Parameterization of G-93 isoprene emission formula for tropical trees Casuarina equisetifolia and Ficus septica

Parameterization of G-93 isoprene emission formula for tropical trees Casuarina equisetifolia and Ficus septica

Atmospheric Environment 141 (2016) 287e296 Contents lists available at ScienceDirect Atmospheric Environment journal homepage: www.elsevier.com/loca...

2MB Sizes 3 Downloads 145 Views

Atmospheric Environment 141 (2016) 287e296

Contents lists available at ScienceDirect

Atmospheric Environment journal homepage: www.elsevier.com/locate/atmosenv

Parameterization of G-93 isoprene emission formula for tropical trees Casuarina equisetifolia and Ficus septica Ishmael Mutanda a, b, Masashi Inafuku b, Hironori Iwasaki b, Seikoh Saitoh b, Masakazu Fukuta c, Keiichi Watanabe d, Hirosuke Oku b, * a

United Graduate School of Agricultural Sciences, Kagoshima University, Korimoto 1-21-24, Kagoshima, 890-0065, Japan Molecular Biotechnology Group, Tropical Biosphere Research Center, University of the Ryukyus, Senbaru 1, Nishihara, Okinawa, 903-0213, Japan Faculty of Agriculture, University of the Ryukyus, Senbaru 1, Nishihara, Okinawa, 903-0213, Japan d Faculty of Agriculture, Saga University, 1-Honjo-machi, Saga, 840-8502, Japan b c

h i g h l i g h t s  Isoprene emission profile of tropical trees deviate from predictions by G-93 model.  We propose a method to optimize the G-93 parameters based on field observations.  Optimized formula predicted 81e96% of diurnal emission compared to 73e76% by G-93.

a r t i c l e i n f o

a b s t r a c t

Article history: Received 2 March 2016 Received in revised form 15 May 2016 Accepted 20 June 2016 Available online 21 June 2016

Tropical trees account for most emissions of isoprene, a reactive biogenic volatile organic compound into the atmosphere. The Guenther 1993 (G-93) model is the most widely used algorithm for predicting isoprene emissions by leaves of terrestrial plants. Several studies have reported on the poor performance of the G-93 model in predicting emissions from tropical tree species. To improve the performance of the G-93 model in tropical regions, we carried out diurnal leaf-scale observations of tropical trees Casuarina equisetifolia and Ficus septica outdoors. We developed an iterative method that uses mutual and repetitive step-by-step optimization of the G-93 parameters for temperature (CT) and light (CL) variables using best fit practices, named “Ping-Pong” optimization. Using temperate tree species (Poplar) for comparison, we show that emissions from C. equisetifolia and F. septica had a diversion from predictions of the G93 formula, especially at high temperatures and high light intensities during mid-day, whilst emissions from temperate tree species were fairly captured by G-93. Results demonstrate that our optimized formulas greatly improved capturing of high light and temperature responses of emission profiles from tropical trees whilst it also performed well for poplar species. Parameterization of the G-93 formula greatly improved its performance on predicting diurnal isoprene emission from tropical trees C. equisetifolia and F. septica; explaining 81e96% of variation up from 73 to 77% explained by default G-93. We propose that there is a need to optimize the G-93 model to more accurately predict regional emissions from tropical ecosystems. © 2016 Elsevier Ltd. All rights reserved.

Keywords: G-93 model Isoprene emission Tropical tree Model parameterization Temperature response Light response

1. Introduction Terrestrial vegetation emits very huge amounts of isoprene, a biogenic volatile organic compound (BVOC) into the atmosphere where it influences atmospheric chemistry (Guenther et al., 2006;

* Corresponding author. E-mail address: [email protected] (H. Oku). http://dx.doi.org/10.1016/j.atmosenv.2016.06.052 1352-2310/© 2016 Elsevier Ltd. All rights reserved.

Arneth et al., 2008). In the presence of NOx, isoprene contributes to formation of tropospheric ozone and a variety of other atmospheric constituents (Trainer et al., 1987; Fehsenfeld et al., 1992), furthermore, its reaction with hydroxyl radicals increase the lifetime of methane, a greenhouse gas (Poisson et al., 2000). Due to its high flux estimated at 500e750 Tg per year (Guenther et al., 2006), and to its high reactivity, isoprene is the most important BVOC emitted mainly by land plants. Considerable research efforts have been channeled towards development of algorithms to predict

