By Chris Hall
Ever since that launch of the first Orbital Carbon Observatory satellite (OCO), I’ve been intrigued by the possibility of being able to directly observe where CO2 in the atmosphere comes from and goes to. Unfortunately, I’ve not seen much information in the press about the results from any of the three satellites launched so far. Maybe I’m just not looking in the right places, or maybe the researchers are unusually shy. In any case, I decided to have a look at some of the data myself. One of the things I wanted to check was how well mixed this so-called “well mixed gas” really is. What follows is a collection of visualizations of OCO-2 data. The cool thing about OCO data is that the satellites provide truly global coverage with a spatial resolution, in the case of the second satellite OCO-2, of 0.5 degrees of latitude and 0.625 degrees of longitude.
The original data is in the form of XCO2, or mole fraction of CO2 in dry air, which is equivalent to the volume fraction. For brief details of “how I did it”, see the Appendix below.
First, let’s look at a movie of the reported CO2 concentration from the monthly OCO-2 data.
Well, that’s not all that helpful. The general pattern is for a gradual increase in global CO2 concentration, mostly originating in the Northern Hemisphere and propagating into the Southern Hemisphere. Note that there is considerable seasonal variability, no doubt driven by the cycle of plant growth and decay each year in the Northern Hemisphere.
Another problem with looking at CO2 data is that there is a temptation to blow up any increase, possibly for dramatic effect, i.e., “we’re all gonna fry”. To try to put things into perspective, here’s the same data but presented as latitude averages. This time, I scaled the concentrations from preindustrial levels (~280 ppmv) to double that value. This should roughly correspond to a global temperature rise of ~1.5° to ~3°C, depending on whose climate sensitivity value to believe.
To make the thing a bit more human in scale, I wanted to be able to calculate the mass of CO2 in the atmosphere per square meter. To do this, I also needed an elevation model for the Earth and a reasonable global gravity model, as both you and the atmosphere weigh more at the North Pole than at the equator. Everything got sampled into a compatible 576 x 360 longitude-latitude grid, which enabled me to convert between XCO2 and mass per square meter on both a grid cell and global scale. The average CO2 concentration map in kg/m2 is shown in Fig. 1.
Well, that’s interesting. It looks more like a map of elevation, and it should, because high elevation sites have less CO2, mostly because they have less air due to the drop off of pressure with altitude. When you add up all the atmosphere that sits over land, it only accounts for 28.8% of the total, despite the fact that land area is about 30.6%.
Now let’s see how much CO2 sits over your head. On average, at the end of 2021, the average value was 6.36 kg/m2. That may sound like a lot, but you need to consider that the average mass of air that the CO2 resides in is about 10,080 kg/m2, which is an average over the whole earth, and is slightly less than the sea level value. It turns out that most of the mass of CO2 in the air really comes from O2 via its reaction with carbon and hydrocarbons, so the average mass of C in the air (neglecting methane, etc.) was 1.73 kg/m2. The increase of CO2 in the atmosphere from the beginning of 2015 to the end of 2019 was a whopping 263 grams of CO2 per m2, or 71.7 g/m2 of C. The increase of CO2 in the atmosphere per m2 over 7 years is approximately the same amount that an adult human exhales in about 7 hours.
Getting back to where CO2 comes from and goes to, there’s a global increase of CO2 that tends to swamp the details. I needed to detrend the data so I could see where the major deviations from the overall global pattern occur. I tried several methods, including subtracting off a linear fit to global average concentration record over the first 84 months (7 years) of the record. I also tried subtracting off the residue (i.e. low frequency component) of the global CO2 concentration average after decomposition using the Hilbert-Huang transform (hht library in R), and both of these methods gave similar results. But it bothered me that there was no physical basis for these detrending schemes.
I then came up with a brilliant idea. Why not assume that there was some magical CO2 concentration in pre-industrial times, let’s say 280 ppmv, and that all was in perfect equilibrium then, with the amount of the gas released into the atmosphere being exactly balanced by plant growth, weathering, etc., so that Earth had achieved that sublime concentration? After that, once we started pumping CO2 into the atmosphere, some, but not all, of that excess would raise the concentration above 280 ppmv (global average of 4.97 kg/m2), but if we stopped emitting CO2, the Earth would gradually settle back down to the Utopian level of 280 ppmv. I assumed that all of the increase from preindustrial levels was due to anthropogenic emissions, and that the restoration to 280 ppmv would be via a simple single exponential time constant. This assumes that the ability to fix CO2 is not somehow “poisoned” by rising concentrations.
