Abstract
Summer snow plays an essential role in Arctic hydrology and in maintaining mass and energy balance of sea ice. However, there are great challenges in retrieving long‐term summer snow depths over Arctic sea ice. Here, we proposed a combined novel five‐variable long short‐term memory (hereafter CN5VLSTM) model based on brightness temperature data to yield warm‐season snow depth estimates. Then, year‐round snow depth estimates were obtained for the first time. The CN5VLSTM model and five additional snow depth methods were assessed during the warm season based on the ice mass balance buoy (IMB), Alfred Wegener Institute (AWI) snow buoy (AWI‐SB) and Multidisciplinary Drifting Observatory for the Study of Arctic Climate (MOSAiC) snow buoy (MOSAiC‐SB). According to the three buoy products, the accuracy of the CN5VLSTM‐derived snow depth was highest among the five snow depth estimates with RMSEs of 10.2, 16.4, and 10.1 cm, respectively. Except for in May, the Arctic snow depth showed mainly a downward trend in warm months, and a significant downward trend was found in the Central Arctic. Excluding the Barents Sea, Kara Sea and Canadian Archipelago, the average year‐round snow depth decreased in the other subregions, and a significant negative trend was observed in the East Siberian and Chukchi Seas. Snowfall was an important factor that was related to the changes in snow depth in the East Siberian and Chukchi Seas. This study can provide new insights into the evolution characteristics of summer snow depth.