Isoprenoid vertical profiles were investigated at the triangular tower (TT34 tower – 2∘35.37′ S, 60∘06.92′ W) on a plateau of the Cuieiras Biological Reserve, a primary rainforest reserve located approximately 60 km northwest of Manaus city, in the central Amazonian Basin, in Amazonas, Brazil (Martin et al., 2010). The vegetation in this area is considered to be a mature terra firme rain forest (Pires and Prances, 1985), with a leaf area index of 4.7 (Malhi et al., 2009). The diversity of tree species is above 200 speciesha-1 (Oliveira et al., 2008). Annual precipitation is about 2500 mm (Fig. 1a), with December–May being the wetter period. Although severe droughts impacted part of the Amazon basin in 2005 and in 2010, those droughts did not affect central Amazonia (Marengo et al., 2008, 2011). However, micrometeorological measurements from 1999 to 2012 showed that from August to September the monthly cumulative precipitation can be less than 100 mm per month (Fig. 1a), characterizing this period as dry season. Average air temperature ranges between 24 ∘C (in April) and 27 ∘C (in September) (Fig. 1e). Soil moisture near the surface is slightly reduced (by 10 %) during the dry compared to the wet season (Cuartas et al., 2012).

The period of this study (from 2 September 2010 to 27 January 2011) represents the second half of the dry season (September 2010–October 2010), the dry-to-wet transition season (November 2010), and the beginning of the wet season (December 2010–January 2011). The whole period of measurements includes the period of low precipitation and when precipitation is increasing (Fig. 1b), and when photosynthetically active radiation (PAR) (Fig. 1d) and air temperature (Fig. 1f) are at their peaks. As October 2010 had more precipitation only at the end of the month, for this study October 2010 is also considered as dry season. This is supported by the fact that the length and intensity of the dry season varies from year to year (da Rocha et al., 2009).

Ambient mixing ratio measurements of isoprene, total monoterpenes and total sesquiterpenes were carried out using a commercial high sensitivity proton-transfer reaction mass spectrometer (PTR-MS, IONICON, Austria). The PTR-MS was operated in standard conditions with a drift tube voltage of 600 V and drift tube pressure of 2.0 mbar (E / N, 136 Td). During each PTR-MS measurement cycle, the following mass-to-charge ratios (m/z) were monitored: 21 (H318O+), 32 (O2+), 37 (H2O-H3O+) with a dwell time of 20 ms each; 69 (isoprene-H+), 137 (total monoterpenes-H+) and 205 (total sesquiterpenes-H+) with a dwell time of 5 s each (Jardine et al., 2011, 2012; Lindinger et al., 1998). The isoprenoid vertical profile was installed with 6 ambient air inlets at different tower heights (2, 11, 17, 24, 30, and 40 m). Air was sequentially sampled during 10 min at each of the 6 heights, resulting in one complete profile every hour. Average mixing ratios were calculated for the daytime period (10:00–16:00, LT) and for the nighttime period (22:00–04:00, LT). Calibration slope (m, ppbv/normalized counts per second (PTR-MS signal)) for isoprene, total monoterpenes, and total sesquiterpenes were obtained twice in the field using the dynamic solution injection technique (Jardine et al., 2010). Solutions of isoprene, α-pinene, and β-caryophyllene standards (> 95 % purity, Merk) in 100 mL of cyclohexane were injected into the mixing vial at 0.5, 1.0, 2.0, and 3.0 µLmin-1 (30 min each flow rate) with a constant dilution flow of 1.0 standard Lmin-1 ultra high purity nitrogen passing through. The linearity of calibrations was significant, being r2 of 0.92–0.97 for isoprene, r2 of 0.98–0.99 for α-pinene, and r2 of 0.90–098 for β-caryophyllene. Sample air isoprenoid mixing ratios were calculated by multiplying the calibration slope by normalized counts per second (PTR-MS signal) (average of two calibration slopes). Calibration slopes obtained on October 2010 were within 10 % relative to those from the calibration carried out in September 2010 (isoprene 7.2 %, α-pinene – 8.2 %, and β-caryophyllene – 2.5 %). For 4–7 days before each isoprenoid profile measurement period, ultra high purity nitrogen was run into the inlet of the PTR-MS for 2 h in order to obtain the background signals. The limit of detection for isoprene was 0.14 ppbv, 0.15 ppbv for total monoterpenes and 0.1 ppbv for total sesquiterpenes. More details about calibration and experimental design can be obtained in Jardine et al. (2011, 2012), in which a subset of these data are already described. While the previous study considered a subset of this data and time period (Jardine et al., 2011, 2012), this study examines the whole data set and focuses on seasonality of mixing ratios and fluxes. Also, this is the first study in central Amazonia that correlates long-term measurements of isoprenoids, light and temperature, and leaf phenology.

