The data used in this work were measured in 2022 and 2023 at the ATTO experimental site, located in the central Amazon, about 150 km northeast of Manaus, in the Uatumã Sustainable Development Reserve (SDR) (Fig. a). The site is characterized by dense tropical forest with a relatively uniform canopy, whose tree crowns generally reach a height of around 37 m . A detailed description of the site is provided by .

We used fast-response data (10 Hz) collected by sonic anemometers installed on two towers (located at 2.148° S, 59.0068° W, and 2.1448° S, 59.0008° W) that are 670 m apart . The 81 m tower (INSTANT Tower) and the 325 m tower (ATTO Tower) are on a plateau, located at approximately 130 m a.s.l. . The sensors used included CSAT-3B models from Campbell Scientific, Inc. (Logan, USA) installed at 50 and 81 m on the INSTANT Tower, and 3D Ultrasonic Anemometer model 4.3830 series from Thies Clima3D (Göttingen, Germany) installed at 100, 127, 151, 172, 223, 247, 274, and 298 m on the ATTO tower, all heights positioned above the ground. The variables measured were air temperature (T) and the three components of wind velocity (u, v, and w). Measurements from both towers were integrated into a single vertical profile, since previous studies did not identify significant differences between them .

Relative humidity and air temperature were obtained by thermohygrometers model IAKM I-Series from Galltec-Mela (Bondorf, Germany), and atmospheric pressure was recorded by barometers model 61302V (Young, USA), both operating at 0.01 Hz at the same heights as the anemometers. Net radiation (Rn) was calculated as the sum of the shortwave radiation components (incoming and reflected), measured by a CMP21 pyranometer, and the longwave radiation (incoming and emitted), recorded by a CGR4 pyrgeometer (both Kipp & Zonen, Delft, Netherlands). These measurements were carried out at 81 m on the INSTANT tower, at 0.01 Hz. CO and CH₄ concentrations were measured once per hour by a gas analyzer (FTIR Spectrometer) installed at the foot of ATTO tall tower, sampling air at five heights: 42, 81, 150, 273, and 321 m . Cloud fraction derived from the METEK MIRA-35C Cloud Radar Profiler and GOES-16 IR Channel 13 (10.3 µm) brightness temperature over the ATTO site were also used as complementary data for the case-study analysis. All analyses considered the nighttime period from 00:00 to 09:00 UTC (20:00 to 05:00 LT; UTC-4), avoiding transition intervals.

The main analyses were conducted for 2022–2023, when the complete sonic anemometer profile (up to 325 m) was available at ATTO. However, measurements of net radiation (Rn) at 81 m and turbulence measurements from the sonic anemometer at 50 m were available for the period 2016–2024 and were used to provide a broader climatological context.

As in other micrometeorological studies , the high-frequency data were initially segmented into 30 min intervals (totaling 18 000 measurements) for the application of quality control (QC) procedures. QC was performed for each wind component and for the temperature measured by the sonic anemometers. The QC process included verification of record integrity, detection of error indicators, identification and removal of spikes, as well as the application of stationarity tests. For a detailed description of the quality control protocol adopted at the ATTO Site, it is recommended to consult and .

After QC, the planar fit method was used for tilt correction. For each six-month period, wind velocity components were used to estimate the anemometer tilt angles and define a mean streamline coordinate system. The tilt-corrected wind vectors were then rotated for each 30 min interval to align the horizontal wind components with the mean wind direction . Subsequently, turbulence statistics were calculated over 5 min windows using Reynolds averaging, after applying linear detrending within each window. This shorter time window reduces the influence of non-turbulent motions, which are relevant under stable boundary-layer conditions .

H was obtained by the eddy covariance method , according to the equation: H=ρcpw′T′‾, where ρ is air density (we assumed 1.225 kg m⁻³), cp is the specific heat of air at constant pressure (J kg⁻¹ K⁻¹). The variables T′ and w′ represent the fluctuations of temperature and of the vertical wind component, respectively. The mean wind speed (Um) was calculated as Um=u2+v2, where u and v are the zonal and meridional wind components, respectively. The potential temperature (θ) was calculated as θ=Th(p0/P)Rd/cp, where Th is the air temperature measured by the thermohygrometer (Kelvin), P is atmospheric pressure (hPa), p0 is the reference pressure (1000 hPa), and Rd is the gas constant (287 J kg⁻¹ K⁻¹). Profiles of H were used to estimate hn, while Um and the vertical gradient of θ were used to discuss its variability.

