The study site is located at Jaru Biological Reserve (Rebio Jaru), 10°11′11.4″S; 61°52′29.9″W, a national conservation area in the Rondônia state, Amazonas, at situated around 290 km from Porto Velho, the capital. The Rebio Jaru has a total area of 353,163 ha (IBAMA, 2007). Inside the reserve, flux tower was installed, owned by the LBA (Large Biosphere-Atmosphere Experiment in Amazonia) and operating since 1999 (Webler et al., 2007). At around 80 m far from the LBA flux tower was installed, in 2016, a permanent plot composed of 25 subplots of 20 × 25 m (Da Silva, 2021). Our soil and plant samples were collected from one of these subplots.

Rebio Jaru Reserve contains a forest characterized by a dense evergreen forest. The canopy average height is 20 ± 6.7 m, with emergent trees of about 44 m. Based on NOAA's 34-year historical series, the site's average annual rainfall is 1,923 mm, with a dry period of 5 months typically occurring between May and September (Sombroek, 2001). In terms of rainfall pattern, this site is considered one of the most seasonal LBA sites, with a longer dry period than the other LBA sites (Restrepo-Coupe et al., 2013).

Disturbed and undisturbed soil samples were collected from a soil profile up to 4 m deep (at 0.05, 0.1, 0.15, 0.2, 0.4, 0.8, 1.2, 1.6, 2, 3, and 4 m). Undisturbed soil samples were used to determine soil water retention curves (SWRCs), while disturbed soil samples were used to determine the soil physical properties and for isotopic analysis. Plant samples for isotopic analyses were collected from 18 individuals from 14 different plant species located in the study area (Supplementary Table 1) (Souza et al., 2022). For isotopic analysis, soil and plant samples were collected in two periods—May 2016 and October 2016. To collect soil samples in these two different periods, we used two different pits <5 m apart. To avoid evaporation and fractionation, for each tree, we used suberized twig segments (i.e., 1–2 cm diameter), from which we removed the bark before storing the sample to avoid mixing between the phloem and xylem water. Plant and soil samples were quickly sealed in vials, tightly wrapped with parafilm, kept refrigerated in the field, and frozen in the laboratory until water extraction. Rainwater samples were collected during rainfall events between September 2015 and October 2016. Seven water samples were collected in the Machado River (March and June), 1.2 km away from the EC flux tower. Rainfall samples were not mixed, so the data presented here refer to event-specific values and not to averaged monthly values.

To characterize the environmental conditions during the experiment, we used precipitation (P) and soil moisture (θ) data obtained from the EC flux tower between September 2015 and October 2016. Precipitation amounts were automatically measured by a rain gauge (Environmental Measurements Ltd.—Arg 100). Soil moisture content was measured using Timing Domain Reflectometers (TDR), installed at 0.2, 0.4, 0.6, 0.8, and 1 m depths. These data were collected and treated by the LBA office at Rebio Jaru. Evapotranspiration data were obtained from MODIS (Muler, 2018). The dry period, that is, months where P < 100 mm (Sombroek, 2001), occurred between May and September (2016) (Figure 1A). The driest month in our sampling period was July 2016, when no rainfall was registered. The lowest soil moisture conditions along the first meter of the soil profile were recorded in September 2015 and September 2016 (Figure 1B). In the present work, isotopic measurements in soil and xylem water were performed in May and October 2016, which are considered, respectively, as the WTD and the DTW transition periods.

The measured soil physical properties comprise: (i) soil texture, determined according to the Bouyoucos method (Bouyoucos, 1927; Gee and Bauder, 1986); (ii) soil dry bulk density, ρ_d, determined from the ratio of the mass of oven-dried soil (at 105°C for 48 h) and the volume of the soil core samples; (iii) soil-water retention curves (SWRCs), obtained using a tension table (for soil suctions of 0, 1, 2, 4, 6, 10, 30, 50, 100, and 500 kPa) and pressure chamber (for 1,500 kPa), and (iv) total porosity (φ), determined by 1 – ρ_d/ρ_s, with the density of soil particles, ρ_s, taken as 2.65 g cm⁻³.

Soil cation exchange capacity (CEC) was obtained by the sum of exchangeable contents of Ca, Mg, Al (extracted by KCl 1 mol L⁻¹), and K (Mehlich⁻¹) following standard methodologies (Teixeira et al., 2017). Soil organic matter was obtained using the Walkley–Black method (Walkley and Black, 1934).

