Daily precipitation and mean temperature data from 17 National Meteorological Observatory stations (Fig. 1), with continuous data from 1960 to 2012 in or around the HRB were used for this study. These stations, which possess high quality data, are maintained and released according to the standards set by the National Meteorological Administration of China (http://cdc.cma.gov.cn/home.do). Monthly observed streamflow data of 16 hydrological stations (Table 1) were collected from Hydrological Bureau of Gansu Province and the Inner Mongolia Autonomous Region, which are also of high quality.

Streamflow series of the upper and middle HRB (the first 13 stations) are used to analyze streamflow Tofacitinib variations, and MG-132 research buy that of the last three stations are only used to detect the inflow changes to the downstream for lacking of long term records. A few missing data were filled based on nearby stations and a correlation analysis

between individual stations. This study aims to detect the presence of trends and abrupt changes of the streamflow time series over the HRB. To analyze driving factors of the streamflow change, trends and abrupt changes in the data of the meteorological series were also tested. Throughout this study, two types of statistical analysis methodology were used: a trend test (Mann–Kendall test) and a change-point test (Pettitt test). A trend test is performed on the hydrological and meteorological data to analyze gradual changes or tendencies. The Mann–Kendall test (Mann, 1945 and Kendall, 1975) is one of the most popular trend detection method used in the world. It is a non-parametric test which can cope with missing values and values below a detection limit. For an independently distributed time

series X(n), null hypothesis (H0) of the Mann–Kendall (MK) test is no trend. In the MK test, the sign (sgn) is used to count the difference between two values (xi and xj) from X(n) Oxalosuccinic acid which is defined as: equation(1) sgn(xj−xi)=1ifxj>xi0ifxj=xi−1ifxj00ifS=0(S+1)/V(S)ifS<0 H0 will be rejected at the significance level of α when the absolute value of Z is bigger than z(1−(α/2)). Serial correlations can affect the results of MK test (Yue et al., 2002 and Hamed, 2009), therefore correlations of the series were computed firstly before a trend test. When only a lag-1 autocorrelation was found to be significant, the MK test of Yue et al. (2002) was used.