Second Conference on Artificial Intelligence

2.5

Calibration of Probabilistic, ensemble precipitation forecasts by an artificial neural network

PAPER WITHDRAWN

S. L. Mullen, Univ. of Arizona, Tucson, AZ; and M. M. Poulton and R. Buizza

A variant of the back propagation is used to post-process rainfall forecasts from the ECMWF Ensemble Prediction System (EPS). Daily, 24-h quantitative precipitation forecasts (QPFs) from the EPS at projections of 1-10 days are used as input for the net. Probabilistic QPFs (PQPFs) are output by the net for several thresholds. Cross validation of two years of EPS forecasts is used for training and validation, and the Ranked Probability Score is minimized. Preliminary results indicate that net can process raw PQPFs from the EPS that are unskillful and unreliable, and produce PQPF's that are skillful and reliable for thresholds up to 20 mm per day.

Session 2, Artificial Neural Networks
Monday, 10 January 2000, 1:30 PM-4:30 PM

Previous paper  Next paper

Browse or search entire meeting

AMS Home Page