UKESM1 is an Earth system model that couples individual component models which simulate the ocean, land surface, atmosphere and cryosphere (Sellar et al., 2020). Each component can also be run on their own (so-called “stand-alone”). The two components of relevance for this study are the land surface model (Joint United Kingdom Land Environment Simulator – JULES) (Best et al., 2011; Clark et al., 2011) and the atmospheric chemistry and aerosol model (United Kingdom Chemistry and Aerosols – UKCA; Archibald et al., 2020). In JULES the land surface is described by dividing it into categories which can be grouped as vegetation (trees, grasses and shrubs) and non-vegetation (urban, bare soil, water, ice) (Sellar et al., 2020). Depending on the configuration, there are between 5 and 13 types of vegetation, termed plant functional types (PFTs). Emissions of isoprene and terpenes are calculated using iBVOC (Pacifico et al., 2011), a module within JULES that reads in the simulated land surface. When running as part of the fully coupled UKESM1, emissions from iBVOC in JULES are passed to UKCA, which simulates their addition to the atmosphere. When UKESM1 is run in atmosphere-only mode where vegetation cover is prescribed (along with sea surface temperatures, sea ice and ocean biogeochemistry), iBVOC can be used to calculate BVOC emissions from the prescribed vegetation and pass these emissions to UKCA. The latter configuration is used in this study.

Each PFT in UKESM1 has an associated emission factor (EFmass) for isoprene (IEFmass) and terpenes (TEFmass) with units of mass of emitted carbon per leaf dry weight (dw) per hour (µgC gdw-1 h-1). These EFmass values represent the emission flux for a given PFT under the standard conditions specified in Pacifico et al. (2011) (30 ∘C, 1000 µmol m-2 s-1 of photosynthetically active radiation (PAR), 370 ppm CO2). We note that subscript “mass” is used here to distinguish these emission factors from those with dimensions of mass per unit area of land surface per unit time (e.g. µg msurface-2 h-1), which we denote as EFarea and which are used in the MEGAN v2.1 scheme discussed later.

In iBVOC, EFmass are combined with other PFT-specific parameters, including photosynthetic activity and external variables including temperature and CO2 concentration, to calculate emissions of BVOCs per PFT per grid cell. The dependencies on temperature, CO2 concentration and photosynthetic activity are given in Pacifico et al. (2011).

iBVOC was first implemented with the original five-PFT set-up in UKESM1 which divides vegetated regions into the categories of broadleaf trees, needleleaf trees, C3 grass, C4 grass and shrubs. IEFmass values (Table 1) and the standard conditions were taken from Guenther et al. (1995).

When running JULES stand-alone over the period 1990–1999, these IEFmass values yielded simulated total isoprene emissions of 535 TgC yr-1 (606 Tgisoprene yr-1), with 9 % coming from C4 grass (Pacifico et al., 2011).

To improve land surface modelling, configurations of JULES with 9 and 13 PFTs were developed with the 13-PFT approach, the current standard in UKESM1 (Table 2) and the configuration used for UKESM1's contributions to CMIP6 (Sellar et al., 2020). Going from the 5-PFT configuration to the 13-PFT configuration, EFmass values were assigned partly from those used in the 5-PFT configuration (e.g. the 13-PFT broadleaf deciduous tree category has the same IEFmass as the 5-PFT broadleaf tree category) and partly from the Organizing Carbon and Hydrology in Dynamic Ecosystem (ORCHIDEE) vegetation scheme (Lathière et al., 2006). Unlike the 5-PFT configuration, the 13-PFT configuration does not appear to have been separately validated against observations or other model estimates, and furthermore, the IEFmass of C4 grass was increased from 8 to 24 µgC gdw-1 h-1 (Table 2).

In the context of this study, the limitation with using ORCHIDEE-derived EFmass values for the 13-PFT configuration in UKESM1 is that the ORCHIDEE scheme does not simulate shrubs as a separate PFT. Rather, the IEFs from shrubs are incorporated into the IEF for C3 and C4 grass ORCHIDEE PFTs. This means that the C3 and C4 grass PFTs in ORCHIDEE are not equivalent to those in UKESM1 and should not be used to provide the IEF values.

