Submitted to: Applied Engineering in Agriculture
Publication Type: Peer reviewed journal
Publication Acceptance Date: 4/1/2002
Publication Date: 5/1/2003
Citation: GARBRECHT, J.D., ZHANG, X.J. GENERATING REPRESENTATIVE SEQUENCES OF DAILY PRECIPITATION FOR AGRICULTURAL SIMULATIONS. JOURNAL OF APPLIED ENGINEERING IN AGRICULTURE. 2003. v. 19(4). p. 423-429. Interpretive Summary: In research it is common to use computer programs to simulate soil water availability, crop growth, nutrient transport, etc. for a range of weather conditions. The daily weather for such applications is often generated by a computer and involves the use of random numbers. The random numbers must have certain properties if historical weather statistics are to be accurately reproduced. This study shows that sequences of random numbers do not always have the desired properties for the generation of weather that closely reflects a desired historical data set. A testing and selection procedure is proposed to ensure that only sequences of random numbers that have the desired properties of uniform distributions are selected for the generation of daily weather for a particular natural resource simulation. With this approach, the historical weather data are better approximated and the resulting analyses of soil water availability, crop growth and nutrient transport are also more representative of the historical weather conditions.
Technical Abstract: Uniform random numbers are often used in chain-dependent daily precipitation models to simulate the stochastic component of daily precipitation. This study shows that relatively short sequences of uniform random numbers, often involved in practical water resources and agricultural applications, are not necessarily uniformly distributed as assumed by many precipitation models. The magnitude of the increased variability introduced by the recurrence of short sequences of random numbers that do not conform to the uniform distribution is illustrated. An approach is proposed to test sequences of random numbers before their use and retain only those that are uniformly distributed and conform to the assumptions of the daily precipitation generation model. Generated daily precipitation with and without the testing of the random numbers shows that, for same simulation duration, the differences between the two can be relevant. The use of tested random numbers results in transitional probabilities and a precipitation distribution that, on average, are in closer agreement with those of the historical daily precipitation data. In addition, this increased reproducibility leads to shorter simulation duration and greater effectiveness in simulating subtle changes in precipitation such as associated with seasonal precipitation forecasts or decade-scale precipitation variations.