months before the nominal onset of the 2014-2016 El Niño this data alone cannot be used for calculating anomalous CO 2 emissions. We have used longer time record from GOSAT, TCCON (Total Carbon Column Observing Network) 28 and NOAA cooperative global air sampling network 29 measurements since January 2013 for defining the baseline. Here we report CO 2 flux anomalies with respect to 2013-2014 as the aim of this study is to estimate anomalous CO 2 release for the whole El Niño period. The ACTM_IAV84 simulation successfully simulated CO 2 growth rate during January 2013 to September 2014 (seen as the differences around the 0-line) as measured by GOSAT (Fig. 2d,e,f), TCCON (Fig. 2g,h,i) and NOAA (Fig. 2j,k,l). For the October 2014 to October 2016 (El Niño) period, the ACTM_IAV84 + GFAS simulation most closely simulated the atmospheric XCO 2 measured by GOSAT and TCCON, and also the NOAA flask observations (Fig. 2). Although the ACTM_IAV84 + GFAS simulation very well describes the time evolution of observed XCO 2 in the tropics and most times for the region north of 10°N (mostly within 0.1 ppm), systematic underestimations of up to ~2.0 ppm are seen in the region south of 10°S by April 2016. The larger variability in model-observation mismatches in the northern latitude band (Fig. 2a,d,g) is probably an effect of strong terrestrial biospheric uptake and release cycle, which are not very well constrained by ACTM inversion system using in situ data only. This issue will be addressed later when flux corrections will be validated using TCCON observation.
Global CO2 flux anomaly. Comparing the 3 ACTM simulations with OCO-2 and other measurements, we find that the global pyrogenic emission from GFAS of about 2.64 PgC, which in itself is subject to considerable uncertainties, is similar to our XCO 2 -based estimation for the 2015-2016 El Niño-induced extra carbon flux from vegetation fires, reduced net primary productivity, and errors in the assumed trends of FFC emissions during the period October 2014 -October 2016. Since the XCO 2 values consist of vertically-integrated information for the whole atmospheric column, simple approximations can be applied for estimating CO 2 flux corrections (in PgC month −1 ) from meridional atmospheric CO 2 burden differences (PgC) at monthly time interval (see Methods). The estimated CO 2 flux corrections are summarized in Table 1. For the ACTM_IAV84 + GFAS fluxes, the anomalous CO 2 emissions aggregated over the 'main El Niño period' (defined by July 2015 to June 2016) are in the range of 2.23-2.55 PgC. Because the ACTM_IAV84 + GFAS simulation generally follows the observed OCO-2 XCO 2 (Fig. 2a-c), we use this as the 'best' prior for CO 2 flux correction. The best prior case introduces less error in the flux corrections as the transport of flux increments are ignored in our calculation method. The 0.32 PgC difference in emissions is due to extrapolation of XCO 2 differences poleward in both hemispheres (Fig. 1d) The range of estimated CO 2 flux corrections is consistent with the empirical calculation of the CO 2 flux anomaly (2.67-2.73 PgC) using its linear relationships with the MEI trend (Table 1) 15 . Using the CO 2 flux anomaly and MEI trend relationship 15 , the CO 2 flux anomaly for the 1997/1998 is estimated at 4.4-5.7 PgC, while that from the atmospheric-CO 2 inversion was 6.7 PgC. A global CO 2 emission anomaly of ~2 PgC is estimated for July 1997 -June 1998 due to fires alone 16 .