Atmospheric stability was evaluated using the Monin–Obukhov stability parameter (z/L), where L represents the Obukhov length, determined as L=-u*3T/(kgw′T′‾), with u* the friction velocity, k the von Kármán constant (0.41), g the acceleration due to gravity, and w′T′‾ the kinematic sensible heat flux. The parameter z/L was calculated considering the variables u*, w′T′‾, and T at 50 m, similar to that performed by . Vertical wind shear (dU/dz) was quantified from third-order polynomial fits applied to the wind speed profiles, from which analytical vertical derivatives were calculated.

After QC and calculation of turbulence statistics, two filters were applied to the dataset of the 5 min. First, to avoid the influence of extreme values, outliers in the H measurements were identified by means of histograms by height layer. Based on the observed distribution, the filtering limits were defined empirically, resulting in the following specific ranges: from 50 to 81 m (-50 to 0 W m⁻²), from 100 to 151 m (-30 to 10 W m⁻²) and from 172 to 298 m (-20 to 10 W m⁻²). Values outside these ranges were considered outliers and removed from the analysis. To test the sensitivity of hn to these filtering limits, we applied a more permissive scenario (-60 to 20, -45 to 25, and -35 to 15 W m⁻²; Scenario 2) and a more restrictive scenario (-30 to 0, -20 to 5, and -10 to 5 W m⁻²; Scenario 1) for the three height ranges. The normalized profiles were very similar, and the estimated hn differed by less than 15 m among scenarios (Fig. S1 in the Supplement), indicating that the results are robust to the selected thresholds. The application of differentiated limits reflects the expected variation of H among the layers, being more intense in the lower layers due to direct interaction with the canopy, and smoother at higher altitudes .

The second filter considered very stable atmospheric conditions, characterized by very low sensible heat fluxes above the canopy, indicative of minimal energy exchange, configuring a strongly decoupled regime . In these situations, the vertical profile of H does not present a typical structure (higher values near the canopy and a decrease with height), but rather an irregular pattern, which makes the reliable determination of hn difficult. So, cases in which the H values at 50 m were between 0 and -6 W m⁻² were excluded, a range identified as the most frequent based on the data distribution under very stable conditions (1≤z/L≤10, such as ).

Subsequently, the data were classified by seasonal period (Dry and Wet) in each year, with 2022 associated with La Niña and 2023 with El Niño (Table ). According to the NOAA Climate Prediction Center, La Niña conditions persisted from late 2021 to early 2023, whereas a strong El Niño developed around May 2023 and lasted until early 2024. In the central Amazon, La Niña typically brings wetter conditions and slightly cooler temperatures, while El Niño tends to produce drier and warmer conditions, particularly during the Dry season, due to reduced convective activity and changes in large-scale atmospheric circulation .

The Dry period was defined as the months of August, September, and October, and the Wet period encompassed the months of January to April. The inclusion of January is justified by the scarcity of data available in April, which could compromise the robustness of the analyses. Considering that January still falls within the region’s Wet season, its inclusion increases the representativeness of the data. Moreover, showed that January remains among the rainiest months for ATTO’s area. This division aligns with regional seasonal precipitation patterns in the Central Amazon described by and, for ATTO specifically, by . From this classification, hourly means of all analyzed variables were calculated at their respective measurement heights.

After filtering and classification, hourly mean profiles of H were obtained from the 5 min data, allowing the estimation of hn for each hour of the nighttime period, separately for the Dry and Wet seasons of 2022 and 2023. Since the sensible heat flux is greater near the canopy and decreases with height, hn was identified as the height at which H was less than 5 % of the H value measured near the canopy (50 m) . This turbulence-based approach is consistent with the fundamental definition of the atmospheric boundary layer. To allow comparison among profiles, the H values at all heights were normalized by the value measured at 50 m (H50m).

In addition to the seasonal and hourly analyses, we selected two nights as case studies to investigate the relationship between variability in hn and the vertical concentration profile of CO and CH₄: one with a shallow hn layer (hn∼80 m) and another with a deeper layer (hn∼150 m). The idea here was to investigate to what extent hn affects the temporal and spatial gas concentration variations along the tower.

