Soil Water Stable Isotopes Reveal Surface Soil Evaporation Loss Dynamics in a Subtropical Forest Plantation

Abstract

Line-conditioned excess (lc-excess), the deviation of the relationship between δD and δ¹⁸O in soil water from that of precipitation, is often used to indicate soil evaporation loss, but the conditions of using lc-excess under the influences of precipitation infiltration or percolation had not been identified. The interaction effects of climate, soil and vegetation on soil evaporation in forests are not well known. We collected soil water at 0–5, 15–20 and 40–45 cm depths and event-based precipitation from 2011 to 2015 in a subtropical forest plantation and calculated the lc-excess. Precipitation on the sampling day and percolation of upper soil water with low lc-excess affected the capacity of the lc-excess to indicate the soil evaporation fractionation signals. Lc-excess of soil water at 0–5 cm depth indicated a reliable soil evaporation loss estimate over 30 days prior to the sampling day. Soil evaporation loss was dominated by the cumulative soil temperature (T_ss) during drought periods and was dominated by the relative soil water content (RSWC) during non-drought periods. High T_ss decreased soil evaporation loss by increasing transpiration and relative humidity. Our results emphasize the importance of sampling the upper-most soil layer when there is no rain and vegetation during drought periods in forests when studying soil evaporation loss dynamics.

1. Introduction

Evaporation from the soil surface can increase non-productive soil water loss and adjust the local climate by cooling the surface or increasing the concentrations of atmospheric water vapor [1,2]. The fraction of evaporation loss to total precipitation affects the availability of water in forest ecosystems, determines plant growth and survival and influences the production of food and paper products [3,4]. Surface soil evaporation accounts for an average of 20–40% of evapotranspiration (ET); it can absorb solar energy and transform it into latent heat flux, decreasing sensible heat flux and cooling the evaporating surface [5,6]. A reliable assessment of soil evaporation loss is critical for evaluating the forest ecosystem water and energy balance and the associated effects of climate change.

The line-conditioned excess (lc-excess; [7]), which is defined as the deviation of the relationship between δD and δ¹⁸O in soil water from that of precipitation, can reflect the integrative soil evaporation loss over a period of time [8,9]. The isotopic enrichment in soil water over that in precipitation is mainly caused by soil evaporation; root water uptake generally does not cause isotopic fractionation, except in some salt tolerant coastal species and woody xerophytes (Figure 1) [10,11]. The isotopic enrichment of soil water leads to a deviation of the relationship between δD and δ¹⁸O in soil water from that of precipitation [7]. The primary advantage of the isotope-based method is that the indication corresponds to a relatively long period (e.g., days and months) that occurred before soil sampling, but it does not require continuous field observations [8,12,13]. Field measurement methods, such as direct-measured methods, including micro-lysimeter [14] and chamber technology [15], and indirect-measured methods which are based on ET-partitioning with eddy covariance [16] and chamber techniques [17] and transpiration measurements, such as sapflow techniques [18], estimate soil evaporation at particular points and have relatively short observational timescales [3,9]. The temporally-integrated estimates of soil evaporation with these methods require high-frequency or continuous observations in situ, which are time-consuming and costly [3].

However, precipitation inputs by infiltration and percolation of water from the upper soil layers may hinder the capacity of lc-excess to indicate the soil evaporation loss (Figure 1) and the conditions in which the lc-excess can be used remain unclear [9,19]. The existing studies found that lc-excess was determined by evaporation fractionation factors and soil residual water storage [7]. Precipitation with an unfractionated signal infiltration may dilute the fractionation effects of evaporation and result in an increase in the lc-excess as a result of mixing with pre-event soil water [9]. Sprenger et al. [19] found that when antecedent precipitation volumes govern the fractionation signal of soil water, there is no linear relationship and, even, a lagged response between soil evaporation and the kinetic fraction signal of soil water [19]. In addition, percolation of water from shallow soil layers where evaporation fractionation is high may increase the evaporation fractionation effects and result in a decrease in lc-excess of water in deeper soils [20].

In addition to the effects of climate and soil, transpiration and variability in the shade provided by vegetation can strongly influence soil evaporation loss by controlling soil water status and the microclimate in forest ecosystems, while knowledge of the effects of vegetation on soil evaporation and, consequently, on lc-excess is limited [21,22,23]. Soil evaporation loss is mainly controlled by atmospheric demand and soil water supply, which influences lc-excess by regulating soil evaporation fractionation factors and soil residual water storage [3]. Atmospheric demand of soil evaporation can be controlled by the soil temperature (T_s) affecting the available energy in the upper layer of the soil and the relative humidity (RH) and wind speed (WS) affecting the vapor gradient between the soil and the atmosphere [24]. High T_s, low RH and high WS may cause high soil evaporation loss and a negative lc-excess [19]. Soil water supply for evaporation can be influenced by soil water content, which is affected by precipitation infiltration, transpiration and percolation [25]. Vegetation coverage alters atmospheric demand by intercepting solar radiation and affecting ground relative humidity [26] and changes the soil water supply by intercepting precipitation and absorbing soil water [22]. During the dry season, a positive relationship between the soil water content and lc-excess of surface soil water in a deciduous forest indicated that reduced soil moisture was largely the result of an increase in soil evaporation; the enhancing evaporation of canopy-intercepted precipitation caused a negative relationship between the lc-excess and soil water content in an evergreen coffee plantation with a high leaf area index [21].

China has the largest area of forest plantations in the world, with plantations comprising about 31.6% of the total forest area [27]. Furthermore, 54.3% of the plantations are distributed in the subtropical region [28]. The inconsistent distribution of temperature and precipitation results from seasonal droughts that occur frequently in this region [29,30]. During drought periods, atmospheric demand for soil evaporation increases, soil water supply decreases and dense vegetation regulates the two factors [31]. Therefore, differences can be expected in soil evaporation drivers between drought and non-drought periods.

