Calibration of precipitation forecasts from NWP models for ungauged locations

Statistical calibration of raw forecasts from numerical weather prediction (NWP) models aims to produce probabilistic forecasts that are as skilful as possible and reliable in ensemble spread. As lead time increases, the underlying skill of the raw forecasts diminishes. Some calibration models are developed based on a joint probability theory, allowing the calibrated forecasts to approach observed climatology when the underlying skill is very low. The observed climatology serves as a baseline accuracy for probabilistic forecasts and needs to be estimated as accurately as possible. However, accurate estimation of climatology requires quality and lengthy data records. There are many locations where forecast calibration is desired, but observations are not available. This study proposes a method to calibrate NWP forecasts at locations without rain gauge observations by (a) estimating the observed climatology from gridded reanalysis data to generate the background climatology fields; and (b) combining the reanalysis climatology with the remaining calibration model parameters transferred from a donor gauge with similar reanalysis climatology. The method is tested on daily precipitation forecasts using the Seasonally Coherent Calibration (SCC) model and evaluated at 50 masked-out gauged locations across Australia. It is shown to produce skillful forecasts up to 7 days ahead, with 2.16 % skill score reduction compared to a benchmark model which is established ideally using training data exclusively from the test sites. Due to the use of reanalysis data with adequate spatial and temporal coverage, the reliance on rain gauge density does not significantly impact the SCC_U forecast skill