The oxygen isotopic composition (H218O/ H216O) of water vapor samples over the tropical oceans has been measured at three locations during distinct meteorological regimes (Lawrence et al, 2004). H218O concentrations were lowest in and downwind from the most highly organized precipitation systems and reflect the exposure of the air to extensive areas of rain (Fig1).

The H218O content of ice cores depends on at least four primary factors, 1: the initial content of the source vapor, 2: the effective condensation height, 3: the efficiency of fractionation during condensation and, 4: precipitation and evapotranspiration at upwind locations. The first two factors have either been overlooked or neglected and are focused on here.

Because heavy isotopes condense preferentially and fall from the atmosphere, their concentration decreases with height. In Figure 2, the measured average isotope contents of the Andean ice cores at the Huascarán and Sajama sites are plotted versus temperature at the height of the glaciers. Present temperatures are measured averages while temperatures at the LGM are estimated from oxygen isotope decreases in the ice cores (Thompson et al., 1995, 1998) and the interpretation of snowline data (Greene et al., 2002).

The data points in Figure 2 are compared to curves that show the decrease with temperature of the oxygen isotope ratio of condensate and water vapor in a Rayleigh distillation model. The Rayleigh curves assume that condensation is due to moist adiabatic ascent, and that all precipitation falls from the air instantly upon condensation. Starting points for the Rayleigh curves assume isotopic equilibrium with vapor evaporating from the sea at 76% relative humidity and 26º C for the present and 23º C at the LGM. For the water vapor and the condensation curve for the present (0BP) we assume that the water vapor over the ocean was derived from seawater with an oxygen isotope value of 0 per mil (SMOW). For the LGM we assume a value of 1 per mil (SMOW) for seawater to account for the storage of ice having a low oxygen isotope value in the polar ice caps. The Rayleigh model serves as a benchmark for fractionation processes in the atmosphere (Broecker, 1997). Despite its simplicity, it matches well with mean soundings of isotope ratios of vapor (Gedzelman, 1988). However, Figure 2 shows that measured oxygen isotope ratios are much below the values for the Rayleigh curves. The four factors given above contribute to the discrepancy. Here we address the first two.

Firstly, d18O of the vapor near sea level is 5‰ or more below equilibrium with seawater in or near large, organized storms (Lawrence et al, 2004). This is close to the discrepancy between the measured ice core d18O values from the Huascarán and Sajama sites at 0BP and the corresponding model condensation curve in Figure 2. Clearly, the H218O content of Andean ice cores is sensitive to the nature of the precipitation systems over the oceanic source region. The average oxygen isotope values of ice at the LGM are closer to the LGM model condensation curve. The closer proximity of these data points to the model curve can be explained by a higher oxygen value of the source water vapor. Less intense and less organized storm activity over the tropical oceans would have resulted in higher oxygen isotope values for the source vapors. Pierrehumbert (2000) anticipated significant changes in storm activity in tropics during the LGM " Whatever the world ocean does, the tropics may respond with some surprising reorganizations of convection".

Second, much snow in the Andes is produced by clouds that tower over the mountains and therefore originates at a greater elevation and consequently a significantly lower temperature than at the ice cap. Both Broecker (1997) and Pierrehumbert (1999) assumed snow formation occurred very close to the Andean ice cap elevations. The precipitation radar (PR) on the NASA TRMM satellite was used to estimate the present-day mean elevations of the radar echo tops over the Huascarán and Sajama ice caps at 8148 m and 8290 m, respectively. These are mean values for the months of November to May for the time period January 1998 to December 2001. The coverage areas encompass 0.5 by 0.5 degree latitude/longitude boxes centered on 9.25 S 77.25 W and 18.25 S 68.75 W, respectively. The icecaps are situated at 6048 m and 6542 m respectively (Thompson et al., 1995, 1998). Assuming a moist adiabatic lapse rate and a mass weighted mean condensation height about 200 m below the mid point of the glacier and cloud top, the effective condensation temperature of the ice core data in Figure 2 could be as much as 5º C lower for present conditions and 6º C lower at LGM. This brings the measured values close to the Rayleigh model curves. If, as expected, convection over the Andes were less intense and less organized with a lower tropopause during the LGM, there would be a smaller reduction of assumed condensation temperature for the LGM ice core data field.

The tropical ice core record therefore reflects changes not only of temperature but also of average degree of organization of tropical rain systems during different climate regimes. Finally, preliminary studies of monthly rainfall distribution in Greenland and of storm tracks of tropical cyclones in the most northerly part of the Atlantic Ocean have been examined. The low isotope values produced in these systems may be carried to the most southerly Greenland ice cores. These tropical cyclones often are absorbed in or undergo transition to extra-tropical cyclones that then produce precipitation in southern Greenland.