288

I. Mutanda et al. / Atmospheric Environment 141 (2016) 287e296

isoprene emissions from vegetation that can be used in regional and global air quality models. Temperature and light are the most important environmental drivers of isoprene emission from plant leaves (Monson and Fall 1989; Monson et al., 1992; Sharkey and Yeh, 2001), as such; many models were developed based on explaining response of isoprene emission to these two factors (Tingey et al., 1979; Guenther et al., 1991, 1993, 1995). Among the models developed, the Guenther 1993 (G-93) (Guenther et al., 1993) gained the widest usage in predicting leaf e scale isoprene emissions because it separated instantaneous isoprene emission from the long-term basal emission rate, and also due to the fact that it partitioned light and temperature responses (Monson et al., 2012). Most existing models for isoprene emission are still empirical, based on leaf-scale and canopy e scale observations and best fit regression practices to minimize differences between observed and predicted emission rates (Monson et al., 2012). Despite a lot of research and breakthroughs over the past years, there are still gaps in our understanding of the biochemical controls over instantaneous isoprene emission rate, preventing development of fully mechanistic emission models. Therefore, there is a need to adjust the existing empirical models so that they more accurately predict regional and global emissions. The G-93 model has been widely used globally to predict emissions from plants. Moreover, it forms the basis for many of its successors that model canopy fluxes like the Guenther et al., 1995, Guenther et al., 1999 and the now widely used MEGAN (Guenther et al., 2006, 2012). The G-93 model simulates isoprene emission by two empirically defined equations for light and temperature responses as well as an emission factor at standard conditions. Parameters of the G-93 model were basically determined by best fit to the laboratory observations of attached leaves from temperate plant species as light intensity and temperature were varied, thereby determining the empirical coefficients that describe responsiveness of emission to light and temperature. The G-93 model has been widely and indiscriminately used to predict isoprene emissions from plants in both temperate and tropical regions (Benjamin et al., 1996; He et al., 2000; Owen et al., 2001; Harley et al., 2004; Kuhn et al., 2004a, 2004b; Padhy and Varshney, 2005; Tani and Kawawata, 2008). However, several reports suggested that there might be differences between emission responses of tropical and temperate plants, and that the G-93 model has limited performance in predicting emissions from tropical ecosystems (Lerdau and Keller, 1997; Lerdau and Throop, 1999; Kuhn et al., 2002; Tambunan et al., 2006). Tropical regions represent the most important source of biogenic isoprene, with tropical broad leaf trees as the major contributors to annual global emissions (Guenther et al., 2006, 2012; Arneth et al., 2008). Despite this, there is still limited data on leaf scale observations on tropical tree species, so the G-93 model with its default empirical coefficients developed using temperate plants has been widely used to predict emissions from tropical forests. Lerdau and Keller (1997) observed that emission from tropical trees tend to deviate from the G-93 behavior, especially the nonsaturation of emission at high photosynthetic photon flux density (PPFD). This prompted the first modification of the G-93 algorithm for application in tropical plants (Keller and Lerdau, 1999), where the temperature coefficient (CT) was revised upwards and the light coefficient (CL) was replaced altogether. Similarly, our previous studies with tropical tree species revealed the same deviations from predictions of the G-93 model especially at periods of high light and high temperature (Tambunan et al., 2006; Oku et al., 2008). Furthermore, our recent study characterized the enzymatic properties of isoprene synthases (IspSs) from tropical trees Casuarina equisetifolia and Ficus septica, and found that Km values of IspSs from these trees were lower than

that of IspS from temperate plant poplar (Oku et al., 2015). This finding suggested that substrate affinity of IspS from tropical trees was higher than that of IspS from the temperate plant. Enzymatic properties of IspS as well as substrate availability are very important factors in leaf-level control of isoprene emission over a shortterm period. It is therefore of interest to study the temperature and light dependence of isoprene emission from these tropical trees. Thus, in this study, we measured diurnal isoprene emission from the tropical trees C. equisetifolia and F. septica, and attempted to optimize the parameters of G-93 for better understanding the temperature and light dependencies of isoprene emission from tropical trees. 2. Materials and methods 2.1. Plant materials Saplings of Populus alba and Populus nigra (1.1 m in height) were purchased from Nihon Kaki Garden Center (Saitama, Japan), and were grown under natural conditions with 2e3 times of irrigation per week. For F. septica, cutting clones were prepared from natural tree growing in the campus of University of the Ryukyus (26150 02.300 N 127450 52.000 E) as described previously (Oku et al., 2008). Isoprene emission from Casuarina equisetifolia was measured from trees in the natural forest on campus. 2.2. Diurnal isoprene emission measurements Isoprene emission from an individual leaf of F. septica, P. alba and P. nigra was measured by a real time isoprene analyzer (KFCL-500, Anatec Yanako, Kyoto, Japan) as described previously (Oku et al., 2014). Mature and healthy fully exposed sun leaves (either first or second node from shoot apical meristem) of a branch were used for measurements. A leaf was clamped into a PLC-4C leaf chamber (Shimadzu, Kyoto, Japan), and isoprene emission, light intensity and leaf temperature were measured simultaneously throughout. In the case of conifer type leaves (sepal) of C. equisetifolia, the leaf rods (photosynthetic needle-twigs) were assembled into flattened layer by use of metal clamp as shown in Supporting Information (SI Fig. S1), and isoprene emission was measured as was the case for broad leaf plants such as Populus or Ficus species. In brief, the measurement system consisted of a real-time isoprene analyzer and a portable photosynthesis and transpiration measuring system (SPB-H4, Shimadzu, Kyoto, Japan). Ambient air was pumped into the chamber with a flow rate of 400 ml/min, and the outlet flow of the chamber was introduced through a 3-way valve into the isoprene analyzer with a flow rate of 100 ml/min. The chemiluminescence produced by isoprene/ozone reaction is monitored with a blue-sensitive photomultiplier tube (Hills and Zimmerman, 1990). The analyzer was calibrated with standard isoprene gas (17.52 ppm) purchased from Tokyo-koatsu Co., Tokyo, Japan as recommended by the supplier. The analyzer has two measurement modes of high-range and low-range, and calibration using the standard gas was first performed with the high-range mode (20 ppm max), followed by manual calibration of lowrange (200 ppb max) according to manufacturer’s instructions. Calibration by this procedure usually results in the minimum detection sensitivity (2 s) of 1.2 ppb for low-range measurements. Isoprene emission was analyzed using the low-range mode, and all measurements were corrected by subtracting the background emission level of isoprene (3e4 ppb) in ambient air. Isoprene emission was recorded every 10 s throughout the day using a portable data station (XL-100, Yokogawa Meters & Instruments, Tokyo, Japan). Light intensity and leaf temperature were measured by built in sensors of PLC-4C, and were recorded every

I. Mutanda et al. / Atmospheric Environment 141 (2016) 287e296

30 s. The log file data was exported in Ascii format and incorporated into excel file for data manipulation. Data of isoprene emission, light intensity (photosynthetically active radiation; PAR) and leaf temperature acquired in 10 min of each measurement were averaged and used for parameterization below.

