Abstract It is common for forecasts like accumulated precipitation or average temperature over a period to be produced by different forecast systems, depending on the forecast lead time and period duration. For example, 24‐hr precipitation accumulations out to 7 days are typically produced by a different system than weekly precipitation accumulations out to 4 weeks. To provide a seamless forecast service over all periods and lead times, outputs from one system need to be post‐processed to be consistent with the other system for the overlapping prediction period, or outputs from both need to be blended together to form a new forecast. This article presents methods for post‐processing forecasts from an ensemble prediction system (EPS) so that its forecasts are consistent with statistically well‐defined forecasts produced by a second forecast system, called the target system. We provide methods for two cases: (a) when the cumulative distribution function of the target system is known or can be reconstructed, and (b) when the target system only provides a mean (i.e., expected) value forecast. These methods are illustrated for precipitation and air temperature using weather forecasts (as the target system) and a subseasonal to seasonal EPS, both from the Australian Bureau of Meteorology. It is found that enforcing consistency of daily EPS forecasts at lead days 1 to 7 results in week 1 (e.g., 7‐day precipitation accumulations) forecasts being substantially more accurate than week 1 forecasts from either the original EPS or the target forecast system.
Taggart et al. (Wed,) studied this question.