Lathière et al. (2006) noted that ORCHIDEE considers high IEF values for grasses and also acknowledged the high degree of uncertainty in this area, as several other studies have found low emissions of isoprene from grasses and that a change to these values would lead to different regional distributions of emissions, a topic explored in Sect. 4.

In the updated version of ORCHIDEE, Messina et al. (2016) also note the inclusion of shrubs in the EF values for the grass PFTs in ORCHIDEE, and it remains unclear whether the ORCHIDEE values for C3 and C4 grass are composed totally or only partially of the EFmass from shrubs. Nevertheless, as UKESM1 simulates deciduous and evergreen shrubs as separate PFTs with their own emission factors, including the IEFmass of shrubs in those for grasses is not correct.

Furthermore, as shrubs are relatively strong isoprene emitters (e.g. Lathière et al., 2006; Guenther et al., 2012) and C4 grasses are not (e.g. Guenther et al., 2012; Loreto and Fineschi, 2015), this approach artificially increases the isoprene production potential from the UKESM1 C4 grass PFTs. This is exacerbated by the fact that large swathes of C4 grassland are in warm regions (e.g. sub-Saharan Africa and eastern Brazil), further increasing isoprene production given its strong temperature dependence (for example, isoprene emissions are 35 % higher in iBVOC at 28 ∘C than 25 ∘C). Shrubs by contrast are typically found in higher-latitude regions where the lower temperatures lead to lower isoprene emissions, despite the relatively high IEF.

The Community Earth System Model version 2 (CESM2) is another Earth system model which includes atmospheric, land, ocean and sea ice models that can be run in stand-alone or coupled configurations (Danabasoglu et al., 2020). The land model component is the Community Land Model version 5 (CLM5) (Lawrence et al., 2019), which also simulates BVOC emissions based on prevailing atmospheric conditions and land surface cover using MEGAN v2.1. The development of MEGAN is described in Guenther et al. (1995, 2006, 2012). Like iBVOC, MEGAN v2.1 includes parameterisations for dependencies on temperature, CO2 and PAR while also describing the impact of leaf age and soil moisture. A full description of the parameterisations is given in Guenther et al. (2012). CLM5 has 16 types of natural vegetation (including bare ground) and eight active crops. Similarly to JULES, vegetation and crops are represented by PFTs, each having specific ecophysiological, phenological and biogeochemical parameters (Lawrence et al., 2019). MEGAN v2.1 combines these parameterisations with PFT-specific emission factors (which for MEGAN have units of µg msurface-2 h-1) to calculate BVOC emissions for a range of BVOCs. Furthermore, unlike ORCHIDEE, MEGAN v2.1 considers grasses and shrubs separately, with emission factors for each. This means that the MEGAN EFs for C3 and C4 grasses are more suitable as a starting point for calculating EF values for iBVOC.

In this study, we use the MEGAN v2.1 EF (Table 3 of Guenther et al., 2012) as it offers an alternative source of EFs. We note that the same EFs for isoprene are used in the more recently released version MEGAN v3.0 (e.g. Zhang et al., 2021). MEGAN v2.1 in CESM2 considers 15 PFTs (excluding bare soil), so we had to lump certain PFTs during the conversion to IEF for iBVOC to match the 13-PFT classification in JULES. Table 3 shows the MEGAN v2.1 (CESM2) PFTs and the corresponding equivalents in iBVOC (UKESM1). Only seven PFTs in MEGAN v2.1 have a direct equivalent in UKESM1, allowing direct calculation of the EFmass; the other eight PFTs were lumped into groups and the Crop 1 PFT in MEGAN v2.1 was used for the C3 and C4 crop and pasture PFTs in UKESM1.

EFs in MEGAN are given in units of mass of species per unit area of land surface per unit time (e.g. µgisoprene m-2 h-1), as opposed to µgC gdw-1 h-1 used in UKESM1 and ORCHIDEE, and are denoted hereafter as IEFarea. Therefore, a conversion must be applied to make these values comparable to the EFs used by iBVOC and ORCHIDEE, which are denoted as IEFmass.

