MetGIS-Home

Loading ...

Forecast Quality

Temperature


For forecast ranges of 36 hours, the mean absolute forecast error of MetGIS predictions is typically in the order of 1 to 1.5 degrees centigrade. Generally about 70 % of the temperature forecasts errors are less than 2°C and nearly 90 % are less than 3°C, but these numbers vary depending on time of the day, season and location. In mountainous regions it is relatively easy to forecast afternoon temperatures in summit regions, with 85 % of the forecast errors below 2°C, whereas predictions of night time temperatures for deep alpine valleys are rather prone to errors. This is because numerical models sometimes don't adequately simulate shallow temperature inversions.

The displayed table shows temperature verification parameters calculated for the winter season 2011/2012 for the observation site Schmittenhoehe (1953 m), a typical alpine station near Zell am See (Austria). For the construction of the given table more than 300 forecasts were checked against observations. Even five-day forecasts are still of quite reasonable quality, although, compared to two-day-forecasts, the mean absolute error increases from 1.15 to 2.32 °C and the correlation coefficient drops from 0.98 to 0.90.
Forecast range [days] 1 2 3 4 5
Forecast error < 1°C [in %]: 60.9 56.1 43.2 36.7 28.6
Forecast error < 2°C [in %]: 86.2 82.4 75.9 60.1 53.5
Forecast error < 3°C [in %]: 95.8 95.8 91.2 79.9 72.3
Mean Absolute Error [°C]: 1.03 1.15 1.36 1.82 2.32
Correlation Coefficient: 0.98 0.98 0.97 0.95 0.90
Verification of MetGIS air temperature forecasts for Schmittenhöhe (1953 m, Austria), for the winter season 2011/2012.