In this study, we analyzed the stable isotopic composition of soil water extracted from soil samples at 0–5, 15–20 and 40–45 cm depths and event-based precipitation in an East Asian monsoon subtropical forest plantation and calculated the lc-excess of collected soil water. Our objectives were: (1) to determine the conditions in which lc-excess can be used to indicate surface soil evaporation loss dynamics, under the influences of precipitation infiltration or percolation of water from the upper soil layers, and (2) to analyze the temporal patterns of lc-excess and the effects of climatic factors, soil characteristics and vegetation cover on its temporal patterns during different time periods: over the whole year and in non-drought and drought periods.

2. Materials and Methods

2.1. Study Site

This study was conducted at Qianyanzhou (QYZ) Ecological Experimental Station (26°44′52″ N, 115°39′47″ E and elevation 102 m) of the Chinese Ecosystem Research Network (CERN), a member of ChinaFLUX, located in Jiangxi Province, South China. This region is controlled by the Western Pacific subtropical high and has a subtropical East Asian monsoon climate [32]. Annual precipitation and the mean air temperature are 1407.3 ± 300.2 mm and 18.1 ± 0.5 ℃ (1985–2020), respectively, according to the meteorological records of CERN.

The study plot (100 × 100 m²) is at the top of the hill, where soil depth is less than 100 cm and the average groundwater level is about 3.43 m. The soil at the site is red earth that is weathered from sandstone, sandy conglomerate, mudstone and alluvium, with pH values from 4.04 to 6.25 [33]. The soil bulk density is approximately 1.57 g/cm³ and the sand, silt and clay contents in soil are 17%, 68% and 15%, respectively [34]. The subtropical plantation at the site was planted in 1985 and the dominant tree species are Masson pine (Pinus massoniana L.), slash pine (Pinus elliottii E.) and Chinese fir (Cunninghamia lanceolata L.) [35]. The growing season of the subtropical forest plantation is from March to October.

2.2. Division of Drought and Non-Drought Periods

In this study site, the monthly maximum precipitation occurred in June and more than half of the annual precipitation (68%) takes place in spring (March–May, 32%) and summer (June–August, 36%), which are the main rainy seasons (Figure 2a). The monthly air temperature showed a unimodal distribution and peaked in July (Figure 2b). The aridity index (AI), defined as the ratio of precipitation to potential evapotranspiration, first decreased and then increased and reached a low value of ≤1.0 from July to October (Figure 2c). This indicated that high temperatures, causing high potential evapotranspiration, and a lack of sufficient precipitation resulted in seasonal droughts frequently occurring from July to October [31,33]. Therefore, the period between July and October was defined as the drought period and other months as the non-drought period.

2.3. Sampling Collection and Measurements

Soil samples were collected 2–3 times per week from July 2011 to March 2015 for δD and δ¹⁸O isotopic analyses. Soil samples at 0–5, 15–20 and 40–45 cm depths were collected with a hollow-stem auger (0.04 m in diameter and 0.25 m in length) and three replicate soil samples at each depth were randomly taken in the 100 × 100 m² sampling plot. All soil samples were stored in a refrigerator, at −15 to −20 °C, until soil water extraction. Precipitation samples were collected after each rain event from July 2011 to March 2015 in polyethylene bottles fitted with funnels and topped with ping-pong balls to prevent evaporation [27]. All precipitation samples were refrigerated at 4 °C before isotope analysis.

Soil water was extracted with a cryogenic vacuum distillation system [36], with extraction time of 0.5–1.5 h depending on soil water content. Extraction efficiency of water from soil samples was > 98.0% for obtaining un-fractionated water samples [36]. Soil water and precipitation were filtered through a 0.45 μm mixed cellulose membrane (Jiuding Gaoke Co. Ltd., Beijing, China) and δD and δ¹⁸O were then analyzed using an Isotopic Ratio Infrared Spectroscopy (IRIS) system (Model DLT-100; Los Gatos Research Inc., San Jose, CA, USA) [37]. The analytical precision of the liquid water isotope analyzer was 0.3‰ for δD and 0.1‰ for δ¹⁸O. After water extraction of the soil samples, they were oven dried (105 °C, 48 h) and the soil water content (SWC) was calculated as SWC (cm³/cm³) = ((fresh weight−dry weight)/(dry weight)) × soil bulk density.

2.4. Calculations of Lc-Excess and Soil Evaporation Loss Fraction

Lc-excess is calculated using the following Equation (1) [7]:

lc-excess = δD–a × δ¹⁸O–b

(1)

where a and b represent the slope and intercept of the monthly local meteoritic water line (LMWL). Precipitation amounts of >5–6 mm could pass through the canopy and litter and infiltrate the soil in this study area, according to the correlations between the isotopic composition of soil water and precipitation collected on the same day. Therefore, δD and δ¹⁸O in precipitation >5 mm were used for fitting LMWL.

The slopes or intercepts of the monthly LMWL had significant differences (p < 0.001) based on the univariate analysis of a general linear model. The reason for this may be that the East Asian monsoon climate leads to different water vapor sources in summer and winter. Therefore, the lc-excess was calculated using the monthly LMWL for decreasing the influences of seasonal variability on precipitation inputs. The lc-excess of precipitation calculated with the monthly LMWL was about 0‰ and had relatively little seasonal fluctuation. A negative lc-excess value indicates that soil water may have undergone some isotopic enrichment due to fractionation during soil evaporation [8].