2.3. Parameterization of G-93 The model G-93 estimates isoprene emission (I) as

I ¼ Is $CT $CL ;

(1)

where I is the emission rate predicted at temperature T (K) and PAR of L (mmol m2 s1), and Is is the basal emission rate (emission factor). The two variables CT and CL are respectively temperature and light coefficients, and are defined by

CT ¼ fexp½C T1 ðT  TsÞðRTsTÞ1  g  f1 þ exp½C T2 ðT  TmÞðRTsTÞ1  g1 0:5  CL ¼ ðaCL1 LÞ 1 þ a2 L2

(2) (3)

where CT1 ¼ 95,000 J mol1, CT2 ¼ 230,000 J mol1, Tm ¼ 314 K, a ¼ 0.0027, CL1 ¼ 1.066, R ¼ 8.314 J K1 mol1 and Ts is the leaf temperature at standard condition (303 K). Our previous study suggested that the variables of CT and CL (shown in bold phase in Eqns (2) and (3): CT1, CT2 for CT and a for CL) need revision to predict regional isoprene emission more precisely (Oku et al., 2008). Thus, the present study aimed at optimization of these parameters based on diurnal variations of isoprene emission, light intensity and leaf temperature measured outdoors. Leaf-level isoprene emission, light intensity and leaf temperature were therefore monitored throughout the day under natural conditions from morning to evening. All measurements were conducted in summer time, and were in 2012 for poplars, in 2014 and 2015 for C. equisetifolia, and in 2011 for F. septica. For tropical trees C. equisetifolia and F. septica, four individual plants were used and measurements on each tree were carried out for a whole day from morning till sunset. In the case of Populus species, two measurements were conducted using Populus nigra whilst one measurement was conducted on Populus alba. Observed data sets of isoprene emission (I), light intensity (PAR) and leaf temperature (T) measured throughout the day were respectively incorporated into two data sheets CT and CL that optimize the above parameters by Visual Basic program. The CT sheet optimizes coefficients for CT (CT1 and CT2) and exports these temporal optimized values to the CL sheet that in turn uses them to optimize coefficient for CL (a). An illustration of the iterative optimization procedure using the CT and CL excel sheets is shown in Supporting Information (SI Fig. S2A) and the formulas used in calculating relevant steps are included in Fig. S2B. For optimization of CT1 and CT2, the CT excel sheet calculates CT on the basis of leaf temperature (T) using (2). As illustrated in steps 1e4 in Figs. S2AeB, estimates of CT1 and CT2 were varied in the range between 50,000 (start) to 400,000 (end) with interval of 500 to screen better combination of these coefficients that minimizes error between observed and predicted emissions by iteration. For each observation, temporal CT was calculated using these CT1 and CT2 combinations and, likewise, initial CL was calculated using (3) and an initial value 0.002 for a (a values are optimized by the CL sheet below). Next, the program calculates Is from these temporal CT and CL values using (1) for each observation (step 1 in Fig. S2B), and the mean basal emission rate (Is) under given conditions was

289

subsequently used to estimate isoprene emissions for all observations (step 2). The best combination of CT1 and CT2 was screened and optimized (steps 3e4) to give the lowest squared error (E) between the estimated (Ie) and observed (I) emission rates. This first screening step found temporal optimized values of CT1 and CT2 which were exported to CL sheet for use in optimization of a (step 5). The CL excel sheet, just like steps 1e3 in the CT sheet, calculates Is with CT optimized in the previous step whilst varying the CL coefficient a to minimize sum of E between observed and estimated emissions (steps 6e8). In the case of CL, a value was varied (step 9) in the range between 0.002 (start) to 0.01 (end) with an interval of 0.00001. This step in CL sheet found temporal optimized a value which was then exported back to CT sheet (step 10) for the next round of optimizations for CT1 and CT2 values. This back and forth iterative optimization and subsequent exportation and updating of new optimized values between the CT and CL sheets ultimately led to convergence of estimates, minimizing the overall error between observed and predicted emission values. The optimization steps 1e10 are terminated when there in no further reduction in sum of squared error. We herein call this iterative method “Ping-Pong” optimization, involving mutual step by step optimization of CT1, CT2 and a. The name “Ping-Pong” is therefore named after the reciprocal back and forth data exchange between CT and CL sheets: CT sheet sends optimized CT values to CL sheet for use in optimization of CL, in turn, the CL sheet returns optimized CL values to CT sheet for next step of CT optimization and so on till no further reduction in error is noted. For a test run of this procedure, data sets of isoprene emission rates under variable temperature and light intensity simulating diurnal change (Fig. S3) were generated by arbitrary combination of CT1, CT2 and a. Seven groups of used parameter combinations are listed in Table S1 (groups A-G), and the temperature and light intensity changes simulating diurnal variation is shown in Fig. S3. These values were used for generating test data of isoprene emission. In order to test the validity of our approach, these arbitrary coefficients were conversely parameterized by “Ping-Pong” optimization as described above on the basis of temperature, light intensity and isoprene emission rates. Good estimations of the parameters of CT1, CT2 and a were obtained by our approach (Table S1). Correlation coefficients between test data and estimated parameters were 0.9997 for CT1, 0.9980 for CT2 and 0.9850 for a, showing the robustness of this approach. The high correlation coefficient obtained between test isoprene emission and predicted emission by the formula using optimized parameters (r2 ¼ 0.99995) further demonstrated the accuracy of this method (Fig. S4A). Although more precise estimation of the parameters was possible by decreasing the screening intervals for CT1 and CT2 (thus increasing iterative steps), the accuracy mentioned above was considered to be practically acceptable in this study. 2.4. Statistical analysis Statistical significance of the differences between means was analyzed by non-parametric method of Mann-Whitney U test. Criterion of statistical significance was p < 0.05. 3. Results Diurnal variation of observed isoprene emissions are shown in Fig. 1, together with the predicted emission rates by the G-93 model and by the optimized method. Isoprene emissions showed large variation during the entire day in response to changing weather conditions but in general, emissions increased and peaked around mid-day when temperature and light intensity were highest. In order to evaluate the performance of the optimized formula in

