Field Validation of the DNDC-Rice Model for Methane and Nitrous Oxide Emissions from Double-Cropping Paddy Rice under Di ﬀ erent Irrigation Practices in Tamil Nadu, India

: Two-year ﬁeld experiments were conducted at Tamil Nadu Rice Research Institute, Aduthurai, Tamil Nadu, India, to evaluate the e ﬀ ect of continuous ﬂooding (CF) and alternate wetting and drying (AWD) irrigation strategies on rice grain yield and greenhouse gas emissions from double-cropping paddy rice. Field observation results showed that AWD irrigation was found to reduce the total seasonal methane (CH 4 ) emission by 22.3% to 56.2% compared with CF while maintaining rice yield. By using the observed two-year ﬁeld data, validation of the DNDC-Rice model was conducted for CF and AWD practices. The model overestimated rice grain yield by 24% and 29% in CF and AWD, respectively, averaged over the rice-growing seasons compared to observed values. The simulated seasonal CH 4 emissions for CF were 6.4% lower and 4.2% higher than observed values and for AWD were 9.3% and 12.7% lower in the summer and monsoon season, respectively. The relative deviation of simulated seasonal nitrous oxide (N 2 O) emissions from observed emissions in CF were 27% and − 35% and in AWD were 267% and 234% in the summer and monsoon season, respectively. Although the DNDC-Rice model reasonably estimated the total CH 4 emission in CF and reproduced the mitigation e ﬀ ect of AWD treatment on CH 4 emissions well, the model did not adequately predict the total N 2 O emission under water-saving irrigation. In terms of global warming potential (GWP), nevertheless there was a good agreement between the simulated and observed values for both CF and AWD irrigations due to smaller contributions of N 2 O to the GWP compared with that of CH 4 . This study showed that the DNDC-Rice model could be used for the estimation of CH 4 emissions, the primary source of GWP from double-cropping paddy rice under di ﬀ erent water management conditions in the tropical regions.


Introduction
Rice cultivation is a major source of atmospheric methane (CH 4 ), one of the significant potent greenhouse gases (GHG) and is responsible for approximately 11% of global anthropogenic CH 4 emissions [1]. Rice paddies are also known to emit high nitrous oxide (N 2 O) fluxes under nitrogen fertilization and specific water management regimes [2,3].
Methane emission from rice fields is the net result of CH 4 production and oxidation in soil and transport of CH 4 gas from soil to the atmosphere through rice plants [4]. Conventional management practices of continuously flooded irrigation in paddy fields enhance anaerobic fermentation of carbon sources supplied by the rice plants and added organic matter and results in high CH 4 production. Water management is one of the most effective options for reducing CH 4 emission from irrigated rice. Recently, midseason drainage and alternate wetting and drying irrigation (AWD) practice have been promoted as a strategy to decrease CH 4 emissions from paddy rice fields [3,[5][6][7][8]. However, it can result in increased N 2 O emissions due to a trade-off between CH 4 and N 2 O [2, 3,9]. Frequent alternations in soil redox conditions under water saving irrigation are known to substantially increase N 2 O emissions by favoring both nitrification and denitrification processes [10]. It can substantially offset the advantages of CH 4 mitigation under water-saving irrigation [11,12]. Irrigation management plays a vital role in determining the trade-off between CH 4 and N 2 O emissions from paddy rice fields.
Water management practices relating to the drying and wetting of soil conditions are known to be important factors for CH 4 emissions from paddy rice soil. CH 4 emissions are highly variable depending on practices, and therefore will lead to high uncertainties in the estimation of the emissions for regional and national scales. Evaluation of regional CH 4 emissions from rice paddy differs largely depending on the techniques, approaches, and databases used for extrapolation [13]. Advances are needed in how to effectively scale the measurements from point sources to a regional scale, and it is beneficial to link the available data on CH 4 emissions to a knowledge of underlying processes, such as through a process-based model, DNDC (Denitrification-Decomposition) model [13]. The DNDC model simulates carbon and nitrogen biogeochemistry in agroecosystems and can estimate CO 2 , CH 4 , N 2 O, nitric oxide, and ammonia simultaneously [14][15][16].
The DNDC model was revised to improve its ability to estimate CH 4 emitted from rice paddies under continuously flooded conditions, midseason drainage, and intermittent irrigation [17]. The revised model (DNDC-Rice) was validated with CH 4 data from paddy rice fields in Japan, China, and Thailand [17][18][19][20]. Smakgahn et al. [18] validated the DNDC-Rice model by using CH 4 emission data from nine paddy fields in Thailand under continuous flooding treatment; the simulated values were positively correlated with the observed values. Using the DNDC-Rice model, simulation of N 2 O fluxes has also been reported [19,21]. Babu et al. [13] reported that the DNDC model is capable of capturing quantitatively the significant aspects of CH 4 and N 2 O production and emission from rice fields under widely different geographical locations in India. However, these studies were conducted under continuous flooding (CF) or midseason drainage conditions. Katayanagi et al. [21] validated the DNDC-Rice model by using CH 4 and N 2 O flux data under CF and AWD management conditions in a pot experiment and discussed that the accuracy of the simulation of gross CH 4 emissions and total global warming potential (GWP) values for the CF and AWD treatments was sufficiently good for practical use of the model. However, there is still limited information on field validation of CH 4 and N 2 O fluxes simulated by the DNDC-Rice model under water-saving irrigations in intensive rice cultivation systems, such as double-cropping paddy rice per year in the tropical regions. Therefore, the objectives of this study were to assess (1) whether CH 4 and N 2 O processes are similarly reflected in the DNDC-Rice model; (2) the reliability of the DNDC-Rice model to predict CH 4

