Effect of Short and Long Term Memory on Trend Significancy of Mean Annual Flow by Mann-Kendall Test

Climate variability and change is threatening water resources around the world. One hundred and fourteen (114) stations from Reference Hydrometric Basin Network (RHBN) around Canada with atleast 30 years continuous data (up to 2011) were selected to study the trend in mean annual runoff fordifferent periods of 30 to 100 years in step 10 years by non-parametric Mann-Kendall test. Effect of short term persistent (STP) and long term persistent (LTP) on this test were made through lag 1 serial correlation (r1 ) and Hurst exponent (H), respectively. r1 for about one third of the total casesconsidered was negative. H, based on equivalent Normal deviate (eNv), was slightly right-skewed with minimum and maximum values of 0.20 and 0.87, respectively. About half of the data sets were anti-persistent (H<0.5). No regional pattern was found for r1 and H. Based on five stations with around100 years data it was shown that r1 and H are unstable for record length, roughly, up to 50 years. r1 and H were highly correlated (r=0.86). H from eNd were smaller than H from original data by around 10% with high correlation (r=0.87). Under classic Mann-Kendall trend test, different time periods ofdifferent stations showed different trend direction and significancy, which admits for abrupt change intrend direction and significancy for different time periods. On overall, more than 60% of cases there were no significant trends (i.e. p-value>0.1). The number of positive and negative trend, were nearly the same, though fluctuating for different time spans. p-value after pre-whitening was highly correlatedwith those of before pre-whitening, for both negative and positive trends. There were about 16% of cases that pre-whitening decreased the p-values of the Mann-Kendall trend test, where nearly all of them were negatively trended. The effect of LTP on Mann-Kendall trend test was minor due toinconsistency of originally significant trend case and significant H of greater than 0.5. For recent 30 years length of record (1982-2011), British Columbia is experiencing positive trend in the west and negative trend in the east. Most parts of the New Brunswick are experiencing the positive trend, whilenegative trend is due to Southeast of Ontario. For the longer duration of 40 years, trend statistics and geographical pattern were changed. While the significant trends are decreased, more significant negative trends are governed over New Brunswick. There is no positive trend in British Colombia inthe past 50 years (1962-2011) while there are both negative and positive trends in New Brunswick which negative trends are switched to positive trends in south east of Ontario. For long duration of > 70 years, there are only positive trends in Southeast of Canada (South New Brunswick and South East of Ontario) while central and east of Canada have experienced a negative trend.

Keywords:

Climate Change , Climate Variability , Hurst Exponent , Hydrology , Lag 1 Serial Correlation , Long-memory Process