The forecasts of dust optical depth (DOD) are compared with the total aerosol optical depth (AOD) provided by the AERONET network for 42 selected dust-prone stations located in Northern Africa, Middle East and Southern Europe that are listed in Appendix A. Evaluation scores are computed in order to assess the performance of the median multi-model on a monthly, seasonal and annual basis.
Since AERONET retrievals include the contribution of different types of particles, not only mineral dust, it is intended to restrict the comparison to situations in which dust is the dominant aerosol type. Threshold discrimination is made by discarding observations with an Ångström exponent 440-870 higher than 0.6 (Pérez et al., 2006). Although observations with an Ångström exponent 440-870 higher than 0.6 are discarded in order to restrict the evaluation to situations in which mineral dust is the dominant aerosol type, other particles are always present (anthropogenic aerosol, products from biomass burning, etc.). Therefore, negative bias can be expected.
The evaluation system is applied to instantaneous forecast values of DOD ranging from the initial day (D) at 15:00 UTC to the following day (D+1) at 12:00 UTC. Rather than time-interpolated, AERONET observations are assigned to the nearest multiple-of-3 hour. In case more than one observation is assigned to the same hour, only the closest-in-time is considered.
Results
The scores of the evaluation are presented in Tables 5, 6 and 7.
Period
|
Mean Bias
|
Root Mean Square Error
|
Correlation coefficient
|
Fractional Gross Error
|
Number of cases
|
Sep 2012
|
-0,14
|
0,37
|
0,2
|
0,7
|
993
|
Oct 2012
|
-0,12
|
0,31
|
0,41
|
0,81
|
991
|
Nov 2012
|
-0,16
|
0,4
|
0,29
|
0,96
|
730
|
Dec 2012
|
-0,12
|
0,26
|
0,51
|
1,05
|
866
|
Jan 2013
|
-0,11
|
0,23
|
0,61
|
0,99
|
961
|
Feb 2013
|
-0,12
|
0,27
|
0,62
|
0,75
|
1075
|
Mar 2013
|
-0,13
|
0,3
|
0,57
|
0,82
|
1561
|
Apr 2013
|
-0,12
|
0,28
|
0,62
|
0,65
|
1566
|
May 2013
|
-0,15
|
0,33
|
0,57
|
0,62
|
1545
|
Jun 2013
|
-0,11
|
0,31
|
0,58
|
0,46
|
1567
|
Jul 2013
|
-0,08
|
0,31
|
0,45
|
0,41
|
1317
|
Aug 2013
|
-0,14
|
0,36
|
0,35
|
0,52
|
1051
|
Sep 2013
|
-0,12
|
0,28
|
0,48
|
0,63
|
1060
|
Oct 2013
|
-0,1
|
0,31
|
0,49
|
0,69
|
1215
|
Nov 2013
|
-0,09
|
0,27
|
0,4
|
0,83
|
858
|
Dec 2013
|
-0,15
|
0,35
|
0,59
|
0,94
|
713
|
Jan 2014
|
-0,11
|
0,25
|
0,53
|
1,02
|
854
|
Feb 2014
|
-0,12
|
0,27
|
0,7
|
0,88
|
1122
|
Mar 2014
|
-0,14
|
0,32
|
0,63
|
0,75
|
1213
|
Apr 2014
|
-0,14
|
0,31
|
0,47
|
0,7
|
1225
|
May 2014
|
-0,12
|
0,38
|
0,53
|
0,66
|
1168
|
Jun 2014
|
-0,1
|
0,29
|
0,67
|
0,49
|
1218
|
Jul 2014
|
-0,1
|
0,29
|
0,62
|
0,54
|
1055
|
Aug 2014
|
-0,15
|
0,36
|
0,55
|
0,63
|
872
|
Table 5. Monthly scores for the median multi-model
Period
|
Mean Bias
|
Root Mean Square Error
|
Correlation coefficient
|
Fractional Gross Error
|
Number of cases
|
autumn 2012
|
-0,14
|
0,36
|
0,3
|
0,81
|
2714
|
Winter 2012
|
-0,12
|
0,26
|
0,59
|
0,92
|
2916
|
Spring 2013
|
-0,13
|
0,3
|
0,59
|
0,7
|
4672
|
Summer 2013
|
-0,11
|
0,33
|
0,49
|
0,46
|
3965
|
Autumn 2013
|
-0,1
|
0,29
|
0,49
|
0,71
|
3133
|
Winter 2013
|
-0,12
|
0,29
|
0,64
|
0,94
|
2689
|
Spring 2014
|
-0,13
|
0,33
|
0,57
|
0,69
|
3706
|
Winter 2014
|
-0,11
|
0,31
|
0,62
|
0,54
|
3094
|
Table 6. Seasonal scores for the median multi-model
Period
|
Mean Bias
|
Root Mean Square Error
|
Correlation coefficient
|
Fractional Gross Error
|
Number of cases
|
2013
|
-0,13
|
0,30
|
0,56
|
0,67
|
14525
|
Table 7. Annual scores for the median multi-model
The evolution of the scores with time is shown in figures 2 to 5
Figure 2. Monthly (left) and seasonal (right) mean bias of the median multi-model
Figure 3. Monthly (left) and seasonal (right) root mean square error of the median multi-model
Share with your friends: |