Removing implausible precipitation extremes from CMIP6 climate projections using a GEV-based framework

Reliable identification of implausible values in hydroclimatological datasets is critical for ensuring the reliability of projections and subsequent analyses. Traditional methods, such as statistical techniques and observational comparisons, often fall short in addressing the complexities of large-scale climate data. In this study, we propose a novel methodology that integrates statistical techniques with a probabilistic framework to systematically detect and remove implausible extreme values while preserving the integrity of the dataset. Our method combines the generalized extreme value (GEV) distribution, the concept of risk of failure, L-moment estimation, and Monte Carlo simulations to establish robust thresholds for the detection of implausible values. The resulting risk-of-failure-based extreme value (REV) framework provides a probabilistically calibrated criterion for distinguishing statistically implausible extremes from physically plausible tail behavior in climate projections. By selecting a probabilistic threshold tied to exceptionally rare events, our approach effectively isolates highly improbable values while minimizing false positives and preserving statistically plausible extremes. The proposed methodology offers a scalable and reliable solution for managing large datasets, with significant implications for climate modeling and hydroclimatic research.

Subjects

Climate change

Climate models

Statistical techniques

Statistics

Time series

DDC Class

551: Geology, Hydrology Meteorology

519: Applied Mathematics, Probabilities

Options

Removing implausible precipitation extremes from CMIP6 climate projections using a GEV-based framework