SURFEX (SURface EXternalisée in French) is a surface modelling platform developed by Météo-France in cooperation with the scientific community. SURFEX simulates the interaction between the surface and the atmosphere by simulating the flux exchange between the soil and the upper atmospheric layer (e.g. latent heat flux, sensible heat flux, CO2 flux, chemical species, and aerosols). The most recent version of SURFEX (SURFEXv9.0) was released in January 2023; however, in this work, we used SURFEXv8.1, which is widely used at present (; ; ).

SURFEX can be run in an offline mode or coupled with an atmospheric model, e.g. the global numerical weather prediction model ARPEGE . Used in an online mode, SURFEX extracts the necessary meteorological data from the global weather prediction model. In offline mode, a forcing file should be prescribed as input to the model. The forcing file should contain spatio-temporal gridded maps of atmospheric variables (air temperature, specific humidity, wind components, pressure, rain rate, and CO2) and radiative variables (solar radiation and infrared radiation). During a model time step, each surface grid box receives the forcing variables listed above; in return, SURFEX computes averaged fluxes for momentum, sensible and latent heat, chemical species, and dust fluxes, etc., and then returns these quantities to the atmosphere by adding radiative terms such as surface temperature, direct and diffuse surface albedo, and surface emissivity .

As shown in Fig. , each grid box in SURFEX is represented by four adjacent tiles: nature, town, sea or ocean, and lakes. The final fluxes are the average of the fluxes calculated over nature, town, sea/ocean, and lake weighted by their respective fraction (Sl, St, Sn, Ss). SURFEX contains four principal surface schemes: ISBA for the nature tile , TEB for town , FLAKE for lakes, and SEA for sea and oceans. SURFEX can also simulate aerosol chemistry and surface processes and can be used for assimilation of surface variables .

To define the surface coverage, SURFEX uses ECOCLIMAP-II, which is a 1 km global database of land cover made by CNRM (Centre National des Recherches Météorlogiques, in French) . It describes the types of surfaces covering the whole earth.

ECOCLIMAP-II provides the fraction data for the 19 patches (nature tile). In addition, it provides land surface parameters relative to each patch, i.e. each vegetation type has a defined soil depth, height of trees, LAI available at 10 d time steps, and vegetation fraction. LAI is represented by a 5-year averaged LAI climatology over the period 2002–2006. In ISBA, the calculation of surface parameters is based on an aggregation process at patch level (i.e. from the 19 land cover types down to the selected number of patches) for each point of the grid according to the horizontal resolution .

The MEGAN model is a global emission platform designed to estimate the net emission of gases and aerosols from terrestrial ecosystems into the atmosphere. It is an updated version of MEGANv2.0, developed by to estimate isoprene flux, and MEGAN2.02, which was described for monoterpene and sesquiterpene emissions by .

MEGANv2.1 (the model's routines and input data can be found at https://bai.ess.uci.edu/megan/data-and-code/megan21, last access: 8 September 2023) includes algorithms that take into account the main known processes controlling biogenic emissions; it makes it possible to estimate the flux of 19 compound classes, which are decomposed into 147 individual species such as isoprene, monoterpenes, sesquiterpenes, carbon monoxide, alkanes, alkenes, aldehydes, acids, ketones, and other oxygenated VOCs . These species can then be lumped into the appropriate categories for the chemical scheme for use in chemical transport models. The stand-alone version of MEGANv2.1 requires as input weather data (temperature, precipitation, solar radiation, wind, and photosynthetic photon flux), atmospheric chemical composition data (CO2 concentration), land cover data (plant functional type distribution and LAI data), and emission factor data.