and N 2 O emissions
Agriculture 2020, 10, 355 3 of 16 from the double-cropping paddy rice system under different irrigation practices to contribute mitigation strategies in tropical rice production. The results of the simulations were validated using the flux data from two-year field observations at Tamil Nadu Rice Research Institute, Tamil Nadu, India.

Experimental Site and Design
The field experiments were carried out from May 2016 until January 2018, comprising four rice-growing seasons, at the Tamil Nadu Rice Research Institute (TRRI), Aduthurai, Thanjavur District, Tamil Nadu, India (11 • 0 N, 79 • 30 E, 19.4 m above sea level). The region has a tropical wet and dry/savanna climate with a pronounced dry season in the high-sun months, and no cold or wet seasons (monsoon season) in the low-sun months. Figure 1 shows daily rainfall and maximum and minimum temperatures from January 2016 until January 2018 measured at the study site. The soil type is alluvial clay with major properties indicated as 13.6% sand, 61.2% silt, and 25.3% clay, 1.1 g kg −1 total N, 19.6 g kg −1 total C, pH 7.5 (1:5 H 2 O), and electrical conductivity (EC) 11.6 m S m −1 [22]. There were two rice-growing seasons per year, summer-hot and dry season (local name-Kuruvai season; from May to September) and monsoon-wet season (local name-Thaladi season; from September to January).
Agriculture 2020, 10, x FOR PEER REVIEW 3 of 16 contribute mitigation strategies in tropical rice production. The results of the simulations were validated using the flux data from two-year field observations at Tamil Nadu Rice Research Institute, Tamil Nadu, India.

