Stationarity Assessment of Precipitation and Temperature Extremes in the Continental United States

Stationarity assessments of annual extremes of monthly precipitation and minimum, average, and maximum temperatures at over 1200 sites in the U.S. are carried out using a nonoverlapping block stratified random sampling approach. The approach uses random partitioning of the time series into several blocks to assess different forms (i.e., weak, strong) of stationarity using nonparametric two-sample and multi-sample hypothesis tests. This approach's assessment of stationarity is compared with those derived from nonparametric Spearman’s rank correlation and variants of Mann-Kendall tests considering seasonality and autocorrelation. Monthly data of precipitation and temperature obtained from the United States Historical Climatology Network (USHCN) for the period 1910-2019 are used for this analysis. Tests (e.g.,  Augmented Dickey-Fuller (ADF) and Kwiatkowski-Phillips-Schmidt-Shin (KPSS)) specifically geared for stationarity assessments in econometrics and time series forecasting are also used for comparative assessment. Discrepancies in assessments from the nonparametric tests, ADF and KPSS, and nonoverlapping random sampling approach are noted in the number of sites. The random sampling approach used in the current study provides a robust assessment of stationarity considering the different characteristics of the hydroclimatic time series.