Among the triplicates of the soil water samples at 0–5, 15–20 and 40–45 cm depths, δD and δ¹⁸O showed no significant differences (p > 0.05) according to multiple comparisons. Soil water content among the triplicates of soil water samples at 0–5 cm depths showed no significant differences, but at 15–20 and 40–45 cm depths showed significant differences. The weighted mean lc-excess was obtained by weighting the soil water content of each sample, using the following equation:

$${\left(\mathrm{lc}-\mathrm{excess}\right)}_{\mathrm{mean}}=\left({\displaystyle \sum _{\mathrm{i}=1}^3\left({\mathsf{\theta }}_{\mathrm{i}}\times \mathsf{\delta }{\mathrm{D}}_{\mathrm{i}}\right)÷{\displaystyle \sum _{\mathrm{i}=1}^3{\mathsf{\theta }}_{\mathrm{i}}}}\right)-\mathrm{a}\times \left({\displaystyle \sum _{\mathrm{i}=1}^3\left({\mathsf{\theta }}_{\mathrm{i}}\times {\mathsf{\delta }}^{18}{\mathrm{O}}_{\mathrm{i}}\right)÷{\displaystyle \sum _{\mathrm{i}=1}^3{\mathsf{\theta }}_{\mathrm{i}}}}\right)-\mathrm{b}$$

(2)

where θ is the soil water content of a sample; i is the number of soil core replicates. Machine learning [38] was used for predicting the missing δD and δ¹⁸O of precipitation events after data quality control. Root-mean-squared errors (RMSE) of prediction, using machine learning, were 16.90 for δD and 2.04 for δ¹⁸O, respectively.

The soil evaporation loss fraction (f = evaporation/precipitation, %) was calculated based on the soil water balance, ignoring the surface runoff and the steady-state isotope mass balance, with the assumption that the isotope ratio of transpiration was equal to that of the soil water and the lc-excess of the effective input water was set to zero [13]. The soil evaporation loss fraction is calculated using the following equation:

$$\mathrm{f}=\raisebox{1ex}{$\left(\mathsf{\delta }{\mathrm{D}}_{\mathrm{S}\mathrm{W}}-\mathrm{a}\times {\mathsf{\delta }}^{18}{\mathrm{O}}_{\mathrm{S}\mathrm{W}}-\mathrm{b}\right)$}\!\left/ \!\raisebox{-1ex}{$\left(\mathrm{a}\times \left({\mathsf{\delta }}^{18}{\mathrm{O}}_{\mathrm{E}}-{\mathsf{\delta }}^{18}{\mathrm{O}}_{\mathrm{S}\mathrm{W}}\right)-\left(\mathsf{\delta }{\mathrm{D}}_{\mathrm{E}}-\mathsf{\delta }{\mathrm{D}}_{\mathrm{S}\mathrm{W}}\right)\right)$}\right.\times 100$$

(3)

where SW refers to soil water, E refers to soil evaporation and a and b represent the slope and intercept of the monthly local meteoritic water line (LMWL). The values of δD_E and δ¹⁸O_E were estimated based on the Craig–Gordon model [39]:

$${\mathsf{\delta }}_{\mathrm{E}}=\frac{({\mathsf{\delta }}_{\mathrm{SW}}-{\mathsf{\epsilon }}^{+})/{\mathsf{\alpha }}^{+}-\mathrm{R}\mathrm{H}\times {\mathsf{\delta }}_{\mathrm{A}}-{\mathsf{\epsilon }}_{\mathrm{K}}}{1-\mathrm{R}\mathrm{H}+{10}^{-3}\times {\mathsf{\epsilon }}_{\mathrm{K}}}$$

(4)

where RH is the relative humidity, set to the average relative humidity over 0, 1, 2, 3, 5, 7, 10, 15, 20, 25, 30, 35, 40 and 45 days prior to the sampling day; α⁺ is the liquid-vapor equilibrium fractionation factor; ε⁺ is the equilibrium isotopic separation between liquid and vapor, calculated as ${\mathsf{\epsilon }}^{+}(‰)=\left({\mathsf{\alpha }}^{+}-1\right)\times {10}^3$; ε_K is the kinetic fractionation factor; δ_A is the δD and δ¹⁸O in ambient atmospheric vapor and is determined using the precipitation-equilibrium assumption:

$${\mathsf{\delta }}_{\mathrm{A}}=\left({\mathsf{\delta }}_{\mathrm{P}}-{\mathsf{\epsilon }}^{+}\right)/{\mathsf{\alpha }}^{+}$$

(5)

δ_P is the δD and δ¹⁸O in precipitation, set to the weighted average of δD and δ¹⁸O in precipitation over 0, 1, 2, 3, 5, 7, 10, 15, 20, 25, 30, 35, 40 and 45 days prior to the sampling day; α⁺ is estimated from empirical relationships by Horita and Wesolowski [40]:

$$\begin{array}{l}\ln \left[{\mathsf{\alpha }}^{+}{}_{\left(\mathrm{D}\right)}\right]=\frac{2.9992}{{\left({\mathrm{T}}_{\mathrm{s}}+273.15\right)}^3}\times {10}^6+161.04\times {10}^{-3}+\frac{794.84}{{10}^6}\times \left({\mathrm{T}}_{\mathrm{s}}+273.15\right)\\ -\frac{1620.1}{{10}^9}\times {\left({\mathrm{T}}_{\mathrm{s}}+273.15\right)}^2+\frac{1158.8}{{10}^{12}}\times {\left({\mathrm{T}}_{\mathrm{s}}+273.15\right)}^3\end{array}$$

(6)

$$\ln \left[{\mathsf{\alpha }}^{+}{}_{\left({}^{18}\mathrm{O}\right)}\right]=\frac{0.35041}{{\left({\mathrm{T}}_{\mathrm{s}}+273.15\right)}^3}\times {10}^6-\frac{1.6664}{{\left({\mathrm{T}}_{\mathrm{s}}+273.15\right)}^2}\times {10}^3-\frac{6.7123}{\left({\mathrm{T}}_{\mathrm{s}}+273.15\right)}-7.685\times {10}^{-3}$$

