Berkeley Earth, a non-profit research organization, is releasing a beta version of its new High-Resolution Temperature Dataset. Representing a significant improvement over the existing, peer-reviewed data set, the Berkeley Earth High Resolution Data Set incorporates machine learning technology to reproduce small-scale temperature variations, allowing for unprecedented spatial resolution relative to existing products. At 0.25° x 0.25° spatial resolution, the High-Resolution Data Product will enable Berkeley Earth to provide city-level time series data for over 8000 cities, in addition to gridded and time series estimates for every country on earth.
This new Berkeley Earth High-Resolution data products, available at global, regional, and city levels, are now available in beta via request at https://berkeleyearth.org/data/
Improvements in Resolution and Accuracy
The new Berkeley Earth High Resolution Data Set improves upon the previous version by providing a 0.25° x 0.25° lat-long resolution (approximately 30 km at the equator), which is four times higher than the previous 1° x 1° resolution. This allows for a more accurate representation of small-scale temperature variations, particularly in areas where geography is changing rapidly, such as coastlines and mountainous terrain. It also does a better job of capturing ocean variations related to currents and other structures. The new gridded data product derives its information from approximately 50,000 weather stations and more than 450,000,000 ocean temperature measurements, providing excellent coverage of the Earth’s surface.
The machine learning component of the new data set extracts weather patterns from modern, high-resolution data and uses that information to better estimate small-scale structures in time periods and places when less information is available. This is analogous to other AI technology that takes blurry images of faces and transforms them into estimated high-resolution images by using an awareness of what features faces are expected to have. This technology allows for a better representation of regional and local climate changes.
The machine learning-derived weather patterns are layered on top of Berkeley Earth’s existing, peer-reviewed Kriging-based interpolation technology, resulting in more accurate spatial reconstructions at both local scales and global averages. In addition, the base resolution at which the Kriging field is sampled has been increased. This allows for direct reconstruction of small-scale detail in places where weather observations have been densely collected.
This weather-pattern recognition technology improves the accuracy of the resulting fields and allows for higher-resolution reconstructions. It also creates some small adjustments to the resulting global average estimates. However, it does not change the broad understanding of long-term climate change. The primary value of the new technology is it allows local scale changes to be better understood.
Continued World-Leading Position
This new data set solidifies Berkeley Earth’s position as leaders in the reconstruction of climate change warming trends from historical weather observations. Berkeley Earth’s peer-reviewed data products incorporate more weather observations from more global weather stations than any other instrumental data product, and produce historical temperature fields with a much higher resolution than the 2° to 5° resolution products produced by NOAA, NASA, and UK’s HadCRU.
The new data set not only allows for a better understanding of small-scale climate change, but will also help to better calibrate and test the next generation of high-resolution weather and climate models.
We expect that high-resolution data, such as that now provided by Berkeley Earth’s High Resolution Data Products, will be increasingly relevant to the work of scientists, economists, and policymakers as they seek to anticipate and respond to the local effects of climate change.
Berkeley Earth’s previous peer-reviewed data product, upon which the high-resolution product was developed, was originally established more than a decade ago. Extensively cited in the recent UN IPCC report, Berkeley Earth’s temperature data products are highly trusted by researchers spanning climate science, policy, and economics, referenced in hundreds of academic papers annually, and forming the basis for leading climate journalism. The new Berkeley Earth High-Resolution Data Set builds upon this foundation and represents a significant step forward in our ability to represent, analyze, and understand local and regional climate patterns.
A beta version is being released in March of 2023 to help facilitate feedback from scientists and other potential users. A description paper for Berkeley Earth’s High Resolution Data Set is expected to be submitted later this year.
The Berkeley Earth High-Resolution Data Set has been funded, in part, with generous support from the Patrick J. McGovern Foundation. The current beta version provides monthly mean temperatures. Global gridded fields are provided since 1850, with some gaps where historical measurements are not available. In addition, local and national time series for monthly average temperature are available, with some locations beginning in the late 1700s. Berkeley Earth hopes to extend this to include monthly-average high and low temperatures in the near future.
Additionally, Berkeley Earth is currently seeking funding to support the development of a daily data product using this technology. Such a data product would have immense advantages for the understanding of heat waves and other extreme events.