Experimental Seasonal Forecast of Fire Risk
Practical Exercise
How to create a probabilistic forecast model, make a forecast using new SST information, and understand the probabilistic nature of the forecast.
Basic Structure:
1. Overview: The exercise aims to teach how a probabilistic forecast model is created and how it can be used to make a probabilistic forecast of fire risk. This will be taught using a hands-on interactive Excel-based exercise to facilitate understanding and empower the user.
2. Learning Goals: Specific understanding and skills to be learned include: understanding data, understanding and interpretation of scatterplots and thresholds, understanding and interpretation of probabilistic forecast models using a decision matrix, how to set their own thresholds, access latest NINO4 data, and understand the probabilistic nature of forecasts and uncertainty involved.
3. Materials and resources: Computer, Microsoft Excel, Internet (optional)
4. Time: 1.5 hours
5. Steps:
a. Open Excel, enable macros (see Appendix I), review data and concepts
b. Create scatterplot (Tengah, two month lag), interpret correlation
c. Make probabilistic forecast model using decision matrix, interpret
d. How to access current NINO4 data
e. Make forecast using decision matrix, then have them make their own forecast.
f. Review probabilistic nature of forecasts. Explore how alternative thresholds change probabilistic forecast with interactive pie charts, quiz on probabilities.
g. Explore Forecast Map Room scatterplot and decision matrix, explain differences in appearance.
6. Questions and Comments. Discuss alternative threshold values (what is important to them?), options (is one or two month lag more useful?) and limitations (small data set). Review any problems to ensure they have all the knowledge and skills necessary to make their own forecasts in the future.
Steps
A. Open Excel. Enable Macros (see Appendix I for instructions). Open Exercise data file to “Raw Data” worksheet. Explore data, review concepts: NINO4 Index, anomalies, lagged relationship, median.
.
B. Open “Scatterplots” worksheet. Explain scatterplot, and interpret data relationship (strength and direction of correlation).
Then have them use Tengah province fires and NINO4 2 month lag data to create their own scatterplot. Once plot is created, format x and y axis labels to indicate median thresholds (right click on axis, choose: format axis>Scale tab>Value crosses at (387 for x axis, 0.40 for y axis). Other stylistic formatting optional depending on time.
C. Open “Make Model” Worksheet. Explain how data can be divided into quadrants based on thresholds, and then data can be displayed in a decision matrix. Emphasize relationship between scatterplot quadrants and decision matrix quadrants (High fires and high NINO4 shown in pink for both scatterplot and decision matrix.) Describe how each decision matrix column adds up to 100%. Make probabilistic forecast model using median thresholds in decision matrix, interpret.
Then let them choose a different threshold, for example 1000 fires, in the yellow interactive box. After entering a new threshold, users must press return before pressing RUN. The decision matrix in the yellow box will also be updated. (Scatter plot thresholds are not updated).
D. How to access current NINO4 data. If internet available, go to Map Room, access data.
Brings you to this page. Show how latest NINO4 Index value is added at bottom of the page.
E. Use “Make Forecast” worksheet to illustrate how to make forecast using decision matrix with new NINO4 data. Make sure time is allocated for them to make their own forecast based on a hypothetical June 2009 NINO4 value, have time for them to present results and raise questions.
Example 1: It is April 2008 and the NINO4 value is -0.98. (Or use current April/May NINO4 value). Walk through together how this information can be used in decision matrix to determine likelihood of high or low fires (Light blue areas illustrate “low NINO4 values quadrants in both scatterplot and decision matrix)
Example 2: Present a hypothetical NINO4 value for June 2009 (or current April/May 2008 value) and have them work this through on their own.
F. Use “Forecast Probabilities” worksheet to explain probabilistic nature of forecasts in more detail. First have them understand probabilities using median threshold. Then allow them to explore probabilities for their own thresholds set interactively. (Interactive thresholds MUST be set on “Make Model” worksheet – this will automatically update the “Forecast Probabilities” worksheet).
Follow this up with quiz on probabilities on bottom of worksheet. Answer is C. (This quiz adapted from Rebecca Morss, LDEO Colloquium Speaker, 3/27/09).
G. Show Predicting Fire Activity Map Room, explain differences in appearance.
Appendix I: Enabling macros in Excel.
Open Microsoft Excel (2003). Go to “Tools” tab, choose Options. Open “Security” tab and click on “Macro Security”. Open “Security Level” tab and choose “Medium” level of security. Close exercise and re-open. When the security warning pops up choose “Enable Macros”. If macros are disabled, the interactive threshold setting component of the exercise will not be available.