Spatial Distribution of CO2 Concentration over South America during ENSO Episodes by Using GOSAT Data

Carbon dioxide (CO2) is one of the most important greenhouse gases; its concentration and distribution have important implications on climate change. The El Niño Southern Oscillation (ENSO) is the Earth’s strongest climate fluctuation on inter-annual time scales and has global impacts. However, to date, there is no research on how ENSO affects the spatial distribution of CO2 concentration. In this study, we used spatial CO2 data from the ENVIronmentSATellite (ENVISAT) and the Greenhouse Gases Observing Satellite (GOSAT), the long duration monthly mean atmospheric CO2 from Mauna Loa Observatory, Multivariate ENSO Index (MEI) from Earth System Research Laboratory to analyze the way that ENSO affects spatial distribution of CO2 concentration in South America, which is affected by ENSO seriously. Our research revealed that monthly CO2 growth rate has a moderate, positive correlation relationship with MEI. We used geostatistics to predict and simulate the spatial distribution of CO2 and found that in south of 12 ̊S, CO2 concentration of ENSO warm episode is lower than the one of ENSO cold and neutral episodes. ENSO impacts CO2 spatial distribution mainly in November, December, January and February; moderate-high concentration zone of ENSO warm episode more concentrates in the northern part of South America.

greenhouse gas (GHG) in the atmosphere and is the greatest contributor to global warming [2].The distribution of CO 2 is closely related to human survival, and international concerns about CO 2 are also getting stronger.
The impacts of CO 2 -induced global warming include: Accelerating Sea Level Rise [3], Longer and More Damaging Wildfire Seasons [4], More Frequent and Intense Heat Waves [5], Costly and Growing Health Impacts [6], Heavier Precipitation and Flooding [7] and so on.Global temperatures in 2015 were the warmest on record, according to data published by meteorologists in the UK and US [8].According to the Earth's CO 2 Home, the average global carbon dioxide (CO 2 ) concentration in December 2015 was 401.85 ppm, which is 3 ppm greater than in December 2014 [9].2015 not only was the warmest year on record, but also broke the record by the largest margin by which the record has been broken [8].ENSO is one of the most important and longest-studied climate phenomena on the planet [10], and the strongest quasi-oscillatory pattern observed in the climate system; it is coupled to numerous climatic systems [11].ENSO can affect global average temperatures; for example, in 1998 temperatures were significantly enhanced by strong El Niño conditions.1998 remained the warmest year until 2005 and 2010 and the temperature of both of these years was enhanced by El Niño periods.The large margin by which 2015 is the warmest year is also attributed to another strong El Niño.However, 2014 was ENSO neutral (from the Multivariate ENSO Index (MEI): http://www.esrl.noaa.gov/psd/enso/mei/and the Oceanic Nino Index (ONI): http://www.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/ensoyears.shtml).The only El Niño events in NOAA's 1950-2015 database comparable in strength to the one now developing occurred in 1982-1983 and 1997-1998 from the records ONI and MEI.
There are many complex relationships between carbon dioxide and ENSO from the viewpoints of climatology, geography, oceanography, ecology, biology and so on.From a simple viewpoint: marine dissolves a large number of CO 2 [12]; the increase in atmospheric CO 2 concentrations due to anthropogenic emissions has resulted in the oceans taking up CO 2 at a rate of about 7 GtCO 2 yr −1 [13] (the report published in 2005).As water temperature increases, its ability dissolving CO 2 decreases [14].And the most obvious feature of the El Niño phenomenon is a wide range of abnormal warm seawater in central and eastern equatorial Pacific.Due to the response of the tropical ecosystem to climate variation, during ENSO warm phase, heterotrophic respiration will increase because in the water-stressed region photosynthesis it will be less, vice versa [15].Therefore, there should be a relationship between CO 2 and ENSO.About the relationship between