This study was conducted at the Field Museum Tamakyuryo (FM Tama), a research forest facility of Tokyo University of Agriculture and Technology located in the western suburb of Tokyo (35°38′18′′ N, 139°22′41′′ E), as shown in Fig. S1 in the Supplement. A 30 m high scaffold flux tower has been constructed on the forest floor at FM Tama (168 ma.s.l.). This site has been extensively used to conduct numerous atmospheric observational and micrometeorological studies (e.g. Matsuda et al., 2015; Xu et al., 2021). Details regarding the FM Tama are described in Matsuda et al. (2015); they are briefly outlined below. The site covers approximately 20 ha. The area around the flux tower is dominated by the deciduous tree Quercus serrata, and the needleleaf tree Japanese cedar is also sparsely scattered throughout the site, creating a mixed forest character. The trees reach heights of approximately 20 m, with the canopy layer distributed between 10 and 20 ma.g.l. Q. serrata began to grow lush foliage around April and shed its foliage by around November to December. In contrast, J. cedar is an evergreen tree and generally does not shed its leaves. The J. cedar trees at FM Tama are estimated to be approximately 50 years old (Naemura et al., 2003), and the Q. serrata trees are also presumed to be roughly the same age.

The wind rose for the observation period described in Sect. 2.2 is shown in Fig. S2. Prevailing winds were predominantly southerly, followed by northerly winds and, easterly winds, with westerly winds being rare. Together with the surrounding land use, this wind pattern suggests that BVOCs measured at the flux tower were mainly influenced by emissions from FM Tama under the sampled conditions. However, possible contributions from nearby vegetation outside the research forest cannot be completely excluded. Naganuma Park, a forest park, lies west of the flux tower, but westerly winds were infrequent during the observation period.

BVOC sampling was conducted intermittently from June 2023 to October 2025. Measurements were typically performed for 1–3 d per month, except during periods when sampling could not be conducted because of equipment maintenance, power outages, or unsuitable weather conditions. Rainy days were excluded from the analysis. In total, BVOC samples were collected on 34 sampling days during the study period. The observation dates, sampling schedules, sampling durations, number of samples, and representative meteorological conditions are summarized in Table S1 in the Supplement. All times are expressed in Japan Standard Time (JST, UTC+9).

Sampling was typically conducted using 30 min integrated sorbent tube samples, during two daytime sampling windows: 10:00–12:00 and 14:00–16:00. Under the standard sampling scheme, four consecutive 30 min samples were collected during each window at each sampling height, resulting in up to eight samples per height per day. During the supplemental drone-based observations described in Sect. 2.6, the tower sampling duration was shortened to 15 or 17 min to match the drone sampling intervals and to fit within the available drone flight time and operational constraints. When samples were missing within the 10:00–16:00 daytime sampling window, daily averages were calculated only from the available observations. Note that these daily averages do not represent full diurnal averages.

The sampling design was intended to obtain multi-height BVOC observations under selected daytime conditions over multiple years, rather than to provide continuous flux measurements. Therefore, the dataset should be interpreted as multi-year intermittent observations. Monthly, seasonal and interannual features discussed in this study represent tendencies observed under the sampled conditions and should not be regarded as a complete climatology of BVOC emissions at the site. The main objectives of the observational design were to accumulate data and insights into tower-based BVOC flux measurements in a suburban forest, to examine the spatial representativeness of fixed-point observations using supplemental drone-based sampling, and to compare the observed isoprene fluxes with MEGAN estimates.

A three-dimensional ultrasonic anemometer (YOUNG, 81000) installed at the 30 m tower top measured horizontal and vertical wind directions and speed, as well as virtual temperature, and data were averaged over 10–30 min intervals. Additionally, an integrated weather sensor (VAISALA, WXT520) was installed at the flux tower heights of 25, 20, 17, and 1 m from June to 30 November 2023, and at 30, 24, 20, 17, 10, and 2 m between 22 February 2024 and October 2025 to measure air temperature, relative humidity, and horizontal wind direction and speed (Climatec CVS-HMP110 thermo-hygrometers and Climatec CYG-91000 anemometers were also used for some periods). The 10 min averages were used to obtain vertical profiles of each parameter. In cases where meteorological data could not be acquired at the flux tower due to maintenance or other reasons, measurements from the nearest Hachioji air quality monitoring station, which is managed by the Tokyo Metropolitan Government, were used.

Solar radiation including photosynthetic photon flux density (PPFD) was measured at 10 min intervals using a pyranometer (PREDE, PCM-01N) at 30 m height on the flux tower. The leaf area index (LAI) around the tower was measured for 1–3 d per month using a plant canopy analyzer (LI-COR LAI-2200). LAI was measured at four-fixed locations around the tower, and the average value of these locations was used. Therefore, LAI represents local canopy conditions around the tower rather than those of the entire FM Tama.

Based on the previously reported BVOC measurement conditions (Ichikawa et al., 2023), isoprene and seven monoterpenes (α/β-pinene, camphene, myrcene, 3-carene, limonene, p-cymene) were selected as target substances for measurement. BVOC samples were collected using an in-house sampling device as shown in Fig. S3. Air samples entering through the atmospheric inlet passed through a dehumidification tube (FUJIFILM Wako Pure Chemical, magnesium perchlorate, elemental analysis grade, 6–14 mesh) and an ozone scrubber cartridge (FUJIFILM Wako Pure Chemical, Presep-C Ozone Scrubber, potassium iodide). The air was then divided into four flow paths using a PTFE manifold. Each flow path was connected to a sorbent tube, with a two-way solenoid valve connected downstream. The opening and closing of the valves were controlled by a four-channel timer, enabling sample collection at different time intervals. All tubing was made of PTFE and the flow rate was set to 100 mLmin-1.

The sampling devices were installed at heights of 30 and 23 ma.g.l., above the main canopy layer, at 17 m within the canopy, and at 3 m near the forest floor, as shown in Fig. 1a. Sampling was generally conducted between 10:00 and 16:00 using 30 min integrated sorbent tube samples; typically, four samples were collected in the morning and four in the afternoon.

For the sorbent tube, Air Toxics (Camsco, multibed type packed with Carbograph and Carbosieve) was used from June 2023 to May 2024, and a manufacturer customized multibed type packed with Tenax TA and Carbotrap B (Camsco) was used from June 2024 to October 2025. Conditioning of the sorbent tubes was performed with a temperature program of 40 °C (hold 1 min) →10°Cmin-1→300°C (60 min) in a thermostatic chamber modified from a gas chromatography oven while high-purity nitrogen gas (> 99.9999 % purity) was flowing at a flow rate of 50 mLmin-1. Immediately after conditioning, the sorbent tubes were tightly closed at both ends with a brass cap using PTFE sealer, placed in stainless-steel containers with activated carbon, and stored in a glass desiccator under vacuum until sampling.

After sample collection, the sorbent tubes were tightly capped at both ends with brass caps, sealed in a desiccator with activated carbon to prevent contamination in an air-conditioned room, and stored until further analysis. Sample measurement was performed within 1 month of sampling.

Field blanks were used to evaluate potential contamination during transport, storage, field handling, and exposure associated with the sampling procedure. For each sampling campaign, blank sorbent tubes were transported to the field together with the sample tubes. During sampling preparation and after sample collection, the field blank tubes were uncapped and recapped at the same time as the sample tubes. During active sampling, the field blank tubes were kept capped and placed near the active sampling tubes. The field blanks were subsequently transported, stored, and analyzed in the same manner as the actual samples. The blank values were subtracted from the measured values of the corresponding samples. As the field blanks were not connected to the active sampling line, possible contamination or losses within the inlet and sampling line were not fully evaluated.

Preliminary recovery experiments were conducted to evaluate potential losses of the target BVOCs associated with the dehumidification tube and ozone scrubber cartridge. The recovery rates ranged from 64 % to 129 % among the target compounds: 87 % for isoprene, 99 % for α-pinene, 101 % for camphene, 102 % for myrcene, 64 % for β-pinene, 79 % for 3-carene, 71 % for limonene, and 129 % for p-cymene. The recovery exceeding 100 % for p-cymene does not necessarily indicate contamination, but likely reflects experimental uncertainty associated with the recovery test, including analytical variability, uncertainty in standard preparation, and quantification uncertainty at low concentration levels. Although most compounds showed recoveries within approximately 70 %–130 %, β-pinene showed relatively lower recovery of 64 %. These results indicate that large systematic losses were not observed for most target compounds, but compound-dependent uncertainty associated with the sampling pretreatment should be considered.

