Stochastic generation of precipitation fraction at high resolution with a multiscale constraint from satellite observations

In this work, we propose a method to generate an ensemble of equiprobable fields of rain occurrence at high resolution (1°/16 and 30 min) using a satellite observational constraint. Satellite observations are used to constrain the spatio‐temporal variations of the precipitation fraction at various scales. Spatio‐temporal averages at scales coarser than 1° and 8 h are deterministically derived from the satellite observations. At finer scales, variations are partially stochastically generated by perturbation of wavelet coefficients obtained through a three‐dimensional discrete Haar wavelet orthogonal decomposition. The proposed method can be viewed either as stochastic weather generation or as stochastic downscaling with a multiscale observational constraint. The observational constraint used here is a high‐resolution precipitation index derived from infrared cloud top temperature. As a proof of concept, the method is used here to generate a 300‐member annual ensemble covering a 12,000 km 2 area in Burkina Faso in West Africa, with a parametrization derived from ground radar observations. The stochastically generated fields aim at reproducing the multiscale statistical properties of the true precipitation field (as observed by a ground radar). The ensemble mean is an optimal – in terms of mean squared error – estimation of the true precipitation fraction, with the uncertainty quantified by the ensemble dispersion.