Analysis of Streamflow Trend in the Susquehanna River Basin, USA - Hydrologic Time Series Analysis: Theory and Practice - page 180

The changes of means and the variability of streamflows are the major
factors that have contributed to the end of streamflow stationarity. To accurately
characterize streamflows, it is of interest to detect these changes. Trend analyses
are commonly used in literature to detect changes in streamflow time series.
Linear and nonlinear models are often employed to express the trend in a time
series in hydrology literature (Shahin et al., 1993). Student's t -test is commonly
used to detect linear trend. However, this method assumes that the time series
is normally distributed. For non-normal data, a nonparametric test such as the
Mann-Kendall test (Mann, 1945; Kendall, 1962) is preferred (Hirsch and
Slack, 1984; Helsel and Hirsch, 1992). The serial correlation in a time series
will impact the ability to evaluate the significance of the test (Kulkarni and
von Storch, 1995; von Storch, 1995; Yue et al., 2002b). Lettenmaier et al.
(1994) found strong upward trends in about half the 1009 investigated streams
in the continental United States for the months of November through April
using the streamflow records from 1948 through 1988. Lins and Slack (1999)
found similar results with lower magnitude streamflow quantiles.
Noticing that many studies did not consider the regional cross-correlation
of streamflow, Douglas et al. (2000) proposed a bootstrap approach to account
for it and detected upward trends in low flows in the Midwestern U.S. An
increase in high streamflows in the conterminous US has been reported by
Groisman et al. (2001). Zhu and Day (2005) reported downward trends in 47
streams across Pennsylvania for the 1971-2001 period. Kalra et al. (2008)
documented increased streamflow in the Mississippi and Missouri regions for
the period 1951-2002. Different trends in streamflows in varying months in
the Colorado River Basin have been documented by Miller and Piechota
(2008). Wu et al. (2008) found that there is no uniform trend in droughts in
Nebraska. The detection and characterization of trends should be studied in a
framework that recognizes and characterizes the dependence structure of
hydroclimatic records (Koutsoyiannis and Montanari, 2007). Bhutiyani et al.
(2008) reported changing streamflow patterns in the rivers of northwestern
Himalaya during the 20 th century. Milly et al. (2005) demonstrated an ensemble
of 12 climate models to simulate patterns of changes in global streamflows.
Changnon and Demissie (1996) examined streamflow changes in the Midwest
region of the United States and investigated the effects of land use and climate
fluctuations. Machiwal and Jha (2006) provided an excellent review of trend
analysis on hydrologic time series, together with the application of other time
series analysis techniques in hydrology and climatology.
Traditionally, a period of time is pre-specified and a trend test is conducted
using the data within the selected period. However, this approach cannot
demonstrate the pattern of change. To effectively and efficiently manage water
resources, water resources engineers and managers need to know not only if
there are trends but also if the trends are abrupt or gradual. The magnitude of
the change and length of the period during which the change occurred have
considerably different implications. For a gradual trend, the change occurs