Atmospheric carbon dioxide

The figure above shows atmospheric CO2 measurements from two ground based stations, Mauna Loa, Hawaii in the Northern Hemisphere and Cape Grim, Tasmania in the Southern Hemisphere, together with a combination of recordings from different satellites. The ground based data is well known, in particular that from the Mauna Loa Observatory set up in 1958 by Charles Keeling. The satellite data is only available since 2003 but allows more complete studies of CO2 concentrations at different locations around the Earth

On this webpage the data from four ground stations and from one set of satellite data is analysed to understand the time and spatial variations of the CO2 values. 

CO2 Ground Station data

Around the world there are multiple ground stations measuring atmospheric CO2 concentrations. Figure 2 shows the locations of sites in the database: Global Monitoring Laboratory of NOAA.

Figure 2. Locations of ground stations measuring atmospheric CO2 concentration around the world in the GML database of NOAA.

Seasonal variations

Data from four ground stations in the GML database have been downloaded and analysed using a minimal data filtering to remove bad recordings. The data for four stations, Barrow, Mauna Loa, Tutuila and the South Pole, which have very different latitudes but similar longitudes, are show in figure 3.

Figure 3. Time series data of atmospheric CO2 concentrations for four GML ground stations: Barrow, Mauna Loa, Tutuila and the South Pole. The data is available for different time ranges but is only displayed from 2000.

Figure 4 displays fits to these data, incorporation terms for the long time scale trends and the annual variations. The long term time dependence trend is nearly linear but a small quadratic term is present in the fit. The annual time dependence has two ad-hoc components, a cosine dependence as a function of months in the year and a cosine term truncated to zero for half the year, these two terms have free amplitudes and phases in the fit. Figure 5 shows details of the annual variation terms in these fits. This figure illustrates the large difference in the annual variations which have amplitudes around 1 ppm for the South Pole and Tutuila, 7 ppm for Mauna Loa and 16 ppm for Barrow.

Figure 4. Data for four ground stations (brown points) and fits (red dashed line). The details of the parameters in the fit are given in the text.

Figure 5. Details of the data and fits in figure 4, showing the annual variations. The data points have the long-term trends subtracted and the curves show the annual variation terms in the fit. (Blue, the cosine term, magenta the truncated cosine term and red the sum of the two annual terms) 

CO2 Satellite data

Satellite data time and spatial coverage characteristics

The satellite data used here was download from the ECMWF website: Carbon dioxide data from 2002 to present derived from satellite observations ( From this site the "Merged and EMMA" with "Column-average dry-air mole fraction of atmospheric Carbon Dioxide (XCO2)" option was employed, using data from 2004-2021. This is referred to as "XCO2_EMMA" data. The amount of data available each year varies from around 0.3 million measurements per year for the period 2004-2013 to 2-3 million per year in the period 2015-2021 as shown in figure 6.

Figure 6. Number of measurement per year available in XCO2_EMMA data.

The total number of measurements per year are very unevenly distributed around the globe because of the satellite coverage. The analysis of the data described below looks at the spatial distribution of CO2 concentrations in 5° X 5° bins of latitude x longitude. This analysis becomes complicated when there are months with zero data in a bin because of the season variations of CO2 levels. Figure 7 shows how the global coverage of the satellite data improves with year. For data before 2009 the data is only over land and in subsequent years the coverage extends, including data over the sea. There is zero data at latitudes >75°N or <75°S. Even in the latest 2021 year there are regions with low coverage over some regions on land in the tropics.

Figure 7. Variations in coverage in latitude and longitude for the available satellite data. The plots show the number of months in a  year with measurements per 5° x 5° grid bin. The plots are for four example years 2008, 2012, 2017 and 2021. Black bins have zero measurements.

Annual variations in CO2 from satellite data

Without any consideration of variable coverage, the data for 18 years is plotted in figure 8 comparing Northern and Southern Hemispheres and in the figure at the top of the page comparing with the ground based data. These plots shows the data sets in the two hemisphere with a difference of about 2 ppm between the yearly average CO2 concentration in the NH compared to the SH for the same year. The two hemispheres clearly have different annual variations.

