Satellites and machine learning improve near-surface temperature records
While weather station records offer a useful climate data record for those places for the period covered, there are a number of locations where either access to daily measurements is not possible, or no air temperature records exist. The EU-supported EUSTACE project extracted measurements from satellites to help estimate surface air temperature everywhere on Earth in a more comprehensive way than was previously possible. Producing new climate data records By combining direct measurements, estimates from satellites and machine learning using a statistical algorithm to create information where none existed, EUSTACE produced a number of data-set products. They constructed a standardised data set of weather station records of surface air temperature measurements for Europe since 1951, with a complete analysis of daily maximum and minimum temperature. Surface skin temperature estimates were produced for all surfaces of Earth, but with new, consistent uncertainty estimates. By using land, ocean and ice surface skin (top of the surface) temperature data from other satellite projects, the team were able to build these rigorous uncertainty estimates. This in turn allowed the team to analyse global daily surface air temperatures (with estimates of uncertainty) since 1850, based on combined information from satellite estimates and in situ data measurements. The project also identified, where possible, surface air temperature measurements (from meteorological stations) with discontinuities caused by non-climatic factors such as changes in the instruments used, the station location and the measurement procedure. “For these products, we tested mock datasets amongst international specialists with specific use cases,” says project coordinator Ms Nick Rayner. “This gave us insights into how climate data are used, data format preferences, and which variables were needed. For example, minimum and maximum temperatures were found to be just as important data as daily mean temperatures.” Many of these user requirements then fed into the design of the EUSTACE data file structure and the user guides which accompanied them. The accuracy of the EUSTACE products was evaluated against a subset of measurements from land stations, ice stations, ships and ice buoys. To maintain the independence of this data, it had been previously excluded from the analysis. “The validation results showed that our methods had worked well, especially our new uncertainty estimates,” says Rayner. The EUSTACE data products have been catalogued in the Centre for Environmental Data Analysis (CEDA archive), with data available via an Open Government Licence, where possible. Tackling societal challenges The team are now looking for opportunities to improve their satellite datasets for more accurate analysis. They are also pursuing the potential of digitisation (data rescue) of more historical data to improve their global analysis of air temperature by better constraining their statistical modelling, particularly when targeting regions currently under-represented. Increased understanding of surface air temperature contributes to a range of societal challenges. For example, it could shed light on the links between surface air temperature and diseases, or the impacts of heat stress. Food security, sustainable agriculture and forestry, marine and maritime inland water research, as well as renewable energy generation could all also benefit. “More broadly, by affording a deeper understanding of past surface air temperature variability, ranges and extremes, EUSTACE helps with efforts to develop climate predictions and projections, alongside adaptation and mitigation planning”, says Rayner.
Keywords
EUSTACE, satellite, weather station, skin surface temperature, data, machine learning, meteorological, mitigation, adaptation, climate, sustainability