Appropriate chamber deployment time for separate quantification of CH 4 emissions via plant and ebullition from rice paddies using a modified closed-chamber method

A modified closed-chamber method for estimating total, plant-mediated, and bubbling ( ebullition ) emissions of CH 4 from rice paddies has been developed to use high-time-resolution CH 4 concentration data ( ~1 Hz ) obtained by a spectroscopic mobile gas analyzer. Here we aimed at determining an appropriate minimum time length of chamber closure for accurate flux measurement by investigating 3255 datasets obtained from a 2-year field survey. To investigate the minimum time length for each chamber measurement, we generated a series of datasets from each measurement: by setting the hypothetical termination time of the chamber closure ahead in 1-min intervals, we obtained various chamber CH 4 concentration time series with different durations of chamber closure, and separately estimated CH 4 emissions via rice plants and bubbling from each. The estimated flux was sensitive to time length with short closure times, but became less sensitive with longer closure. We defined the minimum time length at which the difference in estimated flux between adjacent time windows was small enough ( <10 % of plant-mediated emission ) . The estimated minimum time length differed from one measurement to another, but 10 min was sufficient for >99 % of cases. Detailed analysis showed a positive correlation between minimum time length and frequency of bubbling events; the time length needed to be longer as bubbling events became more frequent. From this relationship, we computed the appropriate chamber-duration time as a function of bubbling frequency. In the absence of ebullition, 4 - 5 min was sufficient, but as the bubbling frequency increased to 2.5 times per minute 15 - 20 min was necessary for accurate pathway-dependent flux measurements.


Introduction
Rice paddies are an important source of atmospheric CH 4 , a highly potent greenhouse gas Ciais et al., 2013 . Most methane produced in rice paddies is transported to the atmosphere either though aerenchymous tissues of rice plants or by upward migration of gas bubbles, i.e. ebullition Komiya et al., 2020 . Recently, we developed a protocol for separately determining plant-mediated flux and bubbling flux from high-time-resolution CH 4 concentration [ CH 4 ] data ~1 Hz obtained by a modified closed-chamber method in combination with a portable spectroscopic gas analyzer Kajiura and Tokida, 2021;Tokida, 2021 . The separate quantifications revealed that plant-mediated emission and ebullition had different sensitivities to temperature and rice growth stages, clearly demonstrating the need to determine CH 4 fluxes via each emission pathway for accurate observation and modeling Kajiura and Tokida, 2021 . High-time-resolution and precise determination of chamber [ CH 4 ] may also enable much faster flux determination than the conventional closed-chamber protocols, in which > 60 of studies waited for ≥ 30 min to manually take 3 or 4 gas samples Sander and Wassmann, 2014 . However, information about appropriate chamber deployment time for accurate CH 4 flux measurement is lacking Minamikawa et al., 2015 . Here, we assessed the minimum chamber deployment time necessary for accurate flux estimation by analyzing 3255 datasets of chamber [ CH 4 ] obtained from rice paddies during 2 years of field surveys.

Dataset 2.1.1. Experimental field and rice cultivation
We conducted field experiments in 2019 and 2020 in rice paddies at the National Agriculture and Food Research Organization NARO in Japan 36 01′28″N, 140 06′30″E . In both years we planted various rice genotypes, including members of the World Rice Core Collection Kojima et al., 2005;Tanaka et al., 2020 and chromosome segment substitution lines CSSL in an elite Koshihikari background Fukuoka et al., 2010;Kobayashi et al., 2018;Takai et al., 2014 . Each plot consisted of 24 hills 1.2 m 0.9 m of one genotype, replicated three to eight times depending on the genotype. In total, we prepared 765 plots in 2019 and 448 plots in 2020. Genotypic differences in CH 4 emissions have been detected, but this topic is beyond the scope of this study; here we used the whole dataset to assess appropriate chamber closure time. We sowed pre-germinated rice seeds in seedling trays in April or May , raised the seedlings in the open field, and then transplanted them at the 5-leaf age at a spacing of 15 cm 30 cm with three seedlings per hill late May to early June . The field was continuously flooded from transplanting until the middle of the grain-filling stage.