The estimation of biogenic fluxes in MEGANv2.1 is based on a simple equation (Eq. ) to calculate the net primary emission flux from terrestrial landscapes (Fi) into the above-canopy atmosphere (µm-2s-1). This equation comprises two significant components: firstly, the emission factor, which represents the emission potential of a specific compound associated with a particular vegetation type, and secondly, the emission activity factor, which reflects how this emission potential responds to variations in environmental conditions and meteorological conditions. 1Fi=γi×∑j=1n(εij×χj), where γi is the dimensionless activity factor of a compound i (this factor is equal to 1 in standard conditions described below), εij is the emission potential (also known as “emission factor”) of a compound i and vegetation type j at standard conditions, and χj is the fractional grid box areal coverage.

The coupling of MEGAN2.1 and SURFEXv8.1 is based on a previous implementation of MEGAN in MESO-NH5.4. MESO-NH5.4 is an atmospheric non-hydrostatic research model designed for studies of physics and chemistry . This coupling involved merging MEGAN routines and linking the required inputs of the biogenic model with the SURFEX parameters.

The present study focuses on the online integration of MEGAN into SURFEX. The ultimate aim of this coupling is to be able to force the coupled model through various climate change scenarios in order to assess the climate change impact on the biosphere and to quantify the effect of these changes on biogenic emissions and therefore on global and local air quality. Additionally, this coupling aims to improve biogenic emission estimations by providing the MEGAN model with detailed vegetation-dependent inputs at patch level. This allows key land surface parameters used by MEGAN, i.e. leaf area index and soil moisture, to be calculated at a more precise scale. Thus, activity factors are individually calculated for each patch. This approach allows for a more accurate representation of biogenic emissions in the context of climate change and of their impact on air quality.

In the coupled model the estimation of biogenic fluxes of various species was carried out based on 16 vegetation types extracted from the ECOCLIMAP-II database . Each vegetation type from ECOCLIMAP-II was mapped to its corresponding type defined in CLM4. Table  represents the mapping used in the coupled model. For most CLM4 PFTs, similar existing vegetation types are defined in ECOCLIMAP-II. However, when considering shrubs, CLM4 classifies them into three distinct categories: evergreen temperate shrub, deciduous temperate shrub, and broadleaf deciduous shrub. Conversely, ECOCLIMAP-II does not provide separate classifications for these three distinct types of shrubs. To overcome this limitation, the three plant functional types corresponding to the different types of shrubs were introduced in ECOCLIMAP-II by assigning the shrub patch to a specific geographical area based on a given latitudinal range. The evergreen temperate shrub type is specified in the coupled model as the shrub patch in tropical regions (-30° < latitude < 30°), the deciduous temperate shrub in temperate regions (-60° < latitude < -30° or 30° < latitude < 60°), and the deciduous boreal shrub in boreal regions (60° < latitude). This approach allows for a more accurate representation of shrubs in the coupled model.

Figure  represents a comparison between the vegetation types used in the MEGAN stand-alone version and the ones defined in ECOCLIMAP-II. For comparison, we have grouped the 16 PFTs into 6 main vegetation types: broadleaf evergreen trees, needleleaf evergreen trees, deciduous trees, shrubs, grassland, and crops. The vegetation spatial distribution and intensity are similar for most vegetation types in ECOCLIMAP-II and CLM4. For shrubs, the substantial difference in vegetation distribution is due to the vegetation height threshold used in ECOCLIMAP-II (2 m) and in CLM4 (10 m). For other vegetation types (e.g. needleleaf trees), the difference in vegetation density between ECOCLIMAP and CLM4 is expected to have a small impact on isoprene emissions, as this specific PFT represents only 1.4 % of the total annual emitted isoprene . For vegetation-related input data, MEGAN can use climatological LAI from the ECOCLIMAP-II database; in this case, the LAI is defined for each vegetation type in a 10 d time step, or the dynamic LAI is estimated for each vegetation type with the vegetation scheme in SURFEX. LAIv, defined as the LAI in a grid cell divided by the vegetation fraction, is considered equal to the current LAI, and LAIp (previous LAI) is defined as the LAI value of 10 d in the past.

