Statistics in Hydrometeorological Forecasting – Contributions and Challenges (Invited)

Statistics are imbedded in almost all aspects of hydrometeorology, from the analysis of the “butterfly effect” to the forecasting and verification of severe local storms, to the creation of the climate “hockey stick”. In fact, many decisions we make daily are the result of a statistical blending of past experience and information gathered to form a belief about a following state of nature.

Numerical weather prediction models are formulated on the basis of analyses of past results and experiments to determine a good formulation for a particular desired result. Models characteristically thought of as producing single value forecasts can be run in an ensemble mode to produce probabilistic results. Ensembles of models produce a range of outcomes, but are characteristically poorly calibrated and need statistical processing.

While there have been many accomplishments in the use of statistics in hydrometeorology, there are major challenges. Simple decision models have been built and used, but decisions are usually based on a complex set of inputs; how to best use meteorological information with other types of information is an almost untapped field. Postprocessing techniques that are easy to implement and still provide quality guidance forecasts are still lacking in many ways, especially for high impact weather situations–situations which likely involve several meteorological variables.

Another challenge is developing a hydrometeorological work force that is well versed in statistics and statistical methods. Statistical and probabilistic theory and practice need to be much more heavily emphasized than in the past. Almost every aspect of forecasting involves statistics, yet the U.S. Government qualifications for meteorologist do not mandate a course in statistics. The AMS guidelines for statistics are stronger than previously, but still lack sufficient emphasis. There is a serous deficiency in the education of future scientists.

Postprocessing of model output is necessary to get the weather down to the street. This has become an important activity of weather services worldwide. In some cases, the results of postprocessing become the actual forecasts disseminated to users. In other cases, postprocessing furnishes guidance to forecasters who produce the final product. Quite a number of techniques for postprocessing exist, and there are many variations of each in their application. Of major concern is how to postprocess the members of ensemble forecast systems to produce well calibrated probability distributions. The trend is to provide grids of forecasts, and in so doing objective analysis of observations and/or forecasts is necessary–such processes are a statistical melding of data, and the method employed depends on the specific objectives for the resulting grid.

This paper will discuss some of the past uses and trends in statistics in hydrometeorology and a forecast for the future.