Experimental Site and Design
The field experiments were carried out from May 2016 until January 2018, comprising four ricegrowing seasons, at the Tamil Nadu Rice Research Institute (TRRI), Aduthurai, Thanjavur District, Tamil Nadu, India (11°0′ N, 79°30′ E, 19.4 m above sea level). The region has a tropical wet and dry/savanna climate with a pronounced dry season in the high-sun months, and no cold or wet seasons (monsoon season) in the low-sun months. Figure 1 shows daily rainfall and maximum and minimum temperatures from January 2016 until January 2018 measured at the study site. The soil type is alluvial clay with major properties indicated as 13.6% sand, 61.2% silt, and 25.3% clay, 1.1 g kg −1 total N, 19.6 g kg −1 total C, pH 7.5 (1:5 H2O), and electrical conductivity (EC) 11.6 m S m −1 [22]. There were two rice-growing seasons per year, summer-hot and dry season (local name-Kuruvai season; from May to September) and monsoon-wet season (local name-Thaladi season; from September to January). The field experiment was set up on four rice-growing seasons. Two water management practices, (1) continuous flooding (CF) and (2) alternate wetting and drying irrigation (AWD), were compared in each growing season with three replications. Specific management conditions are summarized in Table 1, along with the rice season weather summaries (average maximum and minimum temperatures and accumulated rainfall for each rice-growing season). For AWD irrigation, a perforated 25-cm long field water tube was inserted in the soil to observe the water level below the soil surface. Irrigation was applied to re-flood the field when the water level had dropped to about 15 cm below the soil surface in AWD irrigation. Pump irrigation was practiced by using groundwater in all growing seasons. Rice stubbles of previous season were incorporated by ploughing the field before rice cultivation, except the summer season of 2017 when rice stubbles were incorporated soon after the previous season's rice harvest. The field experiment was set up on four rice-growing seasons. Two water management practices, (1) continuous flooding (CF) and (2) alternate wetting and drying irrigation (AWD), were compared in each growing season with three replications. Specific management conditions are summarized in Table 1, along with the rice season weather summaries (average maximum and minimum temperatures and accumulated rainfall for each rice-growing season). For AWD irrigation, a perforated 25-cm long field water tube was inserted in the soil to observe the water level below the soil surface. Irrigation was applied to re-flood the field when the water level had dropped to about 15 cm below the soil surface in AWD irrigation. Pump irrigation was practiced by using groundwater in all growing seasons. Rice stubbles of previous season were incorporated by ploughing the field before rice cultivation, except the summer season of 2017 when rice stubbles were incorporated soon after the previous season's rice harvest.

Gas Sample Collection, Measurement, and Calculation
The gas samples were collected using the closed chamber method. In all rice seasons, the sampling frequency was once every week. Whenever there was a fertilizer application event, however, air sampling was done one day and three days after fertilization [3,9]. Gas samples were obtained using a 50 mL plastic syringe at 0, 15, and 30 min after chamber closure. The collected samples were analyzed using a gas chromatograph (GC 2014, Shimadzu Corporation, Kyoto, Japan) equipped with a flame ionization detector (FID) and an electron capture detector (ECD) to determine the concentrations of CH 4  The global warming potential (GWP) was calculated using the following equation.

The DNDC-Rice Model
The DNDC-Rice model consists of three major submodels that simulate soil climate, crop growth, and soil biogeochemistry. The features and scientific background of the DNDC-Rice model are given by Fumoto et al. [17] and all the input parameters are listed in Fumoto et al. [19]. In this study, the site mode of the DNDC-rice model was tested for CH 4 and N 2 O emissions under different water management practices during four rice-growing seasons.
The DNDC-Rice model incorporated the Modules of an Annual Crop Simulator (MACROS) model of rice physiology [23] into its crop growth submodel. The original codes of MACROS, written in the simulation language Continuous System Modelling Program (CSMP) and provided as text in literature [23], were rewritten in C++ to incorporate into DNDC-Rice. Crop physiology and phenology are simulated on the basis of nitrogen availability and the environments above and below the ground [17]. In a recent revision, the mechanistic description of photosynthesis [24] was added to the crop growth submodel as mentioned by Minamikawa et al. [20]. Methane flux is calculated by the fermentation submodel. Under anaerobic conditions, the model calculates the production of hydrogen (H 2 ) and dissolved organic carbon (DOC), which are used as the electron donors for the subsequent reduction of Mn, Fe, and S oxides and CH 4 production. Nitrous oxide production is calculated by nitrification and denitrification processes. Emission of N 2 O from the soil surface is calculated as a function of soil N 2 O content, air-filled porosity, temperature, and clay content.
A preliminary run of DNDC-Rice is essential to achieve a near-steady state for soil carbon pools before the start of the simulation [17]. We ran the model for a time period of 20 years, with constant inputs of weather conditions and agricultural management practices for double-cropping paddy rice per year practiced at TRRI, Aduthurai, India. The datasets of soil, climate, and crop management practices were collected at the experimental site to run the model. DNDC-Rice can explicitly calculate volumetric soil moisture and matric potential, but not the underground water level. To simulate the irrigation under AWD of this study, therefore, the codes were adjusted to assume a condition so that the field is re-flooded when calculated matric potential is lowered to −20 kPa at the depth of 15 cm.