290

I. Mutanda et al. / Atmospheric Environment 141 (2016) 287e296

Fig. 1. Prediction of diurnal variation of isoprene emission by “Ping pong” optimization method or by G-93 formula in tropical trees, Casuarina equsetifolia (A), Ficus septica (B) and in Populus temperate trees (P. nigra and P. alba) (C). Diurnal measurements from a tree in each species were plotted against prediction of emission fluxes by the optimized formula (right panel) or the G-93 formula (left panel). Emission fluxes were recorded every 10 s and averaged for every 10 min interval throughout the day. Optimized parameters used in predictions are listed in Table 1.

predicting emissions, it was applied to predict diurnal variation of isoprene emission from leaves of C. equisetifolia, F. septica and Populus species, and the efficacy was compared to that of G-93 formula (Fig. 1). Optimization of G-93 parameters for C. equisetifolia

greatly improved the ability to predict diurnal variation whilst the original G-93, without any modification of the parameters, showed limited performance especially for predicting higher emission rates at mid-day. This trend was also noted with the case of F. septica

I. Mutanda et al. / Atmospheric Environment 141 (2016) 287e296

(Fig. 1B). In contrast, G-93 showed good performance in prediction of diurnal variation of isoprene emission from Populus species throughout the day, as well as the modified formula in this study (Fig. 1C).

291

Leaf temperature and light intensity (PAR) variations throughout the entire day of measurements are shown in Fig. 2. In general, temperature and PAR increased from morning till around mid-day before they started decreasing. The G-93 model has been

Fig. 2. Variation in leaf temperature, light intensity (photosynthetic active radiation; PAR) and basal emission rate (Is) during the entire day during measurements of isoprene emissions. The left panels show diurnal variation of leaf temperature and PAR. The right panels show comparison of Is of Casuarina equisetifolia, Ficus septica and Populus (P. alba and P. nigra) estimated by either the G-93 formula or by the “Ping pong” optimized parameters. Data were averaged every 10 min throughout the day.

292

I. Mutanda et al. / Atmospheric Environment 141 (2016) 287e296

widely used to estimate the basal emission rate (emission factor) (Is) under variable conditions of temperature and light intensities (Chang et al., 2005; Tambunan et al., 2006; Wilske et al., 2007; Tani and Kawawata, 2008). The changes in Is estimated by the G-93 and the optimized formula for the focal species are also depicted in Fig. 2. In the case of C. equisetifolia, Is estimated by G-93 was much higher than that estimated by the optimized formula. The difference was more pronounced at higher temperatures during the day time. Average Is for a day was 22.0 by G-93, and 9.4 by the optimized formula. It is worth noting that Is estimation by G-93 showed large variation in a day compared to roughly constant estimations by the modified formula throughout. Similarly, Is of F. septica estimated by G-93 was higher than that by the optimized formula, but to a lesser extent compared to the case of C. equisetifolia. Average Is for F. septica estimated by optimized formula was 10.4, and was 14.4 by G-93. Is of poplar estimated by the optimized formula was 11.6 against 14.1 by G-93. Table 1 lists the mean optimized parameters of G-93 for C. equisetifolia and F. septica estimated by using observed diurnal variation of isoprene emission, PAR and leaf temperature as inputs in “Ping-Pong” optimization, and used for predicting emission fluxes shown in Fig. 1. We currently conducted the same experiment with P. alba and P. nigra, and the results were included in the Table as a reference. CT1 of C. equisetifolia and F. septica was significantly higher than that of poplar, and was almost 2-fold and 1.6-fold higher than that for G-93, respectively. With respect to CL parameter, a of F. septica was significantly lower than that of C. equisetifolia and Populus species, and was almost comparable to that of G-93. No significant difference was noted with CT2 between plant species. Temperature and light dependence of isoprene emission from C. equisetifolia, F. septica and poplar with the parameters listed in Table 1 are shown in Fig. 3. The temperature dependence of C. equisetifolia and F. septica showed large deviation from the profile of poplar or that predicted by G-93. The optimized formula for C. equisetifolia predicted a maximal temperature dependent increase of more than 8-fold of the basal emission rate (Fig. 3A). Similarly, the maximal increase predicted by the optimized formula was almost 5-fold of the basal emission rate for F. septica. In contrast to these temperature dependent variations of tropical trees, the profile for Populus species was largely comparable to that predicted by G-93. Table 2 lists the performance evaluation of goodness of fit between the G-93 and the optimized formula in predicting diurnal emissions based on 10 min interval averaged data in Fig. 1. The normalized mean square error (M score) defined as below provides an overall score of model performance (Guenther et al., 1993).

2 1   M ¼ Eo  Ep $ Eo Ep

(4)

where Ep is the predicted emission rate, Ep is the mean predicted emission rate, Eo is the observed emission rate, Eo is the mean observed emission rate. The M score is a function of the F, t and r

statistical scores (F is a measure of bias of variance, t statistic is a measure of bias of magnitude (Eo  Ep Þ2 and r is the correlation coefficient (Guenther et al., 1993). The standard deviations were estimated by bootstrap method with 1000 iterations. A lower M score indicates better overall model performance. From the M scores, optimization improved model performance for C. equisetifolia and F. septica, but not that much for Populus. Parameterization of the CL and CT coefficients improved the M scores 8-fold in the case of C. equisetifolia and 2-fold in the case of F. septica. Judging by the M-scores, it can be seen from Fig. 3 that the more the deviation from the postulates of the G-93 formula, the more the improvement of model performance achieved by the optimized formula. The correlation coefficients (r2) between observed and estimated emission rates showed that the modified model of G-93 explained 96% of the total variation in diurnal isoprene emission from C. equisetifolia, and the value was much higher than 77% explained by G-93. Optimization of G-93 for F. septica similarly improved the model performance from 73% by original G-93 to 81%. Likewise, the total diurnal variation of 93% explained by the optimized formula for Populus was close to, however slightly higher than 91% obtained by G-93. This was also reflected by the mean squared error (MSE) values that show a big improvement in the goodness of fit by the optimized formula in the cases of C. equisetifolia and F. septica, whilst the relatively low values for Populus were comparable by both G-93 and the optimized formula.

