HOT SPOTS OF SOIL RESPIRATION IN A SEASONALLY DRY TROPICAL FOREST IN SOUTHERN VIETNAM: A BRIEF STUDY OF SPATIAL DISTRIBUTION

. Many studies report asymmetrical spatial distribution of soil respiration caused by presence of areas with significantly higher emission rates (so-called hot spots ). For seasonally dry tropical forest soil respiration was measured on 1 ha plot with 20m, 5m and 1 m scale in the first half of dry season. 457 measurements made in 9 series at 54 sampling points. The results suggest that lognormal spatial distribution model appears to be much more supported rather than the normal one. A statistical method proposed for estimation the mean value and its confidence interval of lognormally distributed data. The mean emission rate E( R S ) for the lognormal distribution amounted to 4.28 µmol m -2 s -1 , the 95% confidence interval is 3.93 to 4.76 µmol m -2 s -1 . However, the standard sample mean can be used as an estimator of the mean of lognormally distributed values of soil respiration if their coefficient of variance remains approximately the same as in our study (CV=0.35). Based on the data obtained and literature sources, recommendations are given on the number of sampling points for estimating the spatial average value with a given accuracy spots of soil respiration in a seasonally dry tropical forest in southern Vietnam: a brief study of spatial distribution.


INTRODUCTION
Tropical forests contain about 40% of the global vegetation carbon and are responsible for about 50% of terrestrial gross prima-ry production. Owing to their large C stocks and budgets, tropical forests can affect the global C balance, and hence potentially play an important role in climate change, despite only covering 12% of the total land surface (FAO 2001). Seasonal forests represent 43% of the total tropical forest area, which is comparable with that of tropical rainforests.
In tropical forest ecosystems, the eddy covariance method has improved our understanding of ecosystem carbon cycle processes (e.g. Yamamoto et al. 2005;Kumagai et al. 2006;Kosugi et al. 2008). However, this method can provide unreliable nocturnal CO 2 fluxes under low wind speed conditions (Baldocchi 2003), which can be used as a proxy for ecosystem respiration; thus, alternative chamber methods for measuring respiration are still important for understanding the processes of carbon flow in relation to climate change (Kume at al. 2013).
Soil CO 2 efflux (syn. soil respiration, R s ) is the sum of multiple processes in rhizosphere, bacteria and fungi, soil fauna, litter decomposing organisms respiration and other, and many factors affect each of these components. Although soil temperature and soil water content are considered as primary drivers of soil respiration, coincidence of secondary factors (such as root biomass, organic carbon content, insect nests (Lopes de Gerenyu et al. 2015)) can cause emission values considerably exceeding surrounding background, which makes soil respiration a process of high spatial and temporal variability. Areas with significantly higher emission rates are generally recognized as hot spots (McClain et al. 2003). Sever-al problems may appear when estimating spatial average value of soil respiration. Due to insufficient sampling, the hot spots can be not taken into account at all, or they could be treated as a measurement artifact and discarded.
Soil respiration hot spots can be considered as ecosystem control points -points (spatial or temporal) where the exchange rate is of sufficient magnitude or ubiquity to affect dynamics of the ecosystem (Bernhardt 2017). With this approach, investigating the properties of soil respiraton hotspots can lead to a better understanding of ecosystem processes.
Aim of this study was evaluation of spatial variability of soil CO 2 efflux in seasonally dry tropical forest, describing magnitude, size and temporal behavior of hotspots and estimation of optimum amount of sampling for further studies.

STUDY AREA
The study has been carried out in semi-deciduous tropical forest of Cat Tien national park (a part of the Dong Nai Biosphere Reserve) located in Dong Nai province, Vietnam (Fig. 1) (Kuznetsov and Kuznrtsova 2011). Forest understory is abundant and evergreen, with minor grass cover. Soil of the study site is very dark brown (7.5YR 2.5/2 by Munsell color system) andisol with negligible color variation among study area. Organic carbon content in upper layer (0-20 cm) is 5.34%, nitrogen content is 0.44%, pH is 5.56 (Okolelova et al. 2014). The measurements of soil respiration were repeated in 9 series with 6-8 days interval during December 2017 to February 2018. Each of the series has been measured in two days: on first day each of 25 points of the 100x100m plot has been measured and rest of the points measured the next day. The measurements of each day were taken between 9 am and 12 pm, one time per measuring point. At each sampling point a PVC collar of 250mm diameter was installed with depth of about 3-5 cm below ground and height of 5 cm above, tree litter was not removed from the collars. First measurements were taken 10 days after the collars installation. Soil respiration was measured with non-steady state closed loop chamber method using custom-made respirometer based on LI-820 infrared gas analyzer (LI-COR inc., USA). The measuring chamber ( Fig. 3) was of hemispherical shape 2.8 l in volume. Air intake was mounted at the top of the hemisphere, and the air return flow was distributed around the base perimeter, which made it possible to avoid the use of a fan and the consequent excessive turbulence inside the chamber. Air-tight connection to the collar was realized using a thin rubber sealing ring. CO 2 concentration values were logged with 1 Hz rate during each 120 seconds-long measurement.