Statistical Analysis
The simulation result of CH 4 and N 2 O fluxes were evaluated by using the root mean square error (RMSE) with the following equation: where Fi is simulated value i, Ai is observed value i, and N is the number of samples. Relative variation between the observed and simulated values were calculated by using the following equation by Katayanagi et al. [21]:

Rice Growth
In all rice-growing seasons, the observed grain yields in the CF and AWD treatments did not show a significant difference ( Table 2). Other studies have also reported no yield losses when implementing AWD irrigation compared to CF [7,9]. The results showed that water-saving irrigation is feasible in double-cropping paddy rice in the tropical region without affecting rice grain yield. There is no necessity to maintain continuous standing water throughout the rice-growing season since irrigated rice had developed adaptability to the intermittently flooded conditions [25]. The DNDC-Rice model overestimated rice grain yield by 24% under CF and 29% under AWD on average over the rice-growing seasons compared to observed ones ( Table 2). In contrast, it apparently underestimated the straw biomass under CF and AWD. To simulate rice growth, DNDC-Rice partitions photosynthetic product to different organs (root, stems, leaves, and panicles) depending on the growth stage, according to cultivar-specific functions that were calibrated for a number of rice cultivars. For the Indian cultivars used in this study (ADT 43 and ADT 46), however, we could not obtain adequate datasets (i.e., biomass of each organ measured at different growth stages) required for calibrating the cultivar-specific functions. Beside the limited data availability, the major objective of this study was to validate the DNDC-Rice model in predicting CH 4 and N 2 O emissions under different irrigation practices. Therefore, we did not conduct further calibration of the cultivar-specific functions in this study. In order to accurately estimate rice grain yield and straw biomass under CF and AWD irrigations in the tropical region, however, the DNDC-Rice model will need calibration of its functions that determine the partitioning of photosynthetic product in cultivars grown in the region of interest.