CO 2 and ENSO, many researchers have done a large number of studies mainly in two aspects: one is utilizing the climate model to predict or simulate the frequency and intensity of ENSO [16]- [19]; the ENSO response to global warming differs strongly from model to model and is thus highly uncertain.Some models simulate an increase in ENSO amplitude, others a decrease, and others virtually no change [20].The other is utilizing historical observations to calculate or summarize the relationship and laws [11] [21]; some scholars use different algorithms to calculate the correlation relationship between CO 2 growth rate of lag times of 3, 6, 9, 12 months and ENSO Index [11] [22].Patra, P. K. et al. estimate fluxes from 64 partitions of the globe using atmospheric CO 2 data from 87 stations worldwide by using time-dependent inverse (TDI) model [23] to analyze the relationship between CO 2 growth rates and ENSO episodes [24], and show that atmospheric CO 2 concentration and ENSO are positive correlation.
According to the summary of previous research, it is easy to find out that there is few research on the relationship between ENSO and distribution of CO 2 , concentrating on using sampling observation data or station observation data.Since it is difficult to observe the global variation of GHG because the direct sampling of gases, especially in the upper atmosphere, requires great effort and cost [25].Launching satellites to collect GHG data can solve this issue quite well.At present, the Thermal And Near infrared Sensor for carbon Observation (TANSO) aboard the Greenhouse gases Observing SATellite (GOSAT), launched in 2009 by Japan, is the first instrument dedicated to record CO 2 and methane (CH 4 ) concentrations (denoted XCO 2 and XCH 4 , in ppm.XCO 2 and XCH 4 are column-averaged dry air mole fractions of atmospheric CO 2 and CH 4 ) [26] [27].GOSAT make it possible to research the spatial distribution of CO 2 , and know the impact of ENSO on CO 2 distribution.Although the SCanning Imaging Absorption spectroMeter for Atmospheric CHartographY (SCIAMACHY) aboard the ENVIronmentalSATellite (ENVISAT), launched in 2002 and lost in 2012 by the European Space Agency (ESA), can also retrieve GHG concentrations, but it is not specially designed for GHG retrieval and is not highly accurate; the single-measurement precision is approximately 1% -2%, with an intermonth scatter of 2.3% [28] [29].Therefore, before 2002, the CO 2 concentrations could not be measured directly using the remote sensing technique, maybe this is one reason why there is few study on the impact of ENSO on CO 2 spatial distribution.
El Niño, Warm phase of ENSO is one of three biggest natural disasters in South America, arid area will become wet, even occur the waterlogging in west coast, and oppositely large-scale droughts will happen in humid areas.In tropical South America, CO 2 flux anomaly is related to two basic meteorological parameters (rainfall and air temperature at 2 m) [24].The concentration and spatial distribution of CO 2 will bring a huge impact on natural environment, such as rainforest, increasing CO 2 will lead to degradation of the Amazon rainforest [30] [31], and influence human life in South America.Facing one of the most severe El Niño, from Mar. 2015 to 2016, spatial distribution and changes of CO 2 will be much different from normal climate.In the present study, we analyzed CO 2 distribution by using GOSAT products from July 2009 to April 2014, got the relationship between CO 2 concentration and ENSO intensity by using the MEI and CO 2 data, from March 1958 to June 2015, from NOAA Earth System Research Laboratory.The objectives were: 1) to determine whether there is a relationship between CO 2 concentration and ENSO, 2) to examine the changes of CO 2 concentration distribution in South America during different phase of ENSO.This study explored the spatial distribution of CO 2 concentration of different ENSO episodes and also predicted the uncertainty distribution of CO 2 concentration in this El Niño which started in 2015, by using the distribution changes of CO 2 concentration in recent event, and explored the CO 2 concentration and distribution change during ENSO episode at a large spatial scale.

Study Area
The study region covers the majority of South America (12.4˚N -53.9˚S, 34.8˚W -81.3˚W, an area of 179.7 × 105 km 2 ) and includes Brazil, Argentina, Chile, Colombia and Bolivia, covering a total of 14 counties (Figure 1).In South America, the forest area is 920 million hectares, constituting 50% of the continent's area, constituting 23% of the world's forests [32].Grassland area is approximately to 440 million hectares, constituting 25% of the continent's area, constituting more than 14% of the world's grasslands [32].Agriculture is very important for South America's economy, since most countries in South America engaged in agricultural production, economic crops absolute dominance.
During ENSO warm phase, rainfall across the east-central and eastern Pacific Ocean, including several portions of the South American west coast, will increase due to El Niño's warm pool feeds thunderstorms above.If an ENSO warm episode occurred, weather it was strong or not, along the coasts of northern Peru and Ecuador, major flooding would be caused by warm and very wet weather from April to October.In opposite, During ENSO cold phase, the coastal regions of Peru and Chile will be very dry, but in the central Andes will be richer rainfall than normal even lead to catastrophic flooding, northern Brazil also will be wetter from December to February.South America is the continent which influenced by ENSO most directly and seriously.[33].Data are reported as a dry mole fraction defined as the number of molecules of carbon dioxide divided by the number of molecules of dry air multiplied by one million (ppm).Carbon dioxide measured at Mauna Loa Observatory was used to explore the relationship between CO 2 concentration and ENSO episode for a long duration.In the present study, we select the longest length of monthly data, from March 1958 to December 2015, to calculate the CO 2 monthly increasing rate, and to find out the relationship with ENSO.
2) GOSAT L3 Global CO 2 Distribution Data The primary purpose of the GOSAT project is to accurately estimate the emissions and molecules of CO 2 in the atmosphere absorb light at particular wavelengths.The GOSAT uses the principle, the quantities of CO 2 in an optical path can be calculated by measuring the amount of light that is absorbed by molecules, to measure the concentrations of GHG in the atmosphere.GOSAT flies at an altitude of approximately 666 km and completes one revolution in about 100 minutes.The satellite returns to the same point in space in three days.Level 3 data products that are generated from Short Wavelength Infra-Red (SWIR) data observed by Thermal and Nearinfrared Sensor for carbon Observation-Fourier Transform Spectrometer (TANSO-FTS) onboard GOSAT with the spatial resolution of 2.5˚ latitude × 2.5˚ longitude.L3 data generated by interpolating and extrapolating the FTS L2 data and estimating the distribution of XCO 2 for each month on a global scale.FTS SWIR L2 CO 2 column abundance products (denoted XCO 2 , in ppm) contain column-averaged mixed volume ratios of CO 2 [34].The XCO 2 represents the ratio of the total number of CO 2 molecules against that of dry air molecules, not only in the neighborhood of the Earth's surface but in the total vertical column extending to the top of the atmosphere [29].The detailed algorithms and information for the TANSO-obtained GHG concentrations can be found from Greenhouse gases observing satellite GOSAT Official website (https://data.gosat.nies.go.jp/GosatUserInterfaceGateway/guig/GuigPage/openTechInfo.do) The

Multivariate ENSO Index (MEI) Data
Bimonthly MEI values (in 1/1000 of standard deviations), starting with December 1949/January 1950.ENSO is the most important coupled ocean-atmosphere phenomenon to cause global climate variability on interannual time scales [35].People attempt to monitor ENSO by basing the Multivariate ENSO Index (MEI) on the six main observed variables over the tropical Pacific.MEI is a multivariate measure of the ENSO signal.These six variables are: sea-level pressure (P), zonal (U) and meridional (V) components of the surface wind, sea surface temperature (S), surface air temperature (A), and total cloudiness fraction of the sky (C) [36].The MEI is computed separately for each of twelve sliding bi-monthly seasons (December/January, January/February, …, November/December).In order to keep the MEI comparable, all seasonal values are standardized with respect to each season and to the 1950-1993 reference period.