Based on the static dilution method described in the U.S. EPA published manual “TO-15A” (EPA, 2019), a fixed amount of each BVOC component standard solution dissolved in methanol (FUJIFILM Wako Pure Chemical, LCMS grade) was spiked into a clean inert stainless-steel vacuum canister (GL Sciences, GL-Scan). The canister was then heated at 60 °C for 2 h to completely vaporize the components. After returning to room temperature, the canister was pressurized with VOC-free high-purity nitrogen gas (>99.9999 % purity) to create a BVOC mixed standard gas. Toluene-d₈ (FUJIFILM Wako Pure Chemical, NMR grade, 99.5 % purity) was used as the internal standard gas and was prepared in the same manner as described above.

BVOCs were analyzed using a gas chromatography–mass spectrometer (GC-MS; Shimadzu, GC-MS QP2010plus) connected to an automated thermal desorption system (PerkinElmer, TurboMatrix 650 ATD). Detailed measurement conditions are described in a previous study (Ichikawa et al., 2023). These devices undergo regular manufacturer inspections to ensure that they are in good condition and free of defects. The lower detection limit (3σ) was calculated from the standard deviation (σ) obtained from repeated measurements of the standard gas (n=5–7).

According to Tani et al. (2024), Q. serrata emits isoprene but does not emit monoterpenes. As shown in the Sect. 3.1 below, volume mixing ratio of monoterpenes remained extremely low compared to isoprene, and sometimes fell below the detection limits. Considering that isoprene emission around the flux tower is primarily attributable to the dominant species Q. serrata, flux calculations were performed for isoprene only.

The aerodynamic gradient method (AGM) adopted in this study to determine isoprene emission flux is a micrometeorological method used to estimate fluxes of trace gases. The flux calculation in AGM is defined by Eq. (1) in accordance with Fick's first law (Erisman and Draaijers 1995; Fuentes and Wang, 1999; Räisänen, et al., 2009; Hayashi, 2010; Matsuda et al., 2010). 1Fobs=-K⋅∂C∂z=-u*⋅κ⋅Cu-Clln⁡zu-dzl-d-Ψhzu-dL+Ψhzl-dL where, Fobs is the observed flux (mgm-2h-1), K is the eddy diffusion coefficient (m2s-1), C represents concentration (µgm-3), z represents height (m), and the subscripts u and l denote two heights above the canopy (u>l). u* is the friction velocity (ms-1), κ is the empirically determined von Kármán constant (=0.41), Ψh is the integrated stability correction for heat, L is the Monin–Obukhov length, and d is the displacement height (m). u* and L were derived from micrometeorological elements (e.g. horizontal and vertical wind speed, virtual temperature). The value of d was derived from the vertical wind profile measured at the FM Tama flux tower by Xu et al. (2021) (April–November: 16 m, December–March: 15 m).

Sampling was conducted primarily between 10:00 and 16:00 because the AGM is less suitable under stable nighttime or early morning conditions with weak turbulence (Erisman and Draaijers 1995; Tani et al., 2024). During the observation period, the average and median friction velocities for the entire period were 0.64 and 0.60 ms-1, respectively, and no data points fell below 0.10 ms-1, a threshold used by Räisänen et al. (2009) to exclude weak-turbulence conditions. These results indicate that turbulence was generally sufficient during the sampling period. However, it should be noted that flux estimates based on AGM are subject to uncertainties associated with the concentration gradient, the eddy diffusion coefficient, the stability correction, and the assumption that the selected measurement height lies within a layer of approximately constant flux.

Supplemental drone-based measurements were conducted using a multi-rotor drone (DJI, Matrice 350 RTK) to evaluate the spatial representativeness of tower-based BVOC measurements under selected observational conditions. The drone was used as a supplementary platform for short-duration horizontal sampling around the tower, rather than as a replacement for continuous tower observations.

