An evaluation of sub-daily precipitation simulations from a new high-resolution regional atmospheric reanalysis ALADIN

This transboundary study evaluates the ability of the new high-resolution atmospheric reanalysis ALADIN to simulate sub-daily precipitation totals in the Czech Republic, Germany and, the Netherlands during 18 warm seasons (April–October) from 2002 to 2019. Specifically, the evaluation is performed separately for two numerical weather prediction (NWP) model runs. The first, ALADIN/Reanalysis, includes the full assimilation of the observed data every 6 h using a digital upper-air filter, which combines the high-resolution ALADIN guess with ERA5, the result of a global 4D variational analysis. The second, ALADIN/Evaluation Run, is also driven by ERA5 but uses no surface data assimilation. The evaluation is performed on an hourly basis with a horizontal grid spacing of 2.3 km (total = 215,169 km²) compared with observational data represented by gauge-adjusted radar precipitation measurements. The results show that simulated seasonal precipitation sums are overestimated by an average of 20 % (ALADIN/Reanalysis) and 23 % (ALADIN/Evaluation Run) compared to observations. The overestimation is usually greater in higher-elevation areas such as the Alps and the Black Forest in southern Germany or the Ore Mountains and the Giant Mountains in the Czech Republic. On the other hand, precipitation sums for the Netherlands appear to be the most accurate, with overestimations of less than 20 % for both NWP model runs. The main reason for higher seasonal precipitation sums is the large number of very low hourly precipitation totals (i.e., < 0.1 mm) in the reanalysis. Once filtered out, the mean precipitation totals are typically underestimated throughout the diurnal cycle, especially in higher-elevation areas. In addition, maximum values of convective precipitation in mountainous regions are simulated in the morning, while their actual occurrence prevails in the afternoon when the warming of the Earth's surface culminates. In contrast, the frequency of high hourly precipitation totals is underestimated in both model runs, as evident from the frequency of 24 mm precipitation totals in the ALADIN/Reanalysis and 29 mm in the ALADIN/Evaluation Run. The underestimation is less evident in the ALADIN/Evaluation run, which can produce higher hourly precipitation totals than the ALADIN/Reanalysis run. Additionally, the results show that reanalysis can more plausibly simulate long-term stratiform precipitation events than purely convective events or stratiform events with embedded convection.