Key NOAA Dataset Upgraded Using AI

Since as early as 1850, the National Oceanic and Atmospheric Administration has been measuring changes in sea surface temperatures (SST), studying the impacts of temperature fluctuations on the ocean ecosystem. In the late 18th century, sailors aboard ships collected samples of sea water using wooden buckets to measure SSTs. Today, SSTs are collected by a variety of platforms including ships, buoys and Argo floats, and are usually taken from within the top meter of the ocean. Sea surface temperature data are an integral part of understanding both marine and terrestrial ecosystems, and NOAA’s Extended Reconstructed Sea Surface Temperature (ERSST) dataset is an important tool for studying long-term changes in the SST. ERSST is widely used in ocean research, weather forecasting and other environmental applications. The newest version has been upgraded using an artificial neural network to improve data quality.

Applying Artificial Intelligence

ERSST is a global monthly analysis of sea surface temperature data derived from the International Comprehensive Ocean–Atmosphere Dataset (ICOADS). This dataset allows scientists to study long-term trends in ocean temperatures on both a global and basin-specific scale based on in situ (taken by an on-site instrument) measurements of water temperature worldwide. ERSST provides a historical record of SSTs across the globe from 1850 to the present, incorporating statistical methods to fill in missing data points and create a more complete picture of global ocean temperature fluctuations. 

ERSST Version 6 (ERSSTv6) has been upgraded from ERSST Version 5 (ERSSTv5) by implementing an interpolation method using a form of artificial intelligence to fill in missing data points, creating a complete picture of global SSTs. An artificial neural network (ANN) is a method in artificial intelligence that teaches computers to learn complex patterns so they can produce insights and predictions. It is a subset of machine learning, called deep learning, that uses interconnected nodes or neurons in a layered structure that resembles the human brain.

Thanks to the implementation of the ANN, the ERSSTv6 now has higher spatial coherence and a lower error in the global oceans from 1850 to 2021 compared with ERSSTv5. In comparison with ERSSTv5, the spatial correlation coefficient (SCC; which measures the similarity of two SST patterns in space) with reference to observations increases by 5%, and root-mean-square-difference (RMSD; which measures the absolute difference between two SST fields) with reference to observations decreases by 0.03°C in ERSSTv6. The new version is further improved by progressively implementing the “nearest neighbor check” method in quality control on ship observations, which checks ship observation against nearby buoy or Argo observations and filter out potential outliers of ship observation, large-scale filter on SST super-observations, and ice-SST proxy algorithm in the ice-covered regions.

The improvements of SCC and RMSD are more pronounced in the tropical Pacific and the Southern Hemisphere oceans between 60°S and 30°S. In comparison with ERSST with the ANN method alone, the quality of ERSSTv6 improves in the statistical metrics of SCC and RMSD by 2%‒11% and 0.01°C‒0.24°C, respectively, in the global oceans. In the ice-covered regions, SST bias and RMSD decrease by 0.67°C and 0.29°C, respectively, a groundbreaking improvement in accuracy that will enable scientists to better understand the impact of even minor temperature fluctuations across the global ocean.

The ability to track SST anomalies makes ERSST a valuable tool for El Niño and La Niña monitoring and forecasting. ERSST is also integrated with land surface air temperature from the Global Historical Climatology Network-Monthly (GHCNm) dataset to create integrated surface temperature analyses in NCEI’s NOAA Global Surface Temperature (NOAA GlobalTemp) product.