SCILOV-10Validation of SCIAMACHY limb

operational BrO product

F. Azam, K. Weigel, A. Rozanov, M. Weber, H. Bovensmann and J. P. Burrows

ESA/ESRIN, Frascati, Italy27-02-2014

Contents: SCIAMACHY ESA vs IUP (datasets)

Validation Strategy

BrO Inter-comparisons

Conclusion/Outlook

1

ESA /DLR vs IUP BrO: main retrieval differences

2

ESA/DLR IUPProcessors: Speed Optimized Precision optimizedPre-processing no yes auxiliary spectral fits for each tangent height independently, (improves quality of spectra)Spectral range 337–357 nm 338-356.2 nm

See for details; Rozanov et al, Atmos. Meas. Tech., 4, 1319–1359, 2011

Regularization Optimal regularization weak statistical parameter using L- curve regularization method (smoothness constrain)

Climatology/a priori Constant a priori Latitude dependent in vmr climatological information

* L-curve: too strong regularization with deteriorating vertical resolution and large smoothing errors

Validation StrategySCIAMACHY limb coverage:

Profiles/day:1500 profiles for Aug 2002 - Apr 2012: above 4.5 million (IUP SCIAMACHY BrO ¼ of the amount, the profiles are retrieved as an average of the four azimuth)

Data versions: ESA/DLR BrO version 5.02 and IUP version 3.2

Sub-sampling:ESA SCIAMACHY Sub-sampling (allows for faster computation):

Distance between two profiles is set larger than 5000 km

A profile is not allowed in the same 5° latitude band as any of 26 profiles before

Each latitude band is limited to 20% more profiles than the average

3

Validation StrategySubsampling results in 3% of the entire datasets well distributed over all latitudes, longitudes and time

ESA – IUP Collocation criteria: time = 0.001 h, distance = 1000 km One randomly chosen profile from each state of ESA is compared to the single IUP profile.

4

6

ESA/DLR v5.02 –IUP v3.2 Profile comparisons

Tropics

Near global

NH mid lat.

SH mid lat.

NH high lat.

SH high lat.

meanrelative

differences

7

Annual cycle and Time seriesTropics

8

NH Time series

9

SH Time series

Profile comparisons have large standard deviation. Very large differences

(30-40%) observed for the high lat.

BrO seasonal cycle from the two dataset seems to be anti-correlated

Probable factors contributing to differences;

i)- different climatologies used in the ESA/DLR and IUP retrieval ii)- differences in regularization used as smoothness constrain

MLS, OSIRIS and SMILES also provide BrO data but that could not be used

for comparisons

MLS: no altitude overlap, BrO measurements start around 30 km OSIRIS: provides BrO amounts as daily zonal means and SMILES: time span is too short, Oct.2009 – Apr. 2010

9

Large differences cannot be accounted for by the statistics used

Data quality of the ESA/DLR and IUP datasets were investigted by examining the averaging kernels and the measurment content of the retrieval

Averaging kernels: Give the sensivity of the retrieval to the true state

Measurement content:The degree of contribution of the retrieval to the true state. Calculated as an integral of the rows of the averaging kernels, when it is 1, no contribution from a priori.

Resolution:The width of the averaging kernel provides information on the vertical resolution

10

11

Tropics (Jan)

BrO Averaging Kernels:

ESA

IUP

12

BrO Averaging Kernels:

Arctic (Jan)

ESA

IUP

13

BrO Averaging Kernels:

Antarctic (Jun)

ESA

IUP

14

Conclusions/Outlook ESA/DLR BrO v5.02 compared with IUP data version 3.2

Profiles comparisons showed very large differences (30-40%) for the high lat. between the two SCIAMACHY datasets

Time series showed clear seasonality in differences for mid and high lat.

Diagnostic study carried out on the quality of the datasets, investigating their averaging kernels and the measurmenent content

ESA/DLR retrieval has the maximum of the retrieval sensitivity ~ 25 km where less than 50% of the information comes from the measurements.

IUP BrO retrieval sensitivity has its maximum around 20 km where the measurement contributes to 60-65% of the retrieval.

15

Outlook/Recommendations For ESA/DLR limb BrO, retireval should be precision optimized, care with the

climatology used in the retrieval, S/N ratio needs to be improved.

Extra Slides

ResultsProfile comparisons: mean relative differences plots with the standard deviation of the bias corrected differences for 20-35 km

Annual cycles:annual cycle vs altitude plots as monthly mean absolute amounts, monthy mean percental difference and the monthly mean percental differences for selected altitudes (20, 24, 27 and 31 km)

Time Series:compared for 20–35 km on a monthly grid. For the selected altitudes (20, 24, 27 and 31 km), comparisons carried out on 30 days running averages if more than 10 collocations are found

Tropics (Jun)

ESA

IUP

L-curve methodA compromise between precision and resolution

Produces too strong regularization with deteriorating vertical resolution and large smoothing errors

4