Sampling design
Air temperature inside the measuring chamber was registered for every CO 2 efflux measurement to estimate molar volume. Soil temperature and volumetric water content were measured along with CO 2 efflux measurement in close proximity to the chamber. Soil temperature was measured at 2-5 cm depth using Checktemp-1 digital thermometer (Hanna instruments, USA). Soil volumetric water content was measured using electrical conductivity meter 5TE (Decagon Devices, Inc., USA). Due to the sensor failure, soil water content was measured only during first 4 series.
The method for calculation of CO 2 efflux rate was similar to one used in commercial soil gas flux system LI-8100A (LI-COR inc., USA). CO 2 concentration values from a first few seconds of a measurement were discarded due to disturbances caused by installation of the chamber. However, CO 2 growth rate was estimated at the moment of chamber installation by approximation of good data with a gradient diffusion equation (Healy et al. 1996) using Table-Curve2D software: when the chamber installed, and Cx' is a parameter that defines the asymptote, all in μmol CO 2 /mol air (ppm); a is a parameter that defines the curvature of the fit (s -1 ). The Cx', a and t 0 parameters yielded by the regression were used to estimate CO 2 concentration growth rate at the moment t = t 0 : Then CO 2 efflux rate was calculated: Where R s is soil respiration rate (μmol * m -2 * s -1 ), Vch is the volume of measuring chamber(m 3 ), Sch is the area covered with the chamber (m 2 ) and Vm is the molar volume for temperature measured inside the chamber (m 3 ).

Statistical methods
To make distribution of R s values from each measurement series comparable, the temporal trend was removed: we calculated the median value in each trial (M i ,i=1,...,9), and the median of all values in the dataset (M all ). Then, for each trial (i=1,...,9) the detrended data were calculated as Hot spots were defined as outliers by Tukey's criteria (Tukey 1977) as it was done by other authors for defining geochemical phenomena (Kessel et al. 1993;Ohashi 2007 emission rates at different spatial locations are independent and follow lognormal distribution, the mean emission rate (E(R S )) was estimated using the method by Shen, Brown and zhi (Shen et al. 2006): where X i =LogRs i is log-transformed data, is the sample mean of X, is the sample sum of squared deviations of x, and n is the size of the sample. The bootstrap confidence interval (CI) for E(R S ) was generated with 10 5 simulated samples.
To estimate the number of test sites required to measure the mean CO 2 emission rate with given precision, we studied the properties of the bootstrap CI by Shen et al. (2006).
If R S is a lognormally distributed random variable, x=logR S is normally distributed with mean μ and standard deviation (SD) σ. The dimensionless parameter σ is related to the coefficient of variation of R S : where CV=SD(R S )/E(R S ) is the coefficient of variation of z.
From CI formulas in (Shen et.al. 2006) it can be shown that the relative width of the CI [c 1 ,c 2 ] for E(R S ) is a function of σ and n only. At the same time, RWCI may be considered as a measure of relative precision of estimation of E(R S ). Thus, if and target RWCI are given, it is possible to estimate the required sample size n.

