Improvements to the representation of BVOC chemistry-climate 1 interactions in UKCA ( vn 11 . 5 ) with the CRI-Strat 2 mechanism : 2 Incorporation and Evaluation 3

We present the first incorporation of the Common Representative Intermediates version 2.2 tropospheric 24 chemistry mechanism, CRI v2.2, combined with stratospheric chemistry, into the global chemistry-climate United 25 Kingdom Chemistry and Aerosols (UKCA) model to give the CRI-Strat 2 mechanism. A rigorous comparison of CRI26 Strat 2 with the earlier version, CRI-Strat, is performed in UKCA in addition to an evaluation of three mechanisms, 27 CRI-Strat 2, CRI-Strat and the standard UKCA chemical mechanism, StratTrop vn1.0, against a wide array of surface 28 and airborne chemical data. 29 30 CRI-Strat 2 comprises a state-of-the-art isoprene scheme, optimised against the MCM v3.3.1, which includes isoprene 31 peroxy radical isomerisation, HOx-recycling through the addition of photolabile hydroperoxy aldehydes (HPALDs) 32 and IEPOX formation. CRI-Strat 2 also features updates to several rate constants for the inorganic chemistry including 33 the reactions of inorganic nitrogen and O(1D). 34 35

surface bias of CS, whose drivers were discussed in Archer-Nicholls et al (2020), and the smaller high bias of ST. On 379 a diel basis, the mechanisms are able to replicate the shape of the diel cycle at the ZF2 site (with similar diel profiles 380 at the ATTO site) but perform less well in Borneo, simulating pronounced diel cycles with a high bias compared to 381 much more muted cycles from observation.

383
An increase of ~1-4 ppb relative to CS is also exhibited by CS2 for monthly mean O3 when both mechanisms are 384 compared to observational data at 10 locations from pole to pole at 4 pressure levels (250, 500, 750 and 900 hPa) (Fig. 385 S4). CS2 reduces the low bias in polar regions but increases CS's high bias in the tropics and Eastern US.

387
Model high biases are also observed from flight data comparisons (Figs. 2(b,f), S6(a)). In the Amazon, where the 388 observed and modelled NO vertical profiles agree well (Fig. S6(e)), there is little difference between the three 389 mechanisms. Each exhibits the greatest high bias at low and a smaller high bias in the free troposphere. CS2 exhibits 390 a high bias of 15-20 ppb for the SEAC 4 RS campaign ( Fig. S6(d))), with perhaps some influence from the low altitude 391 NO2 model high bias. In Borneo, all mechanisms exhibit a roughly consistent high bias of ~20 ppb for ST increasing 392 to 30 ppb for CS2. Interestingly, all the mechanisms simulate a significant low bias for NO2 ( Fig. S6(f)) which may 393 indicate biomass burning events which are not simulated, something which might be expected to promote higher ozone 394 concentrations.

397
Modelled surface OH increases in all locations from ST through CS to CS2 with a significant increase in midday OH 398 from CS to CS2 (Fig. 1). In Borneo, OH is consistently low biased in the three mechanisms but the best comparison 399 is exhibited by CS2 where the mean diel bias compared to ST and CS decreases by 43-50% and 24-40%, respectively 400 over the period considered. The drivers of the HOx change are explored further in Section 5.

402
Surface HO2 was also simulated to increase in all locations from ST to CS to CS2. Significant high bias was simulated 403 in Borneo (the only observational dataset) (Fig. S7) for the CRI mechanisms, including at night. The simulated ratio 404 of HO2 to OH is highly biased in all mechanisms. However, it is best simulated in CS2, indicating that the increase in 405 OH is much larger than that for HO2. It should be noted that none of the mechanisms at present include the 406 heterogenous reactions of HO2 and their inclusion, which will be addressed in future work, should reduce the HO2 407 high bias. explaining the modelled low biases in OH. Indeed, the OH model low bias is greater in the June-July period. This

414
highlights the complexity of model-observation comparisons: the CRI mechanisms may well simulate secondary CO 415 production from isoprene more accurately than ST but other model biases, for example in emissions of CO, NO and 416 isoprene, can lead to the CRI mechanisms appearing worse. Nevertheless, if the CO high bias is reduced in future, we 417 might reasonably assume the modelled OH will improve still further.

420
Modelled isoprene from all three mechanisms was compared to surface observations, flight campaign data and 421 isoprene columns measured by satellite.

424
CS2 yields the best model-observation comparison for surface isoprene on a daily basis in all locations ( Fig. 1 (k-o)).

