Calibration, measurement, and characterization of soil moisture dynamics in a central Amazonian tropical forest

Soil moisture plays a key role in hydrological, biogeochemical, and energy budgets of terrestrial ecosystems. Accurate soil moisture measurements in remote ecosystems such as the Amazon are difficult and limited because of logistical constraints. Time domain reflectometry (TDR) sensors are widely used to monitor soil moisture and require calibration to convert the TDR's dielectric permittivity measurement (Ka) to volumetric water content (θv). In this study, our objectives were to develop a field‐based calibration of TDR sensors in an old‐growth upland forest in the central Amazon, to evaluate the performance of the calibration, and then to apply the calibration to determine the dynamics of soil moisture content within a 14.2‐m‐deep vertical soil profile. Depth‐specific TDR calibration using local soils in a controlled laboratory setting yielded a novel Ka–θv third‐degree polynomial calibration. The sensors were later installed to their specific calibration depth in a 14.2‐m pit. The widely used Ka–θv relationship (Topp model) underestimated the site‐specific θv by 22–42%, indicating significant error in the model when applied to these well‐structured, clay‐rich tropical forest soils. The calibrated wet‐ and dry‐season θv data showed a variety of depth and temporal variations highlighting the importance of soil textural differentiation, root uptake depths, as well as event to seasonal precipitation effects. Data such as these are greatly needed for improving our understanding of ecohydrological processes within tropical forests and for improving models of these systems in the face of changing environmental conditions.


INTRODUCTION
Soil moisture refers to water content present in the soil pore space. It strongly affects land-atmosphere feedbacks because of its central role in hydrological, energy and biogeochemical cycling. For example, evapotranspiration as mediated by vegetation is largely controlled by soil moisture availability, which subsequently regulates local and regional cloud formation and rainfall (Eltahir, 1998). The energy used to evaporate water from soil and plant surfaces affects the partition of incoming radiation, and therefore soil moisture has a direct effect on the surface radiation balance (Lei, Crow, Holmes, Hain, & Anderson, 2018;Negrón-Juárez, Hodnett, Fu, Goulden, & von Randow, 2007;Shuttleworth, 1988;WMO-GCOS, 2016). Thus, knowledge of soil moisture availability and its dependency on environmental and edaphic conditions can be used for predictions of temperature and rainfall patterns (Robock, 2015). Soil moisture is an important control on infiltration and percolation into a soil profile (i.e., the vadose zone), overland flow, interflow generation, and groundwater recharge (Koster et al., 2004;Sheffield & Wood, 2008). It also has strong biogeochemical effects through impacts on microbial activity, soil respiration and nutrient cycling (Austin et al., 2004), as well as forest ecosystem health and services (e.g., C storage and agricultural production).
Changes in soil moisture are driven by many factors, which presents a challenge for predicting soil moisture over space and time (Yoo & Kim, 2004). However, quantifying soil moisture is critical for understanding how changes in land surface and climate affect the functioning of the global system (Robock et al., 2000;Sheffield & Wood, 2008). Accurate measurements of soil moisture are important for validating and improving land surface, for remote sensing of soil moisture (Dorigo et al., 2011) across multiple scales (Brocca, Morbidelli, Melone, & Moramarco, 2007;Famiglietti et al., 1999;Koster et al., 2009), and for identifying trends associated with climate change and disturbance (Robock, Mu, Vinnikov, Trofimova, & Adamenko, 2005). Soil moisture is considered an essential climate variable in the Global Climate Observing System (GCOS; WMO-GCOS, 2016).
The Amazon is the largest continuous tropical forest in the world, representing 53% of the global tropical area (Negrón-Juárez et al., 2018). Because of strong recycling effects, evapotranspiration is responsible for up to 50% of the annual rainfall in the Amazon (Salati, Dallolio, Matsui, & Gat, 1979;van der Ent, Savenije, Schaefli, & Steele-Dunne, 2010). Root uptake of soil water and subsequent plant transpiration provide the largest contribution to water recycling in the Amazon (Kunert et al., 2017). Understanding of soil moisture dynamics in the Amazon is complicated due to the regional climatic differences (Negrón-Juárez, Li, Fu, Fernandes, & Cardoso, 2009;Sombroek, 2001), soil characteristics (Quesada et al., 2010), and floristic composition (ter Steege et al., 2013) that alter the exchange of water in the atmosphere-biosphere continuum (Gimenez et al., 2019). Although challenging, continuous measurements of soil moisture in the Amazon have demonstrated the importance of deep water uptake to transpiration during the dry season (Bruno, da Rocha, de Freitas, Goulden, & Miller, 2006;Negrón-Juárez et al., 2007;Nepstad et al., 1994).
Accurate measurements of volumetric water content (θ v ) are important for a variety of reasons. For example, a major interest of the Next Generation Ecosystem Experiments-Tropics (NGEE-Tropics) project (https: //ngee-tropics.lbl.gov/) is to understand the effect of droughts and dry seasons on forests in the central Amazon and the vulnerability of these forests to water stress and mortality (Fontes et al., 2018;Gimenez et al., 2019;Solander et al., 2020). Because θ v varies significantly with depth and time, high-time-resolution measurements that span the root zone and below are needed. Transpiration demand, for example, is typically higher in the shallow soils than in deeper soils (Broedel, Tomasella, Candido, & von Randow, 2017;Negrón-Juárez et al., 2007), requiring characterization of depth variations in θ v that extend beyond the typical 1-to 2-m measurement zone. Time series of θ v are especially important in monsoonal climates because water used to meet transpiration demand during the dry season Vadose Zone Journal can be supplied from the previous rainy season (Negrón-Juárez et al., 2007), since water can remain plant available in the soil for weeks to months (Dirmeyer, Schlosser, & Brubaker, 2009;Koster et al., 2009;Seneviratne & Koster, 2012). The length of the dry season may also be increasing in the Amazon (Marengo et al., 2017), and assessing the long-term impacts of such a trend requires multi-depth time series measurements of θ v . Because soil moisture has a strong effect on forest composition (le Roux, Aalto, & Luoto, 2013), quantifying changes in θ v is also important for understanding vegetation succession under changing environmental conditions. Accurate measurements of θ v in old-growth forests are also needed to understand water cycling across the soil-plant-atmosphere system. Infiltration, redistribution, evaporation, transpiration, precipitation recycling, percolation, and groundwater recharge frequently occur simultaneously (Reichardt & Timm, 2012), and reliable θ v data in the central Amazon will be critical to validate and benchmark models (Christoffersen et al., 2014;Koyen et al., 2020).
Time domain reflectometry (TDR) is a widely used approach to continuously measure soil moisture, quantified as volumetric water content, θ v (Lekshmi, Singh, & Baghini, 2014). Time domain reflectometry systems measure dielectric permittivity in the soil (K a ), which is highly correlated to water content (Capparelli, Spolverino, & Greco, 2018;Robinson, Gardner, & Cooper, 1999). This relationship is typically not significantly affected by soil temperature, which helps enable robust measurements of θ v (Dasberg & Dalton, 1985;Topp, Davis, & Annan, 1980). In the central Amazon, TDR measurements have frequently used calibrations developed from agricultural soils (Broedel et al., 2017;Teixeira, 2001) or directly applied the Topp model (Topp et al., 1980) without developing the recommended site-specific calibrations. However, agriculture alters soil properties such as structure and bulk density, which result in lower percolation compared with oldgrowth forest soils. Furthermore, studies have shown that the Topp model (Topp et al., 1980) can be particularly problematic for tropical soils. In tropical soil characteristics such as Fe oxide content, particle size distribution (texture) and organic matter content can cause the K a -θ v relationship to differ significantly from the Topp model (Bruno et al., 2006;Kaiser, Reinert, Reichert, & Minella, 2010;Madeiros, Castro, Goldenfum, & Clarke, 2007;Roth, Malicki, & Plagge, 1992;Teixeira, 2001;Teixiera, Schroth, Marques, & Huwe, 2003;Tommaselli & Bacchi, 2001;Weitz, Grauel, Keller, & Veldkamp, 1997). These studies point to the need for a site-specific TDR calibration for old-growth forest soils in the central Amazon.
In this study, our objectives are (a) to develop a sitespecific TDR calibration for the central Amazon, (b) to

