A new weather generator based on spectral properties of surface air temperatures

We present a new weather generator which simulates daily precipitation occurrence and amount, daily minimum air temperature, and diurnal temperature range. Precipitation occurrence is simulated with a two-state, first-order Markov process and wet-day precipitation amounts are randomly drawn from a mixed-exponential distribution. Minimum daily temperature and diurnal temperature range are simulated using spectral methods. Data generated at nine stations in the Southeastern USA are then compared to 1) observed station data, and 2) data generated by a variant of the commonly-used autoregressive weather generator. Precipitation simulation is found to be excellent for our study region. Minimum temperature and diurnal temperature range generated using the spectral weather generator are found to be superior to those generated by the autoregressive weather generator in terms of means and variances across multiple timescales as well as the number of freeze events, skewness of temperature distributions, and the relationship between daily minimum and maximum air temperatures. Both weather generators are found to slightly underestimate interannual variability in precipitation occurrence, minimum air temperature, and diurnal temperature range.