Fluxes of isoprene, total monoterpenes and total sesquiterpenes – for dry, dry-to-wet transition, and wet seasons – were estimated using the average daytime (10:00–14:00, LT) concentration vertical profile throughout the canopy and applying an inverse Lagrangian transport model (ILT) (Raupach, 1989; Nemitz et al., 2000; Karl et al., 2004, 2009). The source/sink distributions throughout the canopy were computed according to Eq. (1): C-CRef=D⋅S, where C is the concentration (gm-3) vector for the six levels, CRef is the concentration (gm-3) at reference height (40 m), D (m) is a dispersion matrix, and S (mgm-2h-1layer-1) is the resulting source/sink vector. D is expressed as a function of Lagrangian timescale and profiles of the standard deviation of the vertical wind speed (σw), which was normalized to friction velocity (u*). Integration over all source and sink terms (S) yielded the canopy scale isoprenoid flux (mgm-2h-1). To parameterize D, we use the Lagrangian timescale (Tl) parameterized according to Raupach (1989) and the vertical profile of the standard deviation of the vertical wind speed scaled to measured friction velocity. The normalized turbulence profile was taken from turbulence measurements inside and above the canopy at this site recorded as part of AMAZE-08 (Amazonian Aerosol Characterization Experiment 2008) (Karl et al., 2009). The friction velocity was averaged for each season using daytime data (10:00–14:00, LT) measured at a tower (K34 tower – 2∘36′32.67′′ S, 60∘12′33.48′′ W) that was 2 km away from the tower where isoprenoid profiles were measured (TT34 tower). The calculation of D was based on the far- and near-field approach described by Raupach (1989). As some model inputs (i.e., σw/u*) were obtained during the wet season at the TT34 tower in 2008 (Karl et al., 2009), changes in canopy structure between the two studies could potentially affect the results of this study. However, previous work carried out at the K34 tower showed that u* along with other averaged turbulence data have quite similar daytime values in both wet and dry seasons (Ahlm et al., 2010; Araujo et al., 2002).

Once fluxes from the isoprenoid vertical profiles were obtained by the ILT, they were compared with the isoprenoid fluxes estimated by the Model of Emissions of Gases and Aerosols from Nature (MEGAN 2.1). Isoprenoid emissions estimated by MEGAN 2.1 are based on a simple mechanistic model that takes into account the main processes driving variations in emissions (Guenther et al., 2012). As described by Guenther et al. (2012), the activity factor for isoprene, monoterpenes and sesquiterpenes (γi) considers the emission response to light (γP), temperature (γT), leaf age (γA), soil moisture (γSM), leaf area index (LAI) and CO2 inhibition (γCO2) according to Eq. (2): γi=CCELAIγPγTγAγSMγCO2, where CCE is the canopy environment coefficient. For the present study, the canopy environment model of Guenther et al. (2006) was used. It has a CCE of 0.57. MEGAN 2.1 was run with variation in light and temperature and LAI. Leaf age of the foliage was estimated by the model based on changes in LAI. Soil moisture and CO2 inhibition activity factors were assigned a value γSM=1 and γCO2=1, respectively, which assumes no variation in these parameters. More details about the model settings can be obtained in Guenther et al. (2012).

Photosynthetic photon flux density (PPFD) and air temperature for all model runs were obtained from the K34 tower measurement time series (The Large-Scale Biosphere-Atmosphere Experiment – LBA). LAI inputs were obtained by satellite observations from NASA MODIS during August 2010 to January 2011. The level-4 LAI product is composited every 8 days at 1 km resolution on a sinusoidal grid (MODIS-NASA, 2015).

The main source of errors for applying the ILT is related to the parameterization of two combined effects: (1) vertical diffusion coefficient which is based on measured σ(w)/u* profiles, and (2) the Lagrangian dispersion time scale (Tl). Moreover, some uncertainties may be due to systematic error sources with respect to (3) chemical losses and (4) the number of source layers. The entire parameterization of combined effect (1) and (2) was tested using data from an earlier study (Karl et al., 2009, 2010), where a comparison with eddy covariance measurements was available. Taking the above conservative error assessment, the combined (effect 1 and 2) uncertainty is +/-30 %.

To account for chemistry (effect 3) we used a simple modification of the diffusion coefficient based on Hamba (1993), relying on the fact that the chemical loss will mainly influence the far field of the parameterization. Based on estimated OH and measured O3 densities (Karl et al., 2009, 2010) calculated VOC fluxes were corrected accordingly. Due to low OH and O3 densities in the canopy (< 5×105 moleculescm-3 for OH and < 10 ppbv for O3) the chemical lifetime for isoprene and monoterpenes is considered large compared to the mixing timescale, leading to a chemistry correction on the order of < 5 % for isoprene and monoterpenes. This systematic error is included, but relies on an estimation of OH for isoprene. The overall uncertainty for isoprene is calculated as 0.3–4 % by varying in-canopy OH densities between 5×105 and 5×106 moleculescm-3. It is noted that an in-canopy OH density of 5×106 moleculescm-3 is extremely unrealistic in such a dense canopy and only serves as a very conservative upper limit. Those assumptions were also considered for sesquiterpene flux estimates. However, a sensitivity test was carried out to show if the increasing ozone concentrations during the dry season could effectively affect sesquiterpene lifetime and then sesquiterpene flux estimates. For this test, sesquiterpene lifetime was changed in the ILT model using a range from 2 min to 8 h (upper limit used for isoprene and monoterpene flux estimates). The lower limit (2 min) is based on the lifetime calculated for β-caryophyllene when it is exposed to 24 h average of 7×1011 moleculescm-3 of ozone (∼ 30 ppb) (Atkinson and Arey, 2003). If all sesquiterpenes that occur in this site have similar reactivity with ozone as β-caryophyllene, the overall uncertainty for sesquiterpene flux estimates is calculated as up to 20 % by varying sesquiterpene lifetime from 8 h to 2 min. It is noted that when considering a lifetime range from 8 h to 10 min, the uncertainty for sesquiterpene flux estimates is calculated as up to 4 %. The 20 % of uncertainty may be important only during the dry season, when ozone mixing ratios can eventually reach 30 ppbv above canopy (40 m) around noontime.