Core Ideas
• We conduct the first TDR soil moisture calibration in an old-growth forest in the central Amazon. • We made interannual 30-min observations of soil moisture up to 14.2-m depth. • The Topp model underestimated volumetric water content by 22-42%. • Dry-season morning-night fluctuation of soil moisture were observed.
quantify the magnitude of deviation of the K a -θ v relationship between the site-specific calibration and the Topp model, and how that changes through the soil profile, (c) to understand how soil moisture varies within the soil profile due to changes in soil characteristics, and (d) to determine the dynamics of soil moisture during seasonal wet and dry periods. Deep soil moisture data are quite rare in the humid tropics, and these data are insightful for understanding how soil texture and stratigraphy, precipitation, and vegetation water use drive temporal variations in soil moisture. This study helps demonstrate that accurate soil moisture measurements are critical for improving understanding water dynamics in the Amazon and to benchmark land surface models.

Site description
The study area is located in the Tropical Silviculture Experimental Station (EEST, per its acronym in Portuguese, also known as ZF2) located at (2.45-2.66 • S, 60.02-60.32 • W), 53 km north of the city of Manaus, Amazonas, Brazil, in the central Amazon ( Figure 1a). The EEST encompasses 21,000 ha, most of which is old-growth forest (Andrade & Higuchi, 2009). The topography of the region is characterized by a sequence of plateaus, slopes, and valleys (Ferraz, Oht, & Salles, 1998). In the EEST, plateaus vary from 90 to 105 m asl, and small valleys vary from 45 to 55 m asl (Ferraz et al., 1998;Renno et al., 2008). The climate of the region is tropical monsoon (Am) in the Köppen classification (RADAMBRASIL, 1978). Our study area is characterized by a mean annual temperature of 27 • C and mean annual rainfall of 2365 mm, with the dry season (rainfall < 100 mm mo −1 ; Sombroek, 2001) from July to September (Negrón-Juárez et al., 2017), and annual mean relative humidity of 84% (Leopoldo, Franken, Salati, & Ribeiro, 1987). The soil types are closely related to topography and can be classified into three main types: Yellow Latossol on the plateaus, Red-Yellow Podzolic on the slopes, and hydromorphic sandy soils in the valleys (Ferraz et al., 1998). Soil texture, organic matter content, soil moisture, soil pH, and soil C and N concentrations vary significantly along topographic gradients in the central Amazon (Luizao et al., 2004). On the plateaus, soils are derived from tertiary sediments of the Barreiras Group and are dominated by kaolinite, quartz, iron oxides and hydroxides, and Al (Chauvel et al., 1992). These soils are clay rich (Broedel et al., 2017). Clay contents are 65-75% in the upper 30 cm of the soil profile and reach 80-90% in the 2-to 4-m layer. By 15 m, clay contents slowly decrease to 57%. The soil is typically poor in P, Ca, Mg, and K (Teixeira, Schroth, Marques, & Huwe, 2014) but at the same time supports a highly productive forest due to multiple P-uptake strategies like investment in fine roots (Lugli et al., 2019). The upper 50 cm has the highest concentrations of organic matter of ∼15 Mg C ha −1 , and by 80 cm, it decreases to ∼2.5 Mg C ha −1 (Marques et al., 2015). Broedel et al. (2017) found that porosity varies from 60.6 to 58.5% at 80 cm and from 49.8 to 47.4% at 14.2 m.
The vegetation is upland old-growth forest (Ferreira, Luizao, & Dallarosa, 2005) characterized by a high diversity of tree species (Carneiro et al., 2005;Higuchi et al., 1997;Saito, Sakai, Nakamura, & Higuchi, 2003), with a mean canopy height of ∼30 m and the tallest trees exceeding 40 m (Lima, Teixeira, Carneiro, Santos, & Higuchi, 2007). Lecythidaceae, Sapotaceae, Fabaceae, Chrysobalanaceae, Burseraceae, Annonaceae, Moraceae, and Euphorbiaceae are the most abundant botanical families in the EEST (Carneiro et al., 2005;Higuchi et al., 1997;Saito et al., 2003;Vieira et al., 2004). Some of the most common species that characterize the area are Dinizia excelsa Ducke (angelim-pedra), Eschweilera coriacea (DC.) S.A. Mori (mata matá), Protium apiculatum Swart (breu-vermelho), Scleronema micranthum (Ducke) Ducke (cardeiro), and Micrandropsis scleroxylon (W.A. Rodrigues) W.A. Rodrigues (piãozinho) (Carneiro et al., 2005;Higuchi et al., 2004;Vieira et al., 2004). The mean density of stems with DBH (diameter at breast height) ≥ 10 cm is 584.3 ± 25.9 trees ha −1 (da Silva et al., 2002;Vieira et al., 2004), with an annual mortality rate of 8.7 trees ha −1 (Higuchi et al., 1997). All walls of the deep pit are covered with hard transparent plastic for protection, and the pit opening is sealed with a door. The pit is equipped with a ladder structure and underground access platforms and ventilation system, which allows technicians to descend with safety. Since the installation of the pit, there were no changes in the experimental site, since it is a protected reserve. The deep pit surroundings are covered with intact tropical forest. Between 2014 and 2018, the sensors installed in the pit stopped working and data were not recorded. The pit was closed and remained isolated. During this period, the pit did not suffer any damage such as wall collapse or lateral leakage. The pit was retrofitted in 2014 as part of the GoAmazon project (Martin et al., 2017). In our research area, previous studies have measured soil texture, porosity, macroporosity, microporosity, soil density (Broedel et al., 2017), root distribution (Chauvel et al., 1992), and hydraulic conductivity (on disturbed soils near Manaus) (Teixeira et al., 2014;Tomasella & Hodnett, 1996). No stones are present in the soils and most roots are near the surface (<0.5 m), with their distribution declining rapidly with depth (Chauvel et al., 1992;Cordeiro et al., 2020;Marques et al., 2015). Bedrock and groundwater at the site are quite deep, and the bottom of the pit is many meters above groundwater and the bedrock interface.
In 2018 and as part of the NGEE-Tropics project, TDR CS655 (Campbell Scientific) sensors were installed horizontally in the north pit wall at 0.8-, 1.6-, 2.4-, 3.2-, 4.8-, 6.4-, 8.8-, and 14.20-m depths (details in Section 2.4). Due to the structural design of the deep pit, it was not possible to install sensors shallower than 0.8 m. Therefore, a second shallow pit was constructed near the deep pit, and CS655 TDR sensors were installed at 0.025-, 0.05-, 0.15-, 0.30-, 0.50-m depths. Measurements were initiated in July 2018, and logging was conducted on 30-min intervals using a CR1000x datalogger (Campbell Scientific). Measurements reported here are from September 2018 (to allow for equili-bration after installation) to January 2020. Electrical problems resulted in missing data from 11 to 19 Sept. 2019.
The CS655 sensors operate under the TDR principle. The sensors have two parallel 12-cm-long stainless-steel rods that form an open-ended transmission line in which the wave propagation velocity depends upon the dielectric permittivity of the medium surrounding the rods. A differential oscillator circuit is connected to the two rods, with an oscillator state change triggered by the return of a reflected signal from one of the rods and the two-way travel time of the wave varies with the dielectric permittivity (K a ) that, in turn, depends largely on soil water content (Campbell Scientific, 2018). The volumetric water content is frequently calculated from K a using the Topp model (Topp et al., 1980), although site-specific calibrations should be used as explained below. Further details on the CS655 sensors can be found in the sensor manual available on the Campbell Scientific homepage (https://www.campbellsci.com/).