In the SURFEX model time step, all surface variables are interpolated and updated for each grid cell. Each tile is treated independently by using a specific scheme. For the nature tile, the surface parameters are calculated following the vegetation-type aggregation process, which merges several vegetation types into a single patch (ranging from 1 to 19).

It is important to clarify that the coupling of SURFEX and MEGAN is online, which means that MEGAN's estimation of biogenic fluxes interacts dynamically with the ISBA scheme (Interaction between Soil Biosphere and Atmosphere). ISBA is the scheme used for the nature tile to compute the exchanges of energy and water between the soil–vegetation–snow continuum and the atmosphere above.

The online implementation of MEGAN was done following the SURFEX conceptual framework, which separates the initialization phase from the temporal evolution phase. This involved setting up specific routines to initialize and interpolate MEGAN-related parameters (e.g. emission factors). The temporal estimation of biogenic emissions was carried out as an integral part of the ISBA scheme. This was achieved by integrating MEGAN routines that estimate the activity factor for each vegetation patch, using vegetation parameters estimated by ISBA, which encompass factors like soil moisture and wilting point at different layers (depending on the soil discretization method), leaf area index, photosynthetically active radiation (PAR), and surface temperature. Figure  shows a global representation of the online implementation of MEGAN in SURFEX.

The coupled SURFEX–MEGAN model was utilized to conduct a global simulation of isoprene emissions in 2019 using ERA5 meteorological forcing. This simulation is referred to as the “reference simulation” (abbreviated as REF).

ERA5 is a reanalysis based on the integrated forecasting system IFS (numerical weather forecasting model and data assimilation system developed jointly by ECMWF and Météo-France) . For the REF SURFEX–MEGAN simulation, the ERA5 forcing file includes hourly reanalysis meteorological fields defined on a 1°×1° spatial resolution grid (re-gridded from the native 31 km × 31 km resolution). Temperature and specific humidity were extracted at 2 m height; wind speed and wind direction were calculated based on zonal and meridian wind components at 10 m height. As there are no available inputs for surface incident diffuse short-wave radiation and CO2 rate, these parameters were assigned values of 0 Wm-2 and 410 ppm respectively. The CO2 concentration value corresponds to the 2019 annual mean CO2 observed at Mauna Loa .

In this study, the calculation of PPFD and temperature for sun and shade leaves at different canopy heights was made using the canopy model integrated in MEGAN; the incoming PAR (photosynthetically active radiation) at the top of the canopy was assumed to be 48 % of the incoming short-wave radiation , and a conversion factor of 4.6 and 4.0 µmol photons J-1 was used to convert PAR to PPFD for diffuse and direct radiation respectively . Unless otherwise stated, in all coupled model simulations the estimation of isoprene flux was based on the isoprene potential map, and the effect of soil moisture deficit and CO2 on BVOC emissions was not taken into account (the γSM and γCO2 factors were assigned a value of 1). This choice enables a better comparison with other emission inventories. Additionally, the impact of the CO2 inhibition factor becomes relevant only when CO2 atmospheric concentrations exceed 400 ppmv significantly .

For simplicity, we have used the ISBA 2-L scheme in the present study. In this scheme, the soil is represented with two layers, and the heat and moisture exchanges between the layers and the atmosphere are modelled with the force-restore method ; this approach is described further in Sect. .

The emission rate of isoprene can be influenced by a variety of meteorological factors, including temperature, solar radiation, and atmospheric humidity. To illustrate the impact of these factors on isoprene emission estimated by SURFEX–MEGAN, two simulations were conducted using two meteorological datasets: IFS forecast dataset (operational real-time weather forecast, forecast grid data) and MERRA. MERRA was undertaken by NASA’s Global Modelling and Assimilation Office. The data were generated with version 5.2.0 of the Goddard Earth Observing System (GEOS) atmospheric model and data assimilation system (DAS) and cover the period from 1979 to the present . The MERRA data are defined on an hourly basis on a grid of 0.625° latitude and 0.5° longitude resolution. However, to avoid considering the effect of spatial resolution on isoprene emission , the MERRA reanalysis meteorological fields were interpolated to align with the reference simulation spatial resolution (1°×1°).