4. Discussion The authors measured diurnal isoprene emission from tropical trees C. equisetifolia and F. septica, and compared the emission profiles with that of Populus species in this study. To enable optimization of G-93 model parameters based on these field observations, we developed an iterative procedure named “Ping-Pong” optimization that successively screen best estimates for the coefficients of CL and CT of G-93. We applied the “Ping-Pong” optimization procedure and successfully optimized a of CL, CT1 and CT2 of CT in this study and results demonstrate that optimization significantly improved model performance in predicting diurnal emissions. Our previous study similarly adjusted the parameters of CL and CT for tropical tree Ficus virgata (Oku et al., 2008). In that study, we further modified the CL formula to simulate the decrease in isoprene emissions observed at higher light intensity. The emission rate levelled off at higher light intensities and this was more clearly rendered at light intensities over 1700 mmol m2 s1 (Oku et al., 2008). In this study, we however attempted to adjust only the parameters of original G-93 because the frequencies of higher light intensities above 1700 mmol m2 s1 are not so high in the subtropical area of Okinawa. Adjustment of parameters was therefore considered to be practically good enough to improve model performance in predicting emissions. In fact, it was found that application of our current optimization procedure to that previous diurnal isoprene emission data set of F. virgata (Oku et al., 2008)

Table 1 Optimized G-93 parameters for C. equisetifolia, F. septica and Populus. Plant

C. equisetifolia F. septica Populus a

Optimized parameters

G-93

CT1

CT2

a

CT1

CT2

a

187,125 ± 14,570a 158,500 ± 15,471a 95,166 ± 9,400b

151,300 ± 46,881a 287,250 ± 94,715a 208,500 ± 73,428a

0.0052 ± 0.0008a 0.0034 ± 0.0006b 0.0056 ± 0.0007a

95,000

230,000

0.0027

Data are mean ± SE of four plants for C. equisetifolia and F. septica, and three plants for Populus. a Data for Populus species includes two measurements from Populus nigra and one measurement from P. alba. a-bDifferent letters within a column represent a statistically significant difference (Mann-Whitney U test, p < 0.05).

I. Mutanda et al. / Atmospheric Environment 141 (2016) 287e296

293

Fig. 3. Temperature (A) and light (B) response of leaf-scale isoprene emissions from C. equisetifolia, F. septica and Populus trees using parameters optimized by “Ping-pong” method as compared to responses postulated by G-93 formula. Lines represent the optimized or G-93 functions of CT (A) and CL (B) using the parameters listed in Table 1. Data are means of 3e4 independent plants for C. equisetifolia, F. septica and Populus. For Populus species, two observations were made from P. nigra and one observation was made from P. alba. Normalized emission rate was the ratio of observed emission rates to that of standard conditions (30  C leaf temperature and light intensity of 1000 mmol m2 s1).

Table 2 Performance evaluation of optimized model and G-93. Plant

C. equisetifolia F. septica Populus

Optimized

G-93

M score

r2

MSE

M score

r2

MSE

0.011 ± 0.003 0.105 ± 0.029 0.014 ± 0.003

0.962 0.810 0.925

28.032 40.525 4.263

0.089 ± 0.018 0.205 ± 0.033 0.015 ± 0.004

0.769 0.730 0.908

185.247 81.1607 5.677

M score depicts overall model performance as defined by Equation (4) in text. Data are presented as M score ± standard deviation. Standard deviations were determined by bootstrap method with 1000 iterations. r2 is the correlation coefficient between observed and predicted values. The mean squared error (MSE) is a measure of how close the fitted line with parameters listed in Table 1 was to observed data points.

further improved the model performance (r2 ¼ 0.853 by our current method against r2 ¼ 0.753 by the previous modified formula for the prediction of diurnal variation). For these reasons, we only optimized the parameters of G-93 to more precisely simulate regional

isoprene emission rate from tropical trees. However, this view may not be necessarily true for those tropical areas where plants experience routinely higher light intensities of over 1700 mmol m2 s1. In such cases, parameterization incorporating