The SWRCs for each soil layer were fitted using the van Genuchten (1980) equation

where θ is the volumetric water content (cm³ cm⁻³), and θ_s and θ_r are the saturated and residual water content (cm³ cm⁻³), respectively; h is the soil suction (hPa), and α (hPa⁻¹) and n are empirical parameters. The van Genuchten parameters required in Equation (1) (θ_s, θ_r, α, n) were found by fitting 9 h-θ data-pairs (one set for each sampled layer) with the excel solver (non-linear least-squares fitting) to find the best adjustment between the observed and the estimated soil moisture. Parameters were calculated for each layer, with r²-values >0.97 for all layers (Supplementary Table 2).

Total plant-available water (TAW), in cm³ cm⁻³, was calculated as the difference between the volumetric water content at the field capacity (θ_FC) and the water content at the permanent wilting point (θ_PWP), with FC taken at a suction, h, of 33 kPa and PWP at a suction of 1,500 kPa (1.5 MPa). From the fitted SWRCs (Supplementary Figure 1) for each soil depth, we derived four classes of pore sizes: macropores (Ma, equivalent pore diameter, ϕ, > 300 μm), mesopores (Me, 300 < ϕ < 50 μm), micropores (Mi, 50 < ϕ < 0.2 μm), and cryptopores (Cr, ϕ < 0.2 μm) (Teixeira et al., 2017).

According to the capillarity equation (Washburn, 1921; Rowell, 1994):

where h is the suction in (Pa) and r is the pore radius in m, for macroporosity, h < 1 kPa; mesoporosity 1 < h < 6 kPa; microporosity 6 < h < 1,500 kPa, and for Cr, h > 1,500 kPa. According to Rowell (1994), soil pores diameters (ϕ) of 50 μm or larger (i.e., Ma+Me; h < 6 kPa) corresponds to the “transmission pores” (Tr), while ϕ between 50 and 0.2 μm (i.e., Mi + Cr; h > 6 kPa) refers to water “storage pores” (St pores). To compare the soil's physical properties along the 4 m soil profile, we computed TAW and pore size distributions as a percentage of the total porosity (φ).

Plant traits such as tree height (H, m) and circumference at breast height (CBH, cm) were measured with a tape measure. Circumference at breast height was transformed into DBH (cm) by dividing CBH by π. The leaf water potential at midday (ψ_md, MPa), which is an indicator of plant water stress, was measured with a Scholander camera (PMS Instruments Co., Albany, NY, USA). These measurements were taken in October 2016, the DTW transition period.

The isotopic ratios of H (δ²H) and O (δ¹⁸O) (Coplen, 2011) of the collected and extracted waters were analyzed by laser absorption spectroscopy using a Picarro Li2130. We used two different standards—PLRM1 (δ²H: 16.9‰ and δ¹⁸O: 1.65‰) and PLRM2 (δ²H: −123.1‰ and δ¹⁸O: −16.52‰)—and one Quality Control (δ²H: 46‰ and δ¹⁸O: 7.25‰), previously calibrated with the V-SMOW (Vienna Standard Mean Ocean Water).

Deuterium excess (d-excess) was calculated according to (Dansgaard et al., 1964):

Soil and plant water samples were extracted using the cryogenic distillation method. To prevent the influence of organic contamination in the isotopic results, we immersed activated charcoal for 48 h in the extracted soil and plant water samples. Besides that, the Picarro CRDS L2130-i instrument used has a built-in software—Post Process Chemcorrect—that flags any sample with organic contamination. If samples flag once, they are re-analyzed; and if they flag two times, isotopic analysis is done in the isotope ratio mass spectrometry (IRMS). During our study, no sample was required to be re-analyzed or transferred to the IRMS.

In the dual-isotope space (δ²H × δ¹⁸O), we used analysis of covariance (ANCOVA) to identify differences among the slopes of Global Meteorological Water Line (GMWL), Local Meteorological Water Line (LMWL), River Water Line (RWL), Plant Water Line (PWL), and Soil Water Line (SWL) on the dual isotopic space. We used Pearson's correlation to test the correlation between plant traits such as DBH, H, ψ_md, soil physical properties (i.e., TAW and soil porosity), and δ¹⁸O and δ¹²H.