The reference simulation uses ERA5 meteorological forcing; however, the version of IFS used in ERA5 is a newer and more advanced version of the IFS that was used in the near-real-time forecasts in 2019 for operations. This improved version of the IFS for ERA5 uses a numerical climatology model for modelling physical processes, while the version used for operational real-time forecasts uses process parameterization schemes that are optimized for fast and real-time execution. The IFS meteorological forcing was extracted from the IFS operational real-time forecasts model with a spatial resolution of 1°×1° and a temporal resolution of 3 h. The S1-MERRA simulation has the highest global annual isoprene in 2019 with a total of 462 Tg, followed by reference simulation (ERA5) and S1-IFS with a total of 443 and 421 Tg respectively. The annual mean isoprene flux difference in 2019 between S1 simulations and reference simulation is shown in Fig. . ERA5 isoprene emissions are higher in both the Amazon and Congo rainforests as well as over Indonesia compared to IFS isoprene estimates. On the other hand, ERA5-based isoprene emissions are lower than MERRA isoprene emissions in eastern Australia but higher in Africa. To investigate the origin of these differences, an analysis of meteorological parameters that drive isoprene emissions was performed focusing particularly on temperature and solar radiation. These parameters influence the emission of biogenic species via two factors, γP and γT, detailed by .

As shown by , the estimate of isoprene flux in MEGAN is temperature dependent, with emissions increasing exponentially with temperature to a maximum that depends on the average temperature of the last 24 h. MEGAN emissions depend also on the amount of light received by vegetation. The isoprene estimate increases almost linearly with PPFD; the rate of increase depends on the average PPFD over the last 24 h. To study the linear dependence between the isoprene flux estimates and PPFD, we examined the correlation between the difference in isoprene estimates and the difference in light (PAR) between the reference (with the ERA5 meteorological forcing) and S1 simulations (with the IFS and MERRA meteorological forcings). Figure  displays the temporal correlation coefficient between isoprene flux differences and light differences for the reference and S1-IFS simulations, as well as for the reference and S1-MERRA simulations. The PAR contributes strongly to the explanation of isoprene discrepancies between the reference and S1 simulations, as the correlation coefficient exceeds 0.8 in regions where isoprene is emitted. Thus, the difference in isoprene emission flux across the three simulations is mainly due the different PAR input used in the simulation’s meteorological forcing file. The correlation study was not conducted on other isoprene meteorological drivers, such as temperature, as the dependence of isoprene on this parameter is exponential.

Figures  and represent the isoprene distribution by region for all tests performed and the mean temperature/PAR relative difference between the MERRA and the ERA5 data inputs. On a regional scale, MERRA temperature and downward radiation received by vegetation are higher in Australia and South America compared to ERA5, resulting in higher isoprene estimates in those regions (+10 % in Australia and +7 % in South America). Conversely, MERRA temperature and radiation inputs are lower in East Africa, resulting in lower isoprene estimates for that region (-3 %).

A regional analysis was also conducted to quantify the impact of using different meteorological datasets on isoprene estimates. Figure  displays the monthly isoprene emissions of the reference and S1-MERRA simulations across regions of the globe shown in Fig. . The isoprene flux absolute difference is mostly pronounced in Australia, South America, and Southern Africa, where S1-MERRA isoprene estimates are higher than the reference simulation. In South America, Southern Africa, and Australia, S1-MERRA monthly isoprene emissions are higher than the reference simulation by a range of 2 %–10 %, 1 %–11 %, and 6 %–15 %. In these regions, although the temperature difference between MERRA and ERA5 is small (less than 0.5°), the photosynthetically active radiation (PAR) difference is significant. In these regions, PAR variations range at -1–9, -2–8, and -1–8 W m-2 respectively. Consequently, the main factor driving monthly variations in isoprene emissions between the reference simulation and the S1-MERRA simulation is PAR.

