I have a new article coauthored with Robert Kaufmann in Climatic Change titled "Natural and Anthropogenic Causes of Climate Change." In the paper, we test for Granger causality between temperature and, as the title says, potential natural and human causes of climate change. We find that both natural and anthropogenic factors cause temperature
change and also that temperature causes greenhouse gas concentration
changes. Although the effects of greenhouse gases and volcanic forcing
are robust across model specifications, we cannot detect any effect of
black carbon on temperature, the effect of changes in solar irradiance
is weak, and the effect of anthropogenic sulfate aerosols may be only
around half that usually attributed to them.
It is very important in Granger causality testing to control for as many other possibly relevant explanatory variables as possible. So, in the paper we always include all the causes we consider in each regression model. We found only one other paper that attempted to do this - Triacca et al. (2013) - which was very recently published after we initially submitted our paper. And the latter paper still does not include anthropogenic aerosols. This was the main reason why we wrote this paper. The paper is also a follow up on our 1997 paper in Nature which pioneered the application of Granger causality testing to climate change.
All models in the paper include all the potential causes, the differences between models in the paper are in terms of:
1. Sources of temperature data - We use both the HADCRUT 4 and GISS3 datasets.
2. Time periods - We look at the full 1850-2011 period as well as the post 1958 period. Regular atmospheric sampling of carbon dioxide started in 1958 at Mauna Loa.
3. Ocean heat content - The ocean stores most of the heat accumulated through global warming. It is important to include it in climate modelling especially when using short time series. But we only have data from 1955 on. So we estimate models with and without this variable.
4. Restrictions on the equality of climate sensitivity across causes - All explanatory variables are converted to radiative forcing in Watts per square metre. If we aggregate all of these together into total radiative forcing we assume that the relative sizes of the effects have been correctly estimated and the dynamics of temperature in response to changes in radiative forcing are equal across all factors. So we estimate both more restrictive models that impose these assumptions and ones that don't. In particular, we allow the size of the effects of anthropogenic aerosols to vary. There is particularly high uncertainty concerning the size of the effects of sulfur and black carbon aerosols.
Another important dimension of the paper is that we use Granger causality tests that are robust to the non-stationary (potentially stochastically trending) nature of the global climate data. These are the Toda-Yamamoto Granger causality tests.
It is very important in Granger causality testing to control for as many other possibly relevant explanatory variables as possible. So, in the paper we always include all the causes we consider in each regression model. We found only one other paper that attempted to do this - Triacca et al. (2013) - which was very recently published after we initially submitted our paper. And the latter paper still does not include anthropogenic aerosols. This was the main reason why we wrote this paper. The paper is also a follow up on our 1997 paper in Nature which pioneered the application of Granger causality testing to climate change.
All models in the paper include all the potential causes, the differences between models in the paper are in terms of:
1. Sources of temperature data - We use both the HADCRUT 4 and GISS3 datasets.
2. Time periods - We look at the full 1850-2011 period as well as the post 1958 period. Regular atmospheric sampling of carbon dioxide started in 1958 at Mauna Loa.
3. Ocean heat content - The ocean stores most of the heat accumulated through global warming. It is important to include it in climate modelling especially when using short time series. But we only have data from 1955 on. So we estimate models with and without this variable.
4. Restrictions on the equality of climate sensitivity across causes - All explanatory variables are converted to radiative forcing in Watts per square metre. If we aggregate all of these together into total radiative forcing we assume that the relative sizes of the effects have been correctly estimated and the dynamics of temperature in response to changes in radiative forcing are equal across all factors. So we estimate both more restrictive models that impose these assumptions and ones that don't. In particular, we allow the size of the effects of anthropogenic aerosols to vary. There is particularly high uncertainty concerning the size of the effects of sulfur and black carbon aerosols.
Another important dimension of the paper is that we use Granger causality tests that are robust to the non-stationary (potentially stochastically trending) nature of the global climate data. These are the Toda-Yamamoto Granger causality tests.
Nice job!
ReplyDelete