Environmental Engineering Reference
In-Depth Information
2. The Assimilation Experiment
The assimilation experiment covered a period of 2 weeks starting at February 8,
2006, representing typical winter conditions. The period was divided into assimilation
windows of 24 h. The species included in the run were sulphur dioxide (SO
2
) and
sulphates SO
4
.
The computations were performed using the SILAM chemistry-transport model
and ECMWF meteorological fields. A description of the model and its evaluation
against data from the European Tracer Experiment (ETEX) is given by [2].
Furthermore, the Eulerian dynamical core, based on a semi-Lagrangian advection
scheme [3], was employed in this work. The discretisation of vertical diffusion
follows the extended resistance analogy scheme described in [4]. Oxidation of
SO2 to sulphates was computed using the linear parametrisation described in [5].
Adaptation of the 4D-VAR assimilation method in chemistry-transport modelling
has been described in detail for instance in [1] and [6]. In this work, emission is
used as a control variable by defining the emission rate as
E
(
x
,
t
)=
α
(
x
)
E
0
(
x
,
t
)
(1)
where α is the correction factor, and
E
0
(
x
,
t
) is the a priori (background) value for
emission. A standard quadratic cost function is used, with control variable being
the pair (
α
(
x
),
c
0
(
x
)), where
c
0
(
x
) is the concentration field at
t
= 0. Given the
solution
c
(
x
,
t
) of the adjoint problem, the sensitivity of the cost function with
respect to
α
is obtained by integrating
c
(
x
,
t
)
E
0
(
x
,
t
) over the assimilation window.
The background covariance matrices for
c
0 and a were assumed to be constant
and diagonal, with σ
2
α
= 10.0 (relative units) and σ
2
c
0
= 10
−12
mol m
−3
.
2.1. Observations
The set of measurements used in assimilation consists of hourly concentrations of
SO2 at total of 456 central European stations made available through the Air-
basedatabase (www.eea.europa.eu) maintained by European Environmental Agency.
For grid cells containing several monitoring stations, one was selected randomly.
This limits the number of stations to 260, and consequently, the number of data
points inside the 24 h assimilation window to ~6,200.
Furthermore, a control set of 50 stations located on the same region was chosen
randomly to evaluate the performance of the assimilation system. Because of the
difficulties in collecting accurate information on the measurement errors in the
dataset, the measurement errors are in this study prescribed with a constant standard
deviation
= 10
−7
mol m
−3
.
σ