Measurement of CH 4 emission
Methane emission was measured by a modified closed-chamber method in which high-time-resolution [ CH 4 ] at intervals of 0.9 s was measured with a mobile gas analyzer G4301, Picarro Inc., CA, USA Kajiura and Tokida, 2021;Tokida, 2021 . The chamber air was circulated through the gas analyzer and back into the chamber at a rate of 1.0 L min -1 . We used an acrylic closed-top chamber basal area of 30 cm 60 cm to enclose four hills of rice plants at the center of each plot. Measurements were made at the panicle formation PF , booting BT , and heading HD stages to cover important CH 4 -emitting periods Tokida et al., 2014 . In total, 3255 chamber measurements were conducted in the daytime during the two years. Development stages varied among genotypes, but plants headed from late July to mid August in most genotypes. At the PF stage, a 60-cm-tall chamber equipped with a fan was used, but at the BT and HD stages, the height of the chamber was increased by adding bottom chambers with 3-cm chamber collar to make double-deck 120 cm or triple-deck 140 cm chambers. We set the chamber a few minutes before each measurement and kept the chamber closed until we obtained a steady [ CH 4 ] increase, not interrupted by a bubbling event, for at least 40 s. Therefore, the actual chamber deployment time increased as bubbling events increased.

Analysis
We first estimated minimum time length TL min necessary to accurately determine each type of flux total flux, F total ; plant-mediated flux, F plant ; bubbling flux, F bubble for each single chamber measurement section 2.2.1 . From the estimated TL min , we calculated the success rate i.e., percentage of accurate flux estimates as a function of chamber deployment time section 2.2.2 . Next, we estimated the frequency of bubbling events section 2.2.3 and investigated its relationship with TL min . From the relationship, we proposed an appropriate chamber deployment time as a function of bubbling frequency section 2.2.4 .

Estimation of the minimum time length for each chamber measurement
We prepared a series of datasets from each single chamber measurement after removing unstable periods gray areas in Fig. 1 , bringing the end point of the chamber closure ahead in intervals of 1 min to create [ CH 4 ] time series with different chamber closure times 0 -1, 0 -2, 0 -3, 0 -4; TL1 -TL4; Fig. 1 . The datasets unlikely included the artificial bubbling stimulated by the chamber deployment because bubbling frequency was not reduced as measurement time progressed. Using each [ CH 4 ] time series, we estimated F total , F plant , and F bubble as in Kajiura and Tokida 2021 with slight modifications Fig. S1 . In brief, F plant , corresponding to a period of steady [ CH 4 ] increase, was determined as the lowest flux intensity showing local maxima in the frequency distribution of the CH 4 flux. F total was estimated from the slope of the linear regression between [ CH 4 ] times series and elapsed time. Finally, F bubble , which was reflected as a "jump" in the [ CH 4 ] time series, was calculated as F total F plant .
The estimated flux generally fluctuated with short time lengths TLs , but the dependency on TL became small with longer TL. Therefore, we defined TL min , the minimum time-length necessary for accurate flux estimation, for each chamber measurement, as the TL at which the estimated flux did not differ substantially from that of the next TL 1 min longer , and judged whether the difference in the flux ∆F i = F i +1 Fig. 1. How to determine TL min minimum time length needed for each flux component F total , F plant and F bubble .
Step 1: Preparation of a set of [ CH 4 ] time series with different TLs TL1 -TL4 .
Step 2: Flux F estimation for each dataset and calculation of the difference in F ∆F between adjacent TLs ∆F i = F i +1 F i . The superscript denotes the TL of the corresponding dataset, and the subscript indicates the type of flux. A grey black "∆F" indicates that ∆F was smaller greater than the threshold 0.30 mg-C m -2 h -1 .
Step 3: Selection of TL min for each type of flux TL at which ∆F is smaller than the threshold . In this example, TL min is the same for F total and F bubble 3 min , but shorter for F plant 2 min .
F i was smaller than a threshold value Fig. 1 . Clearly, a tradeoff exists between the accuracy of the flux estimation and the necessary chamber deployment time: the smaller the ∆F threshold, the longer the TL min would be. We used a moderate threshold 0.30 mg-C m -2 h -1 based on comparisons of fluxes estimated using different thresholds see Fig. S2 . Note that TL min is determined in increments of 1 min.

Success rate as a function of time length
For each chamber measurement and each flux type 2.2.1 , we calculated the percentage of accurate flux estimations success rate as a function of TL Eq. 1 : Success rate TL = n Success TL / n Total -n Dropped TL 1 where n Success TL is the number of successful chamber measurements at TL, i.e., TL > TL min , n Total is the number of total measurements, i.e., 3255. In some measurements, ∆F was greater than the threshold throughout the actual chamber deployment time TL measured , and hence, TL min was not obtained. We dropped the number of such measurements in the Success-rate calculation when TL > TL measured n Dropped TL because judgment whether the measurement was successful or not cannot be made.

Frequency of bubbling events
Bubbling emissions, identified as jumps in the chamber [ CH 4 ] time series, can be detected as relatively large local maxima when [ CH 4 ] is converted into flux Fig. S3 . We identified the local maxima including small one using the peaks function in the splus2R package for R software Constantine and Hesterberg, 2021; R Core Team, 2020 : when the peak heights i.e., the differences between the local maxima and the baseline flux Fig. S3b, red line were > 2 mg-C m -2 h -1 25th percentile of observed peak heights of all data , we defined them as bubbles Fig. S3b, red dots ; small bubbling emissions with peak heights of < 2 mg-C m -2 h -1 were ignored. Then we calculated the frequency of bubbling events by dividing the number of events by TL measured .