We have also investigated the effect of (4) – the number of source layers. If the number of selected source layers is too small, systematic errors of the calculated integrated fluxes arise. We have investigated this effect and found that in the present case, 6 source layers are sufficient to capture > 90 % of the flux. In the present setup, the ILT model does not converge for more than nine layers and the numerical solution becomes unstable. If the ILT model was initiated to only calculate two source layers, the integrated flux would be underestimated significantly (e.g. by up 50 %). With six source layers we estimate a systematic error of < 10 % due to this effect. The combined effect of the systematic errors (3) and (4) is estimated to be 5–6 %.

Random errors of the ILT parameterization for effects (1) and (2) mostly relate to precision. Systematic errors (3) and (4) mostly relate to accuracy of the parameterization. While there could also be combined effects of random and small systematic errors, that are difficult to assess, we chose an overall conservative error estimate that should reflect precision and accuracy for effects (1) and (2), noting that the 30 % should mostly relate to precision. All the uncertainties are 1 standard error.

With respect to uncertainties in model estimates, one of the first quantitative estimates of biogenic VOC emissions (Lamb et al., 1987) included an estimate of uncertainty of 210 % based on the propagation of uncertainties in emission factors, emission algorithms, amount of biomass, and land-use distributions. This “factor of 3” uncertainty has continued to be used as a rough assessment of the uncertainty of biogenic VOC emission model estimates applied on regional scales. A more recent study (Hanna et al., 2005) attempted a comprehensive assessment of each model component and concluded that the 95 % confidence range on the calculated uncertainty in isoprene emission was about 1 order of magnitude, while the calculated uncertainty for monoterpenes and other VOC was only ±20 %. Guenther (2013) suggests that the Hanna et al. (2005) study assigns isoprene a higher uncertainty only because more is known about isoprene, and so there are more parameters, and that the lack of observations for quantifying the uncertainties associated with individual model parameters limits the usefulness of this uncertainty estimation approach and instead recommends evaluations that consider the results of model comparisons with canopy scale observations. These studies indicate that models tend to agree with observations within ∼ 30 % for canopy scale studies with site specific parameters (Lamb et al., 1996) or for regional scale estimates with known land cover (Misztal et al., 2014) and differ by as much as a factor of 2 or more for other regional scale studies (Müller et al., 2008; Warneke et al., 2010).

The standard canopy environment model of MEGAN 2.1 was used to model light penetration into the canopy (Guenther et al., 2006). Model inputs included the above-canopy PAR measured (every 30 min) at 50 m on the K34 tower for the whole period of isoprenoid measurements as well as the estimated surface area density of the canopy (m2m-3), with measurements carried out in March 2004 using a Light Detection and Ranging sensor (LIDAR) in a transect on the same plateau area of this study (Parker and Fitzjarrald, 2004).

The light penetration was modeled for five canopy layers distributed from the canopy top to the ground surface. The thickness of each of the five layers was determined based on the canopy surface area density estimated for every 50 cm from the ground surface to the top canopy (Parker and Fitzjarrald, 2004). The layers were distributed according to a Gaussian curve fit to the canopy surface area densities (from 0.5 to 48 m). Light absorption was calculated as the difference in the model estimate of downward light at the top and bottom canopy levels. This light absorption corresponded to light that passed through the canopy vertically. Reflectance and scattering were not considered.

Leaf phenology was estimated based on the observation of leaf flushing events of the upper crown surfaces of 63 living trees around the K34 tower (∼ 2 km far of TT34 tower). For this approach, it is assumed that the leaf phenology of the upper crown surfaces of trees around both towers is similar. For the monitoring, a system of data acquisition and storage, based on a Stardot (model Netcam XL 3MP) camera with a 1024×768 resolution CMOS sensor, was installed at K34 tower, at 15–20 m above the canopy. The camera viewing angle was south azimuth, perpendicular to the solar transit, centered on 32∘ of depression and pointing out to an area of plateau. Images were logged every 15 s to a passively cooled FitPC2i with heat-tolerant SSD drive. The whole system of data acquisition automatically rebooted after power outages. The images obtained by the camera covered approximately 66∘ horizontally and 57∘ vertically, fitting the forest canopy without including any area of sky in the image. The most distant trees in the image were located 150 m from the camera. The framework was fixed by monitoring the same 63 treetops over 4 months of observation (October 2010–January 2011). The analysis of images was based on the number of treetops that showed leaf flushing within 1 month. For this, one image was selected at every 6 days, and then grouped for each month of this study.

