forecasts based on atmospheric chemistry and transport models are used widely by
government departments as part of air-quality forecast operations. However, in
spite of increasingly advanced dynamic, physical and chemical processes within such
models, forecast errors are always found during real forecast applications.
changing the model itself, two technologies can be applied to improve the
forecast skill by using available observations of atmospheric constituents. The
first is data assimilation (DA), which focuses on the model’s input. By mixing
the model’s initial chemical condition with corresponding observations, DA can generate
more accurate input data for the model to yield a better forecast. The second is
model output statistics (MOS), which applies to the model’s output. It corrects
the model’s input according to the statistic relation found by comparing past
model outputs and their corresponding observations.
order to investigate the performance of these two technologies in improving model
forecasts, Chaoqun MA, a Ph.D. student from the research group led by Professor
Tijian WANG in the School of Atmospheric Sciences, Nanjing University, implemented
three-dimensional variational (3DVar) DA and MOS based on one-dimensional
Kalman filtering, separately and simultaneously, during a forecast experiment
in Hebei Province, China. The results, published in Advances in Atmospheric Sciences, showed that MOS is more effective
and durable than DA.
authors explain their results as follows:
chemistry models are not usually sensitive to their initial chemical conditions,
and therefore the better input generated by DA will soon lose its effect during
the model run. However, unlike DA, MOS retains its effect during the forecast,
which is why MOS fairs better than DA in the comparison.”
Chaoqun MA goes on to explain that “Considering the combined use of DA and MOS
shows little improvement, and DA is usually more computationally complex, it is
recommended to try MOS as the first option when seeking to improve air-quality
forecasts. However, given the 3Dvar DA tested is a rather simple DA approach,
it is possible a different conclusion might be reached after applying more
advanced DA algorithms.”
Ma, C. Q., T. J. Wang, Z. L. Zang, and Z. J. Li, 2018: Comparisons of
three-dimensional variational data assimilation and model output statistics in
improving atmospheric chemistry forecasts. Adv.
Atmos. Sci., 35(7), https://doi.org/10.1007/s00376-017-7179-y.