Several studies have been conducted to quantify the sensitivity of the MEGAN model to meteorology. For example, showed that using different meteorological forcings can lead to different emission estimates where the use of CRU (Climatic Research Unit) meteorology instead of the NCEP (National Center for Environmental Prediction) reanalysis product led to a 10 % decrease with MEGANv2. also detected a difference of the total BVOC MEGANv2.1 estimations between CAMS-GLOB-BIOv1.2 and CAMS-GLOB-BIOv3.1 and explained that the discrepancies are mainly due to the use of different meteorological inputs.

On a global scale, the use of different meteorological forcing has been found to have an impact on the amount of isoprene emissions estimated with the SURFEX–MEGAN model. The use of MERRA meteorology led to a 5 % increase in isoprene emissions, while the use of IFS meteorology resulted in a decrease of 4.8 % in comparison with the reference simulation.

MEGANv2.1 defines two approaches to estimate biogenic fluxes. The first one is based on the use of the biogenic species emission potential maps ϵmap; these gridded maps are made based on a land cover including more than 2000 eco regions each with specific emission factors . The compilation of these maps accounts for the large differences in emission potential between species belonging to the same generalized PFT (e.g. temperate deciduous tree). For other PFTs, including only low isoprene emitters, the use of the PFT-specific emission factor is sufficient (e.g. boreal deciduous and needle trees). The second approach consists of using the 16 generalized plant functional type distributions ϵPFT, along with their specific emission factor .

To compare the two approaches, we estimated global isoprene fluxes during 2019 using emission potential values ϵPFT instead of the emission potential data from the gridded maps ϵmap used in the reference simulation. Figure  shows the mean difference in isoprene emissions between the S2 simulation using ϵPFT and the reference simulation using ϵmap. The total annual isoprene of simulation S2 is 390 Tg; the data indicate that on a global scale, the isoprene emissions have decreased by 12 %. As shown in Fig. , this decrease is particularly pronounced in Australia (-58 %) and Southern Africa (-25 %). A notable increase is observed in Europe (+32 %) and in South America (+19 %), particularly in the northern Amazon. The red dots over islands shown in Fig.  are due to the fact that the isoprene emission factor from the input emission potential map is equal to 0 in these areas.

The results of this sensitivity test are aligned with the findings of . The MEGAN-MACC average annual isoprene emissions dropped by 14 % when using the emission potential values ϵPFT instead of the emission potential map ϵmap. The decrease concerns Australia (-47 %) and Southern Africa (-28 %) and the increase concerns South America (+10 %) and Europe (+18 %).

Figure  represents the annual mean isoprene flux obtained with the S2 sensitivity simulation. The results of this sensitivity test partially explain the differences observed in Sect. . CAMS-GLOB-BIOv3.0 and ALBERI inventories used ϵPFT data to estimate isoprene flux, resulting in lower isoprene emissions compared to other datasets, as annual isoprene flux dropped by 29 % and 21 % respectively compared to the reference simulation. reported a similar decrease rate in isoprene emissions estimated at 30 % of CAMS-GLOB-BIOv3.0, which uses PFT-specific emission potential data and PFT distributions, compared to CAMS-GLOB-BIOv3.1, which uses isoprene emission potential gridded maps.

Prior research has investigated the association between soil moisture and isoprene emissions. The results indicate that isoprene emissions exhibit a three-phased response to drought and declining soil water. In the initial days of drought, plants tend to retain a stable isoprene emission rate; in some instances, the emission rate may even slightly increase . The second stage starts when soil moisture falls below a specific threshold, at which point the rate of isoprene emission begins to decrease. Extended exposure to severe drought leads to a gradual decrease in isoprene emissions; eventually, the emissions become insignificant over time (; ; ; ; ).