The swirling airflow generated by propeller rotation can cause atmospheric turbulence around the aircraft. If the placement of sample intake ports and measurement devices is not carefully considered, this turbulence may affect measurement accuracy and precision. However, it has been noted that many previous studies using drones do not address countermeasures for this swirling airflow (Altamira-Colado et al., 2024). Based on computational results from a fluid model by McKinney et al. (2019) and preliminary tests using the actual aircraft, the sample inlet, dehumidification tube, ozone scrubber cartridge and sorbent tube were mounted on a titanium support extending approximately 60 cm above the drone body, as shown in Fig. 2. This configuration was designed to reduce the influence of propeller induced airflow and aircraft vibration on the sampled air.

On 12 October 2023, 23 July 2024, and 17 June 2025, simultaneous tower and drone measurements were conducted at 30 m height. As shown in Fig. 1b, the drone sampled air at four locations approximately 15–30 m from the tower at different azimuths to evaluate horizontal variability in BVOC volume mixing ratios. The flight position was determined by acquiring positional data via the Global Navigation Satellite System (GNSS) antenna and was visually monitored from the tower. After arriving at each observation point, the drone hovered briefly to stabilize before sampling was initiated using the pump timer function. The stability of the drone at its fixed position during hovering is shown in Fig. S4. The drone equipment mounting method for these observations is shown in Fig. 2a, and each sampling period lasted 15 min. The flight altitude was precisely measured relative to the tree canopy using its onboard altitude sensor.

On 29 July 2025, an additional drone experiment was conducted to examine the sensitivity of gradient-based flux estimates to sampling location and height intervals, as shown in Fig. 1c. Drone samples were collected at 30 and 40 m, and fluxes were calculated using the AGM equation in the same manner as for the tower observations. The configuration of equipment installed for this observation is shown in Fig. 2b, which carried two pumps and other equipment. Due to safety concerns, the drone did not collect samples at altitudes below 30 m, so the altitude differs from that of the tower. After completing the first sampling at a height of 30 m at the observation point, the aircraft ascended vertically to 40 m and performed another sampling. As the drone payload limit prevented the installation of a three-dimensional ultrasonic anemometer, micrometeorological parameters measured at the tower were used for the drone-based flux calculation. This calculation assumes that the tower and drone sampling heights were influenced by similar micrometeorological conditions and that the selected height interval was suitable for a gradient-based flux estimate. As the same tower-derived micrometeorological parameters were used for the tower- and drone-based AGM calculations, differences between the tower- and drone-derived flux estimates mainly reflect differences in the measured concentration gradients and sampling height intervals. However, possible horizontal differences in turbulence caused by local canopy roughness, terrain, or flow distortion could not be evaluated. Therefore, drone-derived fluxes should be interpreted as an exploratory assessment of spatial representativeness and methodological uncertainty under tower-derived turbulence conditions, rather than as an independent quantitative validation of the tower-based fluxes.

We compared the tower-based observed isoprene emission fluxes with model calculations from MEGAN (Guenther et al., 2006, 2012), an extensively used BVOC emission estimation model, to evaluate model–observation agreement under the sampled conditions. MEGAN calculations were conducted at a 10 min time resolution for the daytime period from 10:00–16:00 for each day from 1 June 2023 to 31 October 2025. Thus, the model was run continuously for the daytime sampling window throughout the study period, rather than only for the individual observation days. For comparison with the observations, model outputs corresponding to the actual sorbent tube sampling times were extracted and paired with the observed fluxes. Additionally, PPFD, meteorological variables, and LAI data were obtained as described in Sect. 2.3. The comparison therefore represents model–observation agreement during the sampled daytime periods and should not be interpreted as an evaluation of full-day emission estimates. The estimation of isoprene emission flux Fcal (mgm-2h-1) was performed using the following Eq. (2), with parameters set based on the original literature for MEGAN (Guenther et al., 2006, 2012). 2Fcal=ε⋅γP⋅γT⋅γLAI⋅γage⋅γSM⋅γCO2⋅ρ where, ε is the basal emission rate under standard conditions for the plant functional type (PFT), and we adopted the standard value of 10 mgm-2h-1 for isoprene from broadleaf deciduous temperate trees (Guenther et al., 2012). The standard conditions for MEGAN are described in Guenther et al. (2006, 2012); for example, a leaf temperature of 30 °C, PPFD of 1500 µmolm-2s-1, and a standard canopy structure (LAIstd) of 5. Although a PPFD of 1000 µmolm-2s-1 is typically used as the standard condition at the leaf level in the G93 algorithm (Guenther et al., 1993), MEGAN uses 1500 µmolm-2s-1 at the canopy scale. The γ terms represent activity factors related to environmental and phenological conditions, where γP for light response, γT for temperature response, γLAI for LAI, γage for leaf age, γSM for soil moisture response, and γCO2 for carbon dioxide response. The escape efficiency ρ, which is the proportion of isoprene emitted from the canopy that escapes into the upper atmosphere, was set to 0.96 following Guenther et al. (2006). The settings of the activity factors used in this study are described below.