Method
Centralizing Second Moving Average Method Second moving average method is calculating the second moving average of the first moving average, and then using the first moving average and second moving average to establish the trend model.The average values calculated by one moving average method have the shortcomings of lag bias.Especially, when the time-series data showing in a linear trend, moving average values always lag the observed data.Second moving average method can correct this lag bias, establishment the linear mathematical relationship with time to get the trend values.Second moving average method is applicable to the time-series analysis of the phenomenon which has clear trend change, also retains the advantages of first moving average method.
Centralizing refers calculating the average value of average 1 and average 2. Average 1 is the average from t − n to t, average 2 is the average from t to t + n. n is half of one cycle length of the time-series data.This average is the trend value at time t.Trend value at time t, f(t) can be expressed as follows: ( ) where, n refers to one cycle length of time-series, monthly time data n is 6 months.D i means observation data at time i.The length of trend time-series will be shorten, the beginning and ending n months cannot be calculated.This method is packaged as R's function (decompose()).Decompose a time series into seasonal, trend and irregular components using moving averages [37].We got the trend data from monthly CO 2 concentration data by using function decompose() in R.

Data Processes
Logical process of this study is shown in Figure 2. Monthly CO 2 concentration data from March 1958 to December 2015, in total 694 months, were processed by using software R to remove the seasonal and random characteristics of the data, in this way trend CO 2 data can be extracted.According to R Software, time trend is decomposed by using centralizing second moving average method.Then using monthly trend CO 2 concentration data to calculate the monthly change rate of CO 2 .Do correlation analysis of MEI and CO 2 change rates, if uncorrelated, that means there is no relationship between ENSO and CO 2 concentration increasing rate; if correlation, do the analysis on CO 2 spatial pattern during different ENSO phases.Using GOSAT L3 data to obtain monthly CO 2 concentration data over South America.Kriging interpolation method was utilized to predict the spatial unknown values, since GOSAT does not capture the column value when clouds are present [38].Analysis the spatial correlation of CO 2 concentration respectively, in same month during ENSO neutral phase, warm phase and cold phase.If the space consistency is high, ENSO has no effects on the spatial distribution of the CO 2 .Else, spatial relative difference would be analyzed to get the differences of CO 2 Spatial Patterns during different ENSO phase.

The Relationship between CO 2 Concentration and MEI
CO 2 trend calculated by utilizing centralizing second moving average method is shown in Figure 3.The data without seasonal and random characteristic has an obvious increasing trend.And the increasing rate is not always balancing on one level, higher in some months, lower in other months, but maintained an increasing trend.In order to find out whether there is a relationship between CO 2 increasing rate and MEI, we get CO 2 increasing rate through derivation of trend values.Monthly CO 2 increasing rates and MEI values are shown in Figure 4.
From the whole view, two time series show the synchrony changes, except for from 1991 to 1993.Two time series were correlation analyzed by using EXCEL 2013, there is a significant correlation between both with the correlation coefficient 0.398, and p = 0.000.There is a moderate correlation relationship between CO 2 monthly increasing rate and MEI values from the viewpoint of statistics.
To explain the absence of synchrony around 1991, for a few years after a major volcanic eruption (i.e., when there is an abundance of sulphate aerosols in the atmosphere), heterotrophic respiration decreases due to a lowering of the Earth's surface temperature and the productivity of forest ecosystems increases under enhanced diffuse radiation [22].Both processes lead to a negative anomaly in CO 2 growth rate.In June 1991, in the same tropical latitude and upwind of Mauna Loa, there is a violent eruption of Mount Pinatubo, the second largest eruption of the twentieth century.That may the main reason of the absence of synchrony from 1991 to 1993 [11].Remove the data from June 1991 to June 1993, the correlation coefficient between CO 2 monthly increasing rate and MEI values is 0.472, p = 0.000.
In summary, when MEI is high (ENSO warm episodes), CO 2 increasing rate reaches a peak with the exception of the interval around 1991, when MEI is low, ENSO cold episodes are followed by lower values of CO 2 increasing rate.ENSO and the increasing rate of CO 2 concentration show a positive correlation.