The response of isoprene emission to drought is simulated in MEGAN indirectly via the MEGAN canopy environment model by incorporating the leaf temperature estimate, which is affected by soil moisture. MEGAN also includes a γSM factor which directly simulates the response of isoprene emissions to drought. This factor is derived from soil moisture parameterization experiments conducted by . The γSM is defined as follows: 8γSM=1θ>θ1γSM=(θ-θw)Δθ1θw<θ<θ1γSM=0θ<θw, where θ is the soil moisture (volumetric water content, m3 m−3), θw (m3 m−3) is the wilting point (the soil moisture level below which plants cannot absorb water from soil), Δθ1 (= 0.04) is an empirical parameter, and θ1 = θw + Δθ1 .

The third sensitivity test (S3) was conducted to examine the effect of soil moisture on isoprene emissions. To estimate γSM, MEGAN uses wilting point data calculated in SURFEX from the sand and clay covers given as input to the coupled model following the approaches by and . The sand and clay data are extracted from HWSD (The Harmonised World Soil Database), which is a global soil database developed by the FAO (Food and Agriculture Organisation of the United Nations) in collaboration with IIASA (International Institute for Applied Systems Analysis) in order to provide information on the physical and chemical properties of soils across the world.

In order to accurately estimate soil moisture, a 4-year spin-up period was required to stabilize the soil water content with the ISBA force-restore 2-L scheme. This approach is used to simulate the exchange of energy and water between the surface and the atmosphere and is based on the balance between the forces that drive the exchange of energy and water (radiation, temperature, and precipitation) and the restoring forces that return the system to equilibrium (evaporation, transpiration, and runoff) . The wilting point and soil water content are calculated at different soil layers, depending on the ISBA scheme model used. In the ISBA force-restore 2-L scheme, the soil is represented with two layers. In the present study, to evaluate soil moisture impact on isoprene emissions, we used soil moisture and wilting point data from the second layer, as it most accurately represents the root depth of the vegetation.

The integration of the soil moisture algorithms led to total isoprene emissions of 273 Tg, with a global decrease of 38 % compared to the reference simulation. Figures , , and show the annual mean isoprene difference between the S3 simulation and the reference simulation, the spatial distribution of average γSM over 2019, and the annual mean soil liquid water content estimated at the second ISBA-2L layer as well as the relative wilting point data used in the S3 simulation respectively. The decrease concerns mainly arid and semi-arid regions; the largest decrease can be observed in Australia (-89 %), followed by North Africa and the Middle East (-82 %), Southern Africa (-67 %), and East Africa (-38 %). In South America, emissions are lower by 23 % and the decrease is mainly located in Brazil. Previous studies have analysed the impact of soil moisture on isoprene emissions and have reported varying decrease rates. obtained the lowest decrease rate of 7 %, found a decrease rate of 21 %, and reported the highest decrease rate of 50 %. The discrepancies in the reported values of the decreased isoprene rate can be attributed to the use of different soil moisture and wilting point data. The latter is a critical parameter, as it defines the limit below which the soil moisture activity factor is set to 0; consequently, stressed the importance of using consistent wilting point data with the soil moisture input. In this context, the SURFEX–MEGAN model enhances the precision of γSM calculations by using vegetation-type-dependent soil moisture at a given layer and wilting point data at the same soil layer.

Several limitations associated with the use of this MEGANv2.1 soil moisture parameterization have been identified. Primarily, the parameterization exhibits a significant dependency on the wilting point data, which can show inconsistency with the soil moisture data used. Furthermore, it has been shown that this parameterization substantially reduces isoprene emissions, even under moderate drought conditions, thereby indicating a potential over-sensitivity to drought stress . The introduction of the new MEGAN3 soil moisture factor addresses these shortcomings by providing robust performance under both moderate and severe drought conditions . This enhancement in soil moisture representation is anticipated to be integrated into the forthcoming version of the SURFEX–MEGAN coupled model, thereby offering a more accurate and reliable prediction of isoprene emissions under varying soil moisture scenarios.