Figure  shows the hourly mean values of Rn and stability parameter (z/L) for the Wet and Dry periods in the years 2022 and 2023. Both the Wet and Dry seasons of 2022 presented Rn values with larger magnitude than those observed in the Wet and Dry seasons of 2023 (Fig. a, b). In addition, |Rn| values for the Dry season are greater than those observed in the Wet season, in both years. Cloud cover has a direct relationship with Rn values . Greater cloud cover is associated with lower radiative loss, that is, smaller |Rn| values. Therefore, in the Dry season, where cloud cover is lower, there is greater radiative loss. The same reasoning applies to the year 2023, an El Niño year characterized by reduced cloud cover over the central Amazon . Atmospheric stability also varies systematically between seasons (Fig. c, d). The Wet period is less stable (smaller z/L=2.15), while the Dry period shows stronger stability (z/L=3.10) consistent with enhanced radiative cooling and reduced turbulent mixing. El Niño in 2023 enhanced this effect (Fig. d).

To place the analyzed years in a broader climatological context, Fig.  shows the monthly nocturnal climatology of Rn and z/L for the 2016–2024 period, using the same nighttime interval adopted in this study (00:00–09:00 UTC). The climatology highlights a clear seasonal cycle, with stronger nocturnal radiative cooling during the dry season associated with higher z/L values and therefore more stable atmospheric conditions. In 2023, z/L values were higher than the climatological mean during most months, especially from May onward, whereas 2022 remained closer to, or below, the climatological mean values. These departures from the ATTO climatology indicate that the two analyzed years experienced distinct nocturnal radiative cooling and stability conditions. Since Rn and z/L are relevant controls on turbulence intensity and the vertical structure of the nocturnal boundary layer, these differences may be related to the contrasting hn behavior observed in 2022 and 2023 (discussed in Sect. 3.2).

Figure  presents the normalized vertical profiles of sensible heat (H/H50). The turbulent heat fluxes cease at different heights when comparing the Dry and Wet seasons, which indicates important seasonal variations of hn. During the Wet period, turbulence extends up to approximately 200±33 m in 2022 and 150±24 m in 2023, while in the Dry period this height is reduced to about 120±28 m in 2022 and 100±21 m in 2023.

Figure  presents the hourly mean values of hn for the Wet and Dry periods for the two years analyzed here. In 2022, hn exhibited seasonal variations, being 274 m at the beginning and 151 m at the end of the night during the Wet season, while in the Dry season the values range between 223 m (beginning of the night) and 100 m (end of the night). In 2023, the seasonal variation of hn is smaller between the seasons, ranging between 172 m at 00:00 UTC and 100 m at 09:00 UTC. In addition, regardless of the season or the year, the hn values are higher at the beginning of the night and decrease progressively.

In general, hn shows a gradual decrease throughout the night (Fig. ). However, these changes in hn values can be reflected in the vertical profiles of trace gasesmeasured along the tower. In order to explore this relationship, we refer to Fig. , which shows two case studies: one in which the hn values remained low throughout the entire night (18 August 2022), with values around 81 m, and another in which hn showed a progressive increase from about 100 to 170 m throughout the night (25 August 2022).

The wind profiles (Fig. ) show the occurrence of low-level jets (LLJs) on both nights, but with distinct characteristics. On the night of 18 August, the jet nose (maximum speed in the wind profile) was located at a height that oscillated around 80 to 100 m, while on the night of 25 August the jet nose was located around 170 m. Another important difference is observed in the dominant wind direction during the two nights (Fig. ). During the night of 18 August, the predominant wind direction was from the northeast at the lowest levels and from the southeast at the highest levels. During the night of 25 August, the wind direction ranged between north and northeast from the top of the canopy up to 298 m. According to , the jet nose (maximum speed in the wind profile) can serve as an indicator of hn, which is consistent with what we observe in our case studies.