Sadly, I found that this elegant, simple model had already been espoused by Dr Roy Spencer. Drat, I hate it when that happens. Anyway, I had global emission data for the 84 months of the first 7 full years of OCO-2 data and the results of models with different assumed time constants τ are shown in Fig. 2.
The least squares fit of the model to the data gives a value 38.24 yr for the time constant, which is equivalent to a half-life of 26.5 yr. It is possible to calculate an “instantaneous” value of τ from the data, and aside from a blip in 2015, the value is quite stable, suggesting that, at least for these 7 years, the efficiency of CO2 fixation was relatively constant. It should be noted that this model assumes that all of the increase of CO2 is due solely to anthropogenic emissions. If there are any other unaccounted-for sources, the true time constant will be shorter, so a half-life of 26.5 yr can be reasonably regarded as an upper limit. There may or may not be an “eternal” anthropogenic CO2 reservoir in the atmosphere, but it’s not apparent in any of the data that I’ve seen.
The next step was to take this simple model of CO2 growth and assume that it is evenly distributed across the globe. This was then used as the model to detrend the data, to see where the CO2 sources and sinks reside. This detrended data also has a lot of seasonal variability, which still tends to obscure things. However, you can do a sort of spectral analysis of this data, and the method I applied was the Hilbert-Huang transform. This decomposes a time series into a series of intrinsic model frequencies (IMFs), along with a low frequency “residue”. When you do this exercise over the 84 months of the detrended data, you find that grid cells have a minimum of 2 IMFs and a maximum of 6 IMFs. Places like Siberia and regions around the Arctic tend to have large numbers of IMFs, while Antarctica and the Southern Ocean tend to only vary slowly, resulting in few IMFs. In the following video, I show the results of plotting latitude averages of the detrended data, along with the detrended data minus the first two IMFs, i.e., residues with a maximum IMF value set to two.
As you can see from this video, most of the CO2 concentration variability is in the Northern Hemisphere, especially north of 60°N. This variability “whips” its way towards the southern hemisphere. There is also a persistent “bulge” in the concentration of CO2 that resides in a zone between the equator and about 45°N, and this is particularly apparent in the low pass “residue” plot. This is the region where excess CO2 comes from, and the Southern Ocean along with Antarctica is the predominant area where CO2 “goes to die”, so to speak.
This point is further illustrated in Fig. 3, which shows the mean of the low frequency residue from the global detrended CO2 concentrations. From this figure, we can see that the elephant in the room is clearly China, which acts as a major hot spot that sends CO2 across the Pacific and even into the Atlantic. India is also a significant hot spot, but its emissions bump into the Tibetan Plateau. The US is a very minor hot spot, and despite Canada’s relatively high emissions per capita, the original Dominion barely registers at all. I’m afraid that all the angst about emissions from the 2nd and 3rd Dominions, Australia and New Zealand, is hard to justify from this plot, as both countries are on average below the mean and appear completely featureless. Europe is “meh”, but major petroleum producing areas do register a bit above average. It is also interesting to note that the generally accepted figure for CO2 content is based on measurements on the island of Hawaii, but in Fig. 3, we can see that it averages about 20 g/m2 higher than the global average, which is roughly a bias of 0.3% for this “well mixed gas”.
While Fig. 3 shows the mean of the detrended residues, Fig. 4 shows their standard deviations. This map illustrates where there is the greatest variability in CO2 concentrations over the seven-year period.
The places with the least variability in CO2 concentrations include:
- Antarctica and the surrounding Southern Ocean.
- A zone near the Inter Tropical Convergence Zone (ITCZ) running through the Pacific and Atlantic.
- The Sahara.
- The Tibetan Plateau
- Mountain ranges in the Western USA and Central Mexico.
- Portions of the N. Pacific east of Japan.
Places with very high variability include:
- Russia, especially Siberia.
- The Arctic north of Russia.
- Parts of the Canadian Arctic Archipelago.
- The Canadian Maritime Provinces.
- A small part of W. Europe centered on the Netherlands, the breadbasket of Europe.
- Parts of Sub-Saharan Africa, especially the Congo Basin.
- Southern British Columbia, Alberta, Ontario, and Quebec.
- A hint of variability in the Mississippi Valley.
- Some strange “blobs” in the N. Pacific. More on these later.