425
CS2 reduces the high bias in the diel profiles by 50-60% relative to ST and 20-40% to CS at the Z2F, ATTO and 426 Borneo sites, driven by the elevated OH concentrations

428
In most locations the model simulates, to a greater or less extent, a "twin peak" isoprene profile with a sharp rise 429 around 7:00 LT and a second, smaller peak at 19:00 LT. This was most pronounced in the Amazon dry season (ATTO 430 Sept 2013). The morning peak is likely to be due to a combination of the sharp rise in simulated isoprene emissions 431 which starts at 6:00-7:00 am LT, outweighing the concurrent rise in OH, and an underestimation in the model of the 432 rate of BL height growth which can trap isoprene close to the surface, causing a buildup. By contrast, observed 433 isoprene concentrations exhibit a much slower morning growth reaching a peak in early afternoon. While this "out-434 of-phase" behaviour is unlikely to be the sole driver of model-observation difference, it will play a role since isoprene 435 chemistry occurs on the time scale of ~1-2 hours and atmospheric oxidising capacity varies throughout the day.

437
Over the lowest 80 m at the ATTO site, all mechanisms are high biased in the daytime (9:00-15:00) and nighttime 438 (21:00-3:00) ( Fig. S8 (a-d)) with CS2 exhibiting the smallest bias but produce similar isoprene vertical gradients to 439 observations. The effect of boundary layer height was further considered by looking separately at the periods 6:00-440 8:00 LT and 17:00-19:00 LT ( Fig. S8 (e-h)). In contrast to the daytime and nighttime periods, during the 6:00-8:00 441 period the simulated isoprene gradient is significantly more negative than the observation, indicating less vertical 442 mixing and similar results are seen with the MT profile ( Fig. S8 (m-p)). This is most noticeable in September where 443 the largest morning peak is seen in the diel profile for both species and lends support to the theory that the simulated 444 BL height is not increasing as quickly as in reality, leading to more isoprene and MT being trapped at low altitude. accurately simulated than the morning increase.

449
The major drivers of the remaining model-observation difference are likely to be the concentrations of oxidants reduced the model-observation disagreement significantly and attributed the model high bias in their work to high 456 biases in the emissions of isoprene.

459
Model-observation comparisons of isoprene vertical-profiles extending into the boundary layer and into the free 460 troposphere reveal quite a different story from the surface analysis (Fig 2 (a, e, h)).

462
Despite being high biased at the surface and at low altitude, simulated isoprene vertical profiles over the Amazon and campaigns only estimated detection limits (0.1 ppb in both cases) could be used. This has the effect of biasing the 466 median of the observational data to higher values as very low values are ignored. In the SEAC 4 RS campaign, all data 467 points flagged as below the detection limit were set to zero, mitigating this issue. The enhanced oxidative capacity of 468 CS2 at low altitude results in the lowest simulated vertical concentrations among the three mechanisms but the general 469 low bias above the surface is an issue faced by all mechanisms, suggesting it is not just down to modelling of the 470 chemistry.

473
To consider isoprene on a global scale, monthly modelled isoprene columns for all mechanisms are compared to 474 satellite observations from January, April, July and October 2013 (Wells et al., 2020) (Fig. 3).

476
Significant variation in model bias is exhibited between the mechanisms with ST exhibiting the highest isoprene 477 columns and CS2 the lowest. In South America CS2 exhibits the smallest bias while the ST columns are over double 478 the observed values for April and July. CS and CS2 exhibit the smallest biases in Africa and Southeast Asia 479 respectively. The low biases in North America (~ 0.7-1.5×10 15 molecules cm -2 ), Europe (~ 0.5-2.7 ×10 15 molecules 480 cm -2 ) and Central Asia (~ 0.1-1.1 ×10 15 molecules cm -2 ) are quite consistent across the mechanisms and, in some cases 481 almost equal in magnitude to the observed columns, which suggests the bias is driven more by insufficient emissions 482 rather than the chemistry scheme in these locations. and the potential impacts to aerosols. Changes to CO and HCHO are discussed in the SI Section S5.

569
As in Archer-Nicholls et al (2020), the change to O3 was analysed by considering the sum of odd oxygen, NO2 and its 570 reservoir species, termed Ox, and defined in Eq. (1).

575
Tropospheric O3 burden increases by 8% from 328 Tg in CS to 354 Tg in CS2. Much of the free troposphere exhibits 576 increases of 2-6 ppb (~6-14%) in O3 with large parts of the tropical troposphere increasing by more than 4 ppb ( Fig.   577 1). This increase is driven chiefly by a 1.3% decrease in Ox chemical destruction, resulting in an 12% increase in net 578 chemical Ox production. The sensitivity tests (Table S3)    which further degrade producing mostly closed-shell products and HO2. This rapid reaction pathway for RU14O2 sees 631 its burden decrease by 35% in CS2 compared to CS and tropical low altitude mixing ratios decline by over 30%.

632
Similar declines in the RO2 + NO flux (15%) and RO2 burden (33%) are seen for CS2 relative to the CS2_isoprene when CS2 is compared to ST which simulates even lower low altitude OH than CS (