(7)

where T_s is soil temperature at the evaporation front, set to the average soil temperature at 0–5 cm depth over 0, 1, 2, 3, 5, 7, 10, 15, 20, 25, 30, 35, 40 and 45 days prior to the sampling day. The ε_K is calculated as [41]:

$${\mathsf{\epsilon }}_{\mathrm{k}}{}_{\left(\mathrm{D}\right)}=\mathrm{n}\times (1-\mathrm{R}\mathrm{H})\times (1-0.9755)\times {10}^3$$

(8)

$${\mathsf{\epsilon }}_{\mathrm{k}}{}_{\left({}^{18}\mathrm{O}\right)}=\mathrm{n}\times (1-\mathrm{R}\mathrm{H})\times (1-0.9723)\times {10}^3$$

(9)

where n ranges from 0.5 (saturated soil condition) to 1.0 (very dry soil condition) and is set as 0.75 because the evaporating soil layer was expected to alternate between saturation and being dry over time.

2.5. Ancillary Indicator Measurements

Ancillary meteorological data were used for evaluating the influences of climatic factors on the lc-excess and included precipitation (P) influencing the soil water supply, cumulative soil temperature (T_ss) and heat flux reflecting available energy in the upper layer of the soil and relative humidity (RH) and wind speed (WS), reflecting a vapor gradient between the soil and the atmosphere. Precipitation was monitored with a rain gauge (RM Young Inc., Traverse, MI, USA) and recorded every 30 min by a CR10X data logger (Campbell Scientific Inc., Logan, UT, USA). Air temperature and relative humidity sensors (HMP45C, Campbell Scientific Inc., Logan, UT, USA) and wind speed sensors (A100R, Vector GB Ltd., Birmingham, UK) were installed at 2 m above ground level. The air temperature, relative humidity and wind speed raw data were sampled at 10 Hz and the 30 min mean fluxes were calculated and stored by a CR5000 data logger (Campbell Scientific Inc., Logan, UT, USA).

Soil volumetric water contents (SWC) were continuously measured at 5, 20 and 50 cm depths using three TDR probes (CS615-L, Campbell Scientific Inc., Logan, UT, USA) at a single location in the sampling plot in order to reflect the soil water condition of the soil sampling site to some extent. Soil temperatures were continuously measured at 5, 20 and 50 cm depths using three thermocouple temperature sensors (105T, Campbell Scientific Inc., Logan, UT, USA). Soil heat flux sensors (HFP01, Hukseflux, Delft, The Netherlands) were installed at 3 and 5 cm depths. Soil data were collected every 30 min with a CR10X data logger (Campbell Scientific Inc., Logan, UT, USA). Saturated soil water content (SSWC, cm³/cm³) was simulated using Rosetta 3, which was used for estimating soil hydraulic parameters with hierarchical pedotransfer functions [42]. The relative soil water content (RSWC) was estimated with the equation: RSWC = SWC/SSWC and reflected the soil’s residual water storage.

Evapotranspiration was measured with an eddy covariance flux system, which was comprised of an open-path CO₂/H₂O analyzer (Model Li-7500, Licor Inc., Lincoln, NE, USA) and a 3D sonic anemometer (Model CSAT3, Campbell Scientific Inc., Logan, UT, USA) at 39.6 m [31]. The potential evapotranspiration (PET) was estimated using the Penman–Monteith equation [43]. Transpiration (T) at the ecosystem level was calculated based on the accumulation of the sap flow density of each species, which were measured with the FLGS-TDP XM1000 system (TDP-30 mm, Dynamax Inc., California, CA, USA) at the study plot. The detailed calculation method of transpiration can be found in Song et al. [31,33]. Normalized Difference Vegetation Index (NDVI) was selected as a vegetation variable which is particularly sensitive to the leaf area index (LAI) and chlorophyll density [44]. The MODIS NDVI product (MOD13A3), which has a spatial resolution of 250 m and a time resolution of 16 days, was downloaded from the National Aeronautics and Space Administration (NASA) website (https://modis.gsfc.nasa.gov/data/) (accessed on 20 August 2021).

2.6. Statistical Analyses

Linear regression was used for analyzing the correlations between the lc-excess of the last precipitation event preceding sampling and the lc-excess of soil water and among the lc-excess of soil water at 0–5, 15–20 and 40–45 cm depths and the relationship between f and soil water lc-excess with SPSS 22.0 software. Correlation coefficients between the soil evaporation loss fraction (f) and the ratio of potential evapotranspiration to precipitation (PET/P) over different days prior to the sampling day were calculated for analyzing the period of soil evaporation indicated by f based on the linear regression. Pearson correlations were calculated to describe the relationships among lc-excess, average values of RH, WS, soil water content, T and NDVI and cumulative P, soil temperature and heat flux over seven days prior to each sampling campaign. Pearson correlation coefficients were calculated using the R studio 3.5.0.

Multiple linear regression (MLR) was used to assess which variables over the 7, 15 and 30 days prior to the sampling day explain the temporal variance in lc-excess of sampling soil water, during different time periods: whole year, non-drought and drought periods. Predictors included the average values of RH, WS, soil water content, T and NDVI and cumulative P, soil temperature and heat flux over 7, 15 and 30 days prior to each sampling campaign. The significance of predictors was tested with analysis of variance (p < 0.05). The final MLR model only considered significant predictors. All of the data used in the MLR were standardized to account for differences in the units and ranges of the predictors (z-scores transformation). MLR was conducted with SPSS 22.0 software.

