Application of the Weather Research and Forecasting Model for Air Quality Forecasting Applications in Central California

Open Access
Author:
Rogers, Raphael Edward
Graduate Program:
Meteorology
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
June 23, 2011
Committee Members:
  • Aijun Deng, Thesis Advisor
Keywords:
  • California
  • Air Quality
  • WRF
Abstract:
In air quality studies it is important to have the most accurate meteorological data possible for use as inputs into photochemical models. Observations from observing systems such as radiosondes, wind profilers, and surface stations are sparse and are widely distributed over tens to hundreds of kilometers. Therefore the use of numerical weather prediction models is very important to determine the temperature, moisture, and wind fields. The Bay Area Air Quality Management District (BAAQMD) uses the Pennsylvania State University (PSU)/National Center for Atmospheric Research (NCAR) Mesoscale Model Version 5 (MM5) to create the meteorological datasets for use as inputs into photochemical models such as the Community Multiscale Air Quality (CMAQ) model and the Comprehensive Air Quality Model with Extensions (CAMx) model. However, development on the MM5 has been discontinued, and the BAAQMD is interested in transitioning to the Weather Research and Forecasting (WRF) model when it can perform as well as the MM5. The objective of the thesis is to assist the BAAQMD to transition to the WRF modeling system by finding the optimal WRF model configuration for the Bay Area and Central Valley region for both winter and summer seasons. The investigation starts by determining the optimal set of physics packages to use for the region before four dimensional data assimilation (FDDA) is applied. The FDDA is applied throughout the model integrations to produce dynamic analyses of the meteorology for use in the atmospheric chemistry models. The two atmospheric radiation schemes tested are the Rapid Radiative Transfer Model (RRTM) and the RRTM for global climate models (RRTMG). With the use of the better performing radiation scheme, the land surface models (LSMs), the 5-layer thermal diffusion LSM, Noah LSM, Rapid Update Cycle (RUC) LSM, and the Pleim-Xiu (PX) LSM are compared to determine the best LSM for use in the baseline configuration. Statistical analyses are performed to compare the two radiation schemes with each other and to compare the LSMs with each other. With the better radiation scheme and LSM chosen as the baseline, six experiments are conducted to compare different FDDA strategies for the winter case and three FDDA experiments are conducted for the summer case. The FDDA strategies used are 3D and surface analysis nudging, observational nudging, and multiscale FDDA that is a combination of both analysis nudging and observational nudging within a model’s nested-grid framework. Statistical and qualitative analyses are performed to compare the model output from the FDDA experiments with the observations, and the best experiment is then chosen. Analyses are then created to compare the FDDA and no-FDDA experiments for three sub regions within the 4-km domain: the Bay Area, the Sacramento Valley in the northern Central Valley, and the San Joaquin Valley in the southern Central Valley. The incoming marine flow over the Bay Area is examined, as well as the wind flows in the Sacramento and San Joaquin Valleys. The main surface wind flow in both valleys is studied as well as the divergence of the main surface flow during the day and the convergence during the night. Also the upslope and downslope flow along the slopes of the mountains surrounding the Central Valley will be studied. Statistical analysis is used to compare the FDDA and no-FDDA experiments against the assimilated data and also an independent dataset of observations that are not used for data assimilation in order to evaluate the performance of the model. During this study, an objective method for determining the optimal radius of influence (RIN) for use during observational nudging was also explored. Using temperature errors between the observations and the 4-km model at several observation sites correlation coefficients were calculated. With a coefficient threshold subjectively chosen the correlation coefficients for various pairs of sites are compared with the horizontal distance between the pairs of sites. It is shown that a correlation coefficient above 0.5 exists for a pair of observation sites that are 105 km from each other, and the highest correlation occurs at 63 km. For the 4-km domain, the maximum RIN near the surface is defined at 100 km, and a 50-km RIN is used at the surface, similar to the 63-km error correlation length scale with the highest error correlation. It is shown that both RRTM and the RRTMG schemes produce similar errors for all four meteorological fields (relative humidity, temperature, wind direction, and wind speed). The PX LSM is best for the winter case, but no LSM is significantly better for the summer case. The multiscale FDDA strategy produces best fit-to-observation statistics for all four fields. There is an added value of using the special surface wind observations taken by the BAAQMD observation network. Independent verification using withheld other special observations demonstrates the value of FDDA techniques to improve simulations between the assimilated observations.