15A.1 A Novel Quantile Regression Neural Network Scheme for Postprocessing Medium-range Ensemble Quantitative Precipitation Forecasts

Thursday, 1 February 2024: 1:45 PM
345/346 (The Baltimore Convention Center)
Mohammadvaghef Ghazvinian, PhD, Center for Western Weather and Water Extremes, Scripps Institution of Oceanography, Univ. of California, San Diego, La Jolla, CA; and S. Candido, L. Delle Monache, V. Afzali Gorooh, A. Sengupta, W. Hu, and M. M. Ralph

Generating reliable and skillful medium-range probabilistic quantitative precipitation forecasts (PQPFs) is important for many applications, particularly water resources/flood risk management and forecast-informed reservoir operations. This study proposes a computationally efficient quantile regression - artificial neural network (ANN-QR) scheme to generate reliable, skillful, and spatially detailed PQPFs. We test the performance of our proposed model by postprocessing 24-h accumulated ensemble quantitative precipitation forecasts from NOAA’s Global Ensemble Forecast System version 12 (GEFSv12) reforecasts over the Western United States for lead times 1 to 7 days. The PQPFs obtained from the proposed model is compared with those generated from three other ANN-based schemes that share similar predictors and architecture but differ in the estimated probability distributions and loss functions specification. Those include recently proposed ANN censored, shifted gamma distribution (ANN-CSGD; Ghazvinian et al. 2021;2022), a multiclass classification-based scheme, and a parametric mixture model. The results show that the proposed method compares well with the ANN-CSGD in predicting a range of events, heavy-to-extreme events included, while significantly outperforming the two other schemes. The findings also show that whereas ANN-QR and ANN-CSGD highly improve the prediction of heavy-to-extreme events, each exhibits distinct strengths and limitations in mitigating conditional biases and their relative performance varies when the outcome is conditioned on specific extreme events and geographic locations.
In this presentation, we share the findings, discuss the verification results, offer insights on the advantages and limitations of the algorithms, and explore practical-computational implications from an operational perspective.

References:

Ghazvinian, M., Zhang, Y., Seo D.-J., He, M., Fernando, N., 2021: A novel hybrid artificial neural network - Parametric scheme for postprocessing medium-range precipitation forecasts. Advances in Water Resources, Volume 151,103907, https://doi.org/10.1016/j.advwatres.2021.103907.

Ghazvinian, M., Y. Zhang, T. M. Hamill, D.-J. Seo, and N. Fernando, 2022: Improving probabilistic quantitative precipitation forecasts using short training data through artificial neural networks. J. Hydrometeor., 23, 1365–1382, https://doi.org/10.1175/JHM-D-22-0021.1.


- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner