Correcting Errors in MCS Quantitative Precipitation Forecasts of Intensity, Timing, and Location Using Machine Learning Algorithms

Convection-allowing grid spacing models have been shown to improve simulations of heavy rain with warm-season convection. While these models improve the skill in quantitative precipitation forecasts (QPF), inconsistencies concerning precipitation intensity, timing, and location remain evident. Recognizing the implications of these errors, especially in critical scenarios such as flood predictions, our research focuses on resolving these challenges. Harnessing the Method for Object-based Diagnostic Evaluation (MODE) and its time-domain variant (MODE TD), we have examined mesoscale convective system QPF errors within the High-Resolution Ensemble Forecast members for 107 events occurring in the warm seasons from 2019-2021, and have looked at a broad spectrum of metrics encompassing displacement, geometry, and intensity. Additionally, we have gathered information on atmospheric parameters from the Storm Prediction Center mesoanalysis for each event.

We will present research that makes use of the above error information and environmental conditions to train machine learning (ML) algorithms to correct inaccuracies prevalent in ensemble member forecasts. The result of this research will be an ML-based postprocessor, proficient at pinpointing and mitigating errors in QPF displacement, timing, and magnitude. Expected outcomes include multiple forecast products such as probability-matched means and augmented PQPFs. We will present preliminary results of the use of the ML algorithms to reduce QPF errors.