Skip to main content
Research project

Global Surface Air Temperature (GloSAT)

Project overview

The overarching aim of GloSAT is to develop and analyse an extended and consistent global surface temperature climate record back to the 1780s, based on air temperature observations recorded across land, ocean and ice. This will be achieved through the production of a new, longer, and more reliable record of global surface temperature change. Existing estimates of global mean surface temperature combine air temperature over land and terrestrial ice-covered regions with sea surface temperature readings and take varying approaches for regions with sea-ice. The use of sea surface temperature measurements restricts the start date of the temperature record to around 1850, and the inconsistency of combining water and air temperatures limits confidence in estimates of climate sensitivity (an estimate of the temperature change that will result from a doubling of atmospheric CO2 concentration). The new GloSAT temperature record will give a longer and more consistent picture of global surface air temperature change, and its analysis will improve our understanding of climate change since the late 18th century.

University of Southampton will develop data rescue algorithms, including deep learning based natural language processing (NLP) and document layout analysis (DLA), to automate the processing of scanned historical measurements recorded in ship logs. Document layout analysis will identify columns and rows of measurements in tables of hand writen data. Optical character recognition and natural language processing models will extract meaningful measurements from the text on each page, allowing an order of magnitude more records to be included into the data record for use in climate change models.

The GloSAT project brings together the National Oceanography Centre, the National Centre for Atmospheric Science and the Met Office along with the Universities of East Anglia, Edinburgh, Reading, Southampton and York. Funding for the project is provided via a Large Grant from the UK's Natural Environment Research Council.

Staff

Lead researcher

Dr Stuart Middleton

Associate Professor

Research interests

  • Natural Language Processing
  • Human-in-the-loop NLP: Active Learning, Adversarial Training, Rationale-based Learning, Interactive Sense Making
  • Information Extraction: Few/Zero Shot Learning, Graph-based Models, Behaviour Classification, Geoparsing/Location Extraction, Event Extraction, Argument Mining