Use of split sample approach in evaluation of stochastic daily rainfall generation models

Abstract

This paper reviews the use of split sample approach to test the ability of stochastic daily rainfall generation model to generate rainfall data for the future. The catchment adopted is Kangaroo Valley in New South Wales, Australia. Total data of 101 years long are divided into two sets: Earlier Period (80 years) and Later Period (the subsequent 21 years). The model adopted is the 8x8 Transition Probability Matrices Model, using two variations the Shifted Exponential Distribution and Box-Cox Power Transformation for the eighth class. Model parameters including transition probability matrices, exponential distribution parameters and Box-Cox Power Distribution parameters were computed using the data from the Earlier Period. The comparisons of statistical measures were made against the Later Period. Comparisons were made using daily statistical measures, daily extremes, monthly statistical measures, monthly extremes, annual statistical measures, annual extremes and serial correlation coefficients. The results shown that in general satisfactory statistical comparisons were made between the generated data based on Earlier Period against the statistics of the Later Period. In conclusion, the stochastic daily rainfall generation model can be used to generate synthetic data for planning and forecasting.