Current Status of Time Series Analysis in Hydrological Sciences - Hydrologic Time Series Analysis: Theory and Practice - page 116

Lins and Slack (1999) determined secular trends in the streamflows of
395 climate-sensitive streamgaging stations in the conterminous United States
by the nonparametric Mann-Kendall test. Trends were calculated for the selected
quantiles of discharge [0 th ] to [100 th percentiles] to evaluate the differences
between low-, medium-, and high-flow regimes during the twentieth century.
Two general patterns emerged: (i) trends are most prevalent in the annual
minimum ( Q 0 ) to median ( Q 50 ) flow categories and least prevalent in the
annual maximum ( Q 100 ) category; and (ii) at all but the highest quantiles,
streamflow has increased across broad sections of the United States. The
decreases in streamflow was found only in parts of the Pacific Northwest and
Southeast. Systematic patterns were less apparent in the Q 100 flow.
Hydrologically, these results imply that the conterminous US is getting wetter.
Douglas et al. (2000) evaluated trends in the flood and low streamflows
of the US by using a regional average Kendall's S trend test at two spatial
scales and over two timeframes. The field significance was assessed following
a bootstrap methodology to account for the regional cross-correlation of
streamflows. The flood flow series was found trend-free at 5% level of
significance, but low streamflows showed upward trends with significant
temporal persistence. After removing serial correlation from the series,
significant trends in low flows were apparent but were less in numbers. The
ignorance of regional cross-correlation resulted in statistically significant trends
in all but two of the low flow analyses and in two-thirds of the flood flow
analyses. In addition, it was found that the cross-correlation of streamflow
records dramatically reduces the effective number of samples required for
trend assessment.
Radziejewski et al. (2000) normalized and de-seasonalized the raw series
of good quality streamflow data and subsequently transformed to the Fourier
spectral domain. Keeping the power spectrum preserved, the phase spectrum
was subjected to randomization. After transformation back to the temporal
domain, the data were contaminated with trends and step changes in a controlled
way. Then the detectability of nonstationarity by particular tests as a quasi-
continuous function of magnitude of the contaminating change was evaluated.
A method was devised to compare the tests' performance. The analysis of
detectability versus the magnitude of change provides a new insight into the
properties of the tests.
Zhang et al. (2001) studied the trends computed for past 30-50 years for
11 hydroclimatic variables obtained from the recently created Canadian
Reference Hydrometric Basin Network Database. It was found that the annual
mean streamflow has generally decreased during the periods, with significant
decreases in the southern part of the country. The monthly mean streamflow
has also decreased, with the greatest decrease occurring in August and
September. However, significant increases in streamflow were observed in
March and April. Furthermore, significant increases were identified in lower
percentiles of the daily streamflow frequency distribution over northern British