Evaluation of CLIGEN for storm generation on the semiarid Loess Plateau in China

Abstract

The Stochastic generation of storm patterns is often necessary for driving process-based hydrological and ecological models. CLIGEN is the only weather generator being able to generate internal storm patterns. Its goodness needs to be evaluated for its proper application. This paper aims to find the advantages and limitations of CLIGEN on semiarid areas and provide references for custom-built weather generators for the Loess Plateau. The daily rainfall time series (1957–2002) and breakpoint rainfall data (more than 20 years) on six stations on the Loess Plateau were used to estimate input parameters for CLIGEN and to compare with CLIGEN-generated 50 years of storm data. Precipitation occurrence (wet day and dry day sequence) is well-simulated without significant difference across months and sites. Errors of monthly average number of wet days range from − 0.67 to 1.08 days, standard deviations range from − 1.19 to 0.76 days, and the distributions of continuous number of wet and dry days on the semiarid Loess Plateau are adequately simulated. Daily rainfall amount is not simulated as well as precipitation occurrence. The relative errors of average daily rainfall range from − 12.93% to 8.64% and those of standard deviations range from − 21.35% to 27.46%. During the rain seasons (May–September), among 30 month–location combinations, all the Mann–Whitney tests for the means passed, 47% for squared ranks tests rejected the null hypothesis of equality of standard deviations, and 73% for K–S test suggested that the generated and measured distributions of daily rainfall were different at 0.01 level of significance Three variables to describe internal storm patterns in CLIGEN are storm duration, relative peak intensity, and time to peak. Storm duration was not well-reproduced because none of squared ranks tests and K–S test passed at the significance level of 0.01. The frequency of short duration storms (< 300 min) was over-predicted while frequency of long duration storms (400–1200 min) was significantly under-predicted. The distribution of maximum 5 min rainfall intensity (ip 5 ) was well-simulated for four sites out of the six because all tests passed. However, generated maximum ip 5 for all six sites are around 190 mm/h, which are much larger than the measured (70 to 150 mm/h). Ip 30 is simulated better than ip 5 , suggesting that CLIGEN can reliably generate rainfall erosivity. Time to peak was well-simulated because all the tests passed with P values significantly greater than the significance level of P = 0.01. Improvement for CLIGEN has to be made in terms of the daily rainfall simulation in rainfall-concentrated seasons and storm pattern generation in order to generate reliable rainfall time series on the Loess plateau.