"The assumptions made in constructing ensemble forecast systems (EPS) and the intelligent use of their output will be the primary focus of this module. In order for you to better understand EPS output, we will also look at some necessary background information on statistics, probability, and probabilistic forecasting."
Appropriate Level: Advanced undergraduate and above.
Summary comments: Lots of text, some of which can be dense (when the concepts described are relatively challenging). Simple, mostly static diagrams provide clear illustrations. Assessment exercises (multiple choice and multiple answer questions, with feedback) help. Quite a few ensemble forecast products and verification tools are described. Basic concepts of probability and statistics underly most of these (as befits the topic), and an attempt is made to provide some background about these, but a basic statistics course would be a valuable prerequisite to this material. A COMET webcast by NCEP's Dr. Bill Bua offers a simpler, one-hour introduction to a subset of the material in this module, "Introduction to Ensemble Prediction"; for most users unfamiliar with ensemble forecasting, I recommend viewing the webcast before completing the module. Preassessment. This module starts with a randomized, 15-item pre-assessment quiz, which the user repeats at the end of the module as a measure of what the user has learned. The pre-assessment quiz therefore should provide a preview of at least some of the module's content.
The mostly (entirely?) multiple-answer quiz questions include graph and weather map interpretations and verbal questions about aspects of ensemble forecasting and its applications. The questions, graphs and maps--and hence necessarily the module itself--use language and invoke concepts technical enough to be well out of the reach of beginning students. Here's an example of a graph with very limited explanation offered to help the user interpret it (which of course is the point--the user presumably will learn more about it in the module):
(The user is asked about the resolution and reliability of the two forecasts depicted in the graph.)
Depending on the degree of support provided by the instructor and the module itself, the preassessment suggests that upper division majors should be able to learn something from the module. Taking the preassessment quiz yourself is one way of judging whether the module content as it stands is appropriate in content and level for your students. Warning: although the preassessment strikes me as a useful pedagogical tool (for measuring learning progress and offering a preview of the module content), completing it can be discouraging unless users understand that they aren't expected to do well on it and that they should look forward to learning enough from the module to improve dramatically.
Introduction.
Why Ensembles? A table summarizes the advantages of ensemble forecasts over of single forecasts.
Chaos and NWP Models. A Flash animation of an 84-hour "control" 500 mb height forecast at 12-hour intervals and a second forecast with perturbed initial conditions are superimposed, together with color-filled contours of the difference between them. The accompanying text describes the animation and points out how the difference between the two forecasts tends to grows with forecast time. The idea of chaos and its manifestation in the atmosphere and numerical weather forecasting is introduced, usiing the same flash animation. The overarching content of the module is summarized
Generation. Brief, bulleted summary of possible methods used to create ensemble members, based on uncertainty in data or in the NWP model itself.
Perturbation Sources. Uncertainty in initial conditions, and the range of that uncertainty (one static graphic similar to the animation in Section 2 of the Introduction); uncertainty in model formulation (including dynamical formulation and grid- and subgrid-scale physical parameterizations; illustrated with contour plots of precipitation plus a sounding); and boundary value uncertainty (text only).
Implementations. Singular vectors (used by ECMWF); "breeding" cycle (one static graphic); reference to ensemble Kalman filter (EnKF).
Statistical Concepts. An overview of basic concepts and their relevance to ensemble forecasting, including the idea of probability density functions. Very little math; a few static graphs.
Probability Distributions.
Middleness.
Variability.
Shape.
Using PDFs.
Data Application. A spaghetti plot is introduced, with some basic guidance about how to read it. Measures of central tendency (mean, median, mode) and their limitations applied to interpreting ensemble forecasts.
Exercises. Two multiple-choice questions with feedback and discussion. These strike me as pedagogically helpful.
Summarizing Data. Motivates the need for products that help forecasters digest the vast amount of information in an ensemble forecast.
Products. Thirteen pages of examples with explanation, comparison, and discussion of various ensemble forecast analysis products, illustrated with static graphics.
Product Interpretation. Relatively clear, simplified diagrams show how to intepret the mean and the spread of ensemble forecasts from spaghetti diagrams. One multiple choice question with feedback.
Exercises. Three multiple choice questions with feedback, all involving interpretation of figures.
Verification. The problem of verifying ensemble forecasts.
Concepts. Categorical and probabilistic forecasts and verification of the former. Skill score. Verification of probabilistic forecasts: "reliability" and "resolution".
Tools. Six pages on verification tools. Some of these are relatively dense, but interesting.
Applications. A few examples of how some verification tools are used at NCEP (an animated Talagrand diagram is shown). Comparison of NCEP's EPS to those of other forecast centers (links).
Exercises. Three multiple answer questions with feedback.
Case Applications. Links to case studies and to the module quiz.
