18th Conference on Weather and Forecasting, 14th Conference on Numerical Weather Prediction, and Ninth Conference on Mesoscale Processes

Wednesday, 1 August 2001
Model Trends and Satellite Imagery in Forecasting
Brian Motta, CIRA/Colorado State Univ., Fort Collins, CO; and R. H. Grumm and A. Mostek
Poster PDF (184.7 kB)
The Virtual Institute for Satellite Integration Training (VISIT) has developed training aimed at improving the winter storm forecasting process. Strategies required for the improvement of winter storm forecasts focus on two critical factors. The first factor is the critical need for effective evaluation of model initial-hour forecasts. the second is the related evaluation of model trends.

Model initial hour forecasts often have characteristics reflecting both the model first guess and the assimilated data sets. The first guess is often heavily weighted in the initialization process and often has a significant impact on the characteristics of important meteorological features. Features, which may appear in the remotely sensed data, have weights assigned to them during the data assimilation step. These weights can also impact the model initial hour forecasts. Finally, when the remotely sensed data is assimilated with the first guess, significant features may be "washed out" or under-represented depending on the character of the first guess. Satellite and other remotely sensed data can be useful in evaluating where the first guess may be having difficulty initializing a feature properly. Before using a model's forecast, a thorough evaluation of the initial-hour forecast must be carried out using remotely sensed data.

When assessing model trends, it's also important to examine the role of the first guess in a model's performance. Without the initial and careful examination as described above, it is very difficult to recognize features which may "exist" in the model but not in the real world. To that end, the importance of evaluating forecast trends of models using the AWIPS capability called dprog/dt in concert with real-time data becomes an important approach as difficult forecast situations, particularly those in which the models disagree, provides a better forecast approach than interpreting only the latest run. Large run-to-run difference may be indications of a low-confidence forecast.

Current and future VISIT training in these areas will be shown. Forecaster response to the model initialization training has been very positive. Additional training on the integration and use of model output with observed data in the forecast process may include ensembles, medium-range, and large-scale analysis.

Supplementary URL: