LAUSR

Precipitation generators that are able to simulate long‐term, continuous rainfall records, with the observed statistics preserved, have shown great utility in numerous hydro‐climatic applications. This study adopts precipitation generators from four open‐source or freely available weather generators (WGs), namely the modified Wilks approach, Multi‐site weather Generator of École de Technologie Supérieure (MulGETS), Stochastic Climate Library (SCL) and Weather GENerator (WGEN) to evaluate their performance in simulating rainfall in Taiwan. The evaluation of the precipitation generators, based on various performance metrics, is carried out at two watersheds and the entire island to reflect seasonal weather systems of distinct spatial scales and correlation structure. To further dissect the heavy‐rain distributions, rainfall data that exceed 80 mm/day are extracted and assessed by quantile–quantile (Q–Q) plots. While none of the WGs can always outperform the others, MulGETS has the advantage of preserving most of the observed statistics and producing reasonable Q–Q plots. In contrast, the SCL results exhibit a heavier tail owing to positive biases in the simulated rainfall standard deviation and wet‐day frequency. Hydrological simulations driven by the generated rainfall data reveal more comparable percentiles in the low‐flow regime. Besides, although being a single‐site WG with no spatial correlation preserved, WGEN is useful for applications in small watersheds. Our findings indicate that merging generated rainfall data from different WGs by seasons can provide better support for WG‐related applications.