To convert EFarea to EFmass, we adapt Eq. (5) of Messina et al. (2016) to yield Eq. (1). 1EFmass=IEFarea×1LAIref×1SLW×mCarbonmspecies×1CCE, where LAIref is the reference leaf area index used by MEGAN v2.1 (5 mleaf2 msurface-2), SLW is the specific leaf weight (gdw msurface-2), the factor mCarbonmSpecies accounts for the fact that MEGAN v2.1 considers the mass flux of a given species and iBVOC and ORCHIDEE the mass flux of carbon, and CCE is the MEGAN canopy environment coefficient (0.57).

Equation (1) is valid for emissions which are entirely dependent on PAR, as is the case for isoprene in MEGAN v2.1. Emissions of monoterpenes have a light-dependent fraction (LDF) and a light-independent fraction (LIDF = 1 - LDF). In this case, Eq. (1) needs to be modified to give Eq. (2). 2EFmass=IEFarea×1LAIref×1SLW×mCarbonmspecies×LDFCCE+1-LDF

The LDF varies between species, and we used the values given in Table 4 of Guenther et al. (2012).

There are three main areas of uncertainty in the conversion: the lumping of PFTs, the choice of SLW values and, for terpene emissions, the choice of input TEFarea values.

All UKESM1 simulations used the atmosphere-only configuration of UKESM1 run at a horizontal resolution of 1.25∘ × 1.875∘ with 85 vertical levels up to 85 km (Walters et al., 2019) and the GLOMAP-mode aerosol scheme, which simulates sulfate, sea salt, BC and organic matter but does not simulate currently nitrate aerosol (Mulcahy et al., 2020). Mineral dust is simulated using the bin scheme of Woodward (2001). UKESM1 has the capability to perform simulations using specified dynamics, also called “nudging”, where certain offline meteorological fields from the ERA-Interim reanalysis (temperature and horizontal winds) are input (Dee et al., 2011) and free running with online computed meteorology.

Our evaluation has three sections. Firstly, we perform PD simulations nudged to atmospheric reanalyses to compare the model simulations with different EFmass values to observational data (Table 4). Secondly, we perform free-running simulations using conditions from the pre-industrial (PI) and Shared Socioeconomic Pathway SSP3-7.0, which represents a “regional rivalry” scenario at 2050 (O'Neill et al., 2016) to assess the effect these different EFmass values would have in these two periods (Table 4). Finally, we perform free-running simulations using PD conditions but with LULC from either the PD or specific future LULC scenario featuring wide-scale reforestation/afforestation to assess the impact these EFmass values would have on the response of mass tree planting (Table 5). Runs used the CS2 chemical mechanism (Jenkin et al., 2019; Weber et al., 2021), version 12.0 of the Unified Model (UM) and vn5. of JULES.

Nine 1-year PD simulations were performed for November 2012–October 2013, as this period covers 4 months for which there exist satellite CrIS observations of global isoprene columns (January, April, July and October 2013; Wells et al., 2020). Six runs were performed to evaluate plausible EFmass approaches by comparison of the resulting total global emissions to estimates from other sources and, for isoprene, comparison of simulated column values against measured column values. No_C4_emiss_PD was run to isolate the fraction of emissions from C4 PFTs (see Sect. 4). Finally, two 3-year nudged PD simulations were run with UKESM1 default EFmass and the proposed new EFmass to ensure the trends established in the 1-year runs were not simply caused by the prevailing meteorology and persisted over a longer period. We also performed two UKESM1 simulations using LULC taken from a PD CESM2 simulation (with PFTs lumped as described in Sect. 2) to assess the influence of the underlying simulated LULC on emissions. These simulations, IEF_SLW_UKESM_PD_CESM_LULC and IEF_SLW_CLM5_PD_CESM_LULC (Table 4), used the same IEF values as IEF_SLW_UKESM_PD and IEF_SLW_CLM5_PD respectively.

Nudging of temperature and horizontal wind was used to prevent diverging meteorology affecting BVOC emissions and to replicate as closely as possible the atmospheric conditions experienced when the observations were recorded. Thus, nudging, along with the use of observed sea surface temperature (SST) fields, means that, as far as possible, the changes in EFmass will be the only drivers of emission changes and allows for a more faithful comparison to observational data. Nudging only occurred above ∼ 1200 m in altitude, and thus most of the planetary boundary layer was not nudged.

