1. Manfred Birk, Georg Wagner Remote Sensing Technology Institute (IMF) Deutsches Zentrum für Luft- und Raumfahrt (DLR) Lorenzo Lodi, Jonathan Tennyson Department of Physics and Astronomy University College London Water intensities: ab initio versus experiment

2. Introduction Routes to line intensities in spectroscopic databases Minimal approach: Line intensities are derived from laboratory spectroscopy measurements Disadvantage: not all lines included, precision limited by noise Effective Hamiltonian approach: Quantum mechanical data reduction of experimental line intensities Only useful when experimental data are reproduced within their precision Advantage: Intensities for lines not in the experiment can be calculated when avoiding large extrapolation towards higher quantum numbers Further advantages: Noise of experimental line intensities is reduced, experimental line intensities are checked for consistency Pure ab initio approach: Semi-empirical PES, purely ab-initio DMS Advantage: All transitions covered Disadvantage: Experimental validation mandatory

3. How accurate are ab initio calculations? Intercomparison between ab initio and experimental line intensities required But: Experimental line intensities from different labs often differ outside error margins (see below) Thus: High quality experimental data with known uncertainty needed Best case: At least two experimental data sets from independent labs agree within error margins

4. Ab initio calculation of H 2 O line intensities Lorenzo Lodi, Jonathan Tennyson, and Oleg L. Polyansky, “A global, high accuracy ab initio dipole moment surface for the electronic ground state of the water molecule”, J. Chem. Phys. 135, 034113 (2011). Quality estimate: 2 different PES and DMS  4 combinations  fractional ab initio uncertainty = (largest value/smallest value -1) The authors state line intensity errors of 1% for most lines validated by average agreement with HITRAN 2008 (ab initio/HITRAN=1.01±4.5%, S>1e- 22,11% of HITRAN lines with J<13) and agreement of 15 lines measured with CRDS by Lisak and Hodges (NIST) (ab initio/NIST=1.004±0.6%) But: Systematic line intensity errors in HITRAN 2008 S>1e-22 are very likely Example: Update 2004  2008 2 line intensities changed up to 6% (see below). Other regions??? Thus, agreement of 1% between ab initio and HITRAN is not very conclusive This work: Intercomparison of ab initio and high quality experimental water line intensities

5. DLR measurements – strategy Goal: Accurate data with defined error margins Redundancy is important since hidden systematic errors may depend on line width and optical depths. Chi tests and investigation of residuals of redundant data may help to quantify/identify error sources Line intensities retrieved from many measurements with different optical depths (<4) for redundancy Combination of pure water and air-broadened measurements used for increasing redundancy (width and optical depth independently selectable) Influence of instrumental lineshape function is minimized by choosing high resolution Mostly ambient temperature measurements used

6. DLR measurements – experimental set-up Water/air mixturesPure water

7. DLR measurements – 1 µm region Multireflection cell at 85 m High signal-to-noise by narrow band pass (10000-11000 cm -1 ) Double-sided interferograms Line intensity analysis included ambient and non-ambient temperature measurements Line fitting of individual spectra on micro window basis applying speed-dependent Voigt profile yielding line intensities for each measurement and transition

8. DLR measurements – 1 µm region Line intensities of up to 11 measurements averaged Reference: 1 mb pure water measurement at ambient temperature – 5 mb pure water measurement not used as reference due to insufficient spectral resolution Individual measurements (except reference) scaled for 0% mean measurement omc in averaging Quality check: Temperature/scaling factor (S ref x scaling factor = S meas ) fit from line intensities of individual measurements using averaged line intensities as reference Air-broadened measurements: 7 scaling factors less than 1% off 1, 2 scaling factors between 1 and 2%

9. Linestrength intercomparison in 1 µm region NIST: cavity ringdown by Daniel Lisak and Joseph T. Hodges HIT: HITRAN 2008, mainly experimental data by Robert A. Toth Excellent agreement DLR-NIST, mostly <1% HITRAN 2008 shows bias and large scatter

10. Ab initio vs. experiment 1 µm Only transitions shown with experimental precision < 1% Vib transition# of linesMean diff./%Scatter about mean/  meas Mean(  ab initio )/% 1 2 1  0 0 0 78+4.01.97.2 2 0 1  0 0 0 189-0.31.62.1 3 0 0  0 0 0 76+4.32.32.0 1 0 2  0 0 0 17-8.91.47.1 0 0 3  0 0 0 12-0.71.11.2

