Ensemble‐based statistical interpolation of atmospheric variables near the surface

This study presents a sequential implementation of ensemble optimal interpolation (EnOI) applied to the spatial analysis of near‐surface atmospheric variables. The proposed scheme, ensemble‐based statistical interpolation (EnSI), combines numerical model output with observations, as commonly done using optimal interpolation (OI) in some national meteorological services. However, EnSI extends OI by incorporating a multi‐scale loop, consecutive observation times, and cross‐correlations between variables, improving temporal consistency and potentially enhancing physical coherence between variables. We propose a started Box–Cox data transformation for the spatial analysis of variables with non‐Gaussian deviations from the background. EnSI is validated using 231 intense precipitation events, including a case study reconstructing hourly precipitation and temperature over parts of Scandinavia during the 2023 “Hans” extreme weather event. The data sources include ensemble model output and a dense network of in situ observations from both traditional and private weather stations. The availability of high‐density observations allows estimation of the local variability of temperature and precipitation at spatial scales of 2–3 km. Validation results demonstrate that the multi‐scale loop improves precipitation analysis accuracy and precision compared with a single‐scale approach. Additionally, incorporating consecutive observation times enhances temporal consistency in the temperature analysis, reducing sensitivity to sudden changes in observation availability. Finally, EnSI analysis spread can be tuned to represent local observation representativeness errors better, improving its applicability for real‐time weather monitoring and post‐processing of reanalysis.