Figure 3 summarizes the monthly average volume mixing ratios of the eight target BVOC species at each sampling height during the intermittent sampling campaign. The corresponding numerical values are summarized in Table S3. The averages reported in this section were calculated from available daytime samples collected between 10:00 and 16:00 on the observation days and should therefore be interpreted as summaries of the sampled conditions rather than complete monthly averages.

From November to April, volume mixing ratios of all target compounds were generally low at all heights. In contrast, from May to October, isoprene volume mixing ratios increased markedly, with several samples exceeding 10 000 pptv during July and August. During peak summer observations, isoprene accounted for more than 90 % of the measured BVOC composition at all heights (Fig. S5). Although the detailed mechanisms underlying isoprene emission in plants remain incompletely understood, the widely accepted explanation is that it protects leaves from high temperatures (Sharkey et al., 2008). This study also observed a tendency for isoprene volume mixing ratios to increase during periods of high temperature.

Monoterpene volume mixing ratios were much lower than those of isoprene throughout the campaign. Among the measured monoterpenes, α-pinene, limonene, and p-cymene were relatively abundant, but each generally accounted for only about 1 %–2 % of the total measured BVOC volume mixing ratio during summer observations. Distinct vertical gradients of monoterpenes were not evident. These results are consistent with previous reports that Q. serrata is primarily an isoprene-emitting species and emits few monoterpenes (Tani et al., 2024). Therefore, the following sections focus on isoprene.

Figure 4 shows the vertical distribution of isoprene volume mixing ratios at different heights. The boxes represent the 25th–75th percentile ranges, the whiskers represent the 2nd–98th percentile ranges, and the median and average are overlaid.

During the warm-season observations (May to October), when Q. serrata leaves were densely foliated, isoprene volume mixing ratios were generally highest at 17 m, corresponding to the canopy layer. The distributions at 17 m showed higher upper-percentile values than those at the other heights, indicating that enhanced isoprene volume mixing ratios were frequently observed within the canopy, where Q. serrata leaves were concentrated (the canopy range of 10–20 m). At 23 and 30 m, above the main canopy layer, the distributions also showed extended upper tails during the warm-season, suggesting upward transport of isoprene emitted from the canopy. In contrast, isoprene volume mixing ratios at 3 m near the forest floor were generally lower and showed narrower distributions.

During the cold-season observations (November to April), isoprene volume mixing ratios were low at all heights, and clear vertical gradients were not evident. These results indicate that the vertical structure of isoprene was most apparent during the foliated warm-season period.

Figure 5 shows the daily average isoprene emission fluxes (Fobs) calculated from the vertical gradient of isoprene concentration between 23 and 30 m. Canopy temperature and PPFD during the observation days are also shown in Fig. 5. The daily average values and standard deviations for Fobs and related meteorological parameters are summarized in Table 1. Fobs ranged from -0.05 to 15.30 mgm-2h-1 during the observation period.

Positive isoprene fluxes were mainly observed during the warm-season sampling days, especially from June to September. Fluxes were generally low during the cold-season observations and during early spring. High fluxes were observed on several summer sampling days, particularly in 2025. As the sampling was intermittent and meteorological conditions differed among observation days, these differences should be interpreted as variations among sampled summer conditions rather than indicating an interannual trend. Possible meteorological controls are discussed in Sect. 4.1.

