Performance characteristics of a Kalman filter-based bias correction technique

eCast^™ is a real-time product that succinctly presents point forecasts and estimated uncertainty based on a variety of NWP model output, including ensemble model data from the European Center for Medium-Range Weather Forecasts (ECMWF) and National Centers for Environmental Prediction (NCEP). These short- and medium-range forecasts for cities in the US and Europe are used by traders, brokers, hedge funds and re-insurers. AER performs in-house statistical correction to temperature forecasts based, in part, on the Kalman Filter (KF). The KF is a linear, adaptive, recursive, and optimal statistical technique that can bias correct and downscale the estimate from the current raw grid-point NWP forecast. The updated forecast is a balance between the current forecast state and recent observations in a manner that is determined by specifics of the KF implementation and a set of tuning parameters.

In this study, we will present performance characteristics of the eCast^™ implementation of the KF to improve one- to fifteen-day forecasts of daily maximum surface temperature at twenty-four Chicago Mercantile Exchange (CME) sites. Forecast performance over one year of the ECMWF ensemble mean forecasts with, and without, the KF will be stratified by a set of anticipated predictors of performance, such as wind direction and cloud cover. Areas for future improvements to the KF technique will be discussed.