Conditioning Stochastic Rainfall Replicates on Remote Sensing Data

Abstract

Temporally and spatially variable rainfall replicates are frequently required in hydrologic applications of ensemble forecasting and data assimilation. Ensemble methods can be expected to work better when the rainfall replicates more closely resemble observed storms. In particular, the replicates should capture the intermittency and variability that are dominant features of rainfall events. In this paper, we present a new probabilistic procedure for generating realistic rainfall replicates that are constrained by (or conditioned on) remote sensing measurements. The procedure uses remotely sensed cloud top temperatures to identify potentially rainy regions. The cloud top temperatures are obtained from visible/infrared instruments in geostationary orbit. A multipoint geostatistical algorithm generates areas of nonzero rain (rain clusters) within each cloudy region. This algorithm relies on statistics derived from ground-based weather radar [National Operational Weather Radar (NOWRAD)] data. A truncated multiplicative cascade generates rain rates within each rain cluster. A computational experiment based on summer 2004 data from the Central U.S. indicates that the rainfall replicates simulated by the procedure are visually and statistically similar to individual NOWRAD images and to a large ensemble of NOWRAD images collected throughout the summer simulation period.