Lead: "And you thought weather forecasters had it tough. Hydrologists looking to forecast the next flood or dangerously low river flow must start with what weather forecasters give them--predictions of rain and snow, heat and cold--and fold that into myriad predictive models. Then those models must in turn forecast how rain and any melted snow will flow from rivulet to river while liable to loss to evaporation, groundwater, reservoirs, and farmers' fields. During their century in the forecasting business, hydrologists have developed a modicum of skill, but a newly published study fails to find any improvement during the past 20 years in forecasting river levels out to 3 days."
Comment: The graph certainly indicates that little progress has been made in this forecasting over the last decade. I think the problem may be found in other types of forecasting. Some years ago I looked at the forecasts of economic growth made by economists and found that they were on the average less accurate than simply assuming that historical rates of growth would continue.
The Science article further states:
Troubleshooting hydrologic forecasting to understand why it's been resisting improvement will take "objective study and well-structured verification," says Welles, "not expert opinion or ad hoc experience." BAMS Editor-in-Chief Jeff Rosenfeld agrees. Writing in an accompanying editorial, he finds that the Welles paper makes the point that "forecasting must include verification if it is to be scientific. Every forecast is like a hypothesis, and in science every hypothesis must ultimately be tested."There is a need to validate forecasting models. Where the model depends upon estimates from experts or expert interpretation of its forecasts, there is also a need to validate the work of those experts. That is, if you want the forecasts to be accurate. JAD
No comments:
Post a Comment