There appears to be a disconnect between the mechanisms theoretically supposed to provide water vapor to the upper troposphere in the tropics and subtropics, and the mechanism that is most prominent in GCMs. In many GCMs, there is a strong cancellation between upper tropospheric moistening via the large-scale circulation, and drying due to compensatory subsidence from convection. A substantial net moistening effect arises from vertical transport of moisture by eddies, a mechanism that is completely ignored in most theoretical discussions

A key region showing variability in upper tropospheric humidity (UTH) during ENSO events is the subtropical region of the North Pacific. Although this is a region of time-mean descending air in the upper troposphere, analysis of atmospheric data shows that water vapor is supplied to the upper troposphere by vertical flux within transient eddies. The transient eddy activity in this region is in turn modulated by the interaction of the tropical circulation with the midlatitude circulation. Midlatitude interactions with the tropics are greatest during boreal winter and spring when transient eddy Rossby wave activity with periods between 5 and 30 days is at a peak. During most cold events, tropical convection is confined to the far western equatorial Pacific and westerly winds are found in the upper troposphere throughout the central and eastern equatorial Pacific. This allows the opening of a "westerly duct" in the eastern Pacific and supports anomalously high transient eddy activity in the subtropics. Conversely, during most warm events, deep convection and upper-tropospheric easterlies are found over the central and eastern equatorial Pacific. This closes the westerly duct, diminishes transient eddy activity in the subtropics, and creates extremely dry conditions. We have been able to relate the extremes in the tropical average (30N-30S) time series of upper tropospheric humidity (UTH) to variations in the westerly duct. We find that extremes in the UTH tropical time series over the last 20 years are twice as likely to occur in the boreal winter and spring, when the westerly duct is open, versus boreal summer and fall. This dynamical mechanism provides a negative feedback on interannual times scales. Implications of this mechanism for the water vapor feedback in global warming scenarios is also discussed.