I'm not sure I understand your big spanner settings. - your monthly data used (0,1,1) and found nothing significant but the annual mean used (1,1,1) and found something. What does the settings difference mean in simple English?
The full specification is ARIMA(p,d,q)(P,D,Q) where:
p = non-seasonal autoregressive component
d = non-seasonal difference component
q = non-seasonal moving average component
P = seasonal autoregressive component
D = seasonal difference component
Q = seasonal moving average component.
Annual data is usually specified as just ARIMA(p,d,q). Monthly data are modelled as a seasonal series requiring the full 6-term specification except when using seasonally adjusted data, in which case the specification drops to just ARIMA(p,d,q).
When, say, p=1 this indicates inclusion of a non-seasonal autoregressive term at lag 1. If p=2 then this indicates autoregressive terms at lag 1 and lag 2. The same goes for q.
When P=1 this indicates a seasonal autoregressive term one season back etc etc.
Simple non-seasonal differencing is denoted by d=1, whereas seasonal differencing is denoted by D=1. A value of 2 means double differencing.
As for the temp effect what is going on here is that monthly changes in CO2 are not finding their way into monthly changes in anomaly (and vice versa). But we do see a connection when we use coarse annual data (year-on-year changes); this connection will be working both ways due to positive feedback, placing us in a chicken or egg situation.
Thank you so much for analysing this properly, John! Throughout the lockdowns and afterwards I've been "eyeballing" the Keeling curve, but been too lazy to do something more substantial.
As a (retired) modeller of relatively complex biochemical systems, I have been waiting in vain for a prediction from the climate modellers of the effect a sudden significant drop in CO2 production. All I could find were qualitative statements like "it will take years before such changes will become apparent" blaming some very long residency of CO2 in the atmosphere. Your analysis, with the yearly dip in March and the following "recovery" within a few weeks, shows (again!) that the residency argument is complete b******s. Thanks again!
Why thank you kindly! I didn't cover residency because the article was at the email limit as it was, but I may pick this up next time since it'll be used to wriggle out of the no-show as you state. I did some work on the issue of residency back in the following article (part 1 of 4)...
I'm not sure I understand your big spanner settings. - your monthly data used (0,1,1) and found nothing significant but the annual mean used (1,1,1) and found something. What does the settings difference mean in simple English?
The full specification is ARIMA(p,d,q)(P,D,Q) where:
p = non-seasonal autoregressive component
d = non-seasonal difference component
q = non-seasonal moving average component
P = seasonal autoregressive component
D = seasonal difference component
Q = seasonal moving average component.
Annual data is usually specified as just ARIMA(p,d,q). Monthly data are modelled as a seasonal series requiring the full 6-term specification except when using seasonally adjusted data, in which case the specification drops to just ARIMA(p,d,q).
When, say, p=1 this indicates inclusion of a non-seasonal autoregressive term at lag 1. If p=2 then this indicates autoregressive terms at lag 1 and lag 2. The same goes for q.
When P=1 this indicates a seasonal autoregressive term one season back etc etc.
Simple non-seasonal differencing is denoted by d=1, whereas seasonal differencing is denoted by D=1. A value of 2 means double differencing.
I hope that makes sense!
As for the temp effect what is going on here is that monthly changes in CO2 are not finding their way into monthly changes in anomaly (and vice versa). But we do see a connection when we use coarse annual data (year-on-year changes); this connection will be working both ways due to positive feedback, placing us in a chicken or egg situation.
Thank you so much for analysing this properly, John! Throughout the lockdowns and afterwards I've been "eyeballing" the Keeling curve, but been too lazy to do something more substantial.
As a (retired) modeller of relatively complex biochemical systems, I have been waiting in vain for a prediction from the climate modellers of the effect a sudden significant drop in CO2 production. All I could find were qualitative statements like "it will take years before such changes will become apparent" blaming some very long residency of CO2 in the atmosphere. Your analysis, with the yearly dip in March and the following "recovery" within a few weeks, shows (again!) that the residency argument is complete b******s. Thanks again!
Why thank you kindly! I didn't cover residency because the article was at the email limit as it was, but I may pick this up next time since it'll be used to wriggle out of the no-show as you state. I did some work on the issue of residency back in the following article (part 1 of 4)...
https://jdeeclimate.substack.com/p/atmospheric-residency-and-oomph