The layered sequences of melt features preserved in inland polar ice sheets provide valuable proxy data on past variations in summer temperature. A continuous detailed light-table examination was made on the 2037 m-deep Dye 3 ice core immediately after core recovery. Melt features are products of high air temperatures or solar insolation which occur only at or near the snow surface during summer months. A correlation is made between these features and the extent and intensity of summer temperatures. Care must be exercised to identify and distinguish between all mm-thick radiation crusts and wind crusts contained in the record. The absence or presence of these discrete features serve respectively as indicators of total summer cloud cover and the extent of winter storm activity although they are difficult to differentiate from only light-table observations. In this analysis both thin radiation crusts and wind crusts are not in themselves significant indicators of long-term temperature trends, but may serve as incipient subsurface horizons or barrier crusts for the formation of thicker ice melt features caused by downward melt percolation during elevated surface temperature conditions.

More than 10 000 individual melt features, including ice layers, ice lenses and ice wedges (but excluding ice glands) were measured down to a depth of 1278 m; below this depth transformation of air bubbles to transparent air hydrate inclusions occurs (Shoji and Langway, 1987) and the megascopic melt features become obscure. The melt-feature data extends back to 1883 B.c. or approximately 3900 years B.P., based on the accurate time scale of continuous δ18O measurements (Dansgaard and others, 1985) For the entire core profile investigated the annual melt percentage is 5.7. Individual melt features range in thickness from 1 mm to 100 mm. A mean value for melt-feature thickness was calculated for continuous 30-year time intervals to consider the general long-term summer temperature trends with corrections made for progressive annual accumulation layer thinning due to ice flow.

Since the AMP parameter includes noise from radiation and wind crusts it appears that the simple average of melt-feature thickness per longer time-units is a better indication of air temperature paleodata. The average melt-feature thickness is 1.2 cm. The complete curve obtained shows a higher thickness value of about 1.5 cm for the period 1800 B.C. to 1300 B.C. A lower, almost constant thickness value of about 1cm is shown for the period 1000 B.c. to 1800 A.D., with a slight reduction in thickness recorded around 200 B.C., 400 A.D. and 1600 A.D. These long-term trends are coherent with those recorded in the δ18O profile for the same ice core.