The performance of first-moment and full-distribution bias-correction methods of monthly temperature distributions for seasonal prediction is analyzed by comparing two approaches: the standard all-in-data procedure and the 6-hourly stratification of data. Five models are applied to remove the systematic errors of the CFSv2 forecasts of temperature for the rainy season in the Ethiopian Blue Nile River basin domain. Using deterministic evaluation measures, it is found that the stratification marginally increases the forecast skill especially in regions where the data distribution of temperature is prominently multimodal. The improvement may be attributed to a split of the mixed distribution into a set of unimodal distributions. A necessary condition for this splitting into unimodal distributions is that the amplitude of the diurnal cycle be larger than the interannual variability in the sample. The maximum improvement of stratification is achieved by the first-moment correction model.
This paper evaluates bias-correction methods of monthly forecast distributions of temperature to improve seasonal forecast skill. It is found that marginal skill is gained when bias correction of the diurnal cycle is performed. This paper contributes to the discussion on the value of subdaily model output data.
© 2022 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).