294

I. Mutanda et al. / Atmospheric Environment 141 (2016) 287e296

the modification of CL, in combination with “Ping-Pong” optimization may further improve model performance. This point needs further study in the future. It was shown in this study that parameterization of G-93 improved its performance in predicting the emission rate of tropical trees. Fig. 1 pointed out the limited performance of G-93 in predicting higher emission rates of tropical trees at mid-day, whilst its performance was satisfactory for the prediction of isoprene emission from temperate plants Populus. Observations from our previous study suggested that the parameters of G-93 need revisions to better predict regional isoprene emission since there seem to be differences in temperature and light dependencies of emissions from plants in different ecosystems. Of the two formulas, CT and CL, large deviation of tropical trees from G-93 was noted with CT factor that simulates temperature dependence of isoprene emission (Fig. 3A). In contrast, temperature dependence of poplar species showed good agreement with that predicted by G-93. CT of G-93 postulates that the maximal temperature dependent increase in isoprene emission is almost 2-fold over the basal emission rate at standard conditions of 30  C. Our results show that the maximal temperature dependent increase in isoprene emission was almost 8-fold for C. equisetifolia, and was almost 5-fold for F. septica (Fig. 3). These fluctuation magnitudes are far greater than the postulation of G-93 formula, and may reasonably explain the limited performance of the model at higher emission rates of tropical trees (Fig. 1). A statistically significant difference was also noted with the light dependence of isoprene emission (Fig. 3B and Table 1). The difference was however rendered more clearly at lower light intensity than at higher intensity. This could have been as a result of masking of the relatively smaller differences in CL by the large differences in CT at higher temperatures. As shown in Fig. 2, leaf temperature well correlated with light intensity, and the differences in CL between plant species were rather smaller at high temperatures (Fig. 3). The lower performance of G-93 for high emission rates at high temperature of tropical trees was therefore largely due to low efficiencies in CT of G-93. In this study, we applied the same experimental protocol to Populus species and found for the first time that the temperature dependence of isoprene emission from tropical trees, C. equisetifolia and F. septica was clearly different from that of temperate plants poplar (Fig. 3). These findings may negate the possibility that the observed results could have been due to differences in method or differences in environmental conditions. Even under our conditions, the performance of G-93 was found to be satisfactory for the prediction of isoprene emission rate of poplar, but not for estimation of isoprene emission rates of tropical trees C. equisetifolia and F. septica. However, these findings have been applied to a limited number of tropical trees to date, and also comparison was made to Populus species only in this study. Further studies employing a large number of tropical trees are needed to extend or more generalize this view given the high species diversity in tropical ecosystems. The G-93 formula also showed a tendency to overestimate the basal emission rate at standard conditions of 30  C leaf temperature and PAR of 1000 mmol m2 s1. This trend was clearer for trees of which temperature dependence diverged far from that of temperate plants (Figs. 1 and 3). In the case of C. equisetifolia, the average basal emission rate estimated by G-93 was 2.5-fold higher than that estimated by our optimized formula. Similarly, this value was 1.4 fold higher for F. septica. It is thus likely that basal isoprene emissions of plant species with high temperature responses are overestimated to a greater extent compared to that of plants with low temperature responses. The overestimation of basal isoprene emission rate at 30  C and 1000 mmol m2 s1 PAR by G-93 formula with original parameters was also noted in the cold subarctic ecosystems, leading to the suggestion that in those ecosystems,

standard emissions at 20  C might yield better and more comparable results than adopting the widely accepted 30  C (Holst et al., 2010). The present study demonstrated that temperature response of tropical trees at leaf-level is a strong factor that influences isoprene emission rate. Therefore, more accurate predictions of emission profiles with respect to temperature response, combined with improved inventory data of regional tropical forests may promisingly contribute to improved estimation of isoprene emissions from those regions. There is no storage of isoprene inside the leaf, and isoprene is biosynthesized in situ by the enzyme IspS (Silver and Fall., 1991) probably in response to high temperature from newly assimilated carbon of its substrate dimethylallyl diphosphate (DMAPP) supplied by the methylerythritol phosphate (MEP) pathway (Lichtenthaler et al., 1997). The temperature response of isoprene emission resembles the characteristic temperature effect on enzyme kinetics, which formed the basis for its modelling using equation (2) (Guenther et al., 1993). Plant leaves experience minute to minute changes in light intensity and temperature in the field. To adapt to these rapid changes in environmental conditions, some plants emit isoprene apparently as a means to cope with abiotic stress (Sharkey and Singsaas, 1995; Singsaas et al., 1997; Velikova et al., 2011). Isoprene emission is regulated by both enzyme (IspS) activity and concentration of substrate DMAPP on a short timescale of minutes to hours, and by the gene expression of IspS and enzymes in the MEP pathway on longer time-scales of seasonal variation or during senescence (Sharkey and Monson, 2014). Our recent study demonstrated a high substrate affinity of IspS cloned from tropical trees compared to poplar IspS (Oku et al., 2015). It is therefore a reasonable postulate that the IspS of higher substrate affinity shows greater activity fluctuations within the same range of changes in the substrate concentration as illustrated in Fig. 4. The greater temperature dependent variation of isoprene emission from C. equisetifolia thus may be explained by the enzyme properties of its IspS. It is likely that the substrate affinity of IspS is a critical determinant of temperature dependency of isoprene emission. The degree of temperature dependency of emissions in plants shown in Fig. 3 appear to agree well with the order of Km values of their IspSs for DMAPP measured in our previous study:

Fig. 4. Relationship between substrate concentration, Km and enzyme activity. Differences in substrate affinity between isoprene synthases from tropical tree species and temperate species could account for their differences in isoprene emission profiles in response to changes in temperature and light. A similar change in substrate concentration results in different orders of change in enzyme activities depending on substrate binding affinity.