Top-down isoprene emission estimates over the 0.5∘ region around TT34 tower were obtained by using a grid-based source inversion scheme (Stavrakou et al., 2009) constrained by formaldehyde (HCHO) columns. HCHO is an intermediate product of the isoprene degradation process (e.g. Stavrakou et al., 2014). It is measured by UV-visible sensors, such as on the Global Ozone Monitoring Experiment (GOME-2)/MetOp satellite launched in 2006. The source inversion was performed using the global chemistry-transport model IMAGESv2 (Intermediate Model of Annual and Global Evolution of Species) run at a resolution of 2∘×2.5∘ and 40 vertical levels from the surface to the lower stratosphere (Stavrakou et al., 2014, 2015). The priori isoprene emission inventory is taken from MEGAN-MOHYCAN-v2 (Stavrakou et al., 2014, http://tropo.aeronomie.be/models/isoprene.htm), and includes updates regarding isoprene emission rates from Asian tropical forests. IMAGESv2 uses HCHO columns retrieved from GOME-2 sensor as top-down constraints and estimates the posterior biogenic isoprene emission on the global scale. Note that given the early morning (09:30) overpass time of the GOME-2 measurement, and the mostly delayed production of formaldehyde from isoprene oxidation, the top-down emission estimate is dependent on the ability of MEGAN to simulate the diurnal shape of isoprene emission and on the parameterization of chemical and physical processes affecting isoprene and its degradation products in IMAGESv2. For this study, we use daily (24 h) mean satellite-derived isoprene emissions derived from January 2010 to January 2011. More details can be found in Stavrakou et al. (2009, 2014, 2015) and Bauwens et al. (2013).

Vertical profiles of isoprenoids were analyzed for daytime and nighttime for all the seasons considered in this study. Isoprene (Fig. 2a, b, c) and total monoterpenes (Fig. 2d, e, f) had higher mixing ratios during daytime (10:00–16:00, LT) than during nighttime (22:00–04:00, LT) for all seasons, supporting the findings that emissions of isoprene (Alves et al., 2014; Harley et al., 2004) and monoterpenes (Bracho-Nunez et al., 2013; Kuhn et al., 2002, 2004a; Jardine et al., 2015) from Amazonian plant species, at least at this site, are primarily light-dependent and stimulated by increasing temperature.

During daytime, isoprene had a maximum mixing ratio within the canopy. By comparison, at nighttime maximum values occurred above the canopy, and the vertical profiles were similar to those of nighttime air temperature (Fig. 2j, k, l). As isoprene is not emitted at night, this maximum nighttime abundance of isoprene above the canopy may be due to the daytime residual layer concentrations. In addition, isoprene lifetime increases during nighttime owing to the decrease of OH (hydroxyl radical) concentrations in the dark (Goldan et al., 1995) in light of the low concentrations of nitrogen oxides (NOx) in Amazonia (≤ 3 ppb above the canopy during nighttime in the dry-to-wet transition season) (Andreae et al., 2002). Similar results found at another site in central Amazonia suggested that low isoprene concentrations near the ground after sunset could be due to deposition onto and consumption by surfaces (Yáñez-Serrano et al., 2015). Isoprene up-take in the soil has been suggested previously in central Amazonia (Silva, 2010), possibly because of isoprene microbial consumption (Cleveland and Yavitt, 1997; Gray et al., 2014). As with isoprene, higher mixing ratios of total monoterpenes were observed during daytime, indicating that they are light-dependent, which agrees with the evidence of recent photosynthetic origin of monoterpenes (Jardine et al., 2015; Loreto et al., 1996).

The vertical profile of total sesquiterpene mixing ratios differed from that of isoprene and total monoterpenes for all seasons. Total sesquiterpenes had higher mixing ratios near the ground and at the sub-canopy level (17 m) than above the canopy (Fig. 2g, h, i) (P<0.05). Daytime and nighttime vertical profiles had a similar shape, but total sesquiterpene mixing ratios were higher during the nighttime. Even though sesquiterpene emissions for some plant species are both light- and temperature-dependent (Duhl et al., 2008), results reported here indicate that sesquiterpene emissions are not strongly light-dependent in this site, suggesting that their daily variation is driven primarily by temperature. Since some studies have shown that sesquiterpenes are found in the essential oil stored in Amazonian forest trees (e.g. Lima et al., 2005), emissions from these storage structures would not be expected to be light-dependent. In contrast, the monoterpenes, while also present in the essential oil of Amazonian trees (e.g. Fidelis et al., 2012; Lima et al., 2005), appear to be dominated by emissions that occur with no storage (e.g. Loreto et al., 1996; Jardine et al., 2015), similar to isoprene emission processes. Another reason for the higher total sesquiterpene mixing ratios at nighttime might be because of the reduction of oxidative reactions owing to the decrease of OH concentrations in the dark (Goldan et al., 1995) and low concentrations of nitrogen oxides (NOx) (Andreae et al., 2002), ozone, and nitrate (NO3) in Amazonia (Martin et al., 2010). In addition, ozonolysis of sesquiterpenes during daytime can reduce ambient sesquiterpene concentrations, as previously reported for a subset of these data (Jardine et al., 2011). With daytime ozone mixing ratios up to 40 ppbv (40 m) during the dry season, sesquiterpene lifetime with respect to ozonolysis above the canopy (40 m) can be 2 min during the daytime and 5 min during the nighttime (Jardine et al., 2011). Additionally, sesquiterpene concentrations can build up near the surface, because during nighttime the storage in the forest dominates (80–90 %) and is significantly larger than the turbulent flux (Karl et al., 2004).