It’s clear from Figs. 3 and 4 that all the “action” is in the Northern Hemisphere, specifically between the equator and 45°N. In Fig. 5, I’ve plotted the average CO2 mass anomaly from the detrended and low pass filtered data for three separate bands spanning 15° of latitude. The plots start at 105°E longitude, which is roughly at the western edge of China. Here you can see that China is by far the biggest anomaly, with concentrations generally falling as you go further east.
In the bottom panel of Fig. 5 (equator to 15°N), you can see that concentrations are relatively flat, with a bump up near Venezuela (northern S. America), and a later bump around the Congo Basin. In the second panel (15°N to 30°N), concentrations start high in China, with a gradual decline until we get to the Gulf States and India, where there is a rise in CO2. The exception to this general pattern is about 150° from the starting point, which corresponds to the Sierra Madre mountains of Mexico. In the top panel (30°N to 45°N), there is an early rise in China, followed by a steady decline until the Tibetan Plateau, after which CO2 rises again in the vicinity of western China. Within the USA, there is a sharp dip near the Rocky Mountains, which is similar to the pattern seen in Mexico for the middle panel.
It seems that mountain ranges are regions where CO2 concentrations sharply decline. One might be tempted to think that this is solely due to their lower overall concentration values, but this same pattern is exhibited when molar fractions are plotted instead of kg/m2. Some of this drop might be due to weathering, but Liu et al. (2004) suggested that rainwater is also an important mechanism for soaking up CO2 from the atmosphere, and this idea is compatible with the pattern we see in the USA and Mexico, where the concentration of CO2 drops in the area west of the mountains where one might expect orographic rain, and it rises again sharply in the rain shadow.
Now let’s see how the low pass filtered detrended data evolved over the seven years of OCO-2 data from 2015 to the end of 2021. The video is at a speed of 4 months per second.
The video has four different views: global views centered on 0° longitude and 180°E longitude, and polar views centered on 90°N and -90°N. Besides the persistent excess centered on China, there are some other intriguing features. At the beginning of the movie, there is a strange hot spot “blob” in the Pacific east of Japan, with a smaller region of CO2 deficiency just to the south of the zone of excess. This smaller subsidiary blob later becomes a region of excess concentration. This is illustrated as a single frame in Fig. 6. Another strange transitory oceanic region of CO2 release is shown in Fig. 7, which is centered in the N. Pacific south of Alaska.
I have no idea what these “blobs” of gas represent. They do not correspond to important fishing zones, so I doubt that they are “biologics” in the sense that Seaman Jones explains in The Hunt for Red October. Although there are seamounts in these areas, as is true for much of the Pacific, these areas are not especially active either volcanically or seismically. The best guess I have this that some methane clathrates became liberated from the sea floor for some reason and then quickly oxidized to form CO2. The location and size of these emissions might be controlled by deep ocean currents. I welcome any suggestions as to the mechanism for these seemingly random releases of significant quantities of non-anthropogenic gas.
Similar strange anomalies occur in the Arctic, well away from people and even significant plant growth of any kind. These can be seen in the North Pole projections of the detrended low pass filter video. Note that there are significant positive and negative anomalies near Svalbard and a very intense positive anomaly that pops up near Axel Heiberg Island in the Canadian Arctic Archipelago. Similar large-scale anomalies are totally absent in the Southern Ocean.
Other notable anomalies that I think I do have explanations for involve wildfires. Some of these are “anthropogenic” in the sense of fires being deliberately set for the purpose of clearing land for agriculture (e.g., the Amazon Basin), but many are likely purely natural. Areas of the eastern provinces of Canada as well as large parts of Siberia appear to be susceptible to significant wildfires. The large fire in British Columbia in 2017 is shown in Fig. 8. I can also almost convince myself that in the video, one can see the development of the large Greece and Balkans fires from 2021.
There’s no real “moral to the story” here, other than the fact that natural variations in concentrations caused by plant growth and decay, almost totally in the Northern Hemisphere, completely swamp all other features. By detrending the data and putting it through a low pass filter it is possible to see more subtle details. The “well mixed gas” carbon dioxide takes a while to get mixed and there are long term excesses and deficiencies throughout the planet. Barriers to the mixing appear to be the ITCZ and the Sahara Desert. Antarctica, the Southern Oceans, and to a lesser extent the Tibetan Plateau and high mountain ranges seem to act as CO2 sinks, while most other variations on land tend to correlate with agricultural land use and large-scale hydrocarbon production.
