A stochastic representation of random error associated with parametrized ph
ysical processes ('stochastic physics') is described, and its impact in the
European Centre for Medium-Range Weather Forecasts Ensemble Prediction Sys
tem (ECMWF EPS) is discussed. Model random errors associated with physical
parametrizations are simulated by multiplying the total parametrized tenden
cies by a random number sampled from a uniform distribution between 0.5 and
1.5. A number of diagnostics are described and a choice of parameters is m
ade. It is shown how the scheme increases the spread of the ensemble, and i
mproves the skill of the probabilistic prediction of weather parameters suc
h as precipitation. A choice of stochastic parameters is made for operation
al implementation. The scheme was implemented successfully in the operation
al ECMWF EPS on 21 October 1998.