The 1-year PD nudged runs used time series of anthropogenic and biomass burning emissions to keep the simulated conditions close to those when the observations were recorded. The 3-year nudged PD runs used 2014 time-slice anthropogenic and biomass burning emissions. All these simulations used prescribed LULC from a UKESM1 historical ensemble member performed for CMIP6 (Sellar et al., 2020).

The four free-running simulations performed to investigate how the EFmass changes would affect simulated emissions in the PI and in 2050 under conditions used prescribed LULC from the UKESM1 piControl and SSP3-7.0 runs performed for CMIP6 (Sellar et al., 2019) and time-slice emissions from 1850 and SSP3-7.0 2050 respectively (Table 4).

We also explored how the change to EFmass would affect the response to a specific LULC change with a further set of simulations, which used two time periods from a specific land use scenario featuring wide-scale afforestation and reforestation (“MaxForest”). The Maxforest scenario features a very high degree of reforestation and afforestation over the course of the 21st century and was developed to assess the impact of such LULC with regards to carbon sequestration, among other factors. The scenario gradually expands existing forested regions with suitable tree species and also avoids encroachment on cropland, pastures and urban regions. It can thus be considered a scenario representing a near-maximum plausible level of reforestation/afforestation. The Maxforest scenario was originally developed for CLM5 (Lawrence et al., 2019), and we adapted it for UKESM1 using the same lumping of PFTs as discussed in Sect. 2.3.1. We performed simulations with the default and new EFmass values with LULC from the start of the MaxForest scenario at 2010 (no increase in tree cover) and at 2050, when extensive reforestation was well underway. All these simulations used PD anthropogenic and biomass burning emissions and GHG concentrations, but BVOC emissions were allowed to respond to LULC change. We compare the change in isoprene and terpene emissions between 2010 and 2050 Maxforest land use when the default EFmass values were used to when the new EFmass values were used. We also performed the same experiments in CESM2 (Sect. 3.2) and compare the change in BVOC emissions between 2010 and 2050 Maxforest land use to the UKESM simulations.

In all the runs, CO2 was not emitted but was set to a constant field appropriate for the PI, PD and 2050 under SSP3-7.0 conditions, while the other well-mixed greenhouse gases (WMGHGs) CH4, CFCs and N2O were prescribed with constant lower-boundary conditions (Archibald et al., 2020) appropriate for the PI, PD and 2050 under SSP3-7.0 conditions.

Fields for SSTs, sea ice (SI) and ocean biogeochemistry (DMS – dimethyl sulfide – and chlorophyll) were prescribed for all the runs. The nudged PD runs used observed SSTs and SI and ocean biochemistry from a UKESM1 historical ensemble member. The free-running PI runs used a 30-year mean from the UKESM1 piControl for SSTs, SI and ocean biogeochemistry. The SSP3-7.0 2050 runs used 2050 ocean biogeochemistry and 2045–2055 mean SSTs and SI, all taken from one of the UKESM1 SSP3-7.0 ensemble members.

The free-running Maxforest simulations (Table 5) used 2050 ocean biogeochemistry, 2045–2055 mean SSTs and SI and PD anthropogenic emissions and prescribed concentrations of WMGHGs. The same SSTs and WMGHG concentrations were applied to ensure differences in BVOC emissions were due to LULC only.

All UKESM1 runs used oceanic emissions of CO, C2H4, C2H6, C3H6 and C3H8 from the POET 1990 dataset (Olivier et al., 2003), and all biogenic emissions except isoprene and monoterpenes were based on 2001–2010 climatologies from the MEGAN-MACC (Monitoring Atmospheric Composition and Climate) dataset (Sindelarova et al., 2014) calculated by the MEGAN v2.1 model (Guenther et al., 2012) under the MACC project. Emissions of isoprene and monoterpenes (split into a 2:1 ratio between α-pinene and β-pinene) were calculated interactively using iBVOC. Anthropogenic and biomass burning emission data for CMIP6 are from the Community Emissions Data System (CEDS), as described by Hoesly et al. (2018).

