Scholarlink

<p>Gridded surface air temperature (SAT) data for Antarctica is a crucial foundation for studying climate change in the region. However, significant discrepancies exist between the available Antarctic gridded temperature datasets, particularly regarding the spatial distribution characteristics of long-term temperature trends. In this paper, we develop a new, regularly updated, spatio-temporally complete Antarctic monthly SAT dataset from 1979 onwards, with a spatial resolution of 1° x 1° in latitude and longitude, from multiple sources of in situ observations using deep learning method. Deep learning model was trained with daily SATs from three global reanalysis datasets. The reconstructed Antarctic SATs were successfully validated using data from staffed and automated meteorological stations, demonstrating a closer match with observations, particularly in capturing the patterns of temperature trends. This dataset represents a new advance in the development of Antarctic observational climate dataset and is an important resource that underpins research across diverse scientific disciplines, facilitating a deeper understanding of the Antarctic climate system and its global implications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep learning-based reconstruction of monthly Antarctic surface air temperatures from 1979 to 2023

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年5月23日

查看全文
  • Ziqi Ma,
  • Jianbin Huang,
  • Xiangdong Zhang,
  • Yong Luo,
  • Tingfeng Dou,
  • Minghu Ding

摘要

Gridded surface air temperature (SAT) data for Antarctica is a crucial foundation for studying climate change in the region. However, significant discrepancies exist between the available Antarctic gridded temperature datasets, particularly regarding the spatial distribution characteristics of long-term temperature trends. In this paper, we develop a new, regularly updated, spatio-temporally complete Antarctic monthly SAT dataset from 1979 onwards, with a spatial resolution of 1° x 1° in latitude and longitude, from multiple sources of in situ observations using deep learning method. Deep learning model was trained with daily SATs from three global reanalysis datasets. The reconstructed Antarctic SATs were successfully validated using data from staffed and automated meteorological stations, demonstrating a closer match with observations, particularly in capturing the patterns of temperature trends. This dataset represents a new advance in the development of Antarctic observational climate dataset and is an important resource that underpins research across diverse scientific disciplines, facilitating a deeper understanding of the Antarctic climate system and its global implications.