The annual mean CO 2 residual land fluxes for the main El Niño period are then estimated as  Table S1). The consistency over data screening and transport model cases provide us confidence on the adapted methodology for calculation of flux correction from model-observation XCO 2 differences, and suggest that treatment of the data gaps do not significantly affect the estimation CO 2 flux anomaly (2.48 ± 0.07 PgC; mean and 1-σ standard deviation based on 3 sensitivity cases for WL and AMF). The CO 2 flux anomalies estimated from ACTM and GOSAT XCO 2 differences is 2.65 (=1.70 for GFAS + 0.95 from XCO 2 flux correction) PgC for the IAV84 + GFAS fluxes and period June 2015 to May 2016 (note one month difference with OCO-2) are also found to be in good agreement with those estimated using OCO-2. Figure 3 shows the monthly variations in CO 2 flux corrections along with the number of ~1 km 2 pixels with fire, seen from the MODIS sensor onboard the Terra satellite 30 . The positive CO 2 flux corrections for both GOSAT and OCO-2 show high coincidence with large fire counts, e.g., during September-October of 2014 and 2015, high CO 2 emissions are caused by fires in maritime tropical Asia (mainly Indonesia) and America (mainly Brazil), and emissions during March-April 2015 can be linked to fires in the continental tropical Asia (Thailand and the neighboring countries) 14 . As seen from Fig. 3c, more than 90% of global fires (solid line) occur within the latitude band of 30°S-30°N (broken line), and are emitted as pulse in a one month time window. This result of anomalous XCO 2 increase during the 2015-2016 El Niño can be assigned to CO 2 emissions from the tropical land. Because the signal from the enhanced fires is correlated with drought, the CO 2 observation based study cannot quantitatively discriminate the relative roles of reduction in biospheric uptake due to warmer and drier climate, and emissions from biomass burning. Interestingly, although the time-integrated GFAS emissions are in good agreement with tropical XCO 2 increase, the timing of pulsed CO 2 emissions during the fire events is not well represented. However, as a first guess, we estimate fire emissions to be ~0.76 PgC from the peaks in November 2015 and March 2016 (months following the large fire counts as marked by the dotted lines vertical lines in Fig. 3), which is 30-34% of the total flux anomaly for the main El Niño period. Figure 4 shows the meridional distributions of annual mean a priori fluxes and flux corrections using OCO-2 XCO 2 observations. The flux corrections are found to be greatest at around 35-60°N (Fig. 4b,c), up to 10% of the rate of the total a priori biospheric (non-fossil) fluxes, which are of the order of ±20 gC m −2 yr −1 at these latitudes. In general, the flux corrections at all latitudes are smallest for the ACTM_CYC64 simulation and greatest for the ACTM_IAV84 simulation, but an overall source or a weak sink is observed during October 2014 -September 2015 (Fig. 4b). A clear sink tendency is developed for the period October 2015 -September 2016 for the ACTM_CYC64 case and slightly weaker source for the ACTM_IAV84 or ACTM_IAV84 + GFAS simulations (Fig. 4c). These suggest that the effect of El Niño on CO 2 release from the biosphere has been moderated in the latter part of 2016 compared to that in 2015 (ref. also Table 1). Figure 5 shows the TCCON-ACTM mismatches for the simulations using a priori and corrected fluxes, calculated using individual XCO 2 observations. We find that the best flux corrections are obtained for the best a priori case (ACTM_IAV84 + GFAS), where the root-mean-square (RMS) differences of TCCON-ACTM XCO 2 are below 0.78 ppm for 5 out 6 sites (except for Darwin at 1.07 ppm). A reduction in RMS differences of 70-80% are found for this ACTM case. The simulation case of ACTM_CYC64 also achieved RMS differences close to 1.0 ppm or lower following the flux correction. However, the case of ACTM_IAV84 showed a mean RMS difference of 1.5 ppm after flux corrections are applied. Thus a good a priori ACTM simulation is critical for implementing this method of flux correction using OCO-2 measurements. One of the most encouraging improvement in ACTM -OCO-2 difference is seen for Park Fall. At this site, the differences were largest in July, which are reduced by half to ~1 ppm in 2015 and ~2 ppm in 2016 for the ACTM_CYC64 case (Fig. 5a), suggesting that the CO 2 sinks should be increased in the northern mid-latitude region (green line in Fig. 4b,c). Such seasonal bias is not seen for ACTM_IAV84 case, but an overall reduction in sink in the northern mid-latitudes is suggested (consistent with Fig. 4b,c). Both the seasonal and annual biases are the lowest for the ACTM_IAV84 + GFAS case.

Meridional CO2 flux anomaly and flux validation using TCCON.
Following this validation, we conclude the CO 2 flux anomaly to be 2.4 ± 0.2 PgC for the July 2015 -June 2016 period using the flux corrections obtained for ACTM_IAV84 + GFAS case only. An annual total land and ocean sink of 3.9 ± 0.2 PgC yr −1 during July 2015 -June 2016, for the assumed fossil fuel emissions of 10.1 PgC yr −1 , contrasts the average sink of 6.2 PgC yr −1 during the reference year of 2014. This is in huge contrast to the July 1997 -June 1998 period, when the Earth's surface acted as a net source of CO 2 to the atmosphere. Since the atmospheric growth rate measured by the NOAA/ESRL at Mauna Loa is 3.05 ppm yr −1 for the main El Niño period, the global residual sink of 3.6 (=10.1-3.05*2.12) PgC yr −1 is fairly consistent with our results. The residual sink for 1998 based on Mauna Loa growth rate was 0.5 (=6.7-2.93*2.12) PgC yr −1 .