I. Mutanda et al. / Atmospheric Environment 141 (2016) 287e296

plant species of low IspS Km value showed larger temperature dependent variations than plants with IspS of high Km for DMAPP. This view needs further testing, and may be exploited by gene cloning and comparative studies on IspSs from more tropical trees. In conclusion, the temperature and light response of isoprene emission from the tropical trees C. equisetifolia and F. septica diverged from the emission profiles postulated by the widely used G-93 formula as originally defined, and were also different from the emission profile of temperate plants Populus. The G-93 formula was limited in its simulation of high emission rates of the tropical trees especially around midday under high temperatures and high light conditions of a typical summer day in tropical regions. To improve predictions of isoprene emissions from the tropical trees, we developed an iterative optimization method that uses a converging step-by-step estimation of CT (CT1 and CT2) and CL (a) coefficients of the G-93 formula by back and forth exchange of successive optimized values in two excel sheets, herein named “Ping-Pong” optimization method. Results show that our optimization procedure and parameterization of the temperature and light coefficients improved predictions of isoprene emissions in the tropical trees and also in Populus mainly due to better estimation of parameters of CT. In most studies, parameterization procedures require controlled environments where one variable (eg., PAR) is held constant whilst the other variable (eg., temperature) is varied. The proposed iterative optimization method allowed us to simultaneously optimize both CT and CL parameters based on field observations without the need for controlled environments. Porting the optimization procedure to other programming languages in the future might help to improve its computational efficiency and enable automation to process huge data sets from several observations at the same time. Whilst the procedure has been currently applied to a limited number of species in this study given the high biodiversity nature of most tropical ecosystems, more field observations of tropical species in the future might ultimately allow estimation of ecosystem average values that can then be scaled to estimate regional forests and canopy emissions. There are still a lot of uncertainties in regional and global emission models originating from poor simulations of leaf-scale emissions among other factors. The data and optimization method reported in this study can be useful in improving regional emission estimations. Acknowledgements This study was partly supported by Japan Society for the Promotion of Science (JSPS) Grant-in-Aid for Scientific Research (c) Grant No. 15K07484. We would like to acknowledge the contribution of Yu Kameshima in field observations. Appendix A. Supplementary data Supplementary data related to this article can be found at http:// dx.doi.org/10.1016/j.atmosenv.2016.06.052. References Arneth, A., Monson, R.K., Schurgers, G., Niinemets, Ü., Palmer, P.I., 2008. Why are estimates of global terrestrial isoprene emissions so similar (and why is this not so for monoterpenes)? Atmos. Chem. Phys. 8, 4605e4620. Benjamin, M.T., Sudol, M., Bloch, L., Winer, A.M., 1996. Low-emitting urban forests: a taxonomic methodology for assigning isoprene and monoterpene emission rates. Atmos. Environ. 30, 1437e1452. Chang, K.H., Chen, T.F., Huang, H.C., 2005. Estimation of biogenic volatile organic compounds emissions in subtropical island e Taiwan. Sci. Total Environ. 346, 184e199. Fehsenfeld, F., Calvert, J., Fall, R., Goldan, P., Guenther, A.B., Hewitt, C.N., Lamb, B., Liu, S., Trainer, M., Westberg, H., Zimmerman, P., 1992. Emissions of volatile organic compounds from vegetation and the implications for atmospheric chemistry. Glob. Biogeochem. Cycles 6, 389e430.

295

Guenther, A., Baugh, B., Brasseur, G., Greenberg, J., Harley, P., Klinger, L., Serca, D., Vierling, L., 1999. Isoprene emission estimates and uncertainties for the Central African expresso study domain. J. Geophys. Research-Atmospheres 104, 30625e30639. Guenther, A., Hewitt, C.N., Erickson, D., Fall, R., Geron, C., Graedel, T., Harley, P., Klinger, L., Lerdau, M., Mckay, W.A., Pierce, T., Scholes, B., Steinbrecher, R., Tallamraju, R., Taylor, J., Zimmerman, P., 1995. A global-model of natural volatile organic-compound emissions. J. Geophys. Research-Atmospheres 100, 8873e8892. Guenther, A., Karl, T., Harley, P., Wiedinmyer, C., Palmer, P.I., Geron, C., 2006. Estimates of global terrestrial isoprene emissions using MEGAN (model of emissions of gases and aerosols from nature). Atmos. Chem. Phys. 6, 3181e3210. Guenther, A.B., Jiang, X., Heald, C.L., Sakulyanontvittaya, T., Duhl, T., Emmons, L.K., Wang, X., 2012. The Model of Emissions of Gases and Aerosols from Nature version 2.1 (MEGAN2.1): an extended and updated framework for modeling biogenic emissions. Geosci. Model Dev. 5, 1471e1492. Guenther, A.B., Monson, R.K., Fall, R., 1991. Isoprene and monoterpene emission rate variability: observations with eucalyptus and emission rate algorithm development. J. Geophys. Res. Atmos. 96, 10799e10808. Guenther, A.B., Zimmerman, P.R., Harley, P.C., Monson, R.K., Fall, R., 1993. Isoprene and monoterpene emission rate variability: model evaluations and sensitivity analyses. J. Geophys. Res. Atmos. 98, 12609e12617. Harley, P., Vasconcellos, P., Vierling, L., Pinheiro, C.C.D., Greenberg, J., Guenther, A., Klinger, L., De Almeida, S.S., Neill, D., Baker, T., Phillips, O., Malhi, Y., 2004. Variation in potential for isoprene emissions among Neotropical forest sites. Glob. Change Biol. 10, 630e650. He, C.R., Murray, F., Lyons, T., 2000. Monoterpene and isoprene emissions from 15 Eucalyptus species in Australia. Atmos. Environ. 34, 645e655. Hills, A.J., Zimmerman, P.R., 1990. Isoprene measurement by ozone-induced chemiluminessecence. Anal. Chem. 62, 1055e1060. Holst, T., Arneth, A., Hayward, S., Ekberg, A., Mastepanov, M., JackowiczKorczynski, M., Friborg, T., Crill, P.M., Backstrand, K., 2010. BVOC ecosystem flux measurements at a high latitude wetland site. Atmos. Chem. Phys. 10, 1617e1634. Keller, M., Lerdau, M., 1999. Isoprene emission from tropical forest canopy leaves. Glob. Biogeochem. Cycles 13, 19e29. Kuhn, U., Rottenberger, S., Biesenthal, T., Wolf, A., Schebeske, G., Ciccioli, P., Brancaleoni, E., Frattoni, M., Tavares, T.M., Kesselmeier, J., 2002. Isoprene and monoterpene emissions of Amazonian tree species during the wet season: direct and indirect investigations on controlling environmental functions. J. Geophys. Research-Atmospheres 107. Kuhn, U., Rottenberger, S., Biesenthal, T., Wolf, A., Schebeske, G., Ciccioli, P., Brancaleoni, E., Frattoni, M., Tavares, T.M., Kesselmeier, J., 2004a. Seasonal differences in isoprene and light-dependent monoterpene emission by Amazonian tree species. Glob. Change Biol. 10, 663e682. Kuhn, U., Rottenberger, S., Biesenthal, T., Wolf, A., Schebeske, G., Ciccioli, P., Kesselmeier, J., 2004b. Strong correlation between isoprene emission and gross photosynthetic capacity during leaf phenology of the tropical tree species Hymenaea courbaril with fundamental changes in volatile organic compounds emission composition during early leaf development. Plant Cell Environ. 27, 1469e1485. Lerdau, M., Keller, M., 1997. Controls on isoprene emission from trees in a subtropical dry forest. Plant Cell Environ. 20, 569e578. Lerdau, M.T., Throop, H.L., 1999. Isoprene emission and photosynthesis in a tropical forest canopy: implications for model development. Ecol. Appl. 9, 1109e1117. Lichtenthaler, H.K., Schwender, J., Disch, A., Rohmer, M., 1997. Biosynthesis of isoprenoids in higher plant chloroplasts proceeds via a mevalonate-independent pathway. Febs Lett. 400, 271e274. Monson, R.K., Fall, R., 1989. Isoprene emission from aspen leaves: influence of environment and relation to photosynthesis and photorespiration. Plant Physiol. 90, 267e274. Monson, R.K., Grote, R., Niinemets, U., Schnitzler, J.-P., 2012. Modeling the isoprene emission rate from leaves. New Phytol. 195, 541e559. Monson, R.K., Jaeger, C.H., Adams, W.W., Driggers, E.M., Silver, G.M., Fall, R., 1992. Relationships among isoprene emission rate, photosynthesis, and isoprene synthase activity as influenced by temperature. Plant Physiol. 98, 1175e1180. Oku, H., Fukuta, M., Iwasaki, H., Tambunan, P., Baba, S., 2008. Modification of the isoprene emission model G93 for tropical tree Ficus virgata. Atmos. Environ. 42, 8747e8754. Oku, H., Inafuku, M., Ishikawa, T., Takamine, T., Ishmael, M., Fukuta, M., 2015. Molecular cloning and biochemical characterization of isoprene synthases from the tropical trees Ficus virgata, Ficus septica, and Casuarina equisetifolia. J. Plant Res. 128, 849e861. Oku, H., Inafuku, M., Takamine, T., Nagamine, M., Saitoh, S., Fukuta, M., 2014. Temperature threshold of isoprene emission from tropical trees, Ficus virgata and Ficus septica. Chemosphere 95, 268e273. Owen, S.M., Boissard, C., Hewitt, C.N., 2001. Volatile organic compounds (VOCs) emitted from 40 Mediterranean plant species: VOC speciation and extrapolation to habitat scale. Atmos. Environ. 35, 5393e5409. Padhy, P.K., Varshney, C.K., 2005. Emission of volatile organic compounds (VOC) from tropical plant species in India. Chemosphere 59, 1643e1653. Poisson, N., Kanakidou, M., Crutzen, P.J., 2000. Impact of non-methane hydrocarbons on tropospheric chemistry and the oxidizing power of the global troposphere: 3-dimensional modelling results. J. Atmos. Chem. 36, 157e230. Sharkey, T.D., Monson, R.K., 2014. The future of isoprene emission from leaves,

