About the origin of misforecasts

Author:D.Thaler Created:May 2013 Last changed:2017-06-01

How a weather forecast develops

Working in an operational weather service and regularly performing weather forecasts is always a sort of a small research project. Concerning to Sir Karl R. Popper scientific theories or hypothesis can be tested by four ways:

1. Internal test: Is the theory self-consistent without internal contradictions?
2. Scientific substantiality: Is the theory non-tautological and can it principally be falsified?
3. Comparison with other well established theories: Are there contradictions?
4. Empirical test: How does the theory and its logical conclusions comply with observations?

Other aspects of the evolutionary epistemology read:

1. All knowledge is provisional
2. There is no authoritative source of knowledge
3. Hypothesis and theories cannot be verified but only falsified

Applied to the problem of weather forecasting a rough flow diagram can be developed that shows progress of knowledge during the weather forecasting process:

The circle of progress of knowledge must be more or less arbitrarily interrupted because of a time out or - whatever is earlier - presumptively sufficient quality.

Sources of forecast errors

Forecast business is heavily relying on numerical modelling that is dependent on observation and analysis quality. Therefore starting with analysis and ending with clients decision for weather depending actions there is a lot of possibilities of errors.

1. Observation and analysis errors

   Numerical analysis is based on observations. Observations are erroneous and not always representative for the numerical model in use. Numerical forecast centers spent a lot of efforts to overcome these problems. In spite of all the efforts the numerical analysis has a degree of uncertainty that causes forecast errors.

2. Numerical model errors

   Atmospheric numerical models are not a perfect representation of the earth-ocean-atmosphere system neither in formulation nor in numerical solution of the underlying equations. Necessarily this means forecast errors.

3. Forecaster interpretation errors

   The forecaster is a human forecaster or recently also a more or less elaborated computer program that interprets intelligently direct model output (DMO) by model output statistics (MOS) or related methods. Both the human and the program can do mistakes in interpretation.

4. Forecaster formulation errors

   The human and as well the computer forecaster have to translate the forecast in a final step into the “language” of the client, either “language” in a strict sense or in form of icons and tokens. This process is necessarily fuzzy and therefore generally erroneous.

5. Client perception errors

   After all the forecast has to be read, heard or seen and understood by the target client (human or program). This can be a problem.

6. Client decision errors One of the main reasons for weather forecasting is decision making. For a lot of reasons despite of a correct forecast the decision can be wrong.

Ensemble or probability forecast systems

Due to analysis and model errors major numerical weather forecast centers perform a so called ensemble forecasts. The operational analysis fields are changed in a smart way and the numerical model is started with numerous - say a couple of dozens - initial fields. The more or less diverging solutions are considered to be a measure for the predictability of the atmosphere and the quality of the deterministic forecast. Stochastical variables (in multiple dimensions) are necessary to express the uncertainty of the forecast.

This simply means another measure to mitigate numerical forecast errors but certainly cannot avoid them completely. So the above flow diagram still remains valid.

---

Comments to:info at foehnwall dot at