Newly reconstructed Arctic surface air temperatures for 1979-2021 with deep learning method

Area of Science:

• Climatology
• Arctic climate research
• Data science in environmental science

Background:

• Precise, regularly updated Arctic surface air temperature (SAT) data with comprehensive coverage is essential for monitoring and understanding rapid Arctic climate change.
• Existing datasets may have limitations in spatial and temporal coverage or rely on fewer observational sources.

Purpose of the Study:

• To reconstruct a new, precise, and regularly updated monthly gridded Arctic SAT dataset dating back to 1979.
• To improve the representation of in-situ observed temperatures within the Arctic by incorporating data from drifting ice stations and buoys.
• To provide a valuable resource for various applications related to Arctic climate research.

Main Methods:

• A deep learning method was employed to combine surface air temperatures from multiple data sources.
• Data sources include Global Historical Climatology Network (GHCN) land stations, International Comprehensive Ocean-Atmosphere Data Set (ICOADS) over oceans, Russian North Pole (NP) drifting ice stations, and International Arctic Buoy Programme (IABP) buoys.
• The method ensures improved spatial and temporal coverage based on instrumental observations.

Main Results:

• A new monthly gridded Arctic SAT dataset has been successfully reconstructed, with data available beginning in 1979.
• A daily Arctic SAT dataset is also available, starting from 2011.
• The dataset incorporates crucial in-situ observations from Arctic ice stations and buoys, enhancing data accuracy.

Conclusions:

• The newly reconstructed Arctic SAT dataset represents a significant improvement in observational temperature datasets.
• This dataset will facilitate timely monitoring and enhance the understanding of rapid Arctic climate change.
• It offers a valuable tool for diverse applications in Arctic climate research.