Forecasting Cloud Index with an Advection Model and Data Assimilation

We present a system for forecasting cloud index on the scale of a city with a spatial resolution of approximately 1 km and forecast horizons up to two hours. On these temporal and spatial scales, two dimensional advection models can be useful in forecasting cloud index. These simplified cloud motion models also have the benefit of allowing for easier data assimilation and larger ensembles.

In our ensemble forecast system, we assimilate data every 15 minutes in a Bayesian framework. The assimilated data includes wind fields derived from numerical weather prediction models (NWP) and cloud motion vectors derived from GOES 15 visible imagery and a sparse optical flow technique. The LETKF is used when assimilating wind fields from NWP, and the EnKF is used when assimilating wind vectors derived from sparse optical flow. We perform a deterministic and a probabilistic analysis of the forecasting skill and we present comparisons to alternative methods.