- Met Office, Climate Science, United Kingdom of Great Britain – England, Scotland, Wales (stephen.packman@metoffice.gov.uk)
The UK Met Office is an authoritative and trusted source of UK climate information and the custodian of the national historical climate records. We maintain a long-term time series for temperature, precipitation and other climate variables dating back to 1884 for monthly temperature and 1836 for monthly precipitation. The quality control of climate observations is essential for accurate assessment of the impacts of a changing climate. There are indications that a warmer atmosphere will result in increased intensity of rainfall and in order to monitor this our precipitation observations must be of a very high quality
Quality control of observations is a key challenge for precipitation due to its high temporal and spatial variability. The small-scale features and discontinuities in the field make it difficult to distinguish between erroneous values and true measurements. HadUK-Grid is the primary product used by the Met Office for producing areal statistics for monitoring the UK climate. The current quality control of data used by this product uses a simple range check to remove negative or extreme values and a check based on the distortion of the curvature field which assesses neighbouring station values.
Here we describe a machine learning (ML) based approach to the quality control of monthly rainfall values, with the primary goal of detecting and removing anomalies. The ML model incorporates a range of features including station metadata, spatial relationships with nearby stations, geographical variables such as elevation and proximity to coastlines, and past observations. This gives the model spatial context, allowing it to identify and preserve true extreme rainfall totals while removing the anomalous “bullseyes”. By training a model on historical data with known anomalies, the system learns to detect complex error patterns that traditional methods may overlook. The outcome is a more consistent and reliable set of UK monthly rainfall statistics that can be used to monitor the UK climate.
How to cite: Packman, S. and Carlisle, E.: Using Machine Learning based solution for quality control of UK monthly rainfall, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-637, https://doi.org/10.5194/ems2025-637, 2025.