Verification for forecast users

Users of forecasts have multiple reasons to verify the forecasts they receive from forecast providers. Some of these reasons are outlined on this page.

Verification as a first estimate of residual uncertainty

Verification of single valued, deterministic forecasts provides a first estimate of the residual uncertainty. While forecasts reduce uncertainty about the future, this uncertainty can never be fully eliminated. A comparison of forecasts with their corresponding observations reveals the nature of this uncertainty. A simple scatter plot can already be very informative, and so can an assessment of hits and false alarms. If the extent of the verified record is sufficiently large, and if one may reasonable assume that ‘observed’ uncertainty will apply in the future also then verification can serve as an input to post-processing. Past, observed uncertainty is then superimposed on future forecast to produce an estimate of predictive uncertainty.

Towards forecast informed decision making

An analysis of the number of hits, missed events and false alarms may be used to optimize a decision rule. Also, verification may be used to assess whether or not uncertainty estimates may be viewed as probabilistic forecasts.

Identification of the strengths and weaknesses within a chained forecasting system

Often, forecasts are used in a chained forecasting system where a forecast variable is used as input to a model that produces a forecast of another variable. In these cases, verification can serve to identify strengths and weaknesses of the elements in chained forecasting systems. Weaknesses can then be actively targeted with a view to improvement of these elements.

“My forecast provider has already verified the products I am using.”

This is often true. However, your forecast provider is unlikely to have verified at the spatial and temporal scale on which you are using their forecasts. For example, in flood forecasting, gridded precipitation forecasts are often first transformed to the river basin scale and time-aggregated (or disaggregated) before being fed to a hydrological model. The same is true for observations. In order to know the quality of those forecasts, one would need to verify at those spatial (here: river basin) and temporal scales. In addition, you may be interested in specific events (“exceedence of 14.5C”) that may not be addressed by the verification carried out by your forecast provider.

Also, independent verification may serve as a second opinion of the quality of the forecasts considered. It may even serve to assess whether or not forecast providers have achieved quality criteria or benchmarks.