UBC Theses and Dissertations
Ensemble-averaged, probabilistic, and Kalman-filtered regional ozone forecasts Delle Monache, Luca
This thesis investigates the hypothesis that ensemble methods and Kalman-filter (KF) post-processing can be utilized to improve near-surface real-time ozone forecasts. The ensemble approach combines multiple forecasts to yield ensemble-averaged and probabilistic predictions. In non-linear systems such as the atmosphere, it is well established that the ensemble approach provides a better estimate of future evolution than a deterministic forecast. This approach is extended here for ozone forecasts. KF post-processing is applied to remove ozone-forecast bias; i.e., systematic errors. In this dissertation, the filter is applied in a predictor mode to the raw ozone forecasts from the Community Multiscale Air Quality (CMAQ) 3-D numerical model. An ozone ensemble-forecast system based on a multi-model approach has been analyzed. Moreover, a new ensemble design for air-quality forecasts has been proposed, based on both meteorology and emission perturbations. Ozone ensemble-averaged and probabilistic forecasts resulting from these ensemble methods have been realized and tested (introducing a new reliability index). The following are the main findings of this thesis. An ozone ensemble-forecast system based on a multi-model approach produces an ensemble-averaged prediction more skillful than a single-model approach. Ensemble-averaging is able to compensate for some of the predictive-skill deficiencies in deterministic ozone forecasts, and for part of the initial-condition inaccuracy. In the new ensemble air-quality forecast system proposed, the meteorology perturbation is important to capture ozone temporal and spatial distributions. The emission perturbation is needed to accurately predict the ozone concentration magnitude. The emission perturbations are more important than the meteorology ones to capture high (and rarely measured) ozone concentrations. The KF successfully removes part of the ozone-forecast bias caused by errors in the model. The combination of ensemble averaging (unsystematic-error removal) and Kalman filtering (systematic-error removal) results in the best ozone forecast. Ensemble and KF methods can indeed significantly improve near-surface ozone forecasts, even in the complex coastal mountain setting of the Lower Fraser Valley. There are no intrinsic limitations to these methods that would prevent their application in real time to other pollutants in other geographic settings.
Item Citations and Data