COMMENTS

Abstract

amill et al. (2006, hereafter HWM) present a very interesting way to develop objective forecasts of surface weather variables. Basically, this consists of running some numerical model for a long sample period and developing statistical relationships between the weather elements and the archived output. The specific statistical method they present is an application of analogs in a model output statistics (MOS) framework; another method is presented in a similar paper by Hamill et al. (2004). They also suggest the forecasts, which they call “reforecasts,” can be used to diagnose model bias and to study predictability. The primary intent of the authors is“to stimulate a serious discussion about the value of reforecasts.” I do not want to disappoint the authors and offer this as one “serious discussion.” It is not meant to be critical, although I will present another point of view on some aspects. The crucial question is whether a long record— here 25 yr—of a frozen model with appropriate initial conditions run at a lower resolution than the parent model, for economy, is of more use than a shorter record—maybe 5 yr—of an operational model that has possibly undergone some modest evolution over the period of record. I believe there are two main related considerations in trying to answer that question: 1) what is the ultimate purpose of the user of the reforecasts (e.g., diagnosing operational model behavior, studying predictability, making operational forecasts), and 2) can the stripped-down model furnish results competitive with an operational model?