Figure 8. XCO2_EMMA satellite data comparing the two hemispheres. No correction is made to the data plotted to account for the missing days and months of data in the yearly dataset nor the different latitude/longitude coverage.

Figure 9 shows the amplitude of the annual variations in the CO2 levels as a function of latitude, from fits similar to those used for figures 4 and 5. As in the ground station data the annual variations in the southern hemisphere outside the tropics is about 1 ppm and in the northern hemisphere outside the tropics 6 to 8 ppm. Inside the tropical latitudes the situation is changing and also data over land and sea have significant differences.

Figure 9. Amplitude of annual variations of CO2 levels as a function of latitude, using satellite data obtained and the same methods as for figures 4 and 5.

 Global variations in CO2 levels

Figure 10 shows the CO2 concentrations in 5° x 5° latitude/longitude bins. The data for each year uses an algorithm to correct for months without data by using a step-wise interpolation from the months where the data is available.

Figure 10. Distribution of CO2 concentrations in 5° x 5° latitude/longitude pixels, averaged from 2019-2021. The regions with low data coverage are masked out.

From the plot it can be seen clearly that the Northern Hemisphere has higherCO2 concentrations than the Southern. Figure 11 shows the distribution of CO2 concentration for latitude over the regions of longitude where non-zero data is available. It can be seen that the peak of this distribution is between 10-30 °N.

Figure 11. Distribution of CO2 concentrations vs latitude for the period 2019-2021.

In figure 10 there are indications of "hot spots" of CO2 concentrations in for examples Africa and China. These same hot spots are present in the data shown in figure 12, which give the data only over land, for four 3 year periods between 2010 and 2021. The data in these four different time periods is clearly consistent.

Figure 12. Distribution of CO2 concentrations in 5° x 5° pixels as in figure 10, but plotted only over land. Data is shown for four periods: 2010-2012; 2013-2015; 2016-2018 and 2019-2021.

Background subtraction of global variation data

In order to explore the nature of the hotspots seen in figures 10 and 12, a background subtraction to the data in latitude/longitude bins is performed. This subtraction uses a background average of data over sea regions which takes into account CO2 variations with time and with latitude. Figure 13 shows background time dependence used over sea compared to the average over land. This plot is for the latitude range -45° to +45° and for the time period 2012 to 2021 which are used for the plots in this section to minimize bias from variations in data coverage. In these averages the land and sea values are the same within about 0.1 ppm in CO2 concentration.

Figure 13. Time dependence of CO2 concentrations used for the background subtraction from the sea compared to the average levels over land. The plots is within the latitude range -45° to +45°.

Figure 14. Latitude and longitude dependence of CO2 concentrations, for years as indicated and averaged over all land and all sea regions within the latitude range -45° to +45°.

Figure 14 shows the latitude and longitude dependence of the CO2 levels, showing the same latitude dependence as in figure 11 and negligible longitude dependence. This plot shows that the background subtraction must take into account latitude variations but a constant background over longitude can be used.

Figure 15 shows the CO2 concentrations after background subtraction. The background uses year by year averages in sea regions in 15° latitude ranges. Technically this is implemented with the Interpolation function in Wolfram Mathematica which uses a third order polynomial to interpolate between points on a year vs latitude grid. The figure 15 combines years 2015 to 2021.

Figure 15.  Relative CO2 concentrations after the background subtraction taking into account time dependence and latitude dependence. The data is averaged over the period 2015-2021 and is only shown for the latitude range -45° to +45° where the data coverage variations are relatively small.

In this background subtracted plot, the "hotspot" features visible in the raw plots above remain. Some of the hotspots may correspond to regions burning large amounts of fossil fuel but the hotspots in Africa and South America likely do no. Figure 16 shows these regions in more detail.

Figure 16. Relative CO2 concentrations in South America and Africa from satellite data after background subtraction of CO2 level in the sea at the same latitudes. The data is the same as in figure 15 displayed more clearly.