Environmental Engineering Reference
In-Depth Information
Table 1.
Spatial field performance statistics
Spatial field
Number of
monitors
R-squared
Mean bias
(μg/m
3
)
RMSE
(μg/m
3
)
Control case
855
0.98
−0.025
0.43
Fused control
case
855
0.97
0.51
−0.026
In the control cases, the performance statistics between the fused and the
standard interpolation are almost identical. This is expected because data fusion
should have very little impact in grid cells where monitors are located.
4. Cross Validation
Additional sets of interpolated and fused interpolated spatial fields were created in
order to examine interpolation performance in unmonitored areas. These new
spatial fields were created using an abbreviated set of monitors, in which monitors
were removed from the ambient datasets based on distance to the nearest neighbor
monitor. This allowed us to simulate unmonitored areas by removing monitors
from the ambient datasets feeding the spatial fields. Performance for both the
interpolated and fused interpolated spatial field methods was examined by com-
paring the interpolated concentration in the grid cell where the removed monitors
were located to the ambient measured concentration of the removed monitors.
We prepared and ran three sets of cross validation cases. Since we were inter-
ested in examining interpolation performance in unmonitored areas, we chose to
remove monitors that were relatively isolated from all other monitors. We calcu-
lated the distance between monitors and removed all monitors in the domain
whose closest neighbor monitor was >60, >40, and >20 km from the monitor. This
led to the removal of 116, 218, and 589 monitoring sites in the three cases,
respectively.
Table 2
shows the performance statistics for the “40 km” Interpolated and
fused interpolation cases. The performance statistics were calculated for the grid
cells where the monitors were
removed
.
Table 2.
Interpolation performance for the “40 km” case in the grid cells where monitors were
removed
Spatial field
Number of
monitors
removed
R-squared
Mean bias
(μg/m
3
)
RMSE
(μg/m
3
)
40 km interpolation
218
0.72
0.57
1.62
Fused 40 km
interpolation
218
0.72
−0.64
1.62