Redundancy analysis (RDA) was used to explore the relative contributions of climate, soil and vegetation variables to the temporal variation in lc-excess. During RDA analysis, influencing variables with low correlation between variables and lc-excess were gradually taken out, based on collinearity and the total explanation of variability. RDA analysis was done using R studio 3.5.0.

3. Results

3.1. Seasonal Variability of Climate, Soil and Vegetation

The average annual precipitation from 2011 to 2015 was 1503.4 mm and the monthly maximum precipitation of 236.7 mm occurred in June (Figure 3). During the spring and summer monsoon, cumulative seasonal precipitation contributed 66.8% to the total annual precipitation and the two monsoons were the main rainy seasons. Average annual potential evapotranspiration and transpiration from 2011 to 2015 was 802.3 and 466.2 mm, respectively, and their monthly values showed a unimodal distribution with the peak value (average 138.1 mm for potential evapotranspiration and average 83.4 mm for transpiration) in July. The average relative humidity was 85.1%, reaching a low value of 80.2 and 83.4% during July and August. The average wind speed was 0.65 m/s and the wind was predominantly from the south with a maximum of 0.82 m/s in July. The lower average wind speeds of 0.48–0.59 m/s were observed from September to December.

Annual precipitation from 2011 to 2015 was higher than that from 1985 to 2020 (1407.3 mm) due to higher precipitation amounts in November and December of 2011 to 2015 (Figure 2a). During 1985 to 2020 and 2011 to 2015, the monthly air temperature showed a similar unimodal distribution and peaked in July, and their monthly averages were also similar (Figure 2b). The annual average aridity index (AI) from 2011 to 2015 (2.28) was higher than that from 1985 to 2020 (1.68) due to a higher AI in November and December of 2011 to 2015 (Figure 2c). However, the AI was lower than or close to 1 due to the relatively high temperatures and lack of sufficient precipitation between July and October and that period was defined as the drought period.

Monthly cumulative soil temperature and heat flux from 2011 to 2015 also showed the unimodal distribution; the former peaked in July and August, while the latter peaked in May and June. The annual average relative soil water content (RSWC) at 0–5, 15–20 and 40–45 cm depths was 45.8%, 82.8% and 90.5 %, respectively. The RSWC increased with depth, but the range and the standard deviation both decreased with depth. A lower RSWC usually occurred between July and October, which may be related to strong soil evaporation and vegetation transpiration. Compared with deeper depths, the RSWC within the 0–5 cm soil layer was more sensitive to changes in atmospheric environmental factors and it decreased rapidly in conditions of seasonal drought (from July to October). Monthly NDVI from 2011 to 2015 also showed a unimodal distribution, similar to cumulative soil temperature and peaked in August (average 0.81), with an annual average of 0.71.

3.2. Seasonal Variability in Lc-Excess of Precipitation and Soil Water

Since the temporal variability in δD was consistent with that in δ¹⁸O, δ¹⁸O alone was used to identify the temporal variability in the stable isotope composition of precipitation and soil water. The δ¹⁸O in precipitation varied from −21.6‰ to 1.03‰ and its annual average was −6.03‰, indicating a greater enrichment than that during the drought periods (−8.04‰) and greater depletion than that during the non-drought periods (−5.33‰) (Figure S1). The δ¹⁸O in soil water varied from −13.7‰ to −1.59‰ at 0–5 cm depths, from −14.1‰ to −1.94‰ at 15–20 cm depths and from −12.3‰ to −3.87‰ at 40–45 cm depths. The annual average δ¹⁸O of soil water at 0–5 cm depth (−6.41‰) was similar to that of precipitation and more enriched than that of soil water at 15–20 (−7.46‰) and 40–45 (−7.33‰) cm depths. Variability in δ¹⁸O in precipitation and soil water during the non-drought periods was similar to this for a whole year. During the drought periods, δ¹⁸O of soil water at 0–5 (−7.42‰) and 40–45 (−7.99‰) cm depths was the most enriched, followed by δ¹⁸O of precipitation (−8.04‰) and δ¹⁸O of soil water at 15–20 cm depth (−8.62‰) was the most depleted.

The lc-excess of precipitation varied from −12.7 to 10.6 and the averages for a whole year, the non-drought and the drought periods were 0 (Figure S2). The annual average ± SD (standard deviation) of lc-excess of soil water at 0–5, 15–20 and 40–45 cm depths was −10.15 ± 6.84, −6.84 ± 4.07 and −5.89 ± 4.42, respectively (Figure S2). Correlation coefficients between the lc-excess of the last precipitation event preceding sampling and the lc-excess of sampling the soil water within different days after a precipitation event stopped were used for analyzing the effects of precipitation infiltration on the lc-excess of soil water. However, the correlation coefficient between the same-day lc-excess of soil water at 0–5 cm depth and lc-excess of precipitation was significant (p < 0.05). Thus, precipitation can significantly influence the dynamics of the same-day lc-excess of soil water at 0–5 cm depth.

The annual average of lc-excess of sampling soil water at 0–5, 15–20 and 40–45 cm depths when there was no rain on the same day was −10.73 ± 7.00, −7.06 ± 4.10 and −5.96 ± 4.34, respectively (Figure 4). The lc-excess of soil water in each layer exhibited a noticeable seasonal variation and increased from January to June and then sharply declined from July to October (drought periods). The annual average lc-excess of soil water increased with depth, but the range and the standard deviation both decreased with depth. Correlation coefficients between lc-excess of soil water at 0–5 and 15–20 cm depths (0.73), between lc-excess of soil water at 0–5 and 40–45 cm depths (0.65) and between lc-excess of soil water at 15–20 and 40–45 cm depths (0.83) were significant and were greater than those among them and the lc-excess of the last precipitation event preceding sampling.

