Stochastic Simulation of Daily Climatic Data for Agronomic Models¹

Abstract

Many agronomic models require the input of daily climatic data. Simulated climatic data may be used when long series of historic data are not available or convenient, or when future data are needed. A stochastic weather simulation model was developed and validated for a wide range of climates. The model produces possible daily sequences of precipitation amount, maximum and minimum air temperature, and total solar radiation at the earth's surface for the entire year. A first‐order, two‐state Markov chain is used to simulate the occurrence of wet and dry days. Probabilities are used to simulate the occurrence of trace precipitation amounts on wet days. A two‐parameter gamma distribution conditioned by the precipitation status on the previous day is used to generate greater than trace amounts. Two bivariate normal distributions conditioned by the precipitation status on the current day are used to simulate current temperature deviations from long‐term average temperature curves. A two‐parameter gamma distribution simulates current solar radiation deviations from the calculated maximum clear day radiation on dry days. On wet days, the deviations are simulated with a two‐parameter beta distribution. The model was developed with data from Columbia, Mo. Model validation was done for Columbia, Albuquerque, N.M., Caribou, Maine, Medford, Ore., and Miami, Fla. Various statistical tests were done to detect significant differences in central tendency, dispersion, and distribution. Comparisons were made to the base period used for parameter estimation (ranged from 16 to 20 years) and also to the 80‐year period of record available at Columbia. The validation showed that the model produced climatic data which generally did not differ significantly from the base period at any of the locations. At Columbia, it was determined that the 17‐year base period was not long enough to adequately represent the 80 years of precipitation data.

A statistical comparison of the stochastic weather generators WGEN and SIMMETEO

“…Our results showed that this approach worked well for the climate conditions that were included in this study and were consistent with findings of others (Larsen & Pense 1982, Richardson & Wright 1984, Geng et al 1986, Wilks 1999, Sentelhas et al 2001, Hartkamp et al 2003. However, Schubert (1994), Wallis & Griffiths (1995), Semenov et al (1998), Hayhoe (2000, Puche & Silva (2001) Table 2.…”

“…Generated weather data can be used as input for hydrologic models for watershed planning, evaluation and design purposes (Richardson 1981) and for analysis of water resources (Hayhoe 2000). They can also be used in various types of agricultural man-agement models to assess the risks associated with different alternatives (Larsen & Pense 1982, Hayhoe 1998, O'Leary & Connor 1998. In a real-time mode, generated data can be used as future sequences in crop-simulation models for yield forecasting (Thornton et al 1997, Bannayan & Crout 1999, Georgiev & Hoogenboom 1999.…”

Evaluation of a Single-site Daily precipitation Generator in northwestern of Benin

“…In this study, the daily precipitation generator is a first-order two-state Markov chain (Larsen and Pense 1982, Roldom, J. and Woolhiser, DA 1982, Richardson 1985 and exponential and gamma distribution are used to generate the precipitation amount.…”

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

“…To reduce the number of parameters, Larsen and Pense (1981) fitted three parameter Sine curves to the mean daily maximum and minimum temperatures conditioned on the wet/dry state of the present day. The residuals of these two variables from their mean values for the two types of days were modelled by two bi-variate Normal distributions.…”