In the deeper part of polar ice sheets, air hydrates are the primary air inclusions. They trap most of the ancient air molecules used for paleoclimatic reconstructions in their crystal structure. Furthermore, their number and size was found to correlate with past climate. Microphotographs of ice thick sections of the EPICA Dronning Maud Land (EDML) ice core were obtained in the years 2002 - 2006 during the Antarctic field season, shortly after the sample was drilled, using an optical microscope (Kipfstuhl et al., 2006; Kipfstuhl, 2007). This dataset presents eight ground truth (GT) images with the dimensions of 20 x 10 mm together with masks of hand-labeled air hydrates differentiated into five visual categories. The visual categories are: Above focus (AF), In focus, Below focus (BF), Far out of focus (FOF) and air hydrates that can hardly be distinguished from other image artifacts (X).

A GT-set includes the following images (example for 1285.01C):- 1285.01C_GT_crop: The cropped microphotograph of sample 1285.01C.- 1285.01C_segmented: The automatically segmented microphotograph.- all_labels_1285: Masks of hand-labeled air hydrates including all visual categories.- AF_1285, focus_1285, BF_1285, FOF_1285, X_1285: Masks including air hydrates of only one visual category.The images were labeled and analyzed by Florian Painer at the Alfred Wegener Institute, Bremerhaven and the analysis routines were implemented in Python using open-source libraries. The aim was to develop a semi-automated analysis routine and to provide a high-resolution record of air hydrate properties of the EDML ice core.