Feedbacks and the Importance of Short Time Scales
In my last post, I used a simple time-dependent energy balance model to demonstrate that our traditional methods of shortwave feedback diagnosis from observational data are likely leading to âfalse positivesâ?. Without going into great detail again, suffice it to say that much of the interannual SW variability we see in the tropics is likely to be the result of non-feedback cloud forcing. This will always act to bias any diagnosed feedback (even a negative one) in the positive direction, which would then mislead climate researchers in their development and validation of cloud parameterizations and estimates of climate sensitivity.
The example I showed in that post used simple daily random variations in low cloud coverage, which produced substantial interannual (and even decadal) time scale SST behavior in the modelâs swamp ocean. A typical positive feedback value of -1.4 W m-2 K-1 was diagnosed from the cloud-forced SST variability (based upon 365 day averages), even though no cloud feedback was specified. We are currently working on a new paper describing the results of this investigation.
As was advocated by Aires and Rossow (2003 QJRMS), as well as Stephens (2005 J. Climate), our results suggest the importance of examining short time-scale behavior if we are to have any hope of observationally diagnosing cloud feedbacks. Once the data are averaged to seasonal or annual time scales, there is little hope of determining how much of the signal being analyzed is cloud forcing versus cloud feedback (cause versus effect).
But if we are to perform analyses on short time scales, a new issue arises: Is surface temperature necessarily the best reference for identifying cloud feedbacks in the observational data when short time scale behavior is analyzed?
Surface Temperature or Tropospheric Temperature?
Stephens (2005) pointed out that there is no rigorous justification for choosing surface temperature as a reference when doing feedback analysis. I believe that surface temperature is probably a logical choice when addressing the long-term equilibration to a new climate state (say from the radiative forcing due to increasing CO2) since both the surface and troposphere are expected to warm by roughly the same amount. Really, either one could be used as a temperature reference in this case.
But when we analyze the real climate system on interannual time scales, the signals we see do not represent long-term re-equilibration between two climate states. Instead, we are probably seeing relatively high frequency behavior that represents different states of disequilibrium in the system.
I will argue that, in the case of non-equilibrium short term variability, it is better to examine tropospheric temperature as a reference, rather than surface temperature. And hereâs why. When we look for a cloud feedback signature in observational data, we are implicitly asking, âHow do clouds respond when the surface warms?â? Well, given that most cloud activity is going to be caused by convective transport processes of one kind or another, we can expect that a surface temperature peak will precede the atmospheric response. In the case of our GRL analysis of the composite ISO, the time scale of this ocean cooling/tropospheric warming was one to two weeks.
Thus, on short time scales, the cloud response is probably more related to how the tropospheric environment has been modified as a result of convective heat transport, rather than to the surface temperature at that time. In other words, by the time the troposphere has been modified and the resulting cloud feedback appears, the surface has already been cooled to provide for that vertical heat transport! Any cloud feedback would then likely occur in proportion to the time rate of change of SST, but in direct proportion to tropospheric temperature.
The Infrared Iris Effect
Ever since John Christy and I started doing the global temperature monitoring with MSU back in 1989, I had been intrigued by the large intraseasonal oscillations (ISOs) we saw in tropospheric temperature. I have always believed that there were secrets to the fundamental operation of the climate system contained in those fluctuations. So about a year ago, I finally decided to investigate them.
In our resulting August 9 GRL paper we did not use surface temperature to reference the daily variations to, but instead tropospheric temperatures, which in the tropics vary mostly in proportion to variations in latent heating (precipitation). The tropical average surface signal of the intraseasonal oscillations was very small, while the tropospheric temperature variations were very large.
In something of a âfishing expeditionâ? we examined a variety of satellite observations that could be related to the tropical tropospheric heat budget. For the 15 largest intraseasonal oscillations between 2000 and 2005, we averaged TRMM TMI rainfall and SST, Terra MODIS cloud fractions, CERES reflected SW and emitted LW fields, and AMSU-A tropospheric temperature data to daily time scales, over tropical average space scales. The result was clear evidence in support of Lindzenâs âInfrared Irisâ? hypothesis, at least on the intraseasonal time scales we examined. (Unpublished was an analysis of the 15 next-largest ISOs, which revealed very similar signatures.)
We demonstrated though that composite analysis of 15 ISOs that enhanced rainfall activity and warming of the tropical troposphere leads to enhanced loss of LW radiation to space in the cloudy areas, as measured by the CERES instrument on Terra (see figure below). When the LW flux anomaly was normalized by the latent heat release anomaly (panel D), a linear increase in LW loss with time is seen during the period of above-average rainfall. This change is dominated by the cloudy areas (compare âall-skyâ? to âclear skyâ?).