3.3. Influencing Factors of Temporal Variation in Lc-Excess of Soil Water

The fit of the multiple linear regression model with influencing factors over 7 days prior to the sampling day was higher than those with influencing factors over 15 and 30 days prior to the sampling day for 0–5 cm depth, while the opposite results were obtained for the 15–20 and 40–45 cm depths (

Table 1. Results of multiple linear regression models which explain lc-excess of sampling soil water at 0–5 cm depths when there was no rain on the same day, over different time periods: whole year and non-drought and drought periods.

Classify	7 Days			15 Days			30 Days
Multiple Linear Model	Whole Year (n = 335)	Non-Drought Periods (n = 193)	Drought Periods (n = 142)	Whole Year (n = 335)	Non-Drought Periods (n = 193)	Drought Periods (n = 142)	Whole Year (n = 335)	Non-Drought Periods (n = 193)	Drought Periods (n = 142)
Multiple r2 (adjusted)	0.737	0.752	0.747	0.732	0.764	0.715	0.688	0.719	0.687
RH	0.337 (20.5%)	0.233 (18.9%)	0.325 (17.2%)	0.434 (22.4%)	0.365 (24.6%)	0.448 (23%)	0.399 (17.7%)	0.310 (21.8%)	0.402 (22.2%)
WS	0.185 (11.2%)	0.096 (7.8%)	0.372 (19.7%)	0.217 (11.2%)	0.140 (9.4%)	0.463 (23.8%)	0.174 (7.7%)	n.s.	0.437 (24.2%)
RSWC 0–5	0.368 (22.3%)	0.444 (36%)	0.367 (19.4%)	0.177 (9.1%)	0.188 (12.7%)	0.282 (14.5%)	0.102 (4.5%)	0.222 (15.6%)	0.302 (16.7%)
Tss 0–5	n.s.	n.s.	0.511 (27%)	−0.237(12.3%)	−0.205 (13.8%)	0.406 (20.9%)	−0.608 (27%)	−0.285 (20.1%)	0.343 (19%)
G	0.332 (20.1%)	0.378 (30.7%)	n.s.	0.464 (24%)	0.588 (39.6%)	n.s.	0.545 (24.2%)	0.605 (42.5%)	n.s.
T	0.280 (17%)	n.s.	n.s.	0.294 (15.2%)	n.s.	n.s.	0.425 (18.9%)	n.s.	n.s.
NDVI	−0.146 (8.8%)	−0.081 (6.6%)	−0.315 (16.7%)	−0.11 (5.7%)	n.s.	−0.347 (17.8%)	n.s.	n.s.	−0.323 (17.9%)

7, 15 and 30 days mean climatic, soil and vegetation factors over the 7, 15 and 30 days prior to the sampling day were obtained, respectively. RH is relative humidity; WS is wind speed; RSWC 0–5 is relative soil water content at 0–5 cm depth; Tss 0–5 is cumulative soil temperature at 0–5 cm depth; G is cumulative soil heat flux; T is transpiration. p < 0.05 was considered statistically significant for each parameter. n.s. indicates parameters that were not significant for the model. Percentages in brackets show the relative importance of the explanatory parameter.

, Tables S1 and S2). However, the r² of the multiple linear regression models at 15–20 and 40–45 cm depths were low and 0.43–0.46 and 0.48–0.53 during a whole year, respectively. Therefore, temporal variation in the lc-excess of soil water at 0–5 cm depth was mainly affected by the influencing factors over 7 days prior to the sampling day.

Results of the multiple linear regression models at 0–5 cm depth showed that the effects of the relative soil water content at 0–5 cm depth (22.3%), relative humidity (20.5%) and cumulative soil heat flux (20.1%) were more important in explaining the temporal variation in lc-excess of soil water than transpiration (17.0%), wind speed (11.2%) and NDVI (8.8%), when all data were considered (Table 1). Soil water lc-excess during non-drought periods was explained mainly by relative soil water content at 0–5 cm depth (36.0%) and cumulative soil heat flux (30.7%), partially by relative humidity (18.9%), slightly by wind speed (7.8%) and NDVI (6.6%). During drought periods, 27% of soil water lc-excess variability could be explained by cumulative soil temperature, with wind speed, relative soil water content at 0–5 cm depth, relative humidity and NDVI accounting for 19.7%, 19.4%, 17.2% and 16.7%, respectively.

Redundancy analysis explained 73.6% (Figure 5a), 75.5% (Figure 5b) and 74.6% (Figure 5c) of variability in soil water lc-excess during a whole year and non-drought and drought periods, respectively. The positive interaction effects between climate variables, including P, RH and WS and soil variables including RSWC, T_ss and G explained mainly variability in soil water lc-excess, followed by the effects of soil variables during a whole year and non-drought periods. During drought periods, the positive interaction effects between soil and vegetation variables explained the largest fraction of variability in soil water lc-excess, followed by the interaction effects between climate and soil variables.

3.4. Soil Water Evaporation Loss Fraction

Correlation coefficients between the soil evaporation loss fraction (f) at 0–5 cm depth and the ratio of potential evapotranspiration to precipitation (PET/P) over different days prior to the sampling day were calculated for analyzing the period of soil evaporation indicated by f estimated with the lc-excess. The correlation coefficients were gradually improved with the increase in the number of days prior to the sampling day and the correlation coefficients over 30 days prior to the sampling day (0.37) was the largest (Figure 6a). The soil evaporation loss fraction varied greatly, ranging from 0.30% to 17.8% with the annual average 5.50–7.19% from 2011 to 2015 and they had a strong negative correlation with soil water lc-excess at 0–5 cm depth when there was no rain on the same day (R² = 0.96, p < 0.001) (Figure 6b).

