Binary Forecasts and Observations Yes/ No; True/False; 1/0 Form the basis for many verification techniques Used in many applications of weather data Will the temperature in Phoenix exceed 90 degrees tomorrow? While quite simple, still commonly used.
2 x 2 Contingency Table Columns = obs Rows = Forecast
Finley Tornado Data (1884)
A success? Percent Correct = (28+2680)/2803 = 0.966 !!!!
Maybe not. Percent Correct = (0+2752)/2803 = 0.982 – better!! Critisism from Pierce and Gilbert. (1884) Percent Correct = (0+2752)/2803 = 0.982 – better!!
Alternative Statistics Mention – the never forecast has a POD = 0, FAR = 0, CSI = 1 Single statistics can be misleading. Threat Score = 28 / (28 + 72+ 23) = 0.228 Probability of Detection = 28 / (28 + 23) = 0.549 False Alarm Rate= 72/(28 + 72) = 0.720
Attributes of forecast quality Accuracy Bias Reliability Resolution Discrimination Sharpness
Value If Costs can be associated with false positives and false negative forecasts, one can determine if a forecast has value. The value of a forecast varies by user.
Skill Scores Single value to summarize performance. Reference forecast - best naive guess; persistence, climatology Proper skill scores – reflect forecastor true intent. A perfect forecast implies that the object can be perfectly observed
Generic Skill Score Positively oriented – Positive is good
Exercise 2 (ex2.r) Load verification library verify function frcst.type; obs.type; Documentation example()