The CESM simulations used version 2.2.0 (Danabasoglu et al., 2020) at a 0.9∘ × 1.25∘ horizontal resolution. For the atmospheric component, we employ CAM-chem version 6 (hereinafter CAM6-chem), with a full tropospheric O3–NOx–CO–VOC–aerosol chemistry based on an updated tropospheric chemistry mechanism (MOZART-TS1) (Emmons et al., 2020) with the Modal Aerosol Model with 4 modes (MAM4) (Liu et al., 2016). CAM6-chem has 32 vertical layers and a model top of ∼ 45 km and is coupled to CLM5, which provides BVOCs with the MEGAN v2.1 scheme and handles dry deposition. Our simulations used specified sea surface temperatures and thermodynamic sea ice.

For the two CESM simulations, anthropogenic and biomass burning emissions of reactive gases and aerosols were fixed to a 2010 climatology (2006–2014 average) using data from Hoesly et al. (2018). WMGHGs were incorporated as fixed lateral boundary conditions rather than as emissions from the surface and were also fixed to 2010 values (2006–2014 average) using standard concentration data from CMIP6 (Meinshausen et al., 2017).

For LULC, we performed the same free-running PD and 2050 LULC simulations as in UKESM1 (Table 5) to allow comparison of the BVOC emission responses to the LULC change in UKESM1 and CESM. As with UKESM1, the CESM simulations were atmosphere-only and used PD anthropogenic emissions and prescribed WMGHG concentrations and 2045–2055 mean SSTs, with the only difference between the simulations being the LULC.

Monthly mean isoprene columns derived using the space-borne CrIS technique (Wells et al., 2020) were used as the principal method of evaluation since it all allows the regional changes to be readily assessed. The CrIS is a longwave infrared Fourier transform spectrometer on board a satellite which can measure two isoprene IR absorption features. The absorption data collected by the spectrometer are then combined with an artificial neural network to calculate isoprene columns.

Surface isoprene emission measurements from regions with C4 grass were also used to examine the impact of substantial reduction in C4 grass IEFmass. In lieu of observations from C4-only regions (which are very sparse given the understanding that C4 grasses are weak isoprene emitters), we use observations from savanna, which tends to comprise grasses, woody shrubs and a range of trees, including strong isoprene and monoterpene emitters (e.g. Table 3 of Otter et al., 2002), in varying proportions. We select savanna observations from sites specifically noted as being without dominant isoprene emitters (Central Africa Republic – Klinger et al., 1998; Nylsvley, South Africa – Guenther et al., 1996, and Otter et al., 2002). We note that these sites were likely to have grass species with very low isoprene emissions, below the instrument detection limits (e.g. Harley et al., 2003). Therefore, our compiled observations represent an upper bound for isoprene emissions from C4 grass.

We also used monoterpene emission data measured at the SMEAR II site in Hyytiälä (https://smear.avaa.csc.fi, last access: 19 March 2023) to evaluate the change in terpene emissions.

Table 4 presents the global isoprene and terpenes emissions in our simulations. For isoprene, all nudged PD simulations, except No_C4_emiss_PD, yield total emissions within the range of previous estimates. Simulations using IEFmass from UKESM1 SLW (“IEF_SLW_UKESM_PD”) and CLM5 SLW (“IEF_SLW_CLM5_PD”) yield emissions 13 % and 7 % lower than the UKESM1 control simulation (“Control_1yr_PD”) respectively.

When UKESM1 LULC was replaced with CESM2 LULC, isoprene emissions are 380 Tg yr-1 (IEF_SLW_UKESM_PD_CESM_LULC) and 420 Tg yr-1 (IEF_SLW_CLM5_PD EF). These values are lower than the 457 and 490 Tg yr-1 simulated using the same IEF values and UKESM LULC, yet they are still well within the range of simulated emissions (310–680 Tg yr-1) of Fig. 1 of Messina et al. (2016). This highlights the influence of uncertainty in LULC on BVOC emissions but, as iBVOC is chiefly for use with UKESM1, we will focus on the simulations using UKESM1 LULC.

For terpenes, when the TEFmass is based solely on α-pinene (TEF_AP_PD), total emissions (177 Tg yr-1) are higher than previously published results (Messina et al., 2016) and when the TEFmass is derived from the weighted average of ratio of α-pinene, β-pinene and other monoterpenes” (TEF_all_PD), total emissions (88 Tg yr-1) are at the lower end of estimates. However, when taking a 2:1 ratio of α-pinene and β-pinene TEFarea (TEF_AP_BP_PD), the resulting emissions (130 Tg yr-1) are more in line with other estimates.

