Hourly Rainfall Synthesis for a Network

Generation of synthetic hourly rainfall data for a network is investigated because existing monthly streamflow models are inadequate and because watershed models are available to accurately predict streamflow given the hourly rainfall data in the watershed. The storm model was based on a multivariate Guassian Markov process. A fractional power transformation normalized the rainfall probability distributions. Estimation of parameters was carried out by using a combination of nonlinear least squares and probability plotting. The model reproduced the characteristics of hourly rainfall data with reasonable accuracy. However, some deficiencies in covariance structure and seasonal distribution occurred. The distribution of peak flows and annual volumes for Dry Creek near Cloverdale, CA, derived from synthetic rainfall data compared favorably with the same values derived from historic rainfall data.