Many studies on precipitation trend have been performed in recent years (e.g., Brunetti et al., 2006; Rodrigo et al., 2007; Turkesh, 1996), but spatial analysis of the sign of trends is rarely carried out (Suppiah and Hennessy, 1998; De Luis et al., 2000). In the present paper we analyze the observational precipitation data from some oldest meteorological stations to investigate the possible systemic trend in precipitation across Iran. We then search for the spatial coherence of the precipitation change.
Daily precipitation data from 38 stations were selected from the Iranian Meteorological Organization (IRIMO) files for a 42-year period (1960-2001). The linear regression model is used to determine the signs and magnitudes of annual precipitation trends. The non-parametric Kendall-taw statistic (Sneyers, 1990) is applied to evaluate the statistical significance of trends.
To assess the spatial distribution of precipitation trends, we applied a statistical analysis similar to De Luis et al. (2000) which is based on the Cramer-von Misses non-parametric test (Zimmerman, 1993; Syrjala, 1996). The test is applied to three possible signs of trend: positive, negative and zero. The null hypothesis is that, there are no differences in the spatial distribution of signs of trend, implying that the signs are distributed randomly in the study area and only the local factors are responsible for the spatial variability of trends. If differences are significant, we may conclude that precipitation has evolved differently in different areas.
To compare distribution of signs, each station was categorized by the presence of each of the three trend signs (+, -, or 0). The Cramer-von Misses test is insensitive to differences in the total number of trends in the study area, but sensitive to differences of respective number of stations with a given sign of trend. Therefore, each pair of signs (+ versus -, + versus 0, - versus 0) was first normalized to eliminate the effect of differing sizes in population. The test statistic is based on the differences between two cumulative distribution functions that are calculated along spatial gradients. The spatial gradients used in the present study are those along the mean annual precipitation, altitude (one dimensional), longitude (east-west), latitude (north-south) and those parallel and perpendicular to the dividing heights of the Alborz and Zagros ranges (two dimensional).
The test determines whether the stations with significant local trends have occurred by chance. A modified version of the Cramer-von Misses non-parametric test is applied (De Luis et al., 2000) to test for spatial distribution of precipitation trends. The test statistic is calculated as the squared difference between the two cumulative distribution functions summed over all sampling locations. For application of the Cramer-von Misses test, data must be selected at random in space, but because locations of the stations are fixed, the sign of trend in each station is assumed as the random variable. The null hypothesis is that there are no differences in the spatial distribution of trends. Of the 38 stations studied, 18 showed positive trend, 16 showed negative trend and 4 showed zero trend on a yearly basis.
The results of the application of the Cramer-von Misses test along annual precipitation gradient indicate that stations with negative sign show a different distribution when compared with those with positive sign (99% significant). In contrast positive and zero stations, as well as negative and zero stations seemed to overlap along the precipitation gradient. It seems that semi-arid and humid regions show a significant decrease in annual precipitation, while the arid region shows a positive trend (1960-2001).
The results of the 2-dimensional test application show that on the east-westerly gradient positive trends are distributed differently from negative ones (95% significant). This means that the stations with these signs are not randomly distributed along the east-west direction. On the other hand the negative versus zero trends along the north-south gradient show a significant (90%) difference, indicating that the distribution of these signs is not random is space. Non-significant differences between other pairs along these gradients indicate that these pairs of trend signs are distributed randomly and overlapping in space. The results of the test to trend signs along the gradients parallel and perpendicular to the two main mountain ranges (Zagros and Alborz) show no significant difference, implying that the signs are distributed randomly.