Stochastic generation of annual, monthly and daily climate data: A review

摘要

The generation ofrainfall and other climate data needs a range of models depending on the timeand spatial scales involved. Most of the models usedpreviously do not take into account year to year variations in the modelparameters. Long periods of wet and dry years were observedin the past but were not taken into account. Recently, Thyer and Kuczera (1999)developed a hidden state Markov model to account for thewet and dry spells explicitly in annual rainfall. This review looks firstly attraditional time series models and then at the more complex modelswhich take account of the pseudo-cycles in the data. Monthly rainfall data havebeen generated successfully by using the method of fragments.The main criticism of this approach is the repetitions of the same yearlypattern when only a limited number of years of historical dataare available. This deficiency has been overcome by using synthetic fragmentsbut this brings an additional problem of generating the rightnumber of months with zero rainfall. Disaggregation schemes are effective inobtaining monthly data but the main problem is the large numberof parameters to be estimated when dealing with many sites. Severalsimplifications have been proposed to overcome this problem. Modelsfor generating daily rainfall are well developed. The transition probabilitymatrix method preserves most of the characteristics of daily,monthly and annual characteristics and is shown to be the best performing model.The two-part model has been shown by many researchers toperform well across a range of climates at the daily level but has not beentested adequately at monthly or annual levels. A shortcomingof the existing models is the consistent underestimation of the variances of thesimulated monthly and annual totals. As an alternative,conditioning model parameters on monthly amounts or perturbing the modelparameters with the Southern Oscillation Index (SOI)result in better agreement between the variance of the simulated and observedannual rainfall and these approaches should be investigated further.As climate data are less variable than rainfall, but are correlated amongthemselves and with rainfall, multisite-type models have been usedsuccessfully to generate annual data. The monthly climate data can be obtainedby disaggregating these annual data. On a daily time step ata site, climate data have been generated using a multisite type modelconditional on the state of the present and previous days. The generationof daily climate data at a number of sites remains a challenging problem. Ifdaily rainfall can be modelled successfully by a censoredpower of normal distribution then the model can be extended easily to generatedaily climate data at several sites simultaneously. Mostof the early work on the impacts of climate change used historical data adjustedfor the climate change. In recent studies, stochastic dailyweather generation models are used to compute climate data by adjusting theparameters appropriately for the future climates assumed.