Soil moisture calibration
To test soil moisture calibration approaches, we collected soil samples from the deep pit at the same eight depths where the sensors were installed using a 10-cm-diam. soil sampling auger to collect bulk soil samples 1 m horizontally from the north pit face. Soil samples were placed into plastic bags, sealed, and transported to the laboratory. After this procedure, bulk density samples were collected using 98-cm 3 -diam. Kopecky rings of (DIK-1801, Daiki Rika Kogyo Company). At each of the eight horizontal depths, three replicate bulk density sampling rings were collected (24 samples total). The rings were pressed into the soil, carefully excavated, cleaned to remove excess soil, and then immediately wrapped in plastic film. Additional soil samples were collected at each depth with the auger (∼10 cm horizontally) to determine the maximum rooting depth of live roots. Live roots were distinguished from dead roots based on their brighter color, stiffness and turgidity, and tightly adhered cortex and periderm. Dead roots were mushy or desiccated and brittle, easily broken or separated, and darker brown or black in color. Only one sample per depth was used to determine the presence of roots.
To calculate the field gravimetric water contents and bulk densities to determine volumetric water contents, the 24 Kopecky ring samples were first weighed with an analytical balance with 0.001-g precision (AD 330, Marte Cientifica) to determine wet mass (m wet ). The samples were then oven dried at 105 • C for 48 h, which is recommended for tropical organic soils (ASTM, 2014;O'Kelly & Sivakumar, 2014). After oven drying, the mass of dry soil was determined. Soil bulk density was then calculated based on dry soil per unit volume (m dry ). The gravimetric water content, θ g , was calculated as Volumetric water content was calculated as where ρ bulk = m dry /volume Kopecky , and volume Kopecky = 98 cm 3 . Supplemental Table S1 provides details about the undisturbed soil samples. Soil samples collected for TDR calibration were homogenized following methods developed by the Brazilian Agricultural Research Corporation (EMBRAPA; Donagema, Campos, Calderano, Teixeira, & Viana, 2011). The bulk soil samples were crushed and air dried for 2 wk in a room conditioned with continuous dry air to lower water contents to residual water content levels. The air-dried samples were sieved (2-mm mesh, Telas MM) to produce air-dried fine soil (ADFS; Donagema et al., 2011).
To develop data for TDR calibration, we followed the recommendations of the manufacturer for the model CS655 sensors (Campbell Scientific, 2018). Pairs of short and long cylindrical PVC (polyvinyl chloride) were filled with soil to determine if there were soil volume effects on the calibration measurements. The two cylinders were filled with soils collected from each target sensor depth in the pit. Both cylinders had 20-cm internal diameters, the short ones were 22 cm long, and the longer ones were 37 cm long. Both cylinders were larger than the minimum recommended soil radius around (>7.5 cm) and below (>4.5 cm) each TDR rod following Campbell Scientific (2018). The amount of soil used in each cylinder was fixed and determined by obtaining the apparent density of the field conditions. Prior to calibrations, the soil was rehydrated by placing samples in a 50-L container and carefully mixed with sequentially increasing of water (0, 5, 10, 15, 20, 25, 30, and 40%). The wet soil was then packed into the cylinders to the bulk density observed in the field (see Section 3). The TDR probe rods were then carefully inserted in the cylinders, and measurements of K a were initiated after a ∼1to 2-min period to allow for sensor stability. The measurements were collected using a datalogger model CR800 (Campbell Scientific). Each TDR sensor was calibrated by making K a measurements for multiple water contents for the appropriate soil from the depth where the sensor would be installed. To verify water contents, one sample was collected from each cylinder using Kopecky rings for calculating volumetric water contents using gravimetric water contents and bulk densities as described in Equations 1 and 2. Calibration calculations are discussed in Section 2.5 below.

Field sensor installation
After calibration, each TDR sensor was installed inside the pit wall at the same locations that were used to collect the respective calibration soil for each sensor (one TDR per soil depth). To minimize edge effects of the pit, sensors were installed 1.5 m into the pit wall using a custom-made device (Figure 1b). This device makes two small pilot holes (with a slightly smaller diameter than the sensor rods to assure good soil contact, Figure 1c), and then the sensor was pushed into place. After installation, the holes were backfilled with the previously removed soil from the same depth as the sensor was calibrated (Figure 1d).

Calibration calculations using local soils
The K a values from the short and long cylinders were similar (<2% difference) for a given water content and soil (i.e., no volume effects were observed), and therefore all the data were used to establishing K a -θ v relationships for the local soil calibration. K a values were used to obtain the calibrated model of the volumetric water content (θ v ) using a third degree polynomial (θ v = aK a + bK a 2 + cK a 3 + d, where a, b, c, and d are coefficients), which was solved using SigmaPlot 11 software (Systat Software). The standard error (σ est = σ∕ the fit was calculated where y and y′ are the observed and predicted values, respectively, and N is the number of data (Wilks, 2006). The N was 128 (eight depths, eight moisture increments, two pots). The SEs of the coefficients (Wilks, 2006) of the fit were also calculated. To compare measured θ v values with values using the Topp model (θ v_Topp ), we calculated: (θ v − θ v_Topp ) × 100/θ v (%).