Figure S6 shows the relationships between the isoprene flux and canopy temperature, PPFD, and LAI using the unaveraged dataset (n=263). As isoprene emissions exhibit exponential or saturated nonlinear responses to temperature, PPFD, and LAI (Guenther et al., 1993, 2006; Yu et al., 2017), Spearman's rank correlation coefficient (rs) was used. Canopy temperature (rs=0.71, p<0.001), PPFD (rs=0.49, p<0.001), and LAI (rs=0.58, p<0.001) showed positive monotonic relationships with isoprene flux. These relationships are consistent with the known temperature, light, and phenological controls on isoprene emissions. However, because the dataset was obtained from intermittent daytime sampling, these correlations should be interpreted as relationships within the sampled conditions rather than as a comprehensive characterization of seasonal or interannual controls.

To further examine the combined effects of temperature and light, we also examined the relationship among Fobs, canopy temperature, and PPFD using a three-dimensional scatter plot (Fig. S7). Under high temperature and high PPFD conditions, Fobs tended to be high. However, some data points showed relatively low Fobs even when both canopy temperature and PPFD were high. This result indicates that temperature and light were important drivers of isoprene flux under the sampled daytime conditions, but they did not fully explain the observed flux variability. To analyze the factors involved, we first calculated the 75th percentile (Q3) for both temperature and PPFD from the entire observation dataset (n=263). We then extracted data points simultaneously satisfying temperature ≥ Q3 (approximately 32 °C) and PPFD ≥ Q3 (approximately 1571 µmolm-2s-1). The Fobs distribution of this subset (n=37) was classified as follows. L-Fobs denotes the group below the 25th percentile (approximately 1.5 mgm-2h-1, n=10), and corresponds to the data points marked with black circles in Fig. S7. Although we were unable to identify a clear cause for the occurrence of L-Fobs, other factors, such as canopy-scale source distribution, local light environment, turbulent transport, concentration-gradient uncertainty, and tower-footprint representativeness, may also have contributed to the variability in Fobs.

The tower-based observations showed positive isoprene fluxes mainly during the warm-season sampling days, especially from June to September. The positive relationships between isoprene flux and canopy temperature, PPFD, and LAI indicate that the observed fluxes were influenced by temperature, light availability, and canopy development under the sampled daytime conditions. As the observations were conducted intermittently and did not cover complete diurnal cycles or all meteorological conditions, these relationships should be interpreted as tendencies within the sampled conditions rather than as a complete characterization of seasonal or interannual controls.

Isoprene production is strongly influenced by the availability of synthetic substrates, such as dimethylallyl diphosphate, generated through photosynthetic processes within chloroplasts, and by the activity of isoprene synthase. Photosynthesis depends on light availability, whereas isoprene synthase activity is strongly affected by temperature (Rasulov et al., 2010; Chen et al., 2022). Therefore, isoprene emissions are considered to correlate with temperature and PPFD. These physiological controls are consistent with the positive relationships observed between Fobs, canopy temperature, and PPFD in this study. However, according to Rasulov et al. (2010), the pool size of dimethylallyl diphosphate varies with temperature and decreases under high temperature conditions exceeding a threshold, suggesting that elevated temperatures may influence isoprene emission rates. Guenther et al. (1993) observed the relationship between isoprene emission and leaf temperature in isoprene emitting trees – Sweetgum, Eucalyptus, Aspen, and Velvet bean. They reported that, depending on the tree species, the emission peak occurred at approximately 35 °C, with emission levels decreasing above approximately 40 °C. Exceeding the threshold temperature is thought to be associated with the inactivation of the synthetic substrate. In this study, the highest canopy air temperatures (maximum of 35.7 °C) during sampling approached this range. However, leaf temperature was not measured directly, and canopy air temperature was used as a proxy. Therefore, the possible occurrence of thermal optimum or high-temperature suppression cannot be evaluated quantitatively from the present data.

Leaf phenology also likely contributed to the observed flux variability. Young leaves generally have lower isoprene emission capacity than mature leaves (Kuzma and Fall, 1993), and Q. serrata at FM Tama develops new leaves in spring and maintains mature leaves during summer. The low fluxes observed in early spring are therefore consistent with the expected lower emission capacity of young leaves, whereas higher fluxes during summer are consistent with mature foliage and favorable light and temperature conditions. LAI showed a positive relationship with isoprene flux, although this relationship is expected to saturate when canopy light availability becomes limiting (Yu et al., 2017).

