EVALUATION OF RADIATION AND TEMPERATURE DATA GENERATORS IN THE AUSTRALIAN TROPICS AND SUBTROPICS USING CROP SIMULATION-MODELS

Long-term, historical weather data are needed to conduct crop simulati on analyses. However, the network of weather recording stations which collect all necessary daily weather data (commonly rainfall, solar rad iation, maximum and minimum temperature) for such analyses is sparse. Frequently only rainfall is recorded. Thus, weather data generation te chniques are required for three situations: (i) where only rainfall da ta are available, (ii) where both rainfall and temperature data are av ailable, but radiation is missing, and (iii) where records are otherwi se complete, but techniques are required to fill short periods of miss ing data. Three weather generation techniques are compared, termed her e (i) Bristow and Campbell's method, (ii) TAMSIM and (iii) WGEN. Metho ds (ii) and (iii) were used to generate temperature and radiation data to accompany recorded rainfall records, and methods (i)-(iii) to gene rate a solar radiation record to accompany recorded temperature and ra infall records. Data from four stations in tropical and subtropical Au stralia with long-term complete weather records were used to compare a ctual with generated data sets. Results were evaluated firstly by comp aring the cumulative distribution function (CDF) of generated and actu al values, and secondly by comparing CDFs calculated from the output o f three crop simulation models used with the generated and actual data sets. Generally the distributions of radiation and temperature differ ed significantly. However, when the weather data sets were used by sim ulation models to estimate biomass, only 10 of the 50 CDFs differed si gnificantly. When both temperature and radiation were generated, 30% o f CDFs from TAMSIM and 20% of WGEN differed significantly. When only r adiation was generated, 40% of CDFs generated by the Bristow and Campb ell's method, 10% of WGEN and none of TAMSIM differed significantly. A ll methods simulated the more temperate sites with higher precision th an the wet, tropical site. Simulated yields showed a similar pattern. It was concluded that where both temperature and radiation data have t o be generated, WGEN is appropriate because it contains a stochastic e lement and thus simulates catastrophic events such as frosts. Where on ly radiation generation is required, both WGEN and TAMSIM performed ad equately. Where temperature or radiation data sets are complete except for occasional missing days, TAMSIM was considered to be the most app ropriate. In cases were the objective is not to conduct long-term simu lation analyses, Bristow and Campbell's method appeared more appropria te because of its ability to better simulate the day to day variation in solar radiation.