Intercomparison of Different Automatic Threshold Selection Methods in Modelling Precipitation Extremes via Peak Over Threshold Model

The peak-over-threshold (POT) model is the most extensively used for regional precipitation frequency analysis (RPFA) for estimating extreme precipitation events (EPEs). Yet, choosing proper threshold values is critical and challenging while estimating rainfall quantiles for the Indian subcontinent due to the diverse climatic conditions and physical barriers. This study investigates and compares various threshold methodologies, including graphical, analytical, and multiple threshold methods (MTM) for identifying EPEs. These extracted extreme events with high thresholds followed the Generalized Pareto distribution (GPD), whose shape and scale parameters remain constant and increase linearly with increased threshold values. Therefore, the POT-GPD model was employed in the current work, and the parameters were estimated using L-moments to explore and quantify the heavy tail behavior. In addition, the uncertainty associated with the quantiles was also evaluated using nonparametric bootstrapping techniques and later understanding the spatial variability of the GPD parameters from various methods. The effectiveness of the models is assessed on daily gridded precipitation datasets for the Indian region and validated using synthetic datasets generated through Monte Carlo simulations. Results reveal the importance of combining the MTM and analytical threshold methods for identifying a range of critical thresholds to overcome the subjectivity of graphical methods and quantify the uncertainty. These findings contribute to developing region-specific thresholds, highlighting the importance of modifying thresholds to the regional characteristics rather than relying on a fixed percentile for characterizing the EPEs. The proposed approach is essential for assessing the increasing intensity and frequency of precipitation extremes associated with climate change while allowing for more focused mitigation actions and disaster risk reduction.