Quantifying Uncertainty in Historical Rainfall: A Simulation Study of Station Network Density and Design

Met Éireann provides 1 × 1 km monthly rainfall grids for Ireland from 1941 onwards, generated from more than 400 station observations using a regression-Kriging statistical gridding methodology. Work has been done recently to extend this dataset by generating historical rainfall grids back to 1855, when station coverage was sparse and spatially uneven. A key question in analysing Ireland’s historical climate is how this low density and particular spatial distribution influences the accuracy and reliability of these gridded rainfall products.

To address this question, we use Met Éireann’s post-1941 published rainfall grids as a benchmark to simulate historical station configurations over the 1855–1940 period. Subsampled networks are constructed to represent the varying station counts and spatial distributions consistent with the historic record. The current operational gridding method is applied to these reduced networks, and the resulting fields are evaluated against the full-network reference using quantitative metrics at both national and county scales. The county level analysis informs grid quality assessments where certain regions of the country can be more accurately gridded than others, and reasons for this can further investigated. This framework allows us to estimate reconstruction error, spatial bias, and temporal variability under historical scenarios with reductions in network density.

In addition, alternative gridding methods are implemented and compared to assess their performance under these sparse station distributions. The results offer methodological guidance for constructing historical rainfall grids and establish statistical thresholds for the minimum station density required for acceptable accuracy at county and national scales. These simulations also suggest principles which can be applied to gridding  methodologies in general including importance of stations at elevation and the climatology to include in the gridding model. We present this simulation framework as a powerful tool for analysing the efficacy of historic rainfall networks.