Independent soil moisture data for validation
In order to further validate the laboratory TDR calibrations, we leveraged a complementary study of upper soil θ v that used frequency domain capacitance (FDC) probes installed ∼40 m to the south of the deep pit, on the opposite side and slightly downslope from the K34 tower (G. Spanner, personal communication, 2020). The FDC probes (EnviroSCAN, Sentek) were installed into PVC access tubes to a depth of 1 m, and θ v was measured at 10-, 20-, 30-, 40-, 50-, 70-, and 100-cm depths. The FDC probes had been calibrated in air and under water, and a site-specific calibration was developed by sampling soil adjacent to one of the access tubes. On 17 Oct. 2018, 5-cm-tall metal Kopecky F I G U R E 2 Calibrated model (black line) between observed volumetric water content (θ v_obs ) and the dielectric constant K a : volumetric water content (θ v , m 3 m −3 ) = -(0.088 ± SE 1 ) + (5.7 ± SE 2 ) × 10 −2 K a -(2 ± SE 3 ) × 10 −3 K a 2 +(2.912 ± SE 4 ) × 10 −5 K a 3 (p < .0001, r 2 = .96, adjusted r 2 = .96). SE 1 = 0.021, SE 2 = 0.4, SE 3 = 0.03, SE 4 = 0.445. Curves were fitted using statistical package of SigmaPlot 11 (Systat Software). The volumetric water content obtained from the Topp model (θ v = −5.3 × 10 −2 + 2.92 × 10 −2 K a − 5.5 × 10 −4 K a 2 + 4.3 × 10 −6 K a 3 ) (Topp et al., 1980) is also shown. (b) Comparison of the difference between the calibrated and Topp model with respect to the observations. The colors of the symbols correspond to depths shown in Panel a. Conf, 95% confidence band; Pred, 95% prediction band rings were used to sample soil carefully along the side of the PVC access pipe, using two samples per depth. The θ v was calculated as described in Equations 1 and 2 for each Kopecky ring, and then the resulting values were compared with the in situ sensor-based FDC estimate of θ v measured at the time the Kopecky rings were sampled. A linear regression between FDC-derived θ v and Kopecky measurements was used to develop the site-specific calibration. Subsequently, the shallow CS655 (TDR) measurements were compared with the simultaneous FDC-based θ v measurements for the same depths.

Rainfall data
Thirty-minute data of accumulated rainfall for the same period of measurements of soil moisture were obtained from the EEST pluviometer network (available at http: //lba2.inpa.gov.br/hidrologia/index.php/edicao-atual). Data were collected using a TB4 tipping bucket rain gauge (Hydrological Services America) located at 2.6090 • S and 60.2133 • W, a few meters from the deep pit.

Data archive
Quality assurance and quality controlled soil moisture and precipitation data were archived following protocols designed by the NGEE-Tropics project and available at https://ngee-tropics.lbl.gov/. Calibrated soil moisture and rainfall data used in this study can be accessed at https: //doi.org/10.15486/ngt/1602141.

RESULTS
Bulk densities and live root depths from the deep pit were important for calculations of volumetric water contents and for describing conditions within the pit (see Section 4). Soil bulk densities for the eight deep pit depths (from top to bottom) were 1.17, 1.14, 1.20, 1.14, 1.11, 1.18, 1.30, 1.32 g cm −3 (Supplemental Table S1) and varied with pit soil stratigraphy described in Chauvel et al. (1992) and Broedel et al. (2017). Live roots were found to a depth of 10 m, but not below. A comparison of the local soil calibration and the Topp model is shown in Figure 2, along with the actual measured calibration values. The local soil calibration represented the observed K a −θ v relation well (black line in Figure 2; p < .0001, r 2 = .96, σ est = 0.03), but the Topp model did not. The Topp model reveals a consistent underestimation of water contents (gray line in Figure 2a). Figure 2b shows that for all depths, the calibrated model was closer to observations than the Topp model. It is also noted that for depths >3.2 m, the calibrated model F I G U R E 3 Comparison of calibrated measured volumetric water contents (θ v ) and the Topp model (θ v_Topp , left axis in black). Percentage differences between the Topp model estimates and the calibrated data are also shown (right axis in red) produced large errors for soil moisture <0.2 m 3 m −3 . This will be addressed in Section 4. Differences between the soil moisture derived using the Topp model and the calibration data are plotted in Figure 3 and show that the Topp model underestimates θ v across the whole range of K a . The average underestimation was 28 ± 5% (mean ± SD) and was largest for low soil moisture values (42% for θ v = 0.25 m 3 m −3 ). Even at higher soil water content, the underestimation was never below 22%.
The data from Figure 2 suggest that the local soil-based calibration provides a substantially better representation of soil moisture than the Topp model. To further test the representativeness of the calibrated data, we compared our results with measurements from the FDC sensor (Figure 4). We emphasize that TDRs and FDC sensors were in different locations (FDC sensors installed 1 m from trees, with site-specific differences, and root and macropores differences), and the FDC sensor integrates its signal against a different volume of soil than the TDR probe. In addition, differential disturbance by insect or animal fauna in the upper soil near measurements contribute to spatial variations in θ v and could also influence both the FDC θ v and TDR θ v measurements. Indeed, there was evidence of several voids around the FDC access tube during calibrations that prevented use of the samples from those locations. Despite these issues, the TDR and FDC agreed fairly well, particularly for the 30-and 50-cm layers where FDC θ v overlapped with TDR θ v .
The variability of rainfall and θ v through the entire depth profile with time is shown in Figure 5. From the soil surface to ∼2.5 m, θ v responded quickly to precipitation events and dry downs during periods with little to no pre-F I G U R E 4 Comparison of volumetric water content (θ v ) at 1200 h on 17 Oct. 2018 between the laboratory-calibrated time domain reflectometry (TDR) probes in the soil pit and six frequency domain capacitance (FDC) probes that used an in situ calibration. Data show the mean and the standard deviations of the FDC θ v content located in close proximity to each other, which were installed on the opposite side and slightly downslope from the K34 tower F I G U R E 5 Thirty-minute rainfall and volumetric water content (θ v ) data from September 2018 to January 2020. The shaded gray area represents the dry season (July-September). Data loss occurred from 11 to 19 Sept. 2019 due to electrical problems cipitation. Below 2.5 m, θ v did not exhibit the same strong dry down periods as the shallower soil zone, although percolation events associated with high precipitation periods were observable to 8 m. Deep percolation events occurred on 7 Nov. 2018, 13 Dec. 2018, and 9 Jan. 2019, for example, and rain events of 10 mm or larger are, in general, correlated with these events. However, antecedent moisture contents and frequency of precipitation events also appear to be important controls on deep percolation. The 6-to 7-m zone consistently had the highest θ v values of all the F I G U R E 6 (a) Variability of dry season precipitation and (b) associated soil moisture variations from 10 Sept. to 22 Oct. 2018. θ v is the volumetric water content depths, and the 8-to 10-m layer had the lowest values in the deeper part of the profile. Below 6 m, soil water contents tended to be much less variable with time than in the shallower soils.
A low rainfall period was selected from 10 Sept. to 22 Oct. 2018 (Figure 6a) to show higher frequency θ v dynamics that are not apparent in Figure 5. The near-surface zone (<2.5 m) has strong variations that are not present at deeper depths (Figure 6b), and shallow soil water contents dropped to <0.3 m 3 m −3 at 0.5 m (Figure 6b). For all the sensors shallower than 2.5 m, the time series show three distinct features. First, there is a general dry down starting after the mid-September rain events until the rain events in mid-October. The data also indicate a pattern of decreasing dry down with increasing depth, and by 2.4 m the diurnal oscillations of the shallower soils are nearly dampened out. The second distinct feature is the effect of diurnal oscillations in θ v that are particularly strong in the shallowest sensors and decrease in amplitude with depth (Figure 6b). At 0.05 m, θ v was typically at its daily maximum during the midnight to early morning period and then decreased until the evening. The third feature are the θ v responses to rainfall events. These events often caused a rapid rise in water contents, followed by a slower decay to lower values. Within a few days after the rain events, values resumed the decreasing trend of water contents during the dry season.

