The air quality forecast rote: Recent changes and future challenges

Evaluation of the offline-coupled GFSv15-FV3-CMAQv5.0.2 in support of the next-generation National Air Quality Forecast Capability over the contiguous United States

As a candidate for the next-generation National Air Quality Forecast Capability (NAQFC), the meteorological forecast from the Global Forecast System with the new Finite Volume Cube-Sphere dynamical core (GFS-FV3) will be applied to drive the chemical evolution of gases and particles described by the Community Multiscale Air Quality modeling system. CMAQv5.0.2, a historical version of CMAQ, has been coupled with the North American Mesoscale Forecast System (NAM) model in the current operational NAQFC. An experimental version of the NAQFC based on the offline-coupled GFS-FV3 version 15 with CMAQv5.0.2 modeling system (GFSv15-CMAQv5.0.2) has been developed by the National Oceanic and Atmospheric Administration (NOAA) to provide real-time air quality forecasts over the contiguous United States (CONUS) since 2018. In this work, comprehensive region-specific, time-specific, and categorical evaluations are conducted for meteorological and chemical forecasts from the offline-coupled GFSv15-CMAQv5.0.2 for the year 2019. The forecast system shows good overall performance in forecasting meteorological variables with the annual mean biases of -0.2 °C for temperature at 2 m, 0.4% for relative humidity at 2 m, and 0.4 m s-1 for wind speed at 10 m compared to the METeorological Aerodrome Reports (METAR) dataset. Larger biases occur in seasonal and monthly mean forecasts, particularly in spring. Although the monthly accumulated precipitation forecasts show generally consistent spatial distributions with those from the remote-sensing and ensemble datasets, moderate-to-large biases exist in hourly precipitation forecasts compared to the Clean Air Status and Trends Network (CASTNET) and METAR. While the forecast system performs well in forecasting ozone (O3) throughout the year and fine particles with a diameter of 2.5 μm or less (PM2.5) for warm months (May-September), it significantly overpredicts annual mean concentrations of PM2.5. This is due mainly to the high predicted concentrations of fine fugitive and coarse-mode particle components. Underpredictions in the southeastern US and California during summer are attributed to missing sources and mechanisms of secondary organic aerosol formation from biogenic volatile organic compounds (VOCs) and semivolatile or intermediate-volatility organic compounds. This work demonstrates the ability of FV3-based GFS in driving the air quality forecasting. It identifies possible underlying causes for systematic region- and time-specific model biases, which will provide a scientific basis for further development of the next-generation NAQFC.

Development and application of a hybrid long-short term memory - three dimensional variational technique for the improvement of PM2.5 forecasting

The current state-of-the-art three-dimensional (3D) numerical model for air quality forecasting is restricted by the uncertainty from the emission inventory, physical/chemical parameterization, and meteorological prediction. Forecasting performance can be improved by using the 3D-variational (3D-VAR) technique for assimilating the observation data, which corrects the initial concentration field. However, errors from the prognostic model cause the correction effects at the first hour to be erased, and the bias of the forecast increases relatively fast as the simulation progresses. As an emerging alternative technique, long short-term memory (LSTM) shows promising performance in air quality forecasting for individual stations and outperforms the traditional persistent statistical models. In this study, a new method was developed to combine a 3D numerical model with 3D-VAR and LSTM techniques. This method integrates the advantage of LSTM, namely its high-accuracy forecasting for a single station and that of the 3D-VAR technique, namely its ability to extend improvement to the whole simulation domain. This hybrid method can effectively improve PM_2.5 forecasting for the next 24 h, relative to forecasting with the 3D-VAR technique which uses the initial hour concentration correction. Results showed that the root-mean-square error and normalized mean error were decreased by 29.3% and 33.3% in the validation stations, respectively. The LSTM-3D-VAR method developed in this study can be further applied in other regions to improve the forecasting of PM_2.5 and other ambient pollutants.

Preliminary Tests on the Sensitivity of the FORAIR_IT Air Quality Forecasting System to Different Meteorological Drivers

Since 2017, the operational high-resolution air quality forecasting system FORAIR_IT, developed and maintained by the Italian National Agency for New Technologies, Energy and Sustainable Economic Development, has been providing three-day forecasts of concentrations of atmospheric pollutants over Europe and Italy, on a daily basis, with high spatial resolution (20 km on Europe, 4 km on Italy). The system is based on the Atmospheric Modelling System of the National Integrated Assessment Model for Italy (AMS-MINNI), which is a national modelling system evaluated in several studies across Italy and Europe. AMS-MINNI, in its forecasting setup, is presently a candidate model for the Copernicus Atmosphere Monitoring Service’s regional production, dedicated to European-scale ensemble model forecasts of air quality. In order to improve the quality of the meteorological input into the chemical transport model component of FORAIR_IT, several tests were carried out on daily forecasts of NO2 and O3 concentrations for January and August 2019 (representative of the meteorological seasons of winter and summer, respectively). The aim was to evaluate the sensitivity to the meteorological input in NO2 and O3 concentration forecasting. More specifically, the Weather Research and Forecasting model (WRF) was tested to potentially improve the meteorological driver with respect to the Regional Atmospheric Modelling System (RAMS), which is currently embedded in FORAIR_IT. In this work, the WRF chain is run in several setups, changing the parameterization of several micrometeorological variables (snow, mixing height, albedo, roughness length, soil heat flux + friction velocity, Monin–Obukhov length), with the main objective being to take advantage of WRF’s consistent physics in the calculation of both mesoscale variables and micrometeorological parameters for air quality simulations. Daily forecast concentrations produced by the different meteorological model configurations are compared to the available measured concentrations, showing the general good performance of WRF-driven results, even if performance skills are different according to the single meteorological configuration and to the pollutant type. WRF-driven forecasts clearly improve the model reproduction of the temporal variability of concentrations, while the bias of O3 is higher than in the RAMS-driven configuration. The results suggest that we should keep testing WRF configurations, with the objective of obtaining a robust improvement in forecast concentrations with respect to RAMS-driven forecasts.

