Author(s): Shouhong Wu
Linked Author(s):
Keywords: No Keywords
Abstract: This paper presents improved algorithm for the generation of weekly or monthly statistically dependent hydrological time series based on historical data. Suppose we need to generate a monthly time series of N years. The data generation algorithm includes the following three steps: (1) Generation of N random data for each month based on statistical distribution parameters of the historical data for the month. (2) The generated data for the first month is fixed and permutation of the generated data for the following months is performed sequentially within each month. After the permutation, the data of each month are correlated with the data of the previous months of the same year to mimic the correlation observed in the historical data. (3) Re Hordering of data years so that cross year end monthly data correlations and annual data auto Hcorrelations are in ways similar to those for the historical data. The proposed improvements include: (1) A new linear scheme that makes permutation converge much faster than the scheme proposed by Ilich (2009), and also avoids the disadvantage of dividing by zero. (2) Virtual row swaps and data sorting that do not involve actual row swaps and reduce computation time significantly.
Year: 2009