Impacts of Multiscale Predictors on Random Forest Based Probabilistic Forecasts of Severe Weather Hazards
Earth and Space Science · 2026-11 · 13 (11)
DOI: 10.1029/2026ea005229 ↗Kurz gefasst
Fachartikel in der Zeitschrift Earth and Space Science, online erschienen am 12.11.2026. Verfasst von Dan Kubalek und 2 weiteren Autor:innen. Die Arbeit nennt 74 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Random forests (RFs) have been shown to skillfully predict severe weather hazards using global model and convection‐allowing ensemble (CAE) predictors. However, RF models typically use locally‐defined predictors, potentially missing important information about the large‐scale flow and environment as a whole. This study introduces a method to incorporate multi‐scale flow‐dependent information into RF models using CAE‐based predictors pre‐processed at various spatial scales. The primary goal is to determine if the use of multi‐scale predictors improves probabilistic forecast skill for severe weather hazards for 24‐hr and 4‐hr forecast windows. Using CAE model fields from 2018 to 2019 Hazardous Weather Testbed Spring Forecasting Experiments, we compared a control set of RF models (CTLRF) using only 80 km predictors with an experimental set (EXPRF) using the 80 km predictors as well as predictors spatially smoothed over increasingly larger scales. Results showed EXPRF models had statistically significantly higher Brier skill score's than CTLRF models for several hazards. EXPRF forecasts generally had better resolution than CTLRF and similarly low calibration error. Predictor contributions showed that when severe weather did occur, on average, the meso‐ scale storm‐attribute predictors contributed the most to skill while the meso‐ scale environmental predictors dominated environment‐related contributions to forecast skill. When severe weather did not occur, on average, the higher resolution meso‐ scale storm‐attribute predictors contributed more to forecast skill than the lower resolution meso‐ scale storm‐attribute predictors for 24‐hr forecasts, but not for 4‐hr forecasts. In general, there were contributions to forecast skill from the predictors across all spatial scales.
Zitieren
APA
Kubalek, D., Johnson, A., & Wang, X. (2026). Impacts of Multiscale Predictors on Random Forest Based Probabilistic Forecasts of Severe Weather Hazards. Earth and Space Science, 13(11). https://doi.org/10.1029/2026ea005229
Vancouver
Kubalek D, Johnson A, Wang X. Impacts of Multiscale Predictors on Random Forest Based Probabilistic Forecasts of Severe Weather Hazards. Earth and Space Science. 2026;13(11). doi:10.1029/2026ea005229
BibTeX
@article{Kubalek2026,
author = {Kubalek, Dan and Johnson, Aaron and Wang, Xuguang},
title = {Impacts of Multiscale Predictors on Random Forest Based Probabilistic Forecasts of Severe Weather Hazards},
journal = {Earth and Space Science},
year = {2026},
volume = {13},
number = {11},
doi = {10.1029/2026ea005229}
}
EUVIMED