Vertical profiles of isoprene had higher mean mixing ratios in the dry season, followed by the dry-to-wet transition season and wet season (top panel of Fig. 3a). The reduction of isoprene mixing ratios from the dry season to dry-to-wet transition season was up to 20 % and from dry season to wet season was up to 65 %. During the dry season, the higher mixing ratios and emissions of isoprene have been attributed to the higher insolation and higher temperatures compared to the wet season and, for this reason, higher isoprene concentrations at the top of the canopy are expected. Nevertheless, in contrast to the observations of Yañez-Serrano et al. (2015), who reported maximum daytime mixing ratios of isoprene at the top of the canopy for both dry and wet seasons, this study showed the highest isoprene mixing ratios inside the canopy (11 m) during the dry season, with this maximum moving to the upper canopy during the dry-to-wet transition season (24 m).

Isoprene emissions inferred from concentration vertical profiles were estimated to be highest in the sub-canopy (16 m) during the dry season and in the upper canopy (28 m) during the dry-to-wet transition season and the wet season (Fig. 4a). Even though there were differences in which layer was the highest emitter of isoprene within the canopy, mean isoprene emissions into the atmosphere were about the same for the dry season and the dry-to-wet transition season (1.37±0.7 and 1.41±0.1 mgm-2h-1, respectively). Both of these seasons had higher isoprene emissions than during the wet season (0.52±0.1 mgm-2h-1) (Fig. 4b).

The maximum absorption of PPFD by canopy, calculated based on PPFD penetration profile modeled by the standard MEGAN 2.1 canopy environment model, occurred right above the maximum of estimated surface area density of the canopy, with the absorption of PPFD being higher during the dry season, followed by the wet season and the dry-to-wet transition season (Fig. 3b). This maximum PPFD absorption at the upper canopy agreed with the maximum of isoprene mixing ratios (top panel of Fig. 3a) and emissions (Fig. 4a) during the dry-to-wet transition season. It differed, however, when compared to peaks of isoprene mixing ratios and emissions during the dry season and the wet season.

One reason for this difference could be the isoprene oxidation in the atmosphere and within plant, especially at the top of the canopy. During the dry season the ratio of methyl vinyl ketone + methacrolein + hydroperoxides (MVK + MAC + ISOPOOH) (Liu et al., 2013) to isoprene was higher compared to the dry-to-wet transition and the wet season (data not shown). This higher ratio may indicate an increased oxidative capacity of the atmosphere during the dry season. Moreover, a small source of MVK + MAC + ISOPOOH was observed at the top of the canopy (Jardine et al., 2012). Under conditions of high abiotic stress, as can occur in the dry season, elevated isoprene oxidation rates in plants can be observed and isoprene oxidation products might be directly emitted by plants (Jardine et al., 2012).

Another important factor might be leaf phenology and/or leaf demography. Different tree species have different isoprene emissions rates, and these rates depend upon the leaf ontogenetic stage. Isoprene emitters can flush at different canopy levels seasonally, and changes in within-canopy isoprene vertical profiles would be expected as a result. Moreover, as more leaf flushing was observed at the upper canopy during the wet-to-dry transition and early dry season, this caused leaves in the age group of 3–8 months to reach the highest abundance in late dry season and early wet season (Nelson et al., 2014). The period with the high abundance of leaves in this age group is coincident with the period when gross ecosystem productivity and landscape-scale photosynthetic capacity is most efficient (Restrepo-Coupe et al., 2013). Here, results show maximum isoprene emission at the upper canopy during the dry-to-wet transition season (Fig. 4a), which is coincident with the period of high abundance of healthy efficient leaves at the canopy top (Nelson et al., 2014) and also coincident with the maximum isoprene emission shown in young mature leaves in the dry-to-wet transition season (Alves et al., 2014). Similarly, higher isoprene emissions during the late dry season have also been related to the increase of active biomass in southern Amazonia (Kesselmeier et al., 2002; Kuhn et al., 2004a, b).

Although the isoprene mixing ratios reported here are within the range of previously reported values in central Amazonia for the dry season and the dry-to-wet transition season (Greenberg and Zimmerman, 1984; Rasmussen and Khalil, 1988; Zimmerman et al., 1988) and for the wet season (Yáñez-Serrano et al., 2015), these results are the lowest observed fluxes of isoprene to atmosphere reported for the Amazonia. However, this could be due to features associated with the site of this study, such as the relatively open canopy caused by the proximity to a dirt road and perhaps a relatively low fraction of isoprene emitting species. Isoprene fluxes measured previously at the same tower site during the wet season were similar (Karl et al., 2009).

