300 likes | 387 Views
Probability Forecasts and Verification. Uncertainty. Forecasts have uncertainty that should be taken into account: Observational Uncertainty (initial conditions)
E N D
Uncertainty • Forecasts have uncertainty that should be taken into account: • Observational Uncertainty (initial conditions) • Instrument errors, bias in frequency of measurements, representativeness, reporting errors, random errors, precision errors, conversion errors … • Model Uncertainty • Insufficient spatial resolution, truncation errors, parameterization errors, coding errors, initialization errors …
Deterministic and Probabilistic Forecasts • Deterministic Forecast: Specific Outcome is Predicted • Probabilistic Forecast: Range of Possible Outcomes Accompanied by Confidence Estimate
Probabilistic Verification • Requires Several Forecast Cases – Cannot Verify a Single Probabilistic Forecast • Compare Forecast Distribution to a Distribution of Observations • Rank Probability Score (RPS) • Rank Probability SkillScore (RPSS) • Relative Operating Characteristics (ROC) • Reliability (Not Really Verification)
NMME July 1 start DJF prec forecast RPSS CFSv2
Relative Operating Characteristics Hit Rate = hits/(hits+misses) False Alarm Rate = false alarms/(hits+false alarms)
Relative Operating Characteristics For Each Threshold Build a Contingency Table Summing Over All Forecasts
Reliability • Prediction System is Reliable if Predicted Probabilities Agree with Observed Frequencies • Ex: Climatological Probability has Perfect Reliability (33%, 33%, 33%) • Reliability Diagram Measure whether the Prediction System is Reliable, Overconfident or Underconfident • Reliability Diagram also Shows Resolution and Sharpness
Probabilistic Verification • Requires Several Forecast Cases – Cannot Verify a Single Probabilistic Forecast • Compare Forecast Distribution to a Distribution of Observations • Rank Probability Score (RPS) • Rank Probability SkillScore (RPSS) • Relative Operating Characteristics (ROC) • Reliability (Not Really Verification)