Slide 1. Thank my co-authors.

Slide 2. Carbon dioxide is the ghg which contributes the most to human-induced climate change, and it is also the gas that is increasing the fastest. One minor good news is that it is arguably the gas for which we have the best understanding of its emissions and sinks. Here I show a sketch of the main components driving the global budget of CO2. In the past 20 years, the emissions of CO2 have been dominated by the burning of fossil fuel, with a 15-20% contribution from deforestation and other land use change.
"Only" 45% of the total emissions remain in the atmosphere on average, the rest is taken up by the land and ocean CO2 sinks, in roughly equal proportions.

Talking in PgC = 1 Gt C = 3.67 Gt CO2

Slide 3. I will go into detail of where these numbers come from and their uncertainty in a minute. For now, I just want to say that if you look at each year in detail, the numbers change quite a lot. On the emissions side, the fossil fuel emissions are growing very rapidly. But the numbers change a lot on the sinks side as well, because of variability in the natural system.

Slide 4-5. Let's have a look component by component: The CO2 emissions from fossil fuel burning have increased by 1% per year in the 1990s, and 3% per year since year 2000.

Slide 6. The growth in global emissions is almost entirely accounted for by growth in countries with economies in transition, here I show how this looks when split into Annex B (industrialised countries), and non-annex B (developping countries). There is a lot of information behind this curve, in particular that 25% of the growth in non-Annex B countries comes from the production of goods consumed in the west. However the reason I show it here is because of the consequence or the uncertainty in emissions. Emissions are computed from statistics on energy consumption within each country, which are then converted to CO2 emissions by fuel type and aggregated. The statistics on energy consumption are far better in industrialised countries which have good and overlapping control on energy use, than in countries with economies in transitions because of the nature of transitions. As the emissions grow in time, the uncertainty will grow (in absolute numbers), but it will also grow because of the larger share of global emissions from economies in transition.

Slide 7. At present, the uncertainty is around 6%, for a +/- 1 sigma uncertainty (66% chance that the real value lies in the gray area).

Slide 8. I add to this the estimates from Land use change estimated by Skee Houghton based on a book keeping method. The LUC remained approximately constant through time, though the geographical location changes a lot from the high latitudes in the first 20 year to the tropics since the 1980s. There is a lot of debate in the community regarding how well we can estimate this component.

Slide 9. Here I show the time evolution of LUC from different estimates on top. The agreement is initially very good, but it becomes particularly controvertial in the late 1980s, owing to the fact that the statistics on deforestation area are not consistent with statistics on conversion to agriculture and pastures globally. So there is a fundamental problem with the underlying datasets used to calculate LUC.

Slide 10. When the ff and luc emissions are added together, the global picture is that of increasing emissions through time, and increasing uncertainty.

Slide 11. That was for the emissions side. Now let me go to the partitioning side.

Slide 12-13. Only 45% of the total emissions remain in the atmosphere on average. There is some evidence from both observations and models that this fraction (the airborne fraction) has increased from about 40% in the 1960 to 45% towards the end of the time series, but this would be the topic of another talk. What is notable here, is that there is very large year-to-year variability in the atmospheric growth rate, but in spite of this the uncertainty in this term is smaller than the black line here. The growth of CO2 in the atmosphere is very well known globally, because the atmosphere is well mixed and thus relatively few measurements are required to get an accurate estimate.

Slide 14. I now move on the the sink and their individual uncertainty. The ocean CO2 sink, here takes up about a quarter of the emissions every year. The mean sink in the 1990s has been estimated from observations, the O2/N2 ratio that Andrew Manning talked about and also other methods, such as methods based on CFC penetration in the ocean and ocean inverse methods. The uncertainty in the mean is +/- 0.4 PgC/y. The trend here was estimated by 4 global general circulation models. The models trends and year-to-year variability is relatively close to each other, and the mean absolute deviation of the models is here only +/- 0.2 PgC/y. It is difficult to assign a confidence on the model results at this point.

