Climate models simulate precipitation and its extremes under climate change (nonstationarity). However, they often have systematic biases that require correction before practical use at local scales. Conventional correction approaches for extreme precipitation do not deal with climate change very well, as they lack explicit and continuous nonstationarity treatment (are, in fact, stationary or quasi-stationary) and are challenged by scarce extreme-event data and high uncertainty. We propose a novel bias correction approach for extreme precipitation that explicitly models continuous nonstationarity due to climate change and leverages information from both ordinary and extreme events. Specifically, we introduce nonstationary quantile mapping and propose incorporating the simplified Metastatistical Extreme Value (SMEV) distribution. We demonstrate the superiority of the proposed method through a simulation study and real-world applications using high-resolution-regional and coarse-resolution-global climate models. Nonstationary quantile mapping reflects nonstationarity more realistically, but when relying on extreme-event records only, faces large estimation errors and uncertainty due to data limitations. These issues, common in conventional approaches, are effectively mitigated by using the SMEV distribution. The proposed nonstationary quantile mapping leveraging the SMEV distribution offers lower estimation error, approximate unbiasedness, reduced uncertainty, and improved representation of the entire distribution, especially with typical long datasets. Other methods may perform competitively with short samples, but they exhibit large biases in quantile-quantile matching due to bypassing nonstationarity modelling. The proposed method avoids these biases, aligns better with their theoretical functioning, and enhances the correction of extremes under climate change.
