A Novel Approach for Assessing Non-stationarity in Hydroclimatic Extremes

The escalating frequency of extreme hydroclimatic events, driven by climate variability and change, rapidly alters hydrological patterns and thus renders the traditional assumption of stationarity in hydraulic design and water resource planning obsolete. This study addresses the challenges posed by high spatial and temporal variability of extreme events, particularly in tropical and semi-arid regions, where understanding the processes driving short- and long-term climate changes remains complex. A novel non-overlapping block-stratified random sampling (NBRS) framework is proposed, integrating multiple nonparametric statistical tests to assess non-stationarity (NS) in hydroclimatic extremes. A modified NBRS framework incorporates a nonparametric clustering approach to detect spatial clusters of NS, caused by shifts in mean, variance, and distribution, or combinations of these factors. The NBRS framework distinguishes between weak and strict forms of stationarity and is further enhanced by a modified variant that identifies the stochastic processes influencing NS. A comparative assessment of the NBRS framework and its modified version with conventional trend and change point methods demonstrates its ability to identify time-invariant characteristics, especially in heteroscedastic variables like extreme rainfall and streamflow. This framework is applied to 28 hydroclimatic indices derived from over four decades of data from west-flowing river basins of India, which are characterized by diverse physio-climatic conditions. The modified NBRS approach effectively identifies NS in extreme hydroclimatic indices, elucidating its root causes and significant implications for hydrologic design. The findings reveal that traditional trend and change point tests are less effective in capturing time-invariant characteristics, particularly in heteroscedastic variables such as extreme rainfall and streamflow. Also, the distributional shifts predominantly drive NS in rainfall and streamflow extremes, whereas temperature extremes are influenced by changes in both mean and distribution properties. Valuable insights into the evolving patterns of hydroclimatic extremes under a changing climate can be drawn from this study.

Keywords: Non-stationarity, Hydroclimatic extremes, Statistical analysis, Climate change, Spatial clustering, Extreme weather events.