11. Ab initio vs. experiment 1 µm Bias for entire vibrational bands, cannot be related to experimental error Mean differences of bands up to 9% Larger scatter for 3 0 0  0 0 0 band Isolated outliers with up to 30% difference (see below) Average ab initio uncertainties mostly conservative (exception 3 0 0  0 0 0)

12. Ab initio vs. experiment 1 µm Largest difference for 2 0 1  0 0 0, 12 0,12  11 0,11 S ab initio - S exp = -26.5%,  Smean = 0.6%,  =1.5,  Sab initio = 1.4% P H2O /mbP tot /mbT/K% peak abs  S /% (S-S mean )/% (S-S mean )/  S In av. 4.0199.6296.015.70.64-0.59-0.931 4.0496.8296.513.02.27-0.06-0.031 16.1998.2315.936.310.382.110.201 4.0998.0316.410.72.575.422.001 3.9997.7277.66.12.845.811.931 2.0498.9317.38.51.70-8.63-5.560 2.0501.8277.64.93.6410.472.601 0.8199.3275.92.64.38-3.48-0.821 1.0 295.24.72.46-2.23-0.931

13. Ab initio vs. experiment 1 µm Subband J’’+1 0,J’’+1  J’’ 0,J’’ Ab initio: Resonance at J’’=6, experiment: resonance at J’’=11 Apparently, energy level of resonating states not correctly predicted from PES J‘‘ (  Sab initio )/% (S ab initio – S exp )/% 04.00.0 21.80.3 41.70.1 65630.30.3 78.70.2 81.6-0.6 91.5-0.8 111.4-26.5

14. H 2 16 O linestrength intercomparison in 2 region HIT04: HITRAN 2004, mainly experimental data by Robert A. Toth Lodi: ab initio calculations DLR: 9 pure water, 16 air- broadened measurements, ambient temperature, Voigt profile analysis HITRAN 2004 – DLR differences up to 6% for strong and weak lines Ab initio – DLR agreement 1e-23 010  000, 020  010 Ab initio – DLRHITRAN 2004 - DLR lg(Smin) lg(Smax) %% %unc  ndata %% %unc  ndata -24.5 -24.0 3.660.80235.900.7924 -24.0 -23.5 4.230.23775.460.2475 -23.5 -23.0 2.120.10943.560.1095 -23.0 -22.5 1.190.041051.870.04106 -22.5 -22.0 0.760.021090.910.02108 -22.0 -21.5 0.740.02720.270.0273 -21.5 -21.0 0.850.0250-1.520.0153 -21.0 -20.5 1.130.0252-2.630.0150 -20.5 -20.0 1.140.0251-2.730.0250 -20.0 -19.5 1.620.0251-3.800.0151 -19.5-19.01.740.0153-5.790.0155

15. H 2 18 O linestrength intercomparison in 2 region HIT04: HITRAN 2004, mainly experimental data by Robert A. Toth Lodi: ab initio calculations % Lodi-DLR Ab initio - DLRHITRAN 2004 - DLR lg(Smin) lg(Smax) %% %unc  ndata %% %unc  ndata -24.5 -24.0 2.841.09124.041.1411 -24.0 -23.5 3.820.32443.270.3245 -23.5 -23.0 1.000.1353-1.200.1353 -23.0 -22.5 0.280.0650-3.300.0653 -22.5 -22.0 0.160.0353-4.750.0355 -22.0 -21.5 0.270.0247-5.110.0242 -21.5-21.00.630.0317-5.430.0317

16. Brand new quick-look 3 results obtained by our PhD student Joep Loos Only single pure water measurement analyzed Only 21 transitions, line intensity >1e-19 Ab initio – experiment: mean -0.96%, scatter 0.25% HITRAN 2008 – experiment: mean -2.44%, scatter 0.39%

17. Conclusion Good agreement of ab initio and DLR experimental line intensities: H 2 16 O, 2, S>1e-23: <2% H 2 18 O, 2, S>1e-23: <2% H 2 16 O, 3, S>1e-19: <1% (preliminary) H 2 16 O, 1 µm region, S>1e-22: <2% Agreement of ab initio and DLR data much better than with older HITRAN versions Good agreement of ab initio and DLR data indicates reliability of both data sources But there exist overtones and combination bands with biases between ab initio and experiment up to 9%. The origin for the differences can be unambiguously attributed to the ab initio calculation Furthermore, singular large differences up to 30% (resonances) and vibrational transition specific scatter related to the ab initio calculations exist Ab initio uncertainties were found to be helpful in assessing data quality. In cases of resonances they may be misleading Further intercomparison between high quality laboratory measurements and ab initio calculations are required