Soil Redox Status and Methane Emissions
The field observation results showed that the soil Eh was as low as -150 mV during the early growth period of the summer season, and then it showed an increasing trend toward the end of the growing period (Figure 2). After the start of AWD irrigation, the soil Eh value showed an increasing trend and was always higher than that of CF treatment. The model predicted the season pattern of the soil Eh value well in CF, but it failed for AWD irrigation due to the overresponse of the model to the drying period during the alternate wetting and drying period.
When soil contains O 2 , DNDC-Rice simulates soil Eh (mV) as a function of the soil O 2 concentration, (O 2 ) (mol kg −1 soil), according to the formula, When soil O 2 has been depleted, in turn, soil Eh is simulated using empirical functions that relate soil Eh to reduction of soil Fe and S [17]. To analyze the behavior of simulated soil Eh, we examined simulated (O 2 ) (at the depth of 5 cm) during the AWD irrigation in the summer season of 2016 and found that it was mostly zero during the wetting periods, but increased to about 0.2-0.9 mmol kg −1 soil during the drying periods, which was about 4-20% of the (O 2 ) level during the most aerobic period between rice-growing seasons. Consequently, simulated soil Eh jumped up to around 400 mV during the drying periods, according to the above formula. If the simulated (O 2 ) is reasonable, therefore, it is suggested that the above function of (O 2 ) is not appropriate for simulating soil Eh during AWD irrigation. Unlike earlier versions of DNDC (e.g., Babu et al. [13]), however, soil Eh does not directly affect CH 4 production in DNDC-Rice, where CH 4 production is explicitly limited by the availability of electron donors (H 2 and dissolved organic carbons) in competition with the alternative electron acceptors (Fe, Mn, and S) [17], instead of applying simulated soil Eh as the threshold for CH 4 production. We expect, therefore, that the over-responding soil Eh did not affect the simulated CH 4 emissions, even though it did not match the observed soil Eh.
Agriculture 2020, 10, x FOR PEER REVIEW 7 of 16 therefore, it is suggested that the above function of (O2) is not appropriate for simulating soil Eh during AWD irrigation. Unlike earlier versions of DNDC (e.g., Babu et al. [13]), however, soil Eh does not directly affect CH4 production in DNDC-Rice, where CH4 production is explicitly limited by the availability of electron donors (H2 and dissolved organic carbons) in competition with the alternative electron acceptors (Fe, Mn, and S) [17], instead of applying simulated soil Eh as the threshold for CH4 production. We expect, therefore, that the over-responding soil Eh did not affect the simulated CH4 emissions, even though it did not match the observed soil Eh. The field observation results showed that the seasonal variations of CH4 fluxes were significantly lower in AWD compared to CF treatment in all rice-growing seasons (Figure 3). Under AWD irrigation, reduction in the irrigation water volume led to a lower surface standing water depth and even no standing water above the surface of the soil, which increased oxygen penetration into the soil and led to soil organic carbon being oxidized and suppressed CH4 emissions [26].  The field observation results showed that the seasonal variations of CH 4 fluxes were significantly lower in AWD compared to CF treatment in all rice-growing seasons (Figure 3). Under AWD irrigation, reduction in the irrigation water volume led to a lower surface standing water depth and even no standing water above the surface of the soil, which increased oxygen penetration into the soil and led to soil organic carbon being oxidized and suppressed CH 4 emissions [26]. With respect to the seasonal variability of CH4 fluxes, high flux was often observed during the early growth stage under both the CF and AWD treatments (Figure 3). The higher CH4 emissions during the early rice-growing season were attributed to high soil temperature and low soil redox potential during that period [3,27]. However, the DNDC-Rice model tended to underestimate the CH4 fluxes during the early growth stage. Presumably, this was caused because the model failed to predict the reductive soil conditions at the early growth stage, as indicated by comparing observed and simulated soil Eh (Figure 2). Minamikawa et al. [6] also reported that the underestimates by the model during the early growing season were mainly due to the unsuccessful prediction of the development of reductive conditions at the early growth stage since soil redox status before cultivation is important in determining the subsequent CH4 emission in the model.