Soil moisture calibration
Given the need for accurate θ v measurements, this study examined the effect of calibration on TDR based measurements. We found that a local soil-based calibration approach provided a much more realistic estimate of θ v than the Topp model (Topp et al., 1980) for the soils at our study site. The Topp model yielded systematically low values, and the magnitude of the differences from independently measured, laboratory-based water contents were substantial ( Figure 3). Maximum deviations occurred at the highest water contents. Caldwell, Bongiovanni, Cosh, Halley, and Young (2018) also found greater deviations at higher water contents for their batch equilibration calibration analyses using the Topp model, although the overall shape of the measured K a −θ v relation for their study soils was quite different from that in this study. Although the Topp model can be satisfactorily applied for some soils, our results suggest that this model should not be assumed as representative for clay-rich humid tropical soils without verification. There are a few likely reasons why the third-degree polynomial fit using local soils differs from the Topp model for the study site soils. Bulk densities from the pit produced a polynomial with coefficients of an order of magnitude different than similar bulk densities used by Topp et al. (1980). Soil texture (i.e., high clay contents at the study site) and structure in our site (discussed in Broedel et al., 2017;Marques et al., 2015) are likely reasons for the deviation. High-clay-content soils can cause dispersive waveforms because of high surface areas, which create bound water effects. Dispersive behavior causes permittivity to be out of phase across the TDR frequency bandwidth (Robinson, Jones, Wraith, Or, & Friedman, 2003).
The local soil calibration for the pit soils resulted in a distinctive K a −θ v relation (Figure 2). For θ v ≤ 0.3 m −3 m 3 , the K a −θ v slope is quite steep. The high slope may be related to the stronger matric suctions associated with drier soils (matric potential decrease and strong pressures bind water to the soil) and residual water content effects (especially for the lowest part of the calibration curve), which can affect the electric field applied by the sensor (Bonan, 2016;de Jeu, Holmes, Parinussa, & Owe, 2014;Wang & Schmugge, 1980). Over the range of θ v between ∼0.3 and ∼0.5 m 3 m −3 , the curve has a lower slope where K a increases more rapidly with a given change in θ v . For θ v ≳ 0.5 m 3 m −3 , the slope increases again as the volumetric water content approaches the porosity of soil (Wang, Schmugge, & Williams, 1978). Figure 2b shows that the calibrated model produced large errors for depth >3.2 m when θ v < 0.2 m 3 m −3 , yet these laboratory-created soil moisture values (for our calibration) were never observed in the field for those depths ( Figure 5).