The clearest indication of the significant contribution C4 PFTs make to BVOC emissions in the current UKESM1 set-up comes from the comparison of Control_1yr_PD and No_C4_emiss_PD, where the EFmass of all C4 PFTs is zero, simulations. This reveals that C4 PFTs contribute about 40 % (18 %) to total isoprene (terpene) emissions in the current UKESM1 set-up, far higher than the 1 % (0.3 %) estimated by Guenther et al. (2012), 9 % for isoprene from the original 5-PFT version of iBVOC (Pacifico et al., 2011) and 1 %–2 % for isoprene estimated by Pfister et al. (2008). As previously discussed, this substantial contribution from C4 grasses is also in stark contrast to other studies, which highlight very low emissions of isoprene from C4 grasses (e.g. Loreto and Fineschi, 2015). Overall, this suggests that while the current UKESM1 approach may produce a reasonable value for total isoprene and terpene emissions, these are derived using unrealistic EF for C4 grasses.

With the updated EFmass, the C4 grass PFT contributes 1 %–3 % of total isoprene emissions (based on the chosen SLW) and 0.2 %–0.7 % of total terpene emissions (based on the choice of EFarea), bringing UKESM1 into line with other estimates.

The decreases in C4 PFT EFmass and increases in the EFmass of the broadleaf evergreen tropical tree PFT lead to the contribution to total isoprene emissions from broadleaf evergreen tropical trees increasing from 45 % to 75 % (50 % to 80 % for terpenes). This contribution is greater than the 46 % estimated by MEGAN v2.1 (Guenther et al., 2012). However, the area of this PFT in UKESM1 is 67 % greater than CLM4 (26.0 vs. 15.6×106 km2). (While this difference in area may seem large, the total areas of tropical trees in UKESM and CLM4 are much more similar if the area of CLM4's deciduous evergreen tropical tree PFT, for which there is no direct analogue in UKESM, is included.) On an emissions per unit area basis for the broadleaf evergreen tropical PFT, isoprene emissions in UKESM1 are within 5 % of that from Guenther et al. (2012), while terpene emissions are ∼ 25 % lower. This separately highlights the important issue of uncertainty in land use and land cover and the effect that it can have on model–model comparisons (e.g. in CMIP6) and model–observation comparisons.

Spatially, the new IEFmass led to isoprene emission increases across Amazonia (albeit with a small reduction around Manaus) and the Congo and decreases north and south of the African rainforest, where the simulated C4 grass PFT dominates (Fig. 3a). Terpene emissions increase over the tropics due to increases in the TEFmass of tropical evergreen broadleaf trees, while they decrease at mid and high latitudes (Fig. 3b) from reductions in the TEFmass of needleleaf evergreen and deciduous PFTs (Fig. 2b).

For the PI and future simulations with UKESM1 (Fig. S1), the new EFmass values lead to reductions in total global isoprene (terpene) emissions of 13 % (11 %) and 8 % (8 %) in the PI and 2050 SSP3-7.0 scenarios respectively compared to the default EFmass (Base PI). For both scenarios, isoprene emissions from C4 PFTs decrease by ∼ 90 %, while emissions from broadleaf evergreen tropical trees increase by ∼ 50 %. This leads to emission increases over Amazonia and the Congo but decreases north and south of the Congo (Fig. S1a, c). Terpene emissions from C4 PFTs drop to almost zero and decrease by ∼ 60 % from needleleaf evergreen trees while increasing by around 50 % from broadleaf evergreen tropical trees, driving a tropical emission increase and high-latitude emission decrease (Fig. S1b, d).

We compare the output from the PD simulations to the CrIS observed isoprene columns (Sect. 3.3) for January, April, July and October 2013 (Fig. 4). The use of nudging significantly reduces the difference in meteorology between the simulations and reality, greatly improving the comparability of modelled and observational data. However, the lowest 11 model levels (up to approx. 700–1000 m) are not nudged, so there will be some differences between the model simulation conditions and reality in the boundary layer, although this is tempered by the nudging applied to the higher levels.

For the 4 months considered, IEF_SLW_UKESM_PD and IEF_SLW_CLM5_PD yield lower total isoprene emissions than Control_1yr_PD but show generally slightly higher column biases in the same regions where the control run has a bias, chiefly in western Amazonia.

