"ecPoint"—A New Post-Processing Technique with Multiple Applications

ECMWF has developed a new software suite to deliver post-processed forecast for points, using gridded global model data as input, and using an innovative non-local worldwide calibration mechanism that needs only 1 year of 36h lead re-forecasts for a single Control run. The concept is called ecPoint. Calibration creates non-parametric site-independent distributions, or mapping functions, that describe the relationship between predictor (the gridbox forecast) and predictand (the point measurement) for a range of different gridbox weather types (for example large scale or convective precipitation). For forecasting, the mapping functions are used to post-process ensemble output by first identifying the gridbox weather types in each ENS member at each lead time. To date the concept has mainly been applied to rainfall totals (“ecPoint-rainfall”), with verification showing notable forecast improvements for small through to very large point totals. The first part of the presentation will give a brief overview of the technique and show some verification results.

Next the many potential applications of this work will be discussed. For rainfall these include better flash flood predictions, and naturally improved forecasts in general. Short range ecPoint output can also provide situation-dependant quality control for rainfall observations, and full point-wise climatologies if created from re-analyses. The mapping functions themselves constitute a form of conditional verification, wherein their integral is the average weather-dependant gridbox bias. Indeed for rainfall we have found considerable variations in this bias level, as will be illustrated. This could be pivotal for model development long-term and could be used for improved hydrological model input short term. There is also scope to improve offline-generated land-surface representations, that ECMWF creates and uses, by feeding with the bias corrected rainfall, or even, potentially, do this online as an integral part of a model run. One could also improve monthly rainfall forecasts, or even apply to climate change projections as a much cheaper and simpler alternative to nested high resolution runs. Other related/different applications can be envisaged when using ecPoint with other variables such as temperature or cloud cover.