Comparison of GEOS-5 CO Forecasts with Satellite Measurements

At the science team meeting Mark Parrington from the University of Edinburgh presented an analysis of the GEOS-5 CO forecasts with satellite measurements during the BORTAS-A campaign last summer. His analysis showed that the predicted locations of elevated concentrations generally compared well with measurements but that the levels of CO were often overestimated.

On the left is the CO predicted by GEOS-5 using the IASI averaging kernels. On the right is the CO retrieved from the IASI satellite measurements.

It can be seen from the above plots that the levels of CO are overestimated by the GEOS-5 prediction relative to the IASI measurments and that the geographical spread of the elevated concentrations appears to be too great.  By plotting the difference between the modelled and measured data we can see more clearly that the model predicts too much CO in most ares (see plots below).

Plot showing the percentage difference between modelled and measured CO for satellite measurements reporting less than 12% cloud cover on the left and less than 5% on the right.

The percentage difference was plotted for measurements made when cloud cover was less than 12% and less than 5% to see if the lower values in the IASI CO measurements were due to cloud or aerosol in the pixels used for the calculation. The plots above show that the modelled values are still significantly higher when less cloud is present. The average bias towards higher values remains constant for different levels of cloud cover as the plot below demonstrates. This shows the probability distribution of the difference between the predicted and measured CO. The average difference is 10 % for both cloud conditions considered (less than 5% and less than 13%) and for all data and boreal only data.

This shows the probability distribution of the difference between the predicted and measured CO concentrations. The dotted lines show this just for the boreal region (> 50^oN) with the red lines showing data with less than 5% cloud cover and the black lines less than 13%.

So what we see is that the model seems to reproduce the locations where CO levels will be elevated but that it consistently shows a bias predicting much higher CO levels than those observed. This is not terrible news since our use of the forecasts is to enable us to be in the right place to observe the biomass burning plumes, characterised by high CO. It looks like the forecasts will be able to tell us where to go, just that we shouldn't expect to see concentrations that are quite as high as the model suggests.