Probabilistic quantitative precipitation forecasts by a short-range ensemble weather forecasting system and precipitation calibration.

摘要

This dissertation studies probabilistic quantitative precipitation forecasts (PQPFs) by the National Centers Environmental Prediction (NCEP) short-range ensemble forecasting system. Also, precipitation forecasts are calibrated through artificial neural network techniques to correct the forecast bias.; The NCEP Regional Spectral Model (RSM) ensemble system is used to generate eleven ensemble forecasts twice daily over the southwest United States during winter 2002/03. Forecast quality and potential economic value of 12-km PQPFs are found to depend strongly on the verification dataset, geographic region, and precipitation threshold. Compared to the NCEP Stage IV 4-km precipitation analyses, in general, the daily PQPFs are skillful over the California Nevada River Forecast Center region for thresholds between 1-50 mm, but are unskillful over the Colorado Basin. The 6-hour PQPFs show a diurnal cycle of the southwest precipitation, which may cause the large discrepancy of the forecast skill for the daily PQPFs between the 0000 and 1200 UTC forecast cycles.; The model exhibits a wet bias for all thresholds that is larger over Nevada and the Colorado Basin than over the California region. Mitigation of such biases will pose serious challenges to the modeling community in view of the uncertainties inherent in verifying analyses. Since the RSM is good at discriminating precipitation events over some hydrologic regions, the biases of Quantitative precipitation forecasts (QPFs) and PQPFs may be calibrated. By training PQPFs during the four months, a 3-layered feedfoward artificial neural network could reduce conditional bias and significantly increase brier skills of PQPFs during the rest month. Cross validation of bias correction for PQPFs over each month shows that PQPFs reduce the biases over the California region and keep the sharpness as well, while the resolution term of Brier scores decreases at higher thresholds after calibration for other hydrologic regions. Adjustment of QPFs by using the calibrated PQPFs also reduces root mean square errors in QPFs. Less improvement of heavy precipitation events for PQPFs and QPFs indicates that a larger training sample size is desirable. More methods need to be examined for bias removal to improve the PQPF quality without harming the resolution term.