4. Discussion

4.1. Capacity of Lc-Excess to Indicate Surface Soil Evaporation Loss

The capacity of lc-excess to indicate the soil evaporation fractionation signals can be influenced by the dilution effect on soil evaporation fractionation signals of precipitation infiltration [8,45]. In our study, precipitation significantly influenced the lc-excess of the same-day sampling soil water at 0–5 cm depth. This may be the result of recharging to soil-bound water before the formation of mobile water, or the mixture between soil bound water and mobile water due to hydrologic connectivity during precipitation infiltration (Figure 1) [8,45]. Therefore, precipitation on the sampling day with an unfractionated signal (lc-excess = 0) diluted the evaporation fractionation signal (lc-excess < 0) of soil water at 0–5 cm depth due to the mixing between event precipitation and pre-event soil water and affected the capacity of the lc-excess to indicate soil evaporation fractionation signals [19]. In addition, the largest soil water lc-excess in July with the highest air temperature and high precipitation also showed dependency of the soil water fractionation signals on precipitation infiltration (Figure 4).

The capacity of lc-excess to indicate subsoil evaporation fractionation signals may be influenced by percolation of the upper soil water with high evaporation fractionation signals [9,46,47]. In our study, lc-excess of soil water at 15–20 and 40–45 cm depths can be influenced by water from the upper soil layers with high soil evaporation fractionation signals. The low r² of the multiple linear regression models for 15–20 and 40–45 cm depths (Tables S1 and S2) illustrated that the climate, soil and vegetation variables only partly explained the temporal variation in the lc-excess of soil water at 15–20 and 40–45 cm depths. This may be due to percolation into the subsoil of soil water from shallow horizons with certain evaporative fractionation signals by hydrologic connectivity [13].

The largest and positive correlation between the soil evaporation loss fraction (f) at 0–5 cm depth and PET/P over 30 days prior to the sampling day and the strong negative correlation between the f and lc-excess of soil water at 0–5 cm depth (Figure 6) illustrated that the lc-excess can indicate the integrative soil evaporation loss over 30 days prior to the sampling day. Our results showed that the estimated average f at 0–5 cm depth was 6.25%, indicating that soil water evaporation was a small fraction of the total precipitation. The average ratio of soil evaporation (E) to evapotranspiration (ET) was 13.0% (E/ET = f × P/ET) and was consistent with the existing result of Wei et al. [48], who reported that E/ET ratio was 12.0% in this study area with the verification and improvement to the evapotranspiration (ET) model (PT-Fi model). Above all, our results suggest the lc-excess of soil water at 0–5 cm depth presents a reliable f estimate over 30 days prior to the sampling day and emphasizes, when studying soil evaporation loss dynamics in this area, the importance of sampling shallow soil layers when there was no rain on the same day.

4.2. Interaction Effects of Climate, Soil and Vegetation on Temporal Variation in Lc-Excess

Soil water lc-excess at a 0–5 cm depth exhibited an increase and then decline over the course of a year (Figure 4) and this trend was dominated by relative soil water content, relative humidity and cumulative soil heat flux, followed by transpiration, wind speed and NDVI (Table 1). Soil water lc-excess at a 0–5 cm depth during non-drought periods was dominated by relative soil water content and cumulative soil heat flux, followed by relative humidity, wind speed and NDVI (Table 1). During drought periods, the soil water lc-excess at 0–5 cm depth was mainly controlled by cumulative soil temperature, followed by wind speed, relative soil water content, relative humidity and NDVI (Table 1).

Climate variables, including relative humidity and wind speed reflecting a vapor gradient between soil and the atmosphere explained partially temporal variability in soil water lc-excess during the three periods (Table 1, [49]). The increase in relative humidity increased the lc-excess by decreasing the atmospheric demand of soil evaporation. The increase in wind speed caused a decrease in lc-excess by increasing the atmospheric demand of soil evaporation. However, high wind speed during drought periods increased warm and moist air and decreased the atmospheric demand of soil evaporation due to the East Asian monsoon climate [32]. The positive relationship between wind speed and soil water lc-excess in the multiple linear regression models may be due to a greater effect on the lc-excess of wind speed decreasing atmospheric demand (

Table 2. Pearson correlation coefficients (r) among lc-excess of sampling soil water at 0–5 cm depth when there was no rain on the same day, climate, soil and plant properties over 7 days prior to the sampling day.

