Abstract: The applicability of different statistical approaches for estimating hourly precipitation thresholds across China is evaluated using data from 2464 manned weather station and 40437 unmanned weather station. Two categories of methods are compared: Extreme value theory models (EVT)-namely the Generalized Extreme Value (GEV) distribution and the Pearson type Ⅲ (P-Ⅲ) distribution, and empirical distribution-based techniques, including percentile ranking, percentile interpolation, the Z-index, and square/cube root transformations.
Several important findings emerge from the analysis. First, the statistical characteristics of 1-h precipitation are found to be markedly right-skewed, reflecting the dominance of infrequent but intense rainfall events. Conventional normalization techniques, such as the Z-index or root transformations, are observed to enhance distributional normality only to a limited extent. The effectiveness of these methods improves under high truncation thresholds (e.g., 20 mm), yet in regions with frequent heavy rainfall events, their ability to approximate normality remains insufficient. Second, the spatial distribution of precipitation thresholds estimated by different methods is broadly consistent. High threshold values are mainly observed in South China and eastern North China. In contrast, a pronounced northwest-southeast oriented low-value belt is identified, extending from southern Henan through central Anhui to northern Zhejiang, while low values also dominate western North China and the arid northwest region. Third, systematic differences emerge between two methodological categories due to their underlying statistical principles. Thresholds derived from extreme value theory models are generally higher than those estimated from empirical methods. Among the EVT models, the GEV distribution shows greater sensitivity to the most extreme values, making it particularly suitable for characterizing rare and intense rainfall events. In contrast, the P-Ⅲ distribution, though effective, tends to be less responsive to the extremes. Empirical approaches are advantageous involving short data records or when no distributional assumptions can be justified, and they also demonstrate strong spatial transferability across regions. Finally, the selection of the truncation threshold is shown to be critical and should be determined according to research objectives. Low thresholds (e.g., 0.1 mm) are suitable for generalized climatological analyses, whereas higher thresholds are more appropriate for the study of extreme events. This flexibility underscores the importance of aligning methodological decisions with the intended application, whether for hydrological risk assessment, infrastructure planning, or broader climatological studies. For nationwide threshold estimation at individual stations, percentile interpolation is identified as the most suitable method due to its distribution-free nature and applicability to shorter records (e.g., more than 10 years), making it ideal for regional stations or gridded precipitation products with limited temporal coverage. In contrast, region-specific or seasonal analyses require careful selection of truncation thresholds based on local precipitation climatology to enhance representativeness-lower thresholds are sufficient for general climate studies, whereas higher thresholds are more effective in isolate extreme events, particularly in flood-prone regions.