RESULTS AND DISCUSSION
The natural distribution of detrended soil respiration values displayed an asymmetrical distribution with the tail of higher values (Fig. 4). The obtained normality tests' p-values for time-averaged data are 0.1859, 0.0046, 0.6126, <0.001 for the original data and >0.5, 0.5029, 0.9345, >0.5 for the log-transformed data (L, AD, KS, and JB tests respectively). Although only two tests reject normality of the original data at 5% significance level, lognormality of the data appears to be much more supported.
The mean emission rate E(R S ) for the lognormal distribution amounted to 4.28 µmol m-2 s-1, the 95% confidence interval is 3.93 to 4.76 µmol m -2 s -1 . The relative width of the confidence interval (RWCI) is 0.195. The median rate amounted to 3.95 µmol m -2 s -1 which is 12% lower than estimated mean value. The standard sample average, which suggests normal distribution, amounted to 4.31 µmol m-2 s-1 , with confidence interval in 3.88 -4.74 µmol m-2 s-1 and RWCI of 0.20, which is quite close to the results of lognormal distribution Thus, we may state that the standard sample mean can be used as an estimator of the mean of lognormally distributed values of soil respiration if their coefficient of variance remains approximately the same as in our study (CV=0.35). However, in the case of much more expressed hotspots (i.g. as described for methane emissions, denitrification and other processes by Bernhardt (2017)), these estimates may differ significantly, and the use of our proposed method for estimating the average of the lognormal distribution will be reasonable.
The results of the individual measurement series are shown in Table A.1.
There was no significant overall trend in R S during the measurement campaign (slope <0.03 µmol m -2 s -1 per day). Some hot spots were stable in time, while some appeared only once or twice, and some points oppositely reacted to the same change in conditions. Our data shown no statistical dependence of soil respiration on temperature (linear model R 2 < 0.006, p-value: 0.11) Vitaly K. Avilov, Dmitry G. ivanov et al. HOT SPOTS OF SOiL RESPiRATiON iN ... E(R S ) = exp X + (n 1)S 2 2(n + 4)(n 1) + 3S 2 The spatial dependence analysis didn't show any systematic behavior described by theoretical models (linear, spherical, exponential, Gaussian, and power law). This can serve as the evidence of heterogenous nature of time-averaged CO 2 efflux values or a hypothesis that the scale of spatial variation of soil CO 2 efflux at our site was smaller than double minimal distance between sampling points, i.e. less than 2 meters. Two-sample t-test showed statistically different mean values of two most expressed hotspots with any surrounded sampling point within 2 meter radius.
The time-averaged soil respiration values obtaineded at 54 sampling points produce standard deviation of log-transformed values σ=0.3433 (with corresponding 95% confidence interval CV(Z)=0.3537). In order to estimate the possible range of σ, we fetched data from four other studies of tropical forests in Southeast Asia (Ohashi 2007;Katayama 2009;Kosugi 2007;Adachi 2009) in the form of sample means and sample SDs (20 sets of data in total, with varying conditions and sample sizes). The data were used to calculate rough estimations of CVs (because sample means and sample SDs are very biased estimators of true means and SDs of lognormal distributions), and, in turn, to calculate rough estimations of . The minimal observed σ was 0.2576, the maximum one was 0.5847, with a peak of frequency near to our σ=0.3433. So we decided to explore the range σ Є [0.15; 0.75].
The calculated RWCIs for various and n are shown in Fig. 5. Every value was calculated using 105 simulated samples in the bootstrapping procedure. The graphs may be used for estimating the required sample size if target RWCI is known and is estimated in a pilot study or assumed from literature data.
For σ=0.35, which can be considered as typical, some RCWI values were tabulated (Table 1): These values clearly show that sampling design which includes 5-10 spatial probes may produce estimates very far from the true mean CO 2 emission rates.

02|2019 CONCLUSIONS
Soil respiration rate in a seasonally dry tropical forest is highly variable in space, due to the small size of zones with significantly higher rates (hot spots). Spatial distribution of this parameter rejects normality tests, and lognormal distribution is much more supported by the statistical tests. However, we may state that the standard sample mean can be used as an estimator of the mean of lognormally distributed values of soil respiration if their coefficient of variance remains approximately the same as in our study (CV=0.35).
Hotspots apparently have a heterogenous nature, some of them are more stable than another ones. The stable hot spots can be explained by higher root biomass and leaf area index (LAI), the presence of decaying residues or areas of higher soil gas permeability. Rarely distributed short-term hot spots can be caused by insect activity. The detected hotspots apparently have a size of no more than 2 meters.
Assuming the lognormal nature of the distribution of soil respiration and the variability observed in our study and other literature sources, it may be proposed to make at least 20 spatially independent samples to obtain an estimate of the soil respiration with relative width of 95% confidence interval of no more than 0.35 of the estimated mean.

ACkNOwLEDGEMENTS
The study was funded by RFBR and Russian Geographical Society according to the research project № 17-05-41127 and partially was supported the Presidium of the Russian academy of sciences, Programs № 51 «Climate change: causes, risks, consequences, problems of adaptation and regulation» and № 41 "Biodiversity of natural systems and biological resources of Russia"