R	lc-Excess	P	RH	WS	RSWC 0–5	Tss	G	T	NDVI
7 Days (Whole Year)
lc-excess	1.000	0.298 **	0.414 **	0.307 **	0.656 **	0.332 **	0.665 **	0.386 **	0.165 **
P	0.298 **	1.000	0.339 **	0.080	0.360 **	0.140 *	0.167 **	0.047	0.149 **
RH	0.414 **	0.339 **	1.000	−0.321 **	0.521 **	−0.023	0.065	−0.267 **	0.017
WS	0.307 **	0.080	−0.321 **	1.000	0.047	0.267 **	0.305 **	0.468 **	0.131 *
RSWC 0–5	0.656 **	0.360 **	0.521 **	0.047	1.000	−0.143 **	0.333 **	−0.079	−0.110 *
Tss	0.332 **	0.140 *	−0.023	0.267 **	−0.143 **	1.000	0.688 **	0.851 **	0.813 **
G	0.665 **	0.167 **	0.065	0.305 **	0.333 **	0.688 **	1.000	0.697 **	0.432 **
T	0.386 **	0.047	−0.267 **	0.468 **	−0.079	0.851 **	0.697 **	1.000	0.635 **
NDVI	0.165 **	0.149 **	0.017	0.131 *	−0.110 *	0.813 **	0.432 **	0.635 **	1.000
7 Days (Non-Drought Periods)
lc-excess	1.000	0.325 **	0.524 **	0.284 **	0.817 **	0.494 **	0.711 **	0.487 **	0.267 **
P	0.325 **	1.000	0.415 **	0.073	0.446 **	0.390 **	0.257 **	0.267 **	0.336 **
RH	0.524 **	0.415 **	1.000	−0.131	0.528 **	0.277 **	0.222 **	−0.028	0.183 *
WS	0.284 **	0.073	−0.131	1.000	0.300 **	0.128	0.228 **	0.334 **	0.012
RSWC 0–5	0.817 **	0.446 **	0.528 **	0.300 **	1.000	0.473 **	0.659 **	0.481 **	0.339 **
Tss	0.494 **	0.390 **	0.277 **	0.128	0.473 **	1.000	0.731 **	0.833 **	0.740 **
G	0.711 **	0.257 **	0.222 **	0.228 **	0.659 **	0.731 **	1.000	0.744 **	0.407 **
T	0.487 **	0.267 **	−0.028	0.334 **	0.481 **	0.833 **	0.744 **	1.000	0.614 **
NDVI	0.267 **	0.336 **	0.183 *	0.012	0.339 **	0.740 **	0.407 **	0.614 **	1.000
7 Days (Drought Periods)
lc-excess	1.000	0.282 **	0.199 *	0.388 **	0.642 **	0.648 **	0.678 **	0.625 **	0.065
P	0.282 **	1.000	0.243 **	0.116	0.335 **	0.163	0.127	0.015	0.113
RH	0.199 *	0.243 **	1.000	−0.506 **	0.385 **	0.024	−0.090	−0.336 **	0.288 **
WS	0.388 **	0.116	−0.506 **	1.000	0.033	0.386 **	0.380 **	0.557 **	0.091
RSWC 0–5	0.642 **	0.335 **	0.385 **	0.033	1.000	0.328 **	0.375 **	0.418 **	0.094
Tss	0.648 **	0.163	0.024	0.386 **	0.328 **	1.000	0.835 **	0.758 **	0.427 **
G	0.678 **	0.127	−0.090	0.380 **	0.375 **	0.835 **	1.000	0.776 **	0.055
T	0.625 **	0.015	−0.336 **	0.557 **	0.418 **	0.758 **	0.776 **	1.000	0.125
NDVI	0.065	0.113	0.288 **	0.091	0.094	0.427 **	0.055	0.125	1.000

P is precipitation; RH is relative humidity; WS is wind speed; RSWC 0–5 is relative soil water content at 0–5 cm depths; Tss is cumulative soil temperature at 0–5 cm depth; G is cumulative soil heat flux; T is transpiration. ** and * represent a significant relationship at p = 0.01 and 0.05 levels, respectively.

).

RSWC reflected soil residual water storage, decreased with soil evaporation, transpiration and percolation and played a main role in the temporal variation of soil water lc-excess over the whole year and during non-drought periods [23]. On the one hand, as residual water storage decreased, soil tension increased the equilibrium fractionation and resulted in a decrease in soil water lc-excess [25]. On the other hand, water used for soil evaporation may be decreased due to the increase in transpiration or percolation. Cumulative soil temperature was the primary influencing factor of soil water lc-excess and was positively correlated with lc-excess during drought periods (Table 1). On the one hand, high cumulative soil temperature increased vegetation transpiration (Table 2) and lead to the decline of RSWC [12,26], which decreased the soil water supply of soil evaporation, resulting in an increase in soil water lc-excess. On the other hand, the increasing vegetation transpiration due to high cumulative soil temperature increased the relative humidity (Table 2), which decreased the atmospheric demand of soil evaporation, resulting in an increase in soil water lc-excess [50]. The effects of cumulative soil heat flux on soil water lc-excess was similar to the cumulative soil temperature.

On the one hand, the shade provided by vegetation can generate a microclimate with higher relative humidity (Table 2) and lower radiation input, resulting in an increase in soil water lc-excess [26]. On the other hand, vegetation can enhance the evaporation of canopy-intercepted precipitation and decrease the lc-excess of throughfall [21]. According to multiple linear regression models, NDVI partially explained the temporal variation in soil water lc-excess and the two exhibited a negative relationship (Table 1). We propose that the positive effects of vegetation on lc-excess were explained by transpiration (Table 1 and Table 2), while the negative effects of vegetation on lc-excess can be explained by NDVI, which can decrease lc-excess of throughfall by enhancing the evaporation of canopy-intercepted precipitation when the temperature is fixed [21,22]. Therefore, high transpiration may limit soil evaporation loss during drought periods, while the competition between transpiration and soil evaporation was small during non-drought periods due to there being enough water supply.

5. Conclusions

The results of this study indicate that precipitation events with an unfractionated stable isotope signal on the sampling days diluted the evaporation fractionation signal of soil water and affected the capacity of the lc-excess to indicate the soil evaporation fractionation signal. Percolation of the upper soil water with high evaporation fractionation affected the capacity of the lc-excess to indicate subsoil evaporation fractionation signal by hydrologic connectivity. Lc-excess of soil water at 0–5 cm depth indicated a reliable soil evaporation loss estimate over 30 days prior to the sampling day. Therefore, sampling the upper-most soil layer when there was no rain on the same day was important for studying soil evaporation loss dynamics in this forest ecosystem.

Soil water lc-excess at 0–5 cm depth was mainly driven by relative soil water content during a whole year and non-drought periods and by cumulative soil temperature during drought periods. RSWC decreased with soil evaporation and was affected by precipitation infiltration, transpiration and percolation. In this forest ecosystem, the high cumulative soil temperature increased vegetation transpiration and decreased the soil water supply of soil evaporation, meanwhile the increasing vegetation transpiration increased relative humidity and decreased the atmospheric demand of soil evaporation, resulting in an increase in the soil water lc-excess. Our study highlighted that vegetation was important for surface soil evaporation loss dynamics in this forest ecosystem and high transpiration may limit soil evaporation loss during drought periods.