Error characteristics of temperature forecast in Finland for the period 1979–2011 in relation to various weather patterns

The investigation of weather forecast errors has focussed mainly on assessing the physical causes in numerical weather forecasting, such as errors in the initial state and model deficiencies. The framework and context of this paper are different. A 30 year time series (1979-2011) of 1 and 2 day human forecasts of temperature in Finland were used to investigate the statistical and climatological characteristics of maximum and minimum temperature forecast errors. Error dependence on different flow types and weather regimes was also analysed. The error statistics indicate that the forecast root-mean-square error has halved, and the forecast hit rate within a 2.5 degrees C error threshold has increased from 70% to 85-90% during the 30 year period. Large forecast errors of >5 degrees C are presently very rare, and they are mostly seen as positive biases in extreme cold temperatures and inversions. Otherwise, the forecasts nowadays are mostly unbiased with only small negative biases associated with Foehn events. Forecast error extremes were found during winter, with a secondary peak in spring and under high pressure periods in non-westerly and especially northeasterly airflows, dry air mass and relatively weak winds. The summertime maximum temperature errors are typically associated with opposite weather situations that occur most typically with cyclonic circulations, moist air masses and strong winds.