296

I. Mutanda et al. / Atmospheric Environment 141 (2016) 287e296

canopies and landscapes. Plant Cell Environ. 37, 1727e1740. Sharkey, T.D., Singsaas, E.L., 1995. Why plants emit isoprene. Nature 374, 769e769. Sharkey, T.D., Yeh, S.S., 2001. Isoprene emission from plants. Annu. Rev. Plant Physiol. Plant Mol. Biol. 52, 407e436. Silver, G.M., Fall, R., 1991. Enzymatic synthesis of isoprene from dimethylallyl diphosphate in aspen leaf extracts. Plant Physiol. 97, 1588e1591. Singsaas, E.L., Lerdau, M., Winter, K., Sharkey, T.D., 1997. Isoprene increases thermotolerance of isoprene-emitting species. Plant Physiol. 115, 1413e1420. Tambunan, P., Baba, S., Kuniyoshi, A., Iwasaki, H., Nakamura, T., Yamasaki, H., Oku, H., 2006. Isoprene emission from tropical trees in Okinawa Island, Japan. Chemosphere 65, 2138e2144. Tani, A., Kawawata, Y., 2008. Isoprene emission from the major native Quercus spp. in Japan. Atmos. Environ. 42, 4540e4550.

Tingey, D.T., Manning, M., Grothaus, L.C., Burns, W.F., 1979. The influence of light and temperature on isoprene emission rates from live oak. Physiol. Plant. 47, 112e118. Trainer, M., Williams, E.J., Parrish, D.D., Buhr, M.P., Allwine, E.J., Westberg, H.H., Fehsenfeld, F.C., Liu, S.C., 1987. Models and observations of the impact of natural hydrocarbons on rural ozone. Nature 329, 705e707. Velikova, V., Varkonyi, Z., Szabo, M., Maslenkova, L., Nogues, I., Kovacs, L., Peeva, V., Busheva, M., Garab, G., Sharkey, T.D., Loreto, F., 2011. Increased thermostability of thylakoid membranes in isoprene-emitting leaves probed with three biophysical techniques. Plant Physiol. 157, 905e916. Wilske, B., Cao, K.F., Schebeske, G., Chen, J.W., Wang, A., Kesselmeier, J., 2007. Isoprenoid emissions of trees in a tropical rainforest in Xishuangbanna, SW China. Atmos. Environ. 41, 3748e3757.