Appropriate chamber-deployment time length
The estimated TL min for each chamber measurement showed a strong correlation with the frequency of bubbling events see Results and Discussion . Therefore, the appropriate chamber deployment time for accurate flux estimation may be better defined as a function of bubbling frequency rather than as a constant value. We therefore conducted linear regression analysis of the relationship between TL min and bubbling frequency for each flux type F total , F plant , and F bubble . To achieve homoscedasticity, the data were log-transformed log TL min , log bubbling frequency + 1 , and we obtained 95 and 99 prediction intervals by using the predict function in the stats package in R. The upper limits of the intervals as a function of bubbling frequency can be used as the appropriate time length for accurate flux estimation TL ap . Finally, a common unstable period time-length median value of the unstable period of all data was added to yield the appropriate chamber deployment time-length Chamber-TL ap .

Results and Discussion
A series of datasets generated by sequential earlier termination of chamber closure showed that the estimated fluxes were sensitive to TL with short chamber closure, but became less sensitive with longer TL Fig. S4 . TL min , the time length at which estimated flux stabilized, differed from one chamber measurement to another, but it was < 10 min in most cases Fig. 2 . Consequently, the success rate of accurate flux estimation increased with increasing TL and reached 99.7 -100 at 10 min Fig. 3 . F total and F bubble required longer TL min than F plant ; for example, 8 min was necessary for F total and F bubble to reach 99 success rate, while 4 min was enough for F plant .
Estimated TL min showed a strong dependency on bubbling frequency for F total and F bubble Fig. 4 , presumably because longer TL may be necessary for the slope of the linear regression to stabilize for their estimation Fig. S5 . For F plant , on the other hand, the dependency of TL min on bubbling frequency was much weaker although still statistically significant Fig. 4 . Longer time-period may be necessary to correctly capture the baseline in the time-series flux data i.e., F plant when bubbling emissions occurred frequently. Taking longer time would not underestimate F plant Wassmann et al., 2018 because an increase in chamber [ CH 4 ] within a few hours would hardly affect [ CH 4 ] gradient between the soil and the atmosphere, considering very high [ CH 4 ]

Total Plants Bubbles
found in typical paddy soils > 10 in trapped gas phase Tokida et al., 2013 . On the basis of the dependency of TL min on bubbling frequency Fig. 4 , we propose the appropriate chamber deployment time length Chamber-TL ap , including typical unstable period of 1 min to be 4 -15 min by 95 and 5 -20 min by 99 prediction intervals, depending on bubbling frequency Table 1 . In the absence of a bubbling event, a short chamber closure time e.g., 4 min was sufficient, but if bubbling emission occurred frequently e.g., 2.5 times per minute , a much longer deployment time e.g., 15 min was necessary for accurate flux measurements. In our observations, bubbling events were rare < 0.5 min -1 in many cases during the PF stage Fig. S6 ; hence, Chamber-TL ap can be ≤ 6 min Fig. 2, Table 1 . However, they were much more frequent at the HD stage, as reported previously Tokida et al., 2013;Wassmann et al., 1996 , and longer Chamber-TL ap may be necessary Figs. 2, S6 . The appropriate chamber deployment time length may vary with other factors because bubbling frequency explains less than half of TL min variability Fig. 4 . Potential factors include the intensity of bubbling emissions. If a large bubbling emission occurred, the regression lines to estimate F total would not stabilize for a long time, leading to a longer Chamber-TL ap . Therefore, the results summarized in Table 1 might not be directly applicable to other fields where bubbling intensity is substantially different. Nevertheless, given the large dataset used in this study, Chamber-TL ap proposed here can be used as a guideline. At the same time, users of the new closed-chamber method should assess site-specific chamber deployment time for themselves, as we have done here.

Conclusion
We assessed the appropriate chamber-deployment time length for accurate estimation of F total , F plant , and F bubble from rice paddies by using 3255 high-time-resolution [ CH 4 ] datasets 0.9 s interval obtained during two years of field measurements. The appropriate time length depended on the frequency of bubbling events: 5 min was enough in the absence of ebullition, and at most 20 min was necessary when bubbling occurred frequently. 2.0 2.5 2.5 3.3 4 5 0.5 4.6 6.0 2.6 3.2 4.2 5.5 6 7 1.0 6.7 8.8 3.2 3.9 6.1 7.9 8 10 1.5 9.0 11.8 3.  Organization NEDO . The authors declare no conflicts of interest associated with this manuscript.