Total monoterpenes also showed a strong seasonal variation with maximum mixing ratios during the dry-to-wet season, followed by the dry season and the wet season (middle panel of Fig. 3a). Taking mixing ratios of the dry-to-wet transition season as a reference, total monoterpene mixing ratios showed an increase of up to 20 % from the dry season to the dry-to-wet transition season, and a decrease of up to 50 % from the dry-to-wet transition season to the wet season. Although total monoterpene mixing ratios were somewhat higher in the dry-to-wet transition season than during the dry season, total monoterpene fluxes inferred by the vertical profiles were slightly higher during the dry season (1.47±0.06 mgm-2h-1) compared to the dry-to-wet season (1.29±0.2 mgm-2h-1) (Fig. 4e), indicating that the production is higher in the dry season and losses are also higher, leading to lower mixing ratios. In comparison, emissions from these two seasons were considerably higher than during the wet season (0.36±0.05 mgm-2h-1) (Fig. 4e). This again indicates that higher insolation and air temperature during the dry season and dry-to-wet transition season compared to the wet season increased the atmospheric concentrations of monoterpenes and, considering the enhanced ozone mixing ratios during the dry season, this may influence the seasonal pattern in monoterpene ozonolysis loss rates (Jardine et al., 2015). These results agree with branch level measurements that showed higher monoterpene emissions during the dry-to-wet transition season compared to the wet-to-dry transition season (Kuhn et al., 2004a). However, results reported here differ from those presented for the southern Amazonia, where monoterpene mixing ratios were higher during the wet season than during the dry season (Kesselmeier et al., 2002). Although only a few studies have been carried out with the objective of investigating monoterpene seasonal variations, factors other than light and temperature might influence monoterpene emissions from vegetation, including the oxidative capacity of the atmosphere and leaf phenology (Kesselmeier et al., 2002; Kuhn et al., 2004a).

Total monoterpene mixing ratios and fluxes, during the dry season and the dry-to-wet transition season, were similar to values reported for other sites in central Amazonia (Karl et al., 2007; Yáñez-Serrano et al., 2015). However, the monoterpene comparison of reported studies is a difficult endeavor given that some techniques measured total monoterpenes and others measured some specific monoterpene compounds, and also because monoterpene fragmentation during measurements (PTR-MS) could affect the absolute values of these compounds. Therefore, further efforts are needed in order to characterize the seasonal abundance and the seasonal species-specific composition of monoterpenes in the Amazonia.

Average vertical profiles of total sesquiterpene mixing ratios were higher in the dry-to-wet transition season, followed by the dry season and the wet season (bottom panel of Fig. 3a). Taking mixing ratios of the dry-to-wet transition season as a reference, total sesquiterpene mixing ratios increased up to 30 % from the dry season to the dry-to-wet transition season and decreased by up to 55 % from the dry-to-wet transition season to the wet season. During the dry season and the dry-to-wet transition season, the maximum total sesquiterpene mixing ratios were observed near the ground. During the wet season, the maximum mixing ratio was at 17 m (sub-canopy). According to Jardine et al. (2011), during the daytime many sesquiterpenes (46–61 % by mass) are rapidly oxidized by ozone as they undergo within-canopy ozonolysis and contribute to the scarcity of total sesquiterpenes above and near the top of the canopy. Considering that higher insolation and also higher ozone concentrations were observed during the dry season (ozone daily average of ∼ 23 and ∼ 10 ppbv at 40 m in the dry and wet seasons, respectively), an important fraction of the sesquiterpenes emitted by vegetation could be rapidly oxidized by ozone, leading to significantly lower mixing ratios of total sesquiterpene during the dry season (Jardine et al., 2011), which creates a need to account for sesquiterpene oxidation within the canopy when calculating emission rates.

Another potential reason for higher mixing ratios of total sesquiterpenes near the ground is that emission could come from surface sources including litter, roots and soil microbes and fungi. Silva (2010) presented surface BVOC emissions at this site, and the results suggested that the litter decomposition could be an important source of sesquiterpenes to the atmosphere. Litter production is higher during the dry than during the wet season (Luizão, 1989), which could lead to higher amounts of litter at the end of the dry season. Rain starting to increase in the dry-to-wet transition could contribute to more decomposition of the litter storage, which can potentially increase sesquiterpene emissions during the processes of decomposition of dead organic matter. Although the ecological functional role of these sesquiterpenes is not known, abiotic emissions from the litter have a specific signature that can be similar to the concentration profile in the green leaf content (Austin et al., 2014) and in sufficient concentration BVOCs can have the capacity of attracting and repelling soil organisms to a specific location (Austin et al., 2014). Therefore, higher sesquiterpene emissions from the litter could be a signal to the fauna related to the decomposition process and represent an important step of the biogeochemical cycling.