Our results point to a very clayey soil, with clay content >60% along the entire 4 m profile, and sand content of ~30%, except at 2 and 2.4 m, where sand content decreases to values between 20% and 30% (Table 1). The soil presents a relatively high bulk density (ρ_d) > 1.3 g cm⁻³, with a distinctly denser layer (ρ_d > 1.65 g cm⁻³) occurring at 1.2 m depth (Table 1). Total soil porosity (Φ) is high relatively, with the lowest value (0.4 cm³ cm⁻³) occurring at 4 m depth. The soil porosity is mostly (>50%) formed by Cr pores (i.e., pores with ϕ < 0.2 μm; Table 2) where water is retained at suctions >1.5 MPa (Rowell, 1994). Consequently, this soil presents a high water retention capacity, with θ_PWP > 0.3 cm³ cm⁻³ for most of the soil layers and St pores >70% along the entire 4-m soil profile. The highest values of θ_PWP and St pores occur between 1.6 and 3 m depths (Figure 2; Table 2). Van Genuchten parameters for each layer are presented in Supplementary Table 2.

During our sampling period, the isotopic signal of rainfall water ranged between −82.2 and 36.7‰ for δ²H and between −10.1 and 12.2‰ for δ¹⁸O (Figures 3A,B; Table 3). δ²H_xylem ranged from −46.6 to −27.3‰ and δ¹⁸O_xylem ranged from −5.0 to −0.2‰ in May, while in October, xylem water presented more enriched values for both water isotopes, with δ²H ranging from −35.9 to −9.6 ‰ and δ¹⁸O from −3.7 to 2.4‰. Similar behavior was observed for soil water signals with enriched values in the dry season, ranging from −42.9 to −27.5‰ and −46.3 to −10.8‰ for δ²H; and from −6.2 to −4.2‰ and −6.7 to −2.2 ‰ for δ¹⁸O in May and October, respectively (Table 3). Analysis based on single isotopic data (δ²H and δ¹⁸O; Figures 3A,B, respectively; Table 3) showed dry season rainfall (measured in May, June 2016, and October 2016), and throughfall isotopic water signals being more enriched than the wet season rainfall (measured in March, April, and December 2016, and January and February 2017), for both δ²H and δ¹⁸O_. In May 2016, δ²H_xylem was in the range of δ²H_soil, however, δ¹⁸O_xylem was more enriched than δ¹⁸O_soil (Table 3). In October 2016, δ²H_xylem plotted in the range of the more depleted δ²H_soil, but, as in May 2016, δ¹⁸O_xylem is outside the range of δ¹⁸O_soil.

In the dual isotope space (Figure 4), SWL_wet plotted along the GMWL and LMWL in the wet period (LMWL_wet, p > 0.05; Supplementary Table 3), while PWL_wet, measured in May 2016, followed the LMWL_wet (Δ_slope = 0.24; p = 0.930) and the throughfall water line (Δ_slope = 0.34; p = 0.853; Supplementary Table 3). In October 2016, PWL_dry better agreed with the LMWL_dry (Δ_slope = 1.28; p = 0.37), and deviated from the LWML_wet, in response to the higher enrichment of the δ¹⁸O_xylem in relation to the δ²H_xylem. However, some trees presented an isotopic signal that followed the LWML_wet. These trees were marked in bold, in Supplementary Table 3.

Our isotopic results show that the wet season rainfall was isotopically more depleted than the dry season rainfall, a pattern that can be attributed to the rainfall amount effect (Araguás-Araguás et al., 2000). In the dual isotope space, the agreement between GMWL and LMWL_wet, GMWL and throughfall, and GMWL and SWL_Oct (Supplementary Table 3) suggests that the rainfall of past events would be mainly responsible for the soil water recharge. Considering the hypothesis that to meet the plant-atmospheric water demands of the dry period, trees uptake the more depleted water from the past wet periods, stored in the deeper soil layers; we would expect a more depleted xylem signal as the dry season progresses. However, contrary to this, we found xylem water more enriched in October 2016, end of the dry period/transition to wet period, than in May 2016, the beginning of the dry period. Besides, instead of following the isotopically depleted soil signal of the deeper soil layers, the xylem signal is isotopically more enriched and shows a better agreement with the recent rainfall. This pattern can be observed in dual isotope space (Figure 4) and in the ANCOVA analysis (Supplementary Table 3). These results suggest that instead of using the water stored in the soil layers, to meet the plant-atmospheric water demands of the dry period, trees are using a more recent water, which is the rainfall dropped in the past recent events. In May 2016, beginning of the dry period, trees used water from the past wet season rainfall while in October 2016, the peak of the dry period/transition to the wet period, trees used the more enriched water from the dry period.