Soil moisture dynamics
Temporal and depth variability of θ v at the study site from 10 Sept. to 22 Oct. 2018 revealed that moisture content corresponded to rainfall events differently within distinctive zones in the soil profile, highlighting the importance of measuring soil moisture beyond shallow soils (Figure 6). The upper ∼2.5-m layer was highly dynamic, and the lowest measured water contents were observed in the shallowest part of this zone. The zone between ∼2.5 and ∼7 m had consistently high water contents relative to the shallowest zone and intermediate temporal variability. Below ∼8 m, there was a zone of lower water content, and then ∼12 m and deeper, water contents increased. This zone had generally lower temporal variability than the two shallower zones. Broedel et al. (2017) examined θ v profiles within the same pit using θ v data from January 2003 to January 2006. They showed that observed depth variability in θ v was largely related to changes in soil texture and structure (including macroporosity), and root density. This study extends the work of Broedel et al. (2017) in some key ways, beyond the benefit of having an additional set of data from a later time period. First, Broedel et al. (2017) did not use a local soil calibration in an old-growth forest but a calibration developed from soil moisture observation up to 1.5 m over monoculture and agroforestry systems in the central Amazon (Teixeira, 2001). Second, Broedel et al. (2017) did not make TDR measurements shallower than 80 cm (although some shallow data were measured nearby using a neutron probe). As shown in Figures 5 and 6 (and discussed below), with respect to Broedel et al. (2017), this study was able to examine higher frequency θ v dynamics with better depth resolution in the shallow soils where root densities are highest. Finally, the data from this study provided a higher resolution view of the temporal variability across the entire soil profile, which revealed features and characteristics that were not observed in the Broedel et al. (2017) dataset. For example, the details of pulsed percolation events are much more evident in the newer dataset. This difference is likely related to the K a −θ v relationship developed in this study that revealed finer scale differences, as well as improvements in the TDR sensors.
The following discusses the key characteristics of each depth zone. Seasonal soil moisture decline during the dry season and recovery after rainfall events were readily observed in the upper two soil zones, but their dynamics differ, likely due the textural differences and the decline in root biomass with depth. In the upper ∼2.5-m zone, the effects of dry down are clearly seen by the low θ v contents in Figure 5 during the dry season (e.g., September and October of 2018). The dry down was most pronounced in the upper 0.5-to 1-m depths. The 2018 dry season decrease in moisture at shallow depths was largely driven by plant water use because there was little rainfall to replenish soil water storage, and soil evaporation has been shown to be only a minor part of the site water balance (Broedel et al., 2017;Negrón-Juárez et al., 2007). Root density is also substantially higher in the upper 0.5 m than at deeper depths (Chauvel et al., 1992). Although preferential flow was not measured in this study, measurements from a site in central Amazonia (Tomasella & Hodnett, 1996) showed that macropores in the Manaus clayey soils are concentrated in the first 0.75 m of the soil profile and decrease abruptly at 1.2 m deep, concomitant with the reduction of root density. The effect of macropores can be clearly seen in Figure 5 during the dry period, where there is a layer at ∼0.5 m that drains rapidly after rainfall events and remains drier than contiguous layers. Additional discussion of dry period effects on shallow soil moisture is provided below. When the wet season began in late 2018, water contents increased in the shallow zone, and by December, water contents had increased substantially. Multiple percolation pulses can be seen in the shallow soils during the wet season (blue vertical streaks in Figure 5) resulting from large individual storms and the combination with elevated antecedent moisture conditions. Soil structure and macroporosity likely play an important role in terms of infiltration and percolation in the shallow zone, which has a strongly structured microaggregate characteristic (Chauvel et al., 1992) and higher macroporosity (Broedel et al., 2017) than the deeper soil zones. These features promote rapid infiltration and percolation, resulting in little to no overland runoff at the site despite the large amounts of rainfall.
In the intermediate zone (∼2.5 to ∼8 m), soil moisture is more consistent with time, and during the dry season there are only small drawdown effects (consistent with a substantial reduction in root density). Within this zone, there are depth intervals with distinctly higher or lower θ v values (light to dark blue to light blue transitions in Figure 5). Consistently high θ v values occur between 3 and 4 m and between 5 and 7 m (especially between about 6 and 6.5 m). Differences in water content in these intervals are related to soil texture and structure, and root densities. The microaggregate structure becomes less pronounced with depth in this zone, with a major transition to substantially less structure at ∼3.5 m. There is also an increase in porosity at ∼6.5 m (Broedel et al., 2017;Chauvel et al., 1992;Chauvel, Lucas, & Boulet, 1987), which corresponds to a zone of continuously elevated moisture content. Some fine roots are present in this zone, and it appears that plants can utilize this water during drought periods, although little to no plant water uptake from this zone appears to occur during nonextreme drought periods (Broedel et al., 2017), which was the situation during this study. In contrast with the shallow soils with strong transpirationdriven dry down periods, temporal variability in this intermediate zone appears to be largely controlled by wet season percolation events ( Figure 5). The bottom of this layer at ∼8 m appears to be the lowest depth where deep pulsed percolation events have a substantial impact on temporal variability of θ v . It is also notable that the duration of these distinct pulses (based on the width of the dark blue vertical "spikes" in Figure 5) is only a few days, indicating that redistribution of these relatively deep water additions happens rapidly until about the 6.5-m depth.