To better identify the main drivers controlling the contrasting hn behavior during the two case-study nights, we further analyzed Rn at 81 m, cloud cover, the vertical profile of θ and wind shear (dU/dz) (Fig. ). These variables provide complementary information on the local radiative forcing, cloud modulation of nocturnal cooling, thermal stratification, and mechanical turbulence generation. The night of 18 August 2022 was characterized by strong nocturnal radiative cooling (Fig. a), reduced cloud occurrence (Figs. c and S2), and a pronounced vertical potential temperature gradient (Fig. e). These conditions indicate enhanced thermal stratification near the surface. Although wind shear was observed at some levels (Fig. g), the strong stratification suggests that mechanically generated turbulence aloft was not efficiently coupled downward toward the canopy. This supports the interpretation that the shallow hn observed on 18 August 2022 reflects not only strong thermal stratification, but also canopy-induced suppression of vertical mixing near the forest top. This configuration is consistent with a shallow surface-based nocturnal layer that remained partly decoupled from the air above, similar to the layered nocturnal structure described by . In contrast, the night of 25 August 2022 showed larger cloud cover (Figs. d and S3), weaker radiative cooling during the early part of the night (Fig. b), and weaker thermal stratification (Fig. f). Wind shear extended to higher levels (Fig. h), reaching approximately 150–170 m, close to the observed hn. This suggests that, under weaker stratification, mechanically generated turbulence was more effectively coupled over a deeper layer, contributing to the larger and more variable hn observed during this night.

Figure  shows the CO and CH₄ concentrations for the two case studies. During the night of 18 August, the propagation of an air plume rich in CO and CH₄ that reached the tower was observed. However, this air mass was only detected by the inlets located at the highest heights (321 and 273 m), producing strong vertical differences in the tower CO and CH₄ profiles. The observation of elevated CO and CH₄ concentrations at higher altitudes, together with the different flow direction above the hn (Figs.  and ), suggests that these air masses originated from more distant sources. Air arriving from the SE–S sector (above the NBL, Fig. ) likely transported high concentrations of CO and CH₄, possibly associated with regional biomass burning plumes (Fig. S4). In addition, it is believed that the plume did not penetrate down to canopy level (50 and 81 m) because the shallow hn that night acted as a barrier (Fig. ). Meanwhile, air masses originating from the N–NE (below the NBL, Figs.  and ) exhibited lower CO and CH₄ concentration. We hypothesize that the height of 150 m is in the transition between the two layers, which explains why the CO and CH₄ concentrations coincide partly with those observed at 42 and 81 m. It is also noted that around 07:00 UTC the CO and CH₄ concentrations at the higher heights begin to decrease, reaching values similar to those observed at the lower heights at 09:00 UTC, which could indicate that the tower is leaving the concentration plume, caused by a change in wind direction at higher heights which shifted from SE–S to N–NE. To clarify, the layers remain decoupled, but the flow in both layers comes from the same direction and is no longer influenced by air masses with elevated CO and CH₄ concentration (Fig. S4a).

On the night of 25 August, CO concentrations were higher than those observed on 18 August, indicating a multi-day pattern in which CO levels were generally elevated across the region (Fig. S6), as more often observed in the region . At 00:00 UTC on 25 August the highest CO and CH₄ concentrations were observed at the heights closest to the forest canopy, thereby different than at 18 August (Fig. b). Although plumes typically arrive first above the canopy, this apparent inversion can be explained by the slower response of air within the canopy to atmospheric changes. Elevated air from previous days (before 25 August; see Fig. S5) may have penetrated the canopy, and while the passing plume quickly removed high concentrations from the upper layers, ventilation near the canopy occurred more slowly, resulting in higher concentrations at lower levels. Therefore, at 00:00 UTC, it appears that there is more CO and CH₄ at the lower levels, but this may simply reflect a delayed residual effect of a previous plume that has not yet been flushed out from the lower layers of the NBL. This is consistent with the difference in wind speed between heights, about 3 m s⁻¹ at 50 m and 5 m s⁻¹ at 150 m (Fig. ), suggesting slower air renewal at the lower levels.

One hour later (at 01:00 UTC) there is practically no difference between the CO and CH₄ concentrations at all measurement levels, remaining so until around 03:00 UTC. This behavior suggests that a new plume arrived at 01:00 UTC. It should be noted that at 03:00 UTC the CO and CH₄ concentrations at 321 m start to decrease, while at the lower levels there is a lag in this decrease. From 05:00 UTC the CO and CH₄ concentrations begin to decrease rapidly at the higher heights, while at the lower levels they decrease much slower. Overall, these two case studies highlight that the hn is reflected in the vertical CO and CH₄ concentration profiles, thereby supporting our estimated hn. When there is strong decoupling between the NBL and the layer above, air within the NBL can become trapped: the CO and CH₄ plume may be unable to enter the NBL, as on 18 August, or may be slow to ventilate out, as observed on 25 August.