The rapid retreat of mountain glaciers and ice caps has been cited as proxy evidence of global warming of surface air temperatures. In the past 20 years, while glaciers in the tropics, in particular, have retreated, surface air temperature data show marked upward trends. However, at mid-tropospheric levels, comparable to the elevations of mountain ice caps and glaciers, satellite and radiosonde data indicate little or no change in tropical temperatures. This apparent discrepancy has, to some extent, been reconciled by recognizing that, over short periods of observation, and using imperfect data, one might expect different trends at different levels. On large spatial scales, radiosonde observations suggest significantly different trends since 1979 at the surface and free troposphere, especially in the tropics. However, it is unclear whether, at mountain sites, temperature variations and trends more closely resemble those in the free troposphere at the same elevation or those at the surface at lower elevation.

We have used long-term radiosonde observations to compare the seasonal, interannual, and multi-decadal temperature variations at mountain and low elevation sites. Using radiosonde data in the Comprehensive Aerological Reference Data Set, the most complete archive of global radiosonde data, we have selected pairs of stations with long and reasonably complete records to allow direct comparison between high elevation sites (mainly above 750 m) and low elevation sites within 1000 km. The resulting network of 22 pairs of stations samples some of the world's major continental mountain ranges and high elevation regions, including the Sierra de Guadarrama in Spain, the Anatolian Plateau in Turkey, the Asir Mountains of Saudi Arabia, the Great Escarpment in South Africa, the Yunnan Plateau in China, the Tian Shan Mountains in Kyrgyzstan, the Yablonovyy Khrebet of Siberia, the Rocky Mountains in the United States and Canada, the Sierra Madre in Mexico, the Cordillera de Talamanca in Costa Rica, the Andes in South America, and the highlands of southeastern Brazil. We make four sets of comparisons: surface temperature at low and high elevations; temperature at the elevation of the mountain site and at the same elevation above the low site; and temperatures at 1 (and 2) km above the mountain site and at the same elevation above the low site.

We find the mountain site frequently, but not always, cooler than the low elevation site, at the surface. At the mountain elevation, temperatures are cooler at the mountain site than at the low site at night, but generally warmer in the afternoon, although inferences about diurnal variations are indirect. Temperature differences aloft diminish with height and show little diurnal structure, so that by 2 km above the mountain site, they are generally less than a few degrees. Monthly temperature anomalies at the mountain and low sites show high correlations (r > 0.6) at 1 and 2 km above the mountain, but the correlations are much lower at the surface. Correlations aloft tend to decrease with increasing station separation. Temperature trends, evaluated over the periods 1948-2000 and 1979-2000, reveal some statistically significant differences between the mountain and low sites, particularly at the two surfaces, as well as at the elevation of the mountain site. These differences are consistent with the notion of enhanced warming at mountain locations compared with low elevation surfaces and with the free troposphere in the tropics.