Analysis of Streamflow Trend in the Susquehanna River Basin, USA - Hydrologic Time Series Analysis: Theory and Practice - page 185

Average maximum July temperatures range from 30 o C in the valleys to 28 o C
in the mountains.
To investigate the potential streamflow changes and their character,
streamflow measurements at unregulated streams are warranted as they are
minimally impacted by flow regulations or other human activities. Eleven
long-term gauges distributed across the SRB were identified by SRBC staff as
unregulated streams which are suitable for the purpose of this study (Fig. 9.1).
Salient characteristics of the identified gauge stations are provided in Table
9.1. Most of these streams were also listed as such in the USGS Hydro-
Climatic Data Network (HCDN) (Slack and Landwehr, 1992). In the HCDN
datasets, about 1,600 unregulated gauged watersheds distributed throughout
the United States were identified as being potentially suitable for study of
potential impact of climate change on hydrology. Hydrology literature
extensively employed the HCDN dataset for streamflow trend analyses in the
United States (McCabe and Wolock, 2002; Vogel et al., 1997; Kroll et al.,
2004; Zhang and Kroll 2007, a, b). Daily streamflow data from 1940 to 2006
were downloaded from the USGS website (http://waterdata.usgs.gov/nwis/
sw). Annual daily minimum, median and maximum flows are used for the
detection of streamflow changes. The monthly average baseflow, storm runoff,
total streamflow, and monthly minimum total streamflow are used for detection
of changes as well.
9.3 Methodology
When the hydrologic time series records being investigated are in two non-
overlapping periods with a lengthy gap between them or a known event which
is expected to change the hydrologic time series has occurred at a time during
the record, the step change in hydrologic record can be tested by dividing the
record into two periods and using techniques such as the rank-sum test, two-
sample t -test, and analysis of covariance (Helsel and Hirsch, 1992). These
techniques require a highly specific situation or prior hypothesis of a time of
change. Otherwise, monotonic trend tests such as the Mann-Kendall test are
appropriate. In hydrologic practice, hydrologists and water resources managers
often do not have the prior information or a known event to identify the time.
Thus, many studies of changes in hydrologic time series have focussed on the
monotonic changes over time, which can be tested with the Mann-Kendall
test. However, the Mann-Kendall test with a specified period of record is not
capable of showing the trend pattern, i.e., whether the change is gradual or
abrupt (McCabe and Wolock, 2002).
Zhang et al. (2010a) suggested apply multiple Mann-Kendall tests with
varying beginning and ending times to detect trend patterns in hydrologic
time series. The step-by-step procedure is presented here.
Step 1: Rank the data set according to time T .
Step 2: Set the beginning time T b = 1.