In the upper part of the lower zone (∼8-12 m), θ v is typically lower than that at the bottom of the intermediate zone (6-8 m) and below that water contents increase again. Porosity is lower compared with the intermediate zone, and Broedel et al. (2017) suggested that deep percolation was the main control on θ v in the deep zone. Our survey of living roots indicated that there were none below 10 m, which supports the idea of drainage as opposed to evapotranspiration as the major control in this zone. In general, the strong temporal variations that are apparent in the upper and intermediate zones are substantially damped in the lower zone. However, there was an increase in water content to saturation or near-saturation levels starting in very late December 2018 at ∼14 m that appears to be related to percolation of wet season rainfall. The elevated θ v values at 14 m persisted for >3 mo into April 2019. This event lagged the initial wet up in the shallow and intermediate zones by >1 mo, which gives an indication of the time needed for water to percolate from the intermediate zone to 14 m. Recharge to the groundwater table at 30-35 m typically occurs ∼4 mo after onset of the wet season .
A higher resolution time series of the variations in θ v over most of the 2018 dry season is shown in Figure 6, and it shows quite interesting dynamics during a 21-d period when plant available water was most limited during the study. At shallow depths (<2.5 m), θ v values have high temporal variability, whereas deeper depths show minimal variability. As mentioned in Section 3, the shallow time series have three main characteristics: (a) a distinct dry down trend until the rain events in late October; (b) rapid but short-duration increases in θ v in response to infrequent rain events; and (c) diurnal fluctuations in θ v .
The dry down period reduced θ v at 0.05 m from ∼0.4 to ∼0.27 m 3 m −3 over a 40-d period, which included rain events on seven different days ( Figure 6). The extent of the dry down decreased with depth and by 3.2 m was barely detectible. The larger dry downs in the top 0.5 m are consistent with the highest root densities observed in the pit (Chauvel et al., 1992), and based on the overall moisture profiles, it appears that plant water uptake was dominantly in the upper 0.5 m with minor uptake down to ∼3.2 m. However, during drought events, water uptake might occur from very deep soils (Markewitz, Devine, Davidson, Brando, & Nepstad, 2010) Percolation depths inferred by increases in θ v during and after rain events were primarily limited to the upper 0.5 m, although the 1.6-m sensor did show some rain-event-related increases. The deeper parts of the profile do not show any increases in θ v over the same period, so deep percolation was minimal. Overall, these results suggest that plant water availability was sufficient in the shallower depths to meet transpiration demand. Infrequent rains helped buffer stores of shallow plant available water, and it would be interesting to see if deeper water stores would be accessed if rainfall amounts and frequency were reduced during drought conditions. Continued monitoring at the site will help clarify patterns of θ v and plant water use over interannual periods (Bruno et al., 2006).
The shallower θ v profiles show distinct diurnal variations with higher water contents occurring during the night and early morning period, and lower water contents during the afternoon and early evening. Diurnal characteristics with each sensor depth down to 2.4 m are summarized in Table 1, and results were determined using the average θ v of each daily maximum and minimum for a given depth. The times of the daily minima and maxima were also examined. Only rain-free days were used in the summary, although lingering rainfall effects from the relatively infrequent rains likely account for some of the variability between depths. These were small amplitude diurnal changes in θ v where percentage differences (difference between maximum and minimum/maximum × 100) was ∼2% or less (Table 1). Decreases in amplitude with depth are clearly evident from the percentage difference between maximum and minimum data in Table 1, and by 2.4-m depth, there was only occasional diurnal variability. Temporally, the average maximum θ v values typically occurred between midnight and 0300 h, and there was no clear trend with depth. Average minimum values occurred in the afternoons to early evenings between about 1400 and 2000 h. Average durations between maximum and minimum θ v values ranged from 15 to 19 h. The diurnal pattern is common during the dry season, as shown in Figure 6. However, it is not apparent during the wet season (see Supplemental Figure S1), suggesting that the observed pattern is not an artifact (seasonal temperatures are very similar; Negrón-Juárez et al., 2017) but a true ecohydrological response.
Diurnal changes in θ v have been observed in old-growth forests in the eastern Amazon, Brazil (da Lopes, 2001;da Rocha et al., 2004). Diurnal variations observed at the Tapajos field site by da Rocha et al. (2004) extended down to 2-m depth, and daily variations were a few percent at 0.05 m and decreased in amplitude with depth down to 2 m. They showed a similar timing of peak θ v at night, like in this study; however, daily minima appear to have occurred by 1200 h or earlier based on Figure 3 of their manuscript, a substantially shorter duration between TA B L E 1 Summary of 2018 dry-season diurnal variability of volumetric water content (θ v ) with depth. Average values are from 10 Sept. to 24 Oct. 2018. Only rain-free days were included in the calculations minima and maxima than we observed. They attributed the nocturnal recovery in θ v to either hydraulic lift within the plant root system or flow through the bulk soil (presumably related to the physical processes discussed above). Oliveira, Dawson, Burgess, and Nepstad (2005) measured sap flow on roots at the same site, which suggested upward movement of water in tap roots and release to the soil along lateral roots during the dry season, consistent with the concept of hydraulic lift and hydraulic redistribution. They also noted downward movement of water through roots during the wet season, which bypassed some soil depths, creating near-instantaneous increases of water to depths well below that which could be accounted for by simple soil percolation. We observed similar bypassing during the wet season, although given the lack of root sap flow data it is not clear if the bypassing is related to downward redistribution within the roots or preferential or macropore flow effects, which could result in similar behaviors. A detailed examination of the diurnal variations using the deep pit data is underway and will be the subject of a future study.