These results agree with the results found by Miguez-Macho and Fan (2021), who found that in a global survey of studies based on isotope data, 70% of plants used water from the recent rainfall, while 18% use water from the past rainfall events stored in the soil layers. The plant dependence on the recent rainfall events challenges the deep-RWU as a mechanism of forest resilience to droughts. The fact that instead of relying on the water from past events stored in the soil, the maintenance of transpiration rates relies on a more recent rainwater, could explain the high mortality rates of the larger trees during severe droughts that have been registered in the Amazon forest (Nepstad et al., 2007; Phillips et al., 2010; Esquivel-Muelbert et al., 2020). These results can be of great concern considering the perspectives of increase in length and severity of the dry season rainfall, as a result of local and global climate changes (Wright et al., 2017; Gatti et al., 2021). However, it is important to note that the Amazon forest comprises a very huge and diverse ecosystem, and more research is needed to generalize the behavior found here to all the Amazon forest environments, and for more severe dry periods.

Despite statistical analysis (ANCOVA, methods) indicating the existence of significant similarity between PWL_Oct and LMWL_dry slopes (Δ_slope = −0.11; p = 0.873; Supplementary Table 3), visually it is possible to observe a displacement in both lines so that PWL_Oct plots below the LMWL_dry (Figure 4). This displacement is a result of the pronounced enrichment of the δ¹⁸O in relation to the δ²H (Figures 3A,B), a pattern which can be also observed in the d-excess (Supplementary Figure 2).

The relationship between δ²H and δ¹⁸O, expressed in terms of the d-excess (Equation 3), has been used as an indicator of evaporative enrichment (Sprenger et al., 2016). Some works have considered d-excess <-10‰ as an indicator of evaporative enrichment of the rainfall water (Kern et al., 2020). Despite tree waters being sampled, stored, and analyzed following the same procedure, in October 2016—peak of the dry period—we found a pronounced enrichment in the δ¹⁸O (i.e., d-excess <-10‰) for 7 of the 18 sampled trees; Swartzia ingifolia, Dipterix odorata, Dipterix magnifica, Copaifera multijuga, Cariniana decandra, Astronium lecontei, and Anamalocalyx uleanus (Supplementary Figure 3). Four of these trees presented moderate resistance to drought (Dipterix odorata, Cariniana decandra, Astronium lecontei, and Anamalocalyx uleanus) and one (Swartzia ingifolia) presented low resistance to drought, according to Souza et al. (2022). For the other species, resistance to drought (using πTLP as a proxy) was not determined (Souza et al., 2022). However, other species that also presented moderate resistance to drought according to Souza et al. (2022) did not present a pronounced δ¹⁸O enrichment as the dry season progresses. This is the case for Sagotia brachysepala, Pouteria durlandi, and Licania sprucei (Supplementary Figure 3). Based on this, we did not find a relationship between the pronounced enrichment in the xylem isotopic signal and the plant resistance to drought. We also did not find any correlation between enriched xylem isotopic signal in October 2016 and tree H or DBH (Supplementary Table 4). However, the strong correlation between the enriched signal and the ψ_md, both measured in October 2016 (Supplementary Table 4) suggests that an enriched xylem signal is more pronounced in trees submitted to higher water stress (ψ_md > 2 MPa). According to Martín-Gómez et al. (2017), under low water availability and high evaporative demand, stem water loss via evaporation can create significant isotopic enrichment of stem water (Martín-Gómez et al., 2017). That is because, when leaf transpiration is limited, it reduces the input of non-enriched fresh water, allowing for cumulative evaporative enrichment. However, in our study, given the maintenance of relatively high ET rates and EVI along the dry period (i.e., between May and October 2016)—as presented in Souza et al. (2022)—we conclude that the isotopic enrichment of the xylem signal is more a result of the dry season rainwater uptake than of a limitation of the transpiration rates and stem water flow as the dry season progresses, as proposed to Martín-Gómez et al. (2017).

It is also important to note that we did not find an overlap between the soil water signal and xylem signal for any of the both sampled periods (i.e., in May and in October 2016). Very fine resolution soil moisture measurements performed in central Amazon show a very wet layer developed in the upper 5 cm of the soil layer during droughts (Negrón-Juárez et al., 2020). Further investigation is needed to consider if there is a missing source of water for soil layers that were not sampled and could also explain the enriched xylem isotopic signal.