In an attempt to gain further confidence in the ACTM corrected fluxes we compared the meridional gradients in CO 2 fluxes from two other traditional inversions (Fig. 6). The traditional inversions are: CarbonTracker run from NOAA 31 and Copernicus Atmosphere Monitoring Service (CAMS) 32 . The comparison suggests large differences between the inversion fluxes, and the differences showing strong dependence on a priori FFC CO 2 emissions. Generally, the model assumed stronger FFC emissions also suggest stronger biospheric uptake, with particular distinctions in the northern mid-latitude region 33 . This leads us to conclude that the simple inversion system using XCO 2 observations and ACTM simulations is usable for global CO 2 flux anomaly calculation.

Discussion
The powerful 2015-2016 El Niño has made a large impact on the Earth's natural climate system, which in turn affected the terrestrial ecosystem. We analyzed the column-averaged CO 2 dry mole fraction (XCO 2 ) estimates from NASA's OCO-2 observations collected between September 2014 and October 2016. We have also used the longer measurement records from JAXA's GOSAT, TCCON ground-based XCO 2 and NOAA in situ CO 2 measurements in the analysis. Global simulations using JAMSTEC's ACTM are performed for three combinations of terrestrial and oceanic CO 2 fluxes: CYC64, IAV84 and IAV84 + GFAS, and a common field of emissions from fossil fuel consumption and cement production. The XCO 2 and CO 2 growth rates are slightly overestimated by ACTM_CYC64, but a greater underestimation was found for ACTM_IAV84 while compared with OCO-2 observations. The ACTM_IAV84 simulation successfully simulated CO 2 growth rates during January 2013 to mid-2014. Thus the IAV84 + GFAS simulation produced the smallest model-data mismatch over the tropics when GFAS emissions were added from October 2014 (total emission of 2.64 PgC). We estimate that the El Niño event led to excess CO 2 release to the atmosphere in the range of 2.23-2.55 PgC during July 2015 to June 2016, compared to the reference period of 2014. This CO 2 release would be increased by 0.2 PgC if no increase in FFC emission was assumed.
In year 2015, about 0.76 PgC is emitted from fires, which is in the range of 30-34% of total CO 2 flux anomaly. The OCO-2 based CO 2 flux anomaly of 2015-2016 El Niño is comparable to that is estimated from an empirical relation of CO 2 flux anomaly and ENSO index trends (2.67-2.73 PgC). Our estimated fire-induced CO 2 flux anomalies disagree with those calculated from the GFED4.1 s total fire CO 2 emissions of 1.64, 1.88 and 2.09 PgC  14 .
The flux corrections based on OCO-2 measurements are validated using independent TCCON measurements, which suggest systematic reductions in TCCON-ACTM mismatches for the simulations using corrected fluxes compared to the a priori fluxes. A mean 1-σ standard deviation of 0.7 ppm is achieved for 6 TCCON sites for the period of October 2014 to October 2016 using the corrected fluxes. The flux correction method is applicable to satellite observations with near global coverage to calculate global CO 2 flux anomalies at near real-time when a suitable a priori model simulation of atmospheric-CO 2 is available, e.g., ACTM_IAV84 + GFAS case in this study. Based on our best a priori case, the global total flux anomaly is estimated to be 2.4 ± 0.2 PgC to the atmosphere as an effect of the El Niño, while the Earth's surface acted as a net sink of CO 2 by 3.9 ± 0.2 PgC during the period of July 2015 -June 2016.

Methods
We used the bias corrected measurements of XCO 2 from the 'OCO-2 7 LITE LEVEL 2' files 26 (updated document at http://disc.sci.gsfc.nasa.gov/OCO-2/documentation/oco-2-v7; last accessed: 5 December 2016). These files only include those soundings that have passed the cloud screens and converged (xco2_quality_flag = 0). In addition, only those soundings that have a warn level (WL) less than 12 and air mass factor (AMF) less than 3.5 are used in this analysis (Control case), but no distinction is made for the different viewing modes of nadir, glint  Table S2, which are calculated from individual TCCON data.  32 ; version: v15r4; http://apps.ecmwf.int/datasets/data/cams-ghg-inversions/). or target. All the data for the period extending from 06 September 2014 to 31 October 2016 are combined into 2.5° × 2.5° grid boxes at monthly time intervals for the convenience of analysis. Any grid containing less than 3 OCO-2 soundings (N) or an absolute model (ACTM_IAV84 + GFAS case) -observation XCO 2 difference greater than 9 ppm is set to undefined. The limits for WL and AMF are chosen after testing different cut-off levels for making the gridded dataset. For example, use of AMF < 2.5 or < 3.5 did not produce large number of zonal-mean XCO 2 differences greater than ± 1 ppm at most latitude bands (except at the high latitude edge of the satellite orbit) in all months. Similarly XCO 2 differences greater than ±1 ppm were not found frequently for selection of WL < 6 or WL < 12. Various sensitivities of these data screening parameters are shown in the Supplementary  Information (Fig. S1 and S1).
