| dc.description.abstract | Machine learning (ML) algorithms utilized for post-processing of convection-allowing model/ensemble (CAM/CAE) output has been a major area of research to handle limitations with CAM/CAE forecasts. ML has been used to correct systematic biases, relate observed variables to numerical output, and synthesize extremely large data into probabilistic forecasts. In particular, numerous studies have shown random forests (RFs) to be successful in severe weather forecasting applications utilizing predictors from global scale and/or CAE output. However, predictors used in the RF models are typically fixed and treated independently when training the RF models. This can consequently leave out important information about the large-scale flow pattern that is necessary for assessing severe weather risk. This thesis develops a method for manifesting multiscale flow-dependence into RF models through direct incorporation of CAE-based predictors that are pre-processed at increasing spatial
length scales. The different length scales account for different scales of motion with the goal to improve probabilistic forecast skill for a variety of severe weather hazards for next-day (12-12 UTC) - or 24hr and 4hr (20-00 UTC) forecasts. In order to verify the impacts of the multiscale predictors on the skill of the RF models, a control (CTLRF) and experimental (EXPRF) set of RF models were created. The CTLRF models were trained with only predictors pre-processed to 80 kilometers (km) and the EXPRF models were trained with predictors pre-processed to 80 km in addition to larger, spatially smoothed 80 km predictors.
Both models were verified against the storm prediction center (SPC) reports quantitatively and qualitatively. Results show that the EXPRF models had higher brier skill score’s (BSS) than the CTLRF models for all sub-significant severe weather hazards for both forecast periods, but significantly higher BSS’s when forecasting any severe weather hazard (24hr and 4hr), wind (24hr), hail (24hr and 4hr), and significant winds (24hr). The EXPRF forecasts generally had the best resolution component of the brier score (BS), of which some severe weather hazards were significantly higher than CTLRF forecasts. Furthermore, both models
generally had small calibration error. However, the CTLRF 24hr and 4hr wind forecasts had significantly lower calibration error compared to EXPRF. In general, neither model’s probabilistic forecasts were consistently more reliable than the other. Predictor contributions determined via tree interpreter (TI) showed when severe weather did not occur, on average, the meso-γ scale storm-attribute predictors contributed more to forecast skill than the meso-β scale storm-attribute predictors for 24hr forecasts. Whereas the opposite was true for the 4hr forecasts. When severe weather did occur, on average, the meso-β scale
storm-attribute predictors contributed the most to skill in general. Meanwhile, the meso-γ scale environmental predictors dominated environment-related contributions to forecast skill, but in general, most multiscale predictors still contributed to skill. Through case-studies, it was found that the meso-β and meso-α scale storm-attribute predictors accounts for spatial
uncertainty of simulated storms similar to neighborhood-based CAM forecasts. Meanwhile the environment predictors, in particular the convective environment predictors, had greater sensitivity to smoothing and sometimes did not benefit from losing sharp gradients and local extremes that can be associated with synoptically predictable features. | en_US |