Several high-flux events were observed during the summer 2025 sampling days. According to the Japan Meteorological Agency (2025), the 2025 rainy season ended earlier than usual, and in June and July, record-high temperatures were observed across northern, eastern, and western Japan, with the summer average temperature deviation reaching +2.36 °C, significantly surpassing past records. This was attributed to the Tibetan anticyclone extending over Japan as a result of the upper-level westerlies flowing northward from June onward, influenced by the early and active development of the Asian monsoon. Furthermore, the expansion of the Pacific anticyclone coincided with this, bringing clear skies and rising temperatures. Consequently, the cumulative number of locations recording daily maximum temperatures of 40 °C or higher also set a new record.

The average temperatures at FM Tama between 10:00 and 16:00 during the summers of 2023, 2024, and 2025 were 31.96, 29.34, and 33.19 °C respectively. Meanwhile, the average PPFD was not higher in 2025. Furthermore, no differences were observed in wind direction or wind speed among the observation days during the summer. This suggests that the high Fobs values in 2025 were likely associated with high-temperature sampling conditions. However, because the number of sampling days was limited and meteorological conditions differed among observation days, the present dataset cannot separate interannual variability from day-to-day meteorological variability.

Previous studies have shown that warming and heat-wave conditions can enhance BVOC emissions and influence atmospheric chemistry. Kramshøj et al. (2016) reported enhanced BVOC emissions under experimental warming in a tundra ecosystem, although the climate and vegetation differ substantially from those at FM Tama. Megaritis et al. (2013) predicted that a 2.5 °C rise in temperature in Northern Europe would lead to an increase in BVOC emissions, resulting in a 20 % increase in summer biogenic SOA. Moreover Churkina et al. (2017) estimated that increased BVOC emissions during heatwave events in the Berlin-Brandenburg metropolitan area of Germany contributed up to 60 % of O3 formation, indicating that the impact of BVOC increases due to high temperatures on the atmospheric environment is significant. Furthermore, Wang et al. (2024) warn that during heat waves, plants not only increase the production of O3 precursors but also close their stomata, thereby reducing O3 removal by dry deposition; consequently, the increasing frequency of heat waves poses a major challenge for air pollution control. Therefore, continued BVOC flux observations under high-temperature conditions are important. Nevertheless, the present study should be regarded as providing observational evidence of high isoprene fluxes under selected hot daytime conditions, rather than a quantitative assessment of climate-change-driven trends in BVOC emissions.

The supplemental drone-based measurements provided information on short-range horizontal variability around the flux tower. The tower–drone comparison at a 30 m height indicated that isoprene volume mixing ratios differed by approximately 10 % within 15 m and by up to approximately 30 % at 30 m under the sampled conditions. These results suggest that even near a fixed flux tower, horizontal variability in canopy structure, leaf density, light environment, and turbulent transport can affect the interpretation of fixed-point BVOC observations.

However, the drone-based flux comparison should be interpreted with greater caution than the volume mixing ratio comparison. Drone-derived fluxes were calculated from the 30–40 m height interval, whereas tower-based fluxes were calculated from the 23–30 m height interval. In addition, the drone samples at 30 and 40 m were collected sequentially rather than simultaneously, and micrometeorological parameters measured at the tower were used because the drone could not carry a three-dimensional ultrasonic anemometer. Therefore, differences between tower- and drone-derived fluxes reflect not only horizontal variability in isoprene volume mixing ratios but also differences in the height interval, sampling timing, and assumed turbulence conditions.

Future drone-based BVOC flux studies would benefit from simultaneous multi-height sampling and direct turbulence measurements on or near the drone platform. As drone payload capacity and flight endurance improve through advances in battery technology, longer sampling periods and the installation of additional micrometeorological sensors may become possible. These developments would help reduce uncertainties in drone-based BVOC flux estimates and strengthen their role as a complementary approach to tower-based observations.

