Improving solar irradiance estimates for PV production in Ireland using Bayesian spatio-temporal modelling with open-source data

Accurate solar irradiance estimation is essential for reliable modelling of solar photovoltaic(PV) power production. In Ireland’s mid-to-high latitude maritime climate, solar estimation is shaped by seasonal variation, low winter sun angles, frequent cloud variability, sparse ground-based measurements and discrepancies among satellite products, making solar irradiance dataset selection challenging. We develop a Bayesian spatio-temporal framework for country-specific solar irradiance estimates across Ireland using openly available data. The model produces hourly and sub-hourly estimates with full uncertainty quantification, and is designed to support site evaluation and near-real-time applications. Cross-validation demonstrates statistically robust performance across temporal resolutions with hourly showing highest precision whereas sub-hourly encounters higher errors but improved uncertainty quantification. We compare with reanalysis data(ERA5), nearest-station interpolation and satellite-based products(CAMS). Our model outperforms ERA5 for hourly resolutions, reducing RMSE from 82 W/m² to 36 W/m² highlighting the limitations of coarse-resolution global reanalysis products in Ireland’s persistently cloudy climate. We also present the first evaluation of CAMS in an Irish context. For hourly irradiance, CAMS outperforms ERA5; however, our model further reduces RMSE(40 W/m² vs 35 W/m²). For PV power estimation, our framework remains competitive with CAMS while providing uncertainty unavailable for CAMS or ERA5. Our framework's practical value is demonstrated through comparisons with PV power output from residential, industrial and utility sites in Ireland, showing strong agreement with observed generation and providing an open-source irradiance dataset with full uncertainty quantification for high resolution solar resource assessment and improved estimation of PV system losses(e.g.clipping).