Explore a Multivariate Bayesian Uncertainty Processor driven by artificial neural networks for probabilistic PM2.5 forecasting

Quantifying predictive uncertainty inherent in the nonlinear multivariate dependence structure of multi-step-ahead PM_2.5 forecasts is challenging. This study integrates a Multivariate Bayesian Uncertainty Processor (MBUP) and an artificial neural network (ANN) to make accurate probabilistic PM_2.5 forecasts. The contributions of the proposed approach are two-fold. First, the MBUP can capture the nonlinear multivariate dependence structure between observed and forecasted data. Second, the MBUP can alleviate predictive uncertainty encountered in PM_2.5 forecast models that are configured by ANNs. The reliability of the proposed approach was assessed by a case study on air quality in Taipei City of Taiwan. We consider forecasts of PM_2.5 concentrations as a function of meteorological and air quality factors based on long-term (2010-2018) hourly observational datasets. Firstly, the Back Propagation Neural Network (BPNN) and the Adaptive Neural Fuzzy Inference System (ANFIS) were investigated to produce deterministic forecasts. Results revealed that the ANFIS model could learn different air pollutant emission mechanisms (i.e. primary, secondary and natural processes) from the clustering-based fuzzy inference system and produce more accurate deterministic forecasts than the BPNN. The ANFIS model then provided inputs (i.e. point estimates) to probabilistic forecast models. Next, two post-processing techniques (MBUP and the Univariate Bayesian Uncertainty Processor (UBUP)) were separately employed to produce probabilistic forecasts. The Bayesian Uncertainty Processors (BUPs) can model the dependence structure (i.e. posterior density function) between observed and forecasted data using a prior density function and a likelihood density function. Here in BUPs, the Monte Carlo simulation was introduced to create a probabilistic predictive interval of PM_2.5 concentrations. The results demonstrated that the MBUP not only outperformed the UBUP but also suitably characterized the complex nonlinear multivariate dependence structure between observations and forecasts. Consequently, the proposed approach could reduce predictive uncertainty while significantly improving model reliability and PM_2.5 forecast accuracy for future horizons.

On the Limit to the Accuracy of Regional-Scale Air Quality Models

Regional-scale air pollution models are routinely being used world-wide for research, forecasting air quality, and regulatory purposes. It is well recognized that there are both reducible (systematic) and irreducible (unsystematic) errors in the meteorology-atmospheric chemistry modeling systems. The inherent (random) uncertainty stems from our inability to properly characterize stochastic variations in atmospheric dynamics and chemistry, and from the incommensurability associated with comparisons of the volume-averaged model estimates with point measurements. Because these stochastic variations are not being explicitly simulated in the current generation of regional-scale meteorology-air quality models, one should expect to find differences between the model estimates and corresponding observations. This paper presents an observation-based methodology to determine the expected errors from current generation regional air quality models even when the model design, physics, chemistry, and numerical analysis, as well as its input data, were "perfect". To this end, the short-term synoptic-scale fluctuations embedded in the daily maximum 8-hr ozone time series are separated from the longer-term forcing using a simple recursive moving average filter. The inherent uncertainty attributable to the stochastic nature of the atmosphere is determined based on 30+ years of historical ozone time series data measured at various monitoring sites in the contiguous United States. The results reveal that the expected root mean square error at the median and 95^th percentile is about 2 ppb and 5 ppb, respectively, even for "perfect" air quality models driven with "perfect" input data. Quantitative estimation of the limit to the model's accuracy will help in objectively assessing the current state-of-the-science in regional air pollution models, measuring progress in their evolution, and providing meaningful and firm targets for improvements in their accuracy relative to ambient measurements.

Introducing of modeling techniques in the research of POPs in breast milk - A pilot study

This study used advanced statistical and machine learning methods to investigate organochlorine pesticides (OCPs) and polychlorinated biphenyls (PCBs) in breast milk, assuming that in a complex biological mixture, the pollutants emitted from the same source or with similar properties are statistically interrelated and possibly exhibit non-linear dynamics. The elaborated analyses such as Unmix source apportionment characterized individual source groups, while guided regularized random forest indicated the pollutant dependence on the ortho-chlorine atom attached to the congener's phenyl ring and mother's age. Mutual associations among PCBs were further discussed, but the results implied they were mostly not related to child delivery. PCB congeners -153, -180, -170, -118, -156, -105, and -138 appeared to be compounds of the outmost importance for mutual prediction with reference to their interrelations regarding chemical structure and metabolic processes in the mother's body. Finally, machine learning methods, which provided prediction relative errors lower than 30% and correlation coefficients higher than 0.90, suggested a possible strong non-linear relationship among the pollutants and consequently, the complexity of their pathways in the breast milk.