Milankovitch cycles and what?

Observe the vertical green lines in this diagram: Those events shows when temperature starts to sink when radiance is at top or even before that. If there was a simple connection between radiance and temperature the temperature should start to sink after radiance maximum.

What climatic mechanisms could start a global cooling when the radiance is maximal or heading to maximal, due to the Milankovitch cycles? Could it be a drastic increase of convection?

Constructive answers will be rewarded with bounties

• I think your premise is not 100% justified. Looking at this graph, we see that 1.) the T-amplitudes and S-amplitudes are far from being perfectly correlated 2.) between -200k and -150k yrs, T doesn't react to S at all, 3.) yes, there is maybe some small indication that in some cycles T starts to decline earlier than S, but on that scale that is hard to see, and taking into account 2.) might not be significant. Can you maybe show a zoom-in of one of the events you think fits your premise, together with error bars? May 29 '19 at 11:50
• What? That was not at all what I critisied. I said that the facts you're pointing out, are not facts. They're supported by at most 50% of the peaks in your graph, and in the others you don't see the effect which you're stating that exists clearly. In science one has to be careful what one might see after looking at a graph for longer, therefore, a clearer analysis is required. May 29 '19 at 12:45
• You are taking far too much data from climatedata.info, a well known denialist site. They cherry pick, they fudge, they out-and-out lie. You are not going to learn a thing, @Lehs, if you're primary source of information is denialists. May 30 '19 at 4:00
• the temperature should start to sink after radiance maximum Probably incorrect assumption. Just like the temperature during the day, that is at its heighest some hours after the moment of maximum radiation. Not that that explains what you're observing, I'm just illustrating how hard interpretation is. May 30 '19 at 9:04
• @Lehs: All those processes take seconds to minutes as response times to a changing irradiation. The delay timescale of ?hundreds of years you want to see in this data is therefore not it. As i repeatedly said, go into the data, zoom in. May 30 '19 at 15:11