This bias exacerbation is slightly greater in IEF_SLW_CLM5_PD than IEF_SLW_UKESM_PD and is likely driven by the increase in IEFmass for the tropical broadleaf evergreen trees which are dominant in the region. The biases over central Africa are very similar between the three approaches.

The increased biases with the new IEFmass (e.g. the increase of 0.5–0.7 × 1015 molecules cm-2 over South America in January 2013, Fig. 4a–c) is not necessarily indicative of these new IEFmass values being less accurate than the original IEFmass values which may be performing better due to offsetting issues. Biases in LULC, as highlighted by the comparison of UKESM1 and CLM5 in terms of broadleaf vs. deciduous tropical trees, simulated chemistry and emissions of other species (e.g. NOx) which affect the atmospheric oxidising capacity and thus isoprene concentrations will also contribute to the enhanced model bias. The difference in model bias between simulations with the default and new IEFmass values is noticeably smaller than the difference in the model bias when different chemical mechanisms are used. For example, in April 2013, the mean bias over South America in the Strat-Trop mechanism (Archibald et al., 2020) was 5.7 × 1015 molecules cm-2, and this decreased substantially to 0.6 × 1015 molecules cm-2 when CS2 was used (Fig. 2; Weber et al., 2021)

Given the major change to the IEFmass values of C4 grass, the isoprene emissions from C4 grasses were compared to observations in southern Africa (Sect. 3.3, Fig. 5). The model resolution (∼ 100 × 100 km in the region of relevance) means the grid cells where observations were made contained high fractions of strong isoprene emitters, typically broadleaf evergreen tropical trees and C4 grasses. To isolate the impact of C4 grass emissions, we take the area-weighted mean of emissions from grid cells in the region where C4 grasses comprise > 80 % of the total surface types (vegetation and non-vegetation). We use the 3-year monthly mean for the month when observations were recorded and apply a scaling factor of 2 to account for the fact that isoprene emissions are zero at night. (Emission measurements were only taken during the day, while the use of model monthly average values means that approximately half of the data points going into the model value will been zero, halving the model's average.)

While comparison of these model and observational data should be treated as illustrative rather than definitive for the reasons explained above, it suggests that the reduction in C4 IEFmass may help to reduce the model high bias in C4-grass-dominant regions. We also note that the observed values represent an upper bound since the emissions in some regions will be below the limit of detection (e.g. Harley et al., 2003).

While the primary focus of this paper is correcting the error with the emission factors for C4 grasses, we also performed a comparison of monoterpene emissions measured by the SMEAR II station in Hyytiälä in the boreal forest, with emissions from simulations using the current TEFmass (Control_1yr_PD) and updated TEFmass (TEF_AP_BP_PD) values. We found that the new TEFmass yielded emissions which compared well to observations (Fig. S2).

Tropical broadleaf evergreen trees and C4 grass PFTs are some of the most widespread vegetation types in the tropics. The respective increase and decrease in EFmass values for these PFTs means the response of BVOC emissions to a change in the relative fractions of these species is likely to be quite different when using default and new EFmass values. We explored this further using UKESM1 and CESM and the MaxForest scenarios since this scenario involves, among other changes, increases to tropical broadleaf evergreen tree cover at the expense of C4 grasses in Africa and eastern Brazil (Fig. S3 and Table 5).

When the UKESM1 default EFmass values are used, the extensive reforestation/afforestation in the Maxforest scenario yields a reduction in isoprene emissions relative to 2010 (Fig. 6a). This is due to the decrease in C4 grass coverage shown by emission reductions in regions where C4 grasses are replaced by trees. By contrast, when the updated EFmass values are used, the Maxforest scenario leads to an increase in isoprene emissions in UKESM1 (Fig. 6b) which resembles the response simulated in CESM (Fig. 6c). The similarity between the responses in UKESM1 to the new EFmass and CESM is not surprising since CESM also uses the MEGAN v2.1 scheme for the emissions of isoprene.

