IACR-Long Ashton Research Station, University of Bristol, Long Ashton, Bristol B541 9AF, UK

Daily weather data are required for many impact applications and decision support systems. Substitution of daily weather by 'mean' weather can cause large errors in impact assessments because of the non-linear response of most impact models to their input parameters. A stochastic weather generator is a numerical model which produces synthetic daily time series of a suite of climate variables, such as precipitation, temperature and solar radiation, with certain statistical properties. There are several reasons for the development of stochastic weather generators and for the use of synthetic weather data instead of the use of observed data in decision support systems (DSS).

Weather generators are able to generate weather time series long enough to be used in an assessment of risk in agricultural applications. Observed daily weather is one of the major inputs into mathematical models in agriculture and land use, but the length of the time series is often insufficient to allow a good estimation of the probability of extreme events. Moreover, observed time series represent only one 'realisation' of the climate, whereas a weather generator can simulate many 'realisations', thus providing a wider range of feasible situations. Weather generators are becoming a standard component of DSS. There is a danger, however, that generators will be used 'as supplied', i.e. without sufficient validation being carried out for the sites at which they are applied. Weather generators differ in their structure and complexity, resulting in differences in their performance. The results of the recent comparison of two weather generators, WGEN and LARS-WG, for a diverse range of climates in Europe, North America and Asia will be discussed [1].

Impact assessments are now often made on high-resolution grids or at multiple sites across a region where observed weather records are not available. Several interpolation techniques, such as kriging, thin plate smoothing splines or precipitation-eIevation regression on independent slopes model, have been developed to interpolate the monthly means of climate variables, with the emphasis on the interpolation of rainfall in mountainous areas. However, many impact models require daily weather data and so a different approach is required. Rather than interpolating the climate variables directly, the parameters and distributions of a weather generator for each of the observed sites can be interpolated, with the resulting parameters being used by the weather generator to produce synthetic daily date for the unobserved locations. The methodology for the spatial interpolation of the LARS-WG stochastic weather generator, which enables the production of synthetic daily weather for any location in Great Britain, will be presented.

Weather generators can also be used as a flexible tool in constructing 'realistic' climatic scenarios, which preserve statistical properties and correlations between weather variables, for the assessment of different management strategies. Particular interest has recently arisen from climate change studies [2]. The output from Global Climate Models (GCMs), which are the main tools for predicting the evolution of climate on Earth, cannot be used directly at a site with impact models because of their very coarse spatial resolution. A weather generator can serve as a computationally inexpensive tool to produce site-specific climate change scenarios at the daily time-step. The changes in both climatic means and climate variability predicted by the GCM experiments can be applied to the parameters derived by the weather generator for the current climate at the site in question; new parameters can be used to generate future weather. The effect of the incorporation of changes in variability into climate change scenarios on wheat production and associated risk in Europe will be discussed.

1. Semenov MA, Brooks RJ, Barrow EM, Richardson CW, 1998. Climate Research (in press).
2. Semenov MA, Barrow EM, 1997. Climatic Change, 35, 397-414.