Calibrated Gaussian Kernel Method

1
Calibrated Gaussian Kernel Method

• Accept Weights from other methods
• (Bayesian, Best Model Frequency, Ridge
  Regression)
• Compute statistics from cases where precip is
  observed.
• (Need a PoP prediction, 0 Forecast amounts
  OK, Spatial pooling in dry areas)
• May need to transform precip. (Normal
  Quantile?)
• Compute statistics for A) individual ensembles
  and B) Ensemble mean.
• (Statistical components computed with
  exponential smoothing,
• Need to worry about initial values.)
• Use Gaussian Kernel to forecast conditional PDF,
  PDFPoPPDF

2
Required Statistics

• X()(1-a) X(0) w(a x)
• X() time series estimate of x statistic
• w weighting
• a exponential smoothing coef. lt.05
• Fforecast anomalies, oobs anomalies
• Statistics
• F, FF ,FmFm
• o, oo,
• Fo, Fmo
• F-Fm,