CONCLUSIONS
This study provides the first TDR soil moisture calibration (θ v ) in an old-growth forest in the central Amazon. A calibration based on local soil measurements found that the Topp et al. (1980)) model underestimated the soil moisture content by 22-42%. This large difference suggests that site-specific calibration of TDR sensors for tropical soils are necessary, especially in soils with high clay contents as observed in plateau areas in the central Amazon. Using the improved calibration, we were able to examine temporal θ v dynamics down to 14.2 m. Seasonal and subdaily variations in θ v were observed, with the greatest temporal variability occurring in the top soil layer. A strong dry season reduction in θ v was limited to the top 2.4 m, suggesting that this was the zone of greatest plant water use. Although minor decreases in θ v were observed for the deeper depths dur-ing the dry season, pulsed deep percolation events appear to be responsible for much of the temporal variability in the deeper soil. Continued monitoring at the site will provide a better understanding of interannual soil water dynamics in the central Amazon and will be extremely valuable for characterizing future drought impacts in old-growth forests.

A C K N O W L E D G M E N T S
This research was supported as part of the Next Generation Ecosystem Experiments-Tropics, funded by the USDOE, Office of Science, Office of Biological and Environmental Research, under Contract no. DE-AC02-05CH11231. Laura Borma would like to acknowledge Go-Amazon (2013/50531-2) for retrofitting the pit structure. We would like to thank the Large Scale Biosphere-Atmosphere Program (LBA), coordinated by the National Institute for Amazon Researches (INPA), for the use and availability of data, and for logistical support and infrastructure during field activities. M. Mota and R. Oliveira are graduate students funded by CAPES and FAPEAM.

C O N F L I C T O F I N T E R E S T
The authors declare no conflict of interest.