In addition, we have used selected measurements of XCO 2 from the ground-based Total Carbon Column Observing Network (TCCON) 28 and CO 2 from the NOAA cooperative global air sampling network 29 [Product: obspack_co2_1_CarbonTracker-NRT_v2.0_2016-02-12]. We have used the XCO 2 data from TCCON sites at Lauder (45°S, 170°E) 34 , Reunion Is (21°S, 55°E) 35 , Darwin (12°S, 131°E) 36 , Ascension Is (8°S, 14°W) 37 , Lamont (37°N, 97°W) 38 and Park Falls (46°N, 90°W) 39 40 . ACTM is run at a horizontal resolution of T106 spectral truncations (~1.125 × 1.125°), and 32 sigma-pressure vertical levels, and meteorology is nudged to horizontal winds and temperature from the Japanese 55-year Reanalysis (JRA-55) 41 . The following CO 2 flux tracers are simulated by ACTM with an aim to encompass the observed CO 2 growth rates during October 2014 to February 2016 (Table 1) Note that the ACTM_IAV84 simulation successfully simulated the CO 2 concentrations for the time evolution and tropospheric profiles over Asia for the period 2007-2012 41 . Also shown here that the CO 2 growth rates are well simulated by ACTM_IAV84 at the selected TCCON and NOAA ground-based measurement sites for January 2013 to mid-2014. Thus any differences in time evolution during the period September 2014 to February 2016 of OCO-2 data analysis can be attributed to excess CO 2 releases associated with the El Niño event, relative to the 2014 mean.
Model XCO2 are calculated 43 by convoluting model CO2 profile (CO 2 ACTM ) with that of the a priori profile (CO 2 pri°r ) and column averaging kernels (A i ) of instrumental sensitivity to different layers of the atmosphere (P i , i = 20, 20 and 71 for OCO-2, GOSAT and TCCON, respectively). where the XCO 2 difference is the observed minus model values, area of the grid is latitude dependent and air density is calculated as the air mass overhead each 2.5 × 2.5 grid from ACTM air density. The difference in the burden mismatches between October and September 2014 is assigned to the flux correction for October 2014. For these flux estimations in the control case, missing areas are filled by the mean values of the observed -model differences for the 3 latitude bands. This is done based on an assumption that the mean differences will be transported within the semi-hemispheric regions within months by the rapid zonal mixing. In this simple method, we do not expect to resolve the evolution of flux corrections at less than a 1-month time resolution or the contrast between the continents and between land-ocean. However, this method is applicable for near real-time monitoring of biospheric health of Earth's ecosystem without significant additional investment. This method of flux corrections is valid only for sub-hemispheric scales since the zonal transport circulates air masses several times around each of the 3 broad, zonal bands within one month. This method suffers from the extrapolation of data to the missing observation grid boxes. For example, OCO-2 soundings covered a maximum of 70, 70 and 60% of the 2.5 × 2.5° grid cells in the latitudes bands of 90°S-10°S, 10°S-10°N, 10°N−90°N, respectively. In the latitude bands poleward of 10°, monthly data coverage can be as low as 30% in the winter hemisphere. Data coverage in the tropical latitudes suffers mainly from cloud cover (in addition to the model transport error), sometimes for longer than a month, and are approximated at modelers discretion by choosing not to modify the priors or applying a time correlation. The fraction of missing data area will increase further when analyzed for smaller than 2.5° × 2.5° grid sizes. Note that this method cannot be employed for the in situ measurement network without significant extrapolation in space and for the fact that the ground measurement sites do not cover the majority of the continental source regions 44 .
As opposed to the site-based data analysis 12,13,15 for CO 2 flux anomaly, this method based on differences between the observation-model difference does not require a long time series of data. As shown here, only one year of reference is sufficient, (2014 used in this analysis). Another major advantage of this analysis comes from the near uniform data coverage over the continents of tropical Asia, Australia, South America and Africa, which are very sparsely observed by the in situ measurement networks, providing a true global CO 2 flux signal. The traditional analyses mentioned earlier in the Introduction focused on one site, which is often under the influence of regional or local flux signals.
Finally, we are also able to validate the flux corrections from ACTM -OCO-2 XCO2 differences using an independent set of TCCON observations. The zonal mean flux corrections (Fig. 4) are simulated using ACTM and XCO2 signals added to their respective a priori simulations. The results are presented in Fig. 5, which show clear reduction in ACTM -OCO-2 differences after the corrected flux simulations (Table S2). Flux corrections using ACTM and OCO-2 XCO 2 are also compared with CarbonTracker and CAMS traditional inversion results showing greater influence of fossil fuel a priori emissions on the estimated biospheric flux compared to the differences arising from flux estimation methods (Fig. 6).