The Effects of ENSO on CO 2 Spatial Distribution
CO 2 concentration is increasing every year, in order to analysis the spatial distribution of different periods, the CO 2 concentration data need to be normalized to do comparative analysis.North-south span of South America is nearly 70 degrees, latitudinal variation of CO 2 spatial distribution is an obvious characteristic.Normalized values make it feasible to calculate different periods' spatial distribution characteristics, since the normalized values mean the relative amount in the spatial at one period.Latitudinal variation averages of normalized CO 2 spatial distribution during ENSO warm (one episode), cold (two episodes), and neutral (two episodes) phase were calculated by using ARCGIS 10.2 in the present study.The latitudinal spatial distributions of CO 2 concentration of each ENSO episode were shown in  1. CO 2 concentration curves of ENSO neutral and cold episodes have the consistent trend and two peaks, respectively around 3˚S and 20˚S.There is no apparent inconsistencies feature between ENSO cold episode and ENSO neutral episode.But the peak of ENSO warm episode curve (red point line in Figure 5) approximately to 7˚S, with no two peaks, and high value range, from 3˚S to 20˚S, is very stable.South of 12˚S, the value of ENSO warm episode is lower than ones of ENSO cold and neutral episodes.That means, during ENSO warm phase, south of 12˚S in South America, CO 2 concentration is relatively low, more concentrated in Northern South America, relative northward.
Overall, CO 2 spatial distribution of ENSO warm episode is different from ENSO cold and neutral episodes, but more detailed spatial and temporal characteristics are needed for further study.Monthly CO 2 spatial data during different ENSO phase were analyzed by using spatial correlation analysis.This method can help us to know the differences in places and periods.The results of spatial correlation analysis with ENSO warm episode are shown in Table 1.
From the results of spatial correlation analysis, it is easily to find out that spatial distributions of CO 2 during ENSO warm phase and ENSO cold phase are highly correlated to each other except for November, December, January, February and March, spatial pattern are basically the same in geostatistics.From November to March, CO 2 spatial pattern between ENSO warm episode and ENSO cold episode are not highly synchrony.For ENSO neutral episode, high absence of synchrony is from November to February.We can also know that there is no relationship between spatial correlation coefficient and MEI values, from the box in Table 1.In northern hemisphere winter, the CO 2 spatial distribution of different ENSO episodes is most different.
For figuring out the detailed differences of CO 2 spatial pattern among ENSO warm, cold, neutral episodes, the average CO 2 spatial distribution normalized data of November, December, January and February in different ENSO episodes are shown in Figure 6.Moderate-high CO 2 concentration were expressed as warm color, in opposite, Moderate-low ones were expressed as cold color, and images were stretched by using histogram equalize to enhance the contradistinction.During ENSO warm phase: in November, moderate-high CO 2 concentration concentrate in central part of South American continent, from about 5˚S to 30˚S; in December, moderate-high CO 2 concentration zone move northward, concentrate in northern part; moderate-high CO 2 concentration zone in January and February are almost the same to the one in December.During ENSO neutral phase: moderate-high CO 2 concentration concentrate in central part of South American continent same with ENSO cold episode, and further north than ENSO warm episode; in December, synchronized with ENSO warm episode, moderate-high CO 2 concentration zone move northward; in January, moderate-high CO 2 concentration zone extend to southward, and the distribution in February is almost the same to January.During ENSO cold phase: in November, the pattern is consistent with ENSO neutral episode; from November to December, moderate-high CO 2 concentration zone extend to northward; the spatial distribution in December, January and February are substantially the same pattern.In summary, the distribution in November of ENSO cold episode and ENSO neutral episode are consistent, moderate-high CO 2 concentration located in further north than that during ENSO warm phase; in December, the distribution during ENSO cold phase are consistent with that during ENSO neutral phase, moderate-high CO 2 concentration concentrated in northern part; in January and February, moderate-high CO 2 concentration zone of ENSO cold and neutral episode extend to more south than ENSO warm episode.CO 2 concentration during ENSO warm phase is more nonuniform than the ones during ENSO cold and neutral phase, concentrate in northern part of South American continent.