Slide 15-16. The very last component that I will talk about is also the least constrained and the most variable. It is the uptake of CO2 by the terrestrial biosphere in response to increasing CO2 in the atmosphere (a fertilisation effect) and changes in climate. Here we have normalised the land sink so that it is balanced in the 1990s, and estimated the trends with 5 global vegetation models. The error corresponds to the propagation of errors in the budget in the 1990s, plus one mean absolute deviation from all the models. You can quickly see that this is the weak part of the budget here, as it is large and variable and the uncertainty is also large. It is extremely unlikely that we will ever have an annual estimate of the ocean sink based on observations alone (though there is some potential from inversions).

Slide 18. In spite of all the uncertainties, and very encouragingly (surprisingly even), the comparison of all the emissions (in black) and all the sinks (in red) is reasonably good. All the terms are estimated independently here, except that they are forced to balance in the 1990s. The sinks here include the atmosphere, ocean and land. The best part here is that the trend in total emissions is the same as the trend in the sinks. That indicates that the time scales of uptake of both sinks are well represented in the models. It also indicates that the models respond correctly to the recent climate change and variability, and this is of course very important. The other good news is that the match is better in the second half than in the first, and this can be explained at least in part by the fact that the climatic conditions were better known, particularly since we have good spatial coverage from satellite data for many variables. The less good news here is that there are easily half decades or even decades where the independent estimates do not match.

Our paper argues for a simple concept. That is that there is such a thing as a "minimum useful uncertainty" that annual CO2 budgets must reach. The minimum useful uncertainty would be that the uncertainty in the total sinks be less than the uncertainty in the total emissions. If that were the case, and the red line then deviated from the black line, we would have good indication that there is a problem. The problem could be two things (1) CO2 is emitted and not declared (intentionally or not) or (2) the CO2 sinks are not behaving as they should.

Slide 19. Let's have a closer look at the uncertainty, in order to see where gains can be made. In the emissions side, the uncertainty is growing with time, which makes our target of a minimum useful uncertainty in the sinks even easier. On the sinks side, the uncertainty does not grow with time (and it should not). The gains are to be made particularly for the land sink, and also for the ocean sink (in particular in terms of confidence in the model results).

Slide 20. This is where we stand here, with the uncertainty of the total sinks some 20% larger than the growing uncertainty in the emissions.

Slide 21. Let me now have a look at how the uncertainty in each component could be reduced. for the ff itself, the infrastructure for accounting particularly in countries with economies in transition. The potential is rather low because the uncertainty is already low. The cost would be high (>$10 million/y) because it requires significan expansion of infrastructure.

for the land use change, the basic statistical info need to be improved, and probably the best method would be to have satellite monitoring of deforested area like is already in place in Brazil. Even with perfect lu statistics though, there is still the problem of keeping track of the time scales for decay and re-growth, and that is hard. Thus the potential is medium.

Slide 22. Going into the sinks, well globally we cannot do better than what is already in place for the atmosphere. For the ocean, there is a large potential for improved confidence through the expansion of ocean surveys. The biggest gains though can be made in the improvement of land models. Because the uncertainty is very large, and because there is a lot of available information that has not yet been used to constrain the models. The costs on the sinks side is low (here we are talking of the order of $5 million per year) per component. It is difficult to say exactly how much the uncertainty could be reduced here. In order to reach the minimum useful level, it would have to be cut in half. The potential for reducing the uncertainty in the global is large, and this can all be done with existing tools (in many places with already existing information), and at a cost of perhaps 10 million per year if a good strategy was in place.

Slide 23. Now just to put this in perspective, these 10 million per year can be compared to the benefits of a good monitoring system, if it was to increase the incentives to have accurate declared emissions. Here we ran an integrated assessment model (the PAGE model updated from Stern review), a mid-range scenario mid-range climate sensitivity. We looked at the cost in terms of climate impacts if the emissions are really 5% higher than we think they are. This is the red curve here, nearly 2 billion per year in 2050. We then assumed that the monitoring system is in place by 2020 or 2030, and thus that emissions are accurate passed this date. The gains are 0.5 to 1 billion per year depending when the system is in place. The gains grow in time (as the impacts grow) and they are of xx in 2100.