For isoprene, there is little to differentiate the approaches using SLW from CLM5 or UKESM1. The CLM5 SLW approach yields slightly higher column biases but total PD emissions (491 vs. 457 Tg yr-1; Table 4) which are closer to the median of other estimates (∼ 500 Tg yr-1; Messina et al., 2016). The CLM5 SLW approach allows PFT-specific SLW values to be used to calculate the EFmass of the MEGAN v2.1 PFTs before they are lumped into UKESM PFTs, while using the UKESM1 SLW values means lumping must occur before the EFmass are calculated, potentially increasing uncertainty in the output. Overall, we tentatively recommend using the IEFmass values calculated using the CLM5 values (i.e. IEF_SLW_CLM5_PD). For terpenes, based on total emissions we recommend the TEFmass calculated from the TEFarea of α-pinene and β-pinene in a 2:1 ratio (Guenther et al., 2012), i.e. those used in TEF_all_PD. These EFmass values are given in Table 6.

We do not claim that the new EFmass values are the final word on the matter; rather, we believe that they represent an improvement over those currently used in UKESM1 and provide a clear method for recalculation in the future should revised EFarea values be developed and/or a wider range of PFTs considered.

Accurate modelling of BVOC emissions depends on the parameterisations within the emission module (in this case, iBVOC) and the simulations of external factors, which influence emissions. This study deals with just one part of this framework: biases in these external factors can limit the effectiveness of model–observation comparisons, such as the satellite-derived isoprene columns shown in Fig. 4, and offsetting errors can lead to reasonable results, at a given period in time, or improvements to certain components (e.g. emission factors) yielding reductions in model performance. Nevertheless, progress towards an approach that faithfully captures biosphere–atmosphere interactions requires incremental improvements to all contributing factors. Here we describe some other sources of uncertainty in the simulation of BVOCs and areas where future work would be useful.

iBVOC includes dependencies on CO2, temperature, photosynthetic activity and plant functional type, with the latter the focus of this study. MEGANv2.1 considers the role of leaf age in emissions and leaf temperature over the past 24 and 240 h, while these factors are omitted in iBVOC: assessing the impact of these parameterisations in iBVOC would be worthwhile.

Within the parameterisation of PFT dependency updated in the study, several areas of uncertainty have been identified. The impact of SLW value variations and the multiple options regarding which TEFarea values to use have been quantified with the range of simulations performed in this work. Other areas of uncertainty have not been fully scrutinised due to a relative lack of observational data. Compared to the species which are strong isoprene emitters, observations of emissions from grasses are sparse, hindering further model validation. MEGAN v2.1 also prescribes a single, very small EFarea for all crops and pasture, resulting in negligible emissions from these PFTs. Emissions from longer-lived crops and pasture are likely to tend towards grasses and the projected expansion of these PFTs in some future scenarios, particularly those with increasing population, meaning that capturing emissions from these PFTs may become more important. Further observations would aid in this effort. We also note that the emission factors in MEGAN v2.1 are not perfect and will continue to be refined. For example, Sindelarova et al. (2022) updated emission factors for α-pinene for certain tree PFTs. For consistency, the MEGAN emission factors used in this study are all from MEGAN v2.1, but future development of iBVOC should take into account the latest understanding of emission factors.

Simulation of external factors including land cover (cf. the effect of swapping UKESM and CESM LULC on simulated emissions; Table 4), surface temperature and meteorological conditions (e.g. droughts and floods) also affect BVOC emissions (e.g. Sheil, 2018; Yáñez-Serrano et al., 2020).

The reduced nature of Earth system models requires the aggregation of a wide range of vegetation types, which in reality have varying emission factors, into a small number of PFTs. This oversimplification can lead to unrealistic emissions in certain locations (e.g. the inclusion of shrubs EF in grasses EF) and discrepancies between different modelling approaches (e.g. UKESM1 vs. CESM). Assessment of the impact of using a wider range of PFTs, based on more highly resolved emission factor datasets (e.g. Karl et al., 2009), would be informative.

The expansion of iBVOC to speciate terpenes into separate α-pinene and β-pinene tracers and the addition of new molecules, such as sesquiterpenes, would be beneficial for simulating atmospheric composition. α-pinene and β-pinene display different chemical reactivity, while sesquiterpenes can suppress local O3 and affect SOA formation by producing highly involatile species which can nucleate new particles without sulfuric acid (e.g. Bianchi et al., 2019; Weber et al., 2020).