This CERES-observed change in LW flux was supported by ice cloud measurements from the MODIS instrument on Terra, which showed that the LW increase coincided with decreased ice cloud fraction (see second figure; panel C is cloud top temperature). We had additional, but unpublished, evidence that it was a decrease in thunderstorm anvil clouds that was primarily involved.
Interestingly, when the SW and LW anomalies were summed together for the composite ISO, the net radiative cooling of the ocean-atmosphere system was in direct proportion to tropospheric temperature, supporting the use of tropospheric temperature as a reference for feedback analysis. Interpreted as a feedback parameter, the enhanced cooling rate was a comparatively large 6 W m-s K-1.
It is of some interest what the previous critiques of Lindzenâs original Iris work did not see the effect that we measured. While I am only speculating at this point, I suspect the answer is related to some combination of two effects: (1) Those critiques dealt with a restricted region: Lindzenâs original paper used one-ninth of the longitudinal extent of the tropics; and (2) the variations were related to SST, rather than tropospheric temperature, and so there was likely a significant time lag in the resulting relationships.
The Connection to Long-Term Climate Sensitivity
The critical question is this: What, if anything, do these observations have to do with long-term climate feedbacks and climate sensitivity? Well, when it comes to convection and clouds, all we have is short term behavior to analyze. And as discussed above, if we average to seasonal or annual time scales, we have largely averaged out the relationships that might have indicated to what extent feedbacks, versus non-feedback cloud forcing, were involved.
So, while these new satellite observations do not prove negative cloud feedback in the context of long-term forcing, neither do current âfeedbackâ? interpretations based upon observed interannual variations in the climate system. Recall the three other pieces of evidence from my previous posting, the first two of which were based upon the energy balance model calculations: (1) the CERES-observed monthly variability in reflected SW radiation is much too large to be explained in terms of feedbackâ¦they must be dominated by non-feedback cloud forcing, and any underlying feedback signal is likely being obscured; (2) the low correlation in the resulting relationship between SST and SW also suggests a forcing interpretation, since a feedback relationship should have a very high correlation; and (3) in the case where Pinatubo (rather than cloud forcing) was the dominant source of interannual SST variability, the SW feedback signal was less âmaskedâ?, and the resulting diagnosed SW âfeedbackâ? was, in fact, negative.
For these reasons, I believe that the current satellite results are more likely to be related long-term climate sensitivity than are current (mis)interpretations of âfeedbackâ? from natural, internal variability in the climate system.
The Fundamental Role of Precipitation Processes
I am increasingly convinced that the primary atmospheric control over climate sensitivity is the precipitation process. Our new satellite measurements showing a change in ice cloud (anvil cloud, really) during the period of enhanced tropical rainfall suggests a change in microphysical processes in clouds.
Since any condensate that does not precipitate out must, by definition, be detrained back into the environment as cloud and vapor, precipitation efficiency must somehow be involved in the observed changes. This issue is not just restricted to the tropics, since the same concept applies to precipitation systems everywhere. For instance, the extreme aridity of cold polar air masses in the winter, as well as of the air sinking over the Sahara desert, can be traced to precipitation processes somewhere else. It will be important to repeat our satellite analysis for the extratropical regions.
I believe that precipitation systems act as a thermostatic control mechanism on the climate system. Even though relatively small in their spatial extent, tropospheric air is continuously being recycled through them, and most of the Earthâs natural greenhouse effect is, directly or indirectly, controlled by them. Remember, the control is not just through detrained moisture, but also through the moist convective controls over tropospheric temperature lapse rates everywhere, which in turn affect cloud formation (e.g., marine stratus clouds).
The sensitivity of our climate to precipitation microphysical processes is not a new issue, as it has been previously demonstrated in the modeling studies of Grabowski (2000 J. Climate) and Renno et al. (1994 JGR), and was implicit in Lindzenâs original 1990 BAMS article on potential negative feedbacks in the climate system. But it has been largely ignored. To the extent that any of these microphysical processes have a temperature dependence, climate feedbacks can be expected to result.
I predict that, as we learn more about the temperature dependence of precipitation processes and their resulting control over precipitation efficiency, estimates of climate sensitivity will decline. But at this point, there is too little awareness of this issue in the climate community, and it will take time before a critical mass of climate modelers and diagnosticians forms and starts to take the issue seriously.