Conclusion
In this study, South America as one of zones most influenced by ENSO event was selected as the study area.Long duration CO 2 concentration sample data and monthly spatial remote sensing CO 2 concentration data were used to analyze the effects of ENSO on CO 2 concentration, including increasing concentrations and spatial distribution.Trend CO 2 data were calculated from monthly data by using removal seasonal and random method, centralizing second moving average method, in software R.There is a moderate correlation between monthly increasing rate of CO 2 concentration and MEI.ENSO events have an effect on increasing rate of CO 2 concentration.Comparing the impact of ENSO on CO 2 spatial distribution in south of 12˚S of South America, concentration of ENSO warm episode is obviously lower than concentration of ENSO cold and neutral episodes.Through spatial correlation analysis, the months seriously influenced by ENSO condition can be found out.They are November, December, January, February, when northern hemisphere is winter and ENSO episode peaks [39].
Moderate-high CO 2 concentration of ENSO warm episode is located in more north and nonuniform than that of ENSO cold and neutral episode.The spatial distribution in South America and increasing rate of CO 2 concentration are affected by ENSO significantly.The results of the present study also provide supports to predict the effects of the super ENSO warm episode which started from 2015.

Figure 1 .
Figure 1.Land covers map (GlobCover image from 2011 downloaded from the homepage of the NEO: http://neo.sci.gsfc.nasa.gov/) in the study area.
GOSAT L3 global CO 2 distribution data were used in the present study to explore the spatial distribution changes of CO 2 of different ENSO episodes.The GOSAT L3 global CO 2 distribution has been published from June 2009 to July 2015, due to product version change, June 2014, December 2014, January 2015, February 2015 have not been published.Therefore, the CO 2 spatial distribution data during July 2009 to April 2010 (ENSO warm episode), July 2010 to April 2011 (ENSO cold episode), July 2011 to April 2012 (ENSO cold episode), July 2012 to April 2013 (ENSO neutral episode), July 2013 to April 2014 (ENSO neutral episode) were selected to analysis the spatial distribution of CO 2 under different ENSO events.

Figure 3 .
Figure 3.Time plot of monthly CO 2 concentration and trend.

Figure 4 .
Figure 4. Time plot of monthly CO 2 concentration and trend.

Figure 5 .
July 2009 to April 2010 is ENSO warm episode, July 2010 to April 2011 and July 2011 to April 2012 are ENSO cold episodes, July 2012 to April 2013 and July 2013 to April 2014 are ENSO neutral episodes, according to the MEI shown in Table

Figure 5 .
Figure 5.The latitudinal spatial distributions of CO 2 concentration of each ENSO episode normalized value.

Figure 6 .
Figure 6.Spatial distribution of CO 2 concentration from November to February of different ENSO episodes.

2.2. Data 2.2.1. CO 2 Concentration Data
1) NOAA ESRL Data Monthly mean atmospheric carbon dioxide at Mauna Loa Observatory, Hawaii measured as the mole fraction in dry air, on Mauna Loa (lat.19.54˚N; long.155.58˚W; height, 3397 m; interval, 1958-2016) constitute the longest record of direct measurements of CO 2 in the atmosphere.They were started by C. David Keeling of the Scripps Institution of Oceanography in March of 1958 at a facility of the National Oceanic and Atmospheric Administration

Table 1 .
MEI values and spatial correlation coefficient of spatial distribution of CO 2 during ENSO warm, neutral and cold phase.