In contrast to the mixing ratios, the source-sink distribution analysis made from the vertical profiles of total sesquiterpenes indicated that the main source of these compounds is the canopy (24 m) (Fig. 4g), and the integration of sources and sinks showed that the highest total sesquiterpene emission rates going into the atmosphere was during the dry-to-wet transition season (0.77±0.1 mgm-2h-1), followed by the dry season (0.38±0.2 mgm-2h-1), and the wet season (0.34±0.2 mgm-2h-1) (Fig. 4h). However, although Nemitz et al. (2000) have suggested that limitations on the σw/u* parameterization close to the ground do not affect the net flux above the canopy, here we strongly suggest future studies focus on better characterizing the turbulence and oxidation processes at this site, in order to verify the source-sink distribution of sesquiterpenes within the canopy and the emissions from the canopy to atmosphere. This should include speciated sesquiterpene measurements in order to account for their specific reactivity with ozone and other oxidants.

Relative emissions can be calculated as emissions normalized to standard conditions of above-canopy PAR of 1500 µmolm-2s-1 and temperature of 30 ∘C. Based only on light, temperature and LAI variation, relative emissions estimated by MEGAN 2.1 were maximum during the dry season for isoprene, α-pinene, and β-caryophyllene (Fig. 4c, f, i), when the highest light and temperature were observed. This prediction differs from the ILT flux estimates (Fig. 4b, e, h), which showed similar emissions between the dry and the dry-to-wet season for isoprene and total monoterpenes and maximum emission during the dry-to-wet season for total sesquiterpenes. The overall uncertainties related to ILT flux was calculated as ±36 % and MEGAN estimates are considered to be in agreement with observations when they are within ∼ 30 %. However, more observation studies are needed in order to evaluate the degree of observation-modeling agreement and to improve model approaches, especially for total monoterpenes and total sesquiterpenes, which could present larger uncertainties due to the lack of information about atmospheric concentrations and reactivity of monoterpene and sesquiterpene chemical species in Amazonia.

To compare the seasonal variation of isoprenoid emissions with changes in environmental (light and temperature) and biological (LAI) factors in more detail, monthly fluxes of isoprenoids were compared to PAR at 51 m, air temperature at 51 m, and LAI (satellite observations – MODIS) (Fig. 5). The highest fluxes of isoprene and total monoterpene were observed when PAR was at its highest (October 2010) (Fig. 5b, c), and when there is high abundance of healthy efficient leaves (Nelson et al., 2014). The similarity in the behavior of isoprene and monoterpene emissions is supported by the evidence of the photosynthetic origin of monoterpenes (Jardine et al., 2015; Loreto et al., 1996). Interestingly, in September 2010 total monoterpene emissions were higher than isoprene emissions. This could be related to the higher source of monoterpenes in the upper canopy compared to isoprene during this month. When there are more young leaves at the upper canopy during the first half of the dry season (Nelson et al., 2014), high emissions of monoterpenes can be expected. Total sesquiterpene fluxes tracked neither PAR nor air temperature, having the highest emission when PAR and air temperature were decreasing (November 2010) (Fig. 5d).

Predictions from MEGAN 2.1 again differed from measured emissions (Fig. 5b, c, d), showing a reduction in emissions from September 2010 to January 2011. Major quantitative differences between ILT and MEGAN estimates can be shown for isoprene in September, when ILT estimates represented only 4 % of the MEGAN estimates; for total monoterpenes in December, when ILT estimates accounted for 14 % of the MEGAN estimates; and for total sesquiterpenes in November, when ILT estimates were 232 % higher than MEGAN estimates. These differences may be related to local effects, especially leaf phenology and changes in the atmospheric oxidative capacity over the seasons. In order to evaluate the potential effect of leaf phenology on emissions, leaf flushing, PAR, isoprene and total monoterpenes at canopy scale were compared in Fig. 6. They closely tracked each other during the 4 months of measurements. For the period of this study, the analysis of canopy images for every 6 days from October 2010 to January 2011 showed a decrease in leaf flushing from the end of the dry season to the wet season, which was similar to the decrease of isoprene and total monoterpene emissions and PAR. Results from 28 months (October 2010–January 2013) of canopy imaging have shown that the highest number of treetops with leaf flushing occurred during the wet-to-dry transition season (June–July), accounting for 35–50 % of treetops with leaf flushing, followed by a subsequent decrease until the end of the wet season (Tavares, 2013) (Fig. 6). Correspondingly, the results of the present study suggest that lowest emissions might be expected in the June–July time period. These results agree with those presented by Barkley et al. (2009) using remote sensing, suggesting that seasonal changes in isoprene emissions may be strongly affected by leaf phenology in the Amazonia.