Wildfires are the likely cause of many large CO2 emissions, especially in the northern parts of the Northern Hemisphere. There are many transient positive and negative concentration anomalies in the Pacific and Arctic, and I have no explanation for them except they must be somehow related to degassing and absorption controlled by large scale ocean currents. As for large scale obviously anthropogenic emissions, China is the sore thumb. The Southern Hemisphere barely registers.
Appendix
This is a brief description of how I accessed the data, the steps used to put things into single georeferenced packages and what tools were used. This is for nerdy types who might want to tackle this sort of thing on their own.
I downloaded monthly data from the OCO-2 satellite, which covers the period from the beginning of 2015 until the first two months of 2022, just a bit over seven years. This forced me to learn something about how to handle NETCDF files, and with a little effort I was able to stitch the 86 spatial “XCO2” (molar fraction of CO2 of dry air) data arrays together to form a single 3-dimensional array, with the indices representing latitude, longitude and time.
Although daily data is available for OCO-2, I chose monthly data instead for a couple of reasons. One, the size of the downloaded data is a lot smaller, and two, there is complete data for every part of the Earth in the monthly datasets, while there is missing data from daily files. The downside is that gases in the atmosphere can travel quite large distances in the space of days or weeks, meaning that any patterns that might exist in the daily data could be hopelessly smeared out, obscuring any information about possible sources and sinks. I hope to show, however, that this is not the case, and we can definitely see spatial patterns for both CO2 sources and sinks.
I wanted to convert the OCO-2 data into a measure of the mass of CO2 per square meter so that I could get a better feel for how much C and CO2 we are talking about on a more human scale. This presented a little problem as I had to interpolate the OCO-2 data into the mid-points of their grid cells. The original data goes from -90°N to 90°N (361 latitude values), which is a bit awkward as the endpoints have zero area, so concentrations would be infinite. This is the classic fence post vs. fence rail problem, and it reduced the number of latitude values to 360. I shifted the longitude data, but I didn’t have to decrease the number of points, because as Arkady Darrell says in Isaac Asimov’s Second Foundation, “a circle has no end”.
The tools I used for this study were a moderately old laptop, the program RStudio, the R language, the MATLAB work-alike programming language Octave, GLE Graphics Layout Engine, the video editor Shotcut, and a very neat Java program provided by NASA called Panoply, which lets you make nice pictures and movies of NETCDF datasets.
Embedded Videos:
Raw data in ppmv: https://youtu.be/L1Dic2813zk
Raw ppm latitude averaged from 280 to 560 (pre-ind to x2): https://youtu.be/1fOvYYe9UyI
Model detrended low pass (residues) mass per sq. m.: https://youtu.be/IlNj4Vd-w0k
Model detrended mass and residues, low pass latitude averaged: https://youtu.be/LRXHuLqtTwc
References
IEA-EDGAR CO2, a component of the EDGAR (Emissions Database for Global Atmospheric Research) Community GHG database version 7.0 (2022) including or based on data from IEA (2021) Greenhouse Gas Emissions from Energy, www.iea.org/statistics, as modified by the Joint Research Centre. (https://edgar.jrc.ec.europa.eu/dataset_ghg70).
KNMI Climate Explorer and Global Carbon Project, Friedlingstein et al, 2019, https://doi.org/10.5194/essd-11-1783-2019.
Lesley Ott, Brad Weir, OCO-2 GEOS Level 3 daily, 0.5×0.625 assimilated CO2 V10r, Greenbelt, MD, USA, Goddard Earth Sciences Data and Information Services Center (GES DISC), Accessed: [Data Access Date], doi: 10.5067/Y9M4NM9MPCGH
Lesley Ott, Brad Weir, OCO-2 GEOS Level 3 monthly, 0.5×0.625 assimilated CO2 V10r, Greenbelt, MD, USA, Goddard Earth Sciences Data and Information Services Center (GES DISC), Accessed: [Data Access Date], doi: 10.5067/BGFIODET3HZ8
Mauna Loa CO2 Data: Data processed by www.woodfortrees.org. Data from NOAA Earth System Research Laboratory http://www.esrl.noaa.gov/gmd/ccgg/trends/ Time series (esrl) from 1958.2 to 2023.12
Liu, C.J., Ilvesniemi, H., Kutsch, W., Ma, X.Q., Westman, C.J. and Kauppi, P., 2004. An estimate on the rainout of atmospheric CO_2. Journal of Environmental Sciences, 16(1), pp.86-89.
Panoply was developed at the NASA Goddard Institute for Space Studies. More information about Panoply is available at www.giss.nasa.gov/tools/panoply.