The unintended usage of precipitation reanalysis products: Downscaling of daily precipitation time series to hourly values using reanalysis products for parameter estimation

High-resolution precipitation time series are required for numerous applications in hydrology. For data-scarce regions, precipitation reanalysis products (PRP) are a promising data source. We validated two PRP for Slovenia and identified biases, which disable a direct usage of the PRP. However, the PRP were used for the parameter estimation of a cascade model to disaggregate daily time series, which exist for long periods for the whole country. The so assimilated data benefits from the advantages from both datasets: the daily rainfall amounts from the observations and the high-resolution temporal structure from the PRP. The disaggregated time series show a superior representation of the observed high-resolution point and areal precipitation time series in comparison to the PRP themselves, and their usage is recommended instead. The developed concept can be transferred to other data-scarce regions.

In more detail, from the latest PRP two are most promising due to their spatial and temporal resolution: ERA5-Land (raster width ~9 km width, temporal resolution of 1 h) and REA6 (6 km, 1 h). ERA5-Land and REA6 are evaluated in space and time by continuous and event-based characteristics as well as precipitation extreme values for five recording stations and 20 catchments in Slovenia. Both PRP show underestimations of dry spell duration, wet spell amount and average intensity, while wet spell duration is overestimated. For extreme values with 1 h duration both PRP lead to underestimations, whereby the bias increases with the return period. The identified biases are larger for ERA5-Land than for REA6. The PRP time series were used for the parameter estimation of a micro-canonical cascade model to disaggregate observed daily values to hourly values. The so estimated parameters differ from station-based estimations, e.g. probabilities for the generation of dry time steps (P(1/0), P(0/1)) are underestimated. Nevertheless, starting from observed daily rainfall amounts the disaggregated time series show a superior representation of the high-resolution precipitation characteristics in comparison to ERA5-Land and REA6. This conclusion is based on all studied precipitation characteristics.