37 Evaluation of Air Quality Maps Using Cross-Validation …
241
while the optimal analysis reduces the error to 5.4 ppbv for O 3 and 4.6 µg/m
3 for
PM 2.5 . Correspondingly, the fractional bias for the model O 3 and PM 2.5 is respectively
6.5% and 22% and with an optimal analysis reduces to 0.3% for O 3 and 2% for PM 2.5
(see [8, 9]).
37.3 Conclusions
We have developed a framework using cross-validation to evaluate the analysis error
variance and optimize the analysis by obtaining the optimal observational weights.
This method does not require a forecast initialized by the analysis. The method is
quite general and could be used in many different contexts of off-line analysis and
assimilation.
Questions and Answers
Questioner: Shuzhan Ren
Questions: Can this optimal ratio of σ
2
o /σ
2
b be put into an analytical form? Are there
any plans to do data assimilation of concentrations and surface emissions at the same
time?
Answer. Conducting a series of cross-validation analyses for each ratio σ
2
o /σ
2
b is
costly, but we precompute this off-line using monthly statistics. We have not yet
developed a scheme to evaluate the minimum of var(O c -A) by a gradient search, but
this would be useful when we will conduct this procedure in assimilation mode. In
the future, we plan to perform data assimilation to adjust concentrations as well as
emissions. The cross-validation procedure could then be used to obtain the optimal
analysis, while the best forecast could be obtained by adjusting, in addition, the
emissions.
References
1. D.L. Crouse, P.A. Peters, P. Hystad, J.R. Brook, A. van Donkelaar, R.V. Martin, P.J. Villeneuve,
M. Jerrett, M.S. Goldberg, C.A. Pope III, M. Brauer, R.D. Brook, A. Robichaud, R. Ménard,
R.T. Burnett, Associations between ambient PM 2.5 , O 3 , and NO 2 and mortality in the Canadian
Census Health and Environment Cohort (CanCHEC) over a 16-year follow-up. Environ. Health
Perspect. (Open Access) 123, 1180–1186 (2015). https://doi.org/10.1289/ehp.1409276
2. D.L. Crouse, L. Pinault, A. Balram, P. Hystad, P.A. Peters, A. van Donkelaar, R.V. Martin,
R. Ménard, A. Robichaud, P.J. Villeneuve, Urban greenness and mortality in Canada’s largest
cities: a national cohort study. The Lancet, Planet Earth 1, e289–e297 (2017), www.thelancet.
com/planetary-health
241
while the optimal analysis reduces the error to 5.4 ppbv for O 3 and 4.6 µg/m
3 for
PM 2.5 . Correspondingly, the fractional bias for the model O 3 and PM 2.5 is respectively
6.5% and 22% and with an optimal analysis reduces to 0.3% for O 3 and 2% for PM 2.5
(see [8, 9]).
37.3 Conclusions
We have developed a framework using cross-validation to evaluate the analysis error
variance and optimize the analysis by obtaining the optimal observational weights.
This method does not require a forecast initialized by the analysis. The method is
quite general and could be used in many different contexts of off-line analysis and
assimilation.
Questions and Answers
Questioner: Shuzhan Ren
Questions: Can this optimal ratio of σ
2
o /σ
2
b be put into an analytical form? Are there
any plans to do data assimilation of concentrations and surface emissions at the same
time?
Answer. Conducting a series of cross-validation analyses for each ratio σ
2
o /σ
2
b is
costly, but we precompute this off-line using monthly statistics. We have not yet
developed a scheme to evaluate the minimum of var(O c -A) by a gradient search, but
this would be useful when we will conduct this procedure in assimilation mode. In
the future, we plan to perform data assimilation to adjust concentrations as well as
emissions. The cross-validation procedure could then be used to obtain the optimal
analysis, while the best forecast could be obtained by adjusting, in addition, the
emissions.
References
1. D.L. Crouse, P.A. Peters, P. Hystad, J.R. Brook, A. van Donkelaar, R.V. Martin, P.J. Villeneuve,
M. Jerrett, M.S. Goldberg, C.A. Pope III, M. Brauer, R.D. Brook, A. Robichaud, R. Ménard,
R.T. Burnett, Associations between ambient PM 2.5 , O 3 , and NO 2 and mortality in the Canadian
Census Health and Environment Cohort (CanCHEC) over a 16-year follow-up. Environ. Health
Perspect. (Open Access) 123, 1180–1186 (2015). https://doi.org/10.1289/ehp.1409276
2. D.L. Crouse, L. Pinault, A. Balram, P. Hystad, P.A. Peters, A. van Donkelaar, R.V. Martin,
R. Ménard, A. Robichaud, P.J. Villeneuve, Urban greenness and mortality in Canada’s largest
cities: a national cohort study. The Lancet, Planet Earth 1, e289–e297 (2017), www.thelancet.
com/planetary-health