MEGAN generally overestimated the observed tower-based isoprene fluxes during the April–October sampling periods, with the largest MBE and RMSE during summer. As the comparison was based on time-synchronized paired data from intermittent daytime sampling, the results should be interpreted as model–observation differences under the sampled conditions rather than as a complete seasonal model evaluation. The larger summer discrepancy suggests that model uncertainties became more apparent under high-temperature and high-light conditions, when γT and γP were relatively large.

The larger summer discrepancy is important because summer is the period when isoprene emissions and photochemical activity are both enhanced. In urban and suburban environments, uncertainty in BVOC emission estimates can influence simulations of O3 and SOA formation, where biogenic precursors interact with anthropogenic nitrogen oxides. Therefore, reducing model–observation discrepancies under summer high-temperature and high-light conditions is important for improving BVOC emission estimates and their application to air quality modeling.

Both the observed fluxes and the MEGAN estimates include substantial uncertainties. The observed fluxes were estimated using the AGM, which depends on the measured concentration gradient, eddy diffusivity, stability correction, displacement height, and the assumption that the selected measurement heights were within a layer of approximately constant flux (Erisman and Draaijers 1995). In forest canopies, especially within or near the roughness sublayer, this assumption can be difficult to fully satisfy. Chemical loss or transformation of reactive BVOCs within and above the canopy may also influence the relationship between their emission and measured flux. Therefore, part of the model–observation discrepancy may reflect uncertainty in the observed flux estimates themselves.

Uncertainty in the basal emission rate (ε) is another likely source of discrepancy. MEGAN uses PFT-based standard emission factors, but these values may not fully represent the dominant species and local canopy conditions at FM Tama. Langford et al. (2017) raised concerns that ε in MEGAN was calculated based on a very limited number of leaf-level observations, resulting in considerable uncertainty. Tani et al. (2024) also emphasized that differences in measurement methods can cause variability in reported ε even for the same species. For Q. serrata, Okumura et al. (2008) reported leaf-level basal emission rates of 42.9 nmolm-2s-1 for sunlit leaves and 20.5 nmolm-2s-1 for shaded leaves. These values indicate that the emission capacity of Q. serrata can differ substantially between different leaf light environments. In the present study, we used the PFT-based standard emission factor in MEGAN to provide a baseline comparison with the tower-based flux observations. A full sensitivity analysis and validation using species-specific and sun/shade-specific emission factors would require additional assumptions regarding canopy structure, leaf area distribution, sunlit and shaded leaf fractions, and scaling from leaf-level emission capacity to canopy-scale fluxes. Such an analysis is therefore beyond the scope of the present study, but it represents an important next step for improving the MEGAN evaluation at this site.

The temperature response factor γT may also contribute to the discrepancy. In the present study, Fcal showed strong relationships with γT and γP, reflecting the structure of the MEGAN algorithm. Previous studies suggest that temperature response parameterizations may vary among ecosystems and species. Emmerson et al. (2020) measured isoprene using the leaf cuvette method across four Eucalyptus species at temperatures ranging from 293–328 K in 5 K increments, reporting that the temperature-dependent peak occurred at higher temperatures than those reported in the original MEGAN study. DiMaria et al. (2023) argued that ecosystem-specific γT parameterizations are needed to improve the representation of temperature sensitivity. At FM Tama, the high-temperature summer sampling conditions may have amplified uncertainties in the temperature response function and in the representativeness of canopy air temperature for leaf temperature.

Canopy-scale representativeness of the input variables is another important issue. MEGAN was driven by observed meteorological variables and LAI, but the actual foliar environment within the canopy can vary vertically and horizontally. Sunlit and shaded leaves experience different light and temperature conditions, and these differences may not be fully represented by tower-level PPFD, air temperature, or local LAI measurements. The drone observations also indicated short-range horizontal variability in isoprene volume mixing ratios near the tower. These factors may partly explain why a uniform correction to MEGAN is insufficient to reproduce the observed flux variability.

Overall, the MEGAN comparison in this study should be interpreted as an evaluation under intermittent daytime sampling conditions. The results indicate potential sources of model–observation discrepancy, including basal emission factors, temperature response functions, canopy microclimate, spatial representativeness, and AGM-based flux uncertainty. However, because the observations were limited to selected daytime periods, the comparison does not represent a complete seasonal or annual model evaluation.