Part IX:   Shifting operational forecast paradigm

    From the discussion in Part 2, one can see that there is a fundamental shift in NWP practice and philosophy from the single forecast-based paradigm to the ensemble-based one (Fig. 14). Four main differences are summarized below. First, the former views a forecast as a deterministic process expressed by a single value hoping to have a single best shot, while the latter as a stochastic process (within the range of uncertainty) expressed by probabilistic distribution hoping to address forecast uncertainty alone with a most probable solution. In other words, besides providing a most probable and improved single solution an ensemble also provides

Figure 14

extra forecast uncertainty information comparing to a single-value forecast (see Part 4). Therefore, there is nothing to loss but purely gain (complete) when the NWP paradigm shifts from the single-value forecasting to the ensemble forecasting. Besides the omitting of forecast uncertainty, uncertainty in observation and ICs is also ignored in single-forecast based NWP, while, in ensemble-based NWP, observation should also be expressed by a distribution in data assimilation process and multiple analyses would be created to form an ensemble of ICs for model integration to consider uncertainties in both observation and data assimilation procedure. As science and technology advances, ensemble forecasting should become a main tool in weather, climate and water forecasting. In order to smoothly complete this paradigm transition, training and education is an urgent issue.  Meteorologists and end-users need to work together to develop strategies of how effectively and correctly using uncertainty and probabilistic information in decision-making process to benefit the entire society the best. The recent study report published by the U.S. National Research Council of National Academy of Sciences  - Completing the Forecast: Characterizing and Communicating Uncertainty for Better Decisions Using Weather and Climate Forecasts - is a good example of this effort (NRC, 2006). UCAR COMET classroom training courses (http://www.comet.ucar.edu/class/index.html) and local Weather Forecasting Offices’ ensemble training workshops are all good formats in this training effort.

Secondly, traditional single-forecast based NWP is a one-way system: observation determines model prediction but with no feedback from prediction to observation. However, in reality, error also exists in observation and impacts forecast accuracy in a flow-dependent way, i.e. it may affect a forecast significantly in one time (or over one region or for one weather system) but not much on another time (region or system). Therefore, a two-way system is desired: based on the estimation of potential forecast error over a region of interest, observation over a certain upstream region might need to be adjusted and improved accordingly too. Ensemble forecasting provides a bridge to make this two-way system possible. In an ensemble system with good spread-skill relation (see Part 6), the ensemble should be able to identify those weather systems associated with potential large errors and could also be used to trace the errors back to locate possible source regions in upstream using ensemble spread information. To improve the forecast, extra observations might be made over those source regions being identified. Or opposite action can be considered: less observation such as satellite data can be used in data assimilation when weather is calm to save resources. This process is known as adaptive or target observation technique, which is another new frontier of NWP (Palmer et. al., 1998; Bishop and Toth, 1999; Pu and Kalnay, 1999; Szunyogh et. al., 2000; Bishop et. al. 2001; Majumdar et. al., 2002). Obviously, this two-way approach is more sounding both scientifically and economically. Interactive two-way NWP system has been emphasized as one of the main goals in the on-going international joint research project GIFS (global interactive forecasting system).

Thirdly, although the estimation of forecast error distribution in the single-forecast paradigm is also possible via historical forecast data (see Part 2), it’s, however, not flow dependent and doesn’t reflect the true predictability or “error of the day”, while the dynamical ensemble spread is flow-dependent, does reflect “error of the day”, and, therefore, provides more situation-relevant information for better decision-making. No doubt, all the above three changes are reflecting a step forward in science and technology.

Lastly, since it’s hard for human to think nonlinearly, forecasters will eventually not be able to keep up with NWP model’s thinking someday when model forecast is accurately enough with continuing improvement of model and data quality. At that point, forecasters will have to mainly act like broadcasters or messengers by passively passing a model forecast to public or end-users with not much value added by forecasters’ human role if single deterministic forecast is provided. While in the new era of ensemble forecasting, forecaster’s human role will remain important and actively play a key role in the process of forecast-making in the following two folds. Although an ensemble provides multiple possible solutions, only one of them will eventually realize in reality. Therefore, a forecaster can act as an interactive “live” post-processing of the raw ensemble forecasts. For example, he might use other available data and the newest observations as well as his experiences to weigh each ensemble solution or filter out some “unlikely” members to possibly narrow forecast uncertainty. At the same time, forecasters should give proper physical interpretation of each distinct possible ensemble solutions to end-users for them to make better decisions and, therefore, provide an enhanced and value-added service to user community.

In one word, the core of switching from single-forecast paradigm to ensemble-forecast paradigm is to provide better service to user community to meet the variety of needs of our customers by producing a more accurate and complete rather than an overly simplified forecast to truly reflect the complex nature of weather, climate and water systems.

Contact  Jun Du