In order to verify if the seasonal trend of the isoprene emissions observed in this study can also be observed in a 0.5∘ grid cell around TT34 tower, isoprene emissions estimated based on tower vertical profile concentrations are compared with estimates constrained by satellite measurements of HCHO in Fig. 7. The top-down estimates have a seasonal cycle that is similar to the bottom-up approach. Compared to the dry season, fluxes decrease by 40 % during the wet and the wet-to-dry transition season from April to July (Stavrakou et al., 2015), in qualitative agreement with the conclusions drawn in Barkley et al. (2009). The inferred dry season isoprene flux is about twice that of the wet-to-dry season. It peaks in September and gradually drops from October to January (Fig. 7), as a result of decreasing temperature and solar radiation, affecting the oxidation of isoprene leading to HCHO formation. The ground-based estimates exhibit a much stronger month-to-month variation, with flux estimates of 5 times higher in October compared to September and December. The small increase of the flux between December and January is not observed by the satellite observations. Despite these differences, partly due to reduced representativeness when comparing local measurements with flux estimates from a 0.5∘ grid cell, this comparison shows that both large (satellite) and small (ground-based) scales agree that there are enhanced isoprene emissions during the dry season followed by a reduction towards the wet season.

The results reported here are associated with a small footprint area. This together with the huge biodiversity of tropical rainforests makes it impossible to generalize these results to the regional scale. Moreover, although some previous reports have suggested significant seasonal variations of BVOCs based on in situ measurements in different sub-regions of Amazonia, when those investigations (summarized in Table 1) and this study were compared, high variability is apparent among values of mixing ratios and fluxes. This variability could be due to the following: (1) different methodologies, (2) sampling in different seasons, (3) sampling in different regions (e.g., south, north, west, eastern Amazonia), (4) sampling in different ecotones of the same region, (5) different statistical analyses, and (6) perhaps due to small data sets that are not statistically significant to characterize emissions of a specific site.

Although the canopy scale isoprenoid emission measurements presented here differed from those modeled by MEGAN 2.1 (Figs. 4, 5), which assume that variations are driven primarily by light, temperature and leaf area, in terms of seasonal variation, MEGAN 2.1 estimates of isoprene emission agreed fairly well with the satellite-derived isoprene emission, which suggests that other factors at this site could influence isoprene emissions locally. As already mentioned, leaf phenology may cause important effects on local emissions. As MEGAN 2.1 was driven with local variations in PAR and air temperature, and with regional variations of LAI (satellite observations at 1 km resolution), this regional variation in LAI may not represent the local effect of LAI variation on local emissions, since vegetation in Amazonia is phenologically distinct due to the huge biodiversity of this ecosystem (Silva et al., 2013). Furthermore, as the canopy structure might vary seasonally due to leaf phenology/demography, the pattern of light penetration/absorption and then leaf temperature may change as well; thus, this, together with the differences in emissions among species and among leaf ontogenetic stages, could have an important impact on seasonal changes of local emissions.

Besides the effects of light, temperature and leaf phenology/demography, some efforts have been made to include effects of CO2 variation (Arneth et al., 2007; Guenther et al., 2012) as well as the link between photosynthesis and emission (Grote et al., 2014; Morfopoulos et al., 2013, 2014; Unger et al., 2013) into isoprene emission models at regional and global scales. However, the current regional and global BVOC emission models predict much smaller seasonal variations (Guenther et al., 2006, 2012; Müller et al., 2008; Unger et al., 2013) compared to the measurements in Amazonia (Table 1). Furthermore, satellite observations indicate that the current understanding of the processes controlling seasonal variations is insufficient, and models do not simulate the unexpected shutdown of isoprene emission in the Amazonia during the wet-to-dry transition season (Barkley et al., 2009).

Many recently published studies have used the MEGAN model and the majority have focused on improving our understanding of isoprene emissions. Although other models have been developed on the basis of known biochemical processes (Grote et al., 2014; Morfopoulos et al., 2014; Unger et al., 2013), the general framework and processes simulated are similar. The biochemical basis of isoprene production and release must be further understood to develop mechanistic explanations for variation in isoprene emission (Monson et al., 2012), which may reduce uncertainties associated with the responses to environmental factors.

Seasonal variation of isoprene emissions might be explained by the change in energy supply from photosynthesis throughout the seasons (e.g. Grote et al., 2014). This is supported by the generally strong correlation between isoprene emission and gross photosynthetic capacity reported for Amazonian tree species (Kuhn et al., 2004b), and by the fact that higher demography of healthy efficient leaves (Nelson et al., 2014) coincides with the period of most efficient landscape-scale photosynthesis and photosynthetic capacity (Restrepo-Coupe et al., 2013). However, more measurements are needed to examine this relationship which should follow PAR variation. Additionally, since canopy structure may explain some variation in biomass growth over tropical landscapes due to differences in the pattern of light penetration and absorption by the canopies (Stark et al., 2012), measurements of canopy structure may also help to explain some of the differences in isoprenoid emissions among the Amazonian sub-regions.

Therefore, at least for the Amazonian rainforest, models currently do not fully capture seasonal variations in isoprenoid emissions, especially for monoterpenes and sesquiterpenes, which are less investigated compared to isoprene. The scarcity of measurements in Amazonia prevents the development and evaluation of accurate model approaches. Thus, this study strongly encourages future in situ measurements in Amazonia, including at leaf level, in order to verify changes driven by seasonal variations in leaf area, leaf age, phenology and emission response to soil moisture, and the short-term and long-term temperature and light environment.