Under CF conditions, the average rate of observed and simulated CH4 fluxes was 0.75 and 0.70 kg C ha −1 d −1 in the summer and 1.17 and 1.29 kg C ha −1 d −1 in the monsoon season, respectively ( Table  3). The RMSE values for the simulated CH4 fluxes in the CF were 0.81 and 1.23 kg C ha −1 d −1 in the With respect to the seasonal variability of CH 4 fluxes, high flux was often observed during the early growth stage under both the CF and AWD treatments (Figure 3). The higher CH 4 emissions during the early rice-growing season were attributed to high soil temperature and low soil redox potential during that period [3,27]. However, the DNDC-Rice model tended to underestimate the CH 4 fluxes during the early growth stage. Presumably, this was caused because the model failed to predict the reductive soil conditions at the early growth stage, as indicated by comparing observed and simulated soil Eh (Figure 2). Minamikawa et al. [6] also reported that the underestimates by the model during the early growing season were mainly due to the unsuccessful prediction of the development Agriculture 2020, 10, 355 9 of 16 of reductive conditions at the early growth stage since soil redox status before cultivation is important in determining the subsequent CH 4 emission in the model.
Under CF conditions, the average rate of observed and simulated CH 4 fluxes was 0.75 and 0.70 kg C ha −1 d −1 in the summer and 1.17 and 1.29 kg C ha −1 d −1 in the monsoon season, respectively ( Table 3). The RMSE values for the simulated CH 4 fluxes in the CF were 0.81 and 1.23 kg C ha −1 d −1 in the summer and monsoon season, respectively. The average observed daily CH 4 fluxes and RMSE values in this study fall within the simulated flux range from 0.09 to 1.4 kg C ha −1 d −1 and RMSE values from 0.16 to 1.17 kg C ha −1 d −1 from paddy fields in Japan and China [17]. Although the model underestimated the early seasonal emissions, the agreement between the average daily observed and simulated CH 4 fluxes was good under CF conditions in all rice-growing seasons (Table 3, Figure 3). The DNDC-Rice model reproduced the suppressive effect of AWD treatment on CH 4 emission well in all rice-growing seasons (Table 3, Figure 3). Under AWD conditions, the average rate of observed and simulated CH 4 fluxes was 0.35 and 0.36 kg C ha −1 d −1 in the summer and 0.79 and 0.83 kg C ha −1 d −1 in the monsoon season, respectively. The RMSE values for the simulated CH 4 fluxes in the AWD were 0.40 and 0.93 kg C ha −1 d −1 in the summer and monsoon season, respectively. According to our knowledge, this is the first report of validation of the DNDC-Rice model under water-saving AWD irrigation in double-cropping paddy rice under field conditions, although other studies have used the DNDC-Rice model to estimate CH 4 emissions under mid-season drainage and intermittent irrigation [6,17,19,20]. The results of their studies stated that the DNDC-Rice model represents a valuable tool for estimating CH 4 emission from paddy rice soil under mid-season drainage and intermittent irrigation.
The previous study, conducted by Katayanagi et al. [21], validated the DNDC-Rice model for tropical rice paddies in Philippine under AWD irrigation management in a pot experiment. Their result showed that the model simulated the temporal variability of CH 4 fluxes for CF and AWD pots well with the average observed daily CH 4 fluxes of 4.49 and 1.22 kg C ha −1 d −1 , respectively, and the RMSE values of 1.76 and 1.86 kg C ha −1 d −1 . The simulated RMSE values for the simulated CH 4 fluxes under CF and AWD irrigation practices in this study were comparable to the values from rice soil in the Philippines. The results highlighted that the DNDC-Rice model is suitable for estimation of CH 4 fluxes not only for conventional water management techniques also for water saving conditions in double-cropping paddy rice in major rice growing areas in the tropical region.

Nitrous Oxide Emissions
The field observation results showed that the seasonal variations of N 2 O fluxes were relatively higher in AWD compared to CF treatment in all rice-growing seasons (Figure 4). Under continuously flooded conditions, the consistently low soil Eh (Figure 2) resulted in complete denitrification, and consequently reduced N 2 O emission [3]. Ussiri and Lal [28] discussed that prolonged flooding promotes the development of strong anaerobic conditions in soils, reducing any N 2 O produced in the paddy fields to N 2 . The increase in N 2 O emissions from AWD treatments under N fertilization was due to the abundant N supply and the suitable soil moisture conditions due to successive moist and dry periods during the rice-growing season.
Under CF conditions, the seasonal variability of N 2 O fluxes was simulated reasonably by the DNDC-Rice model in all rice-growing seasons (Figure 4). The DNDC-Rice model simulated near zero N 2 O emission from the flooded rice soils throughout the rice-growing season and peak emