A GNN-based approach for reconstruction of synoptic chart isobars

The Met Office contains a large archive of historic synoptic charts, which contain pressure field contour lines (isobars). We show that existing approaches to digitise contour lines are not robust enough to process these, and propose a novel approach that makes use of Graph Neural Networks (GNNs) to automatically perform the digitisation and reconstruction of isobars, as well as associating the contours with pressure values, for synoptic charts within the range of 1960–1975. Our approach results in the automatic transcription of 10 out of the 40 test maps requiring absolutely no manual editing, and the majority require very little editing (Median IoU of 98.2% for the geometry of the contour lines, 97.7% if including the associated pressure labels). Even without correction, the average mean absolute error for the reconstructed pressure field is 0.188hPa, and the average mean absolute error for the reconstructed wind field is 0.596m/s.

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-16
DOI
https://doi.org/10.5281/zenodo.24291081
Primary Topic
Handwritten Text Recognition Techniques
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

A GNN-based approach for reconstruction of synoptic chart isobars

Oliver Gibson
Zenodo (CERN European Organization for Nuclear Research)
Handwritten Text Recognition Techniques
preprint

A GNN-based approach for reconstruction of synoptic chart isobars

Oliver Gibson
preprint en

Abstract

The Met Office contains a large archive of historic synoptic charts, which contain pressure field contour lines (isobars). We show that existing approaches to digitise contour lines are not robust enough to process these, and propose a novel approach that makes use of Graph Neural Networks (GNNs) to automatically perform the digitisation and reconstruction of isobars, as well as associating the contours with pressure values, for synoptic charts within the range of 1960–1975. Our approach results in the automatic transcription of 10 out of the 40 test maps requiring absolutely no manual editing, and the majority require very little editing (Median IoU of 98.2% for the geometry of the contour lines, 97.7% if including the associated pressure labels). Even without correction, the average mean absolute error for the reconstructed pressure field is 0.188hPa, and the average mean absolute error for the reconstructed wind field is 0.596m/s.

Zenodo (CERN European Organization for Nuclear Research)
University of Cambridge (GB)
Handwritten Text Recognition Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon