We describe the generation of synthetic sequences of precipitation and maximum and minimum daily temperatures at two locations, in western and eastern Africa respectively. The sequences are generated at the monthly time scale and incorporate both explicitly modelled annual-to-decadal variability, based on the observational record, and long-range (i.e., climate change) trends, as inferred from an ensemble of global climate models. Annual-to-decadal variability is modelled as a first-order vector autoregressive (VAR) process, and the simulations are temporally downscaled to monthly time resolution using a nonparametric resampling scheme. The modelled sequences reproduce well the observed covariances as well as serial autocorrelation in individual variables. The simulations are intended to drive agricultural or other applications models to investigate responses to a range of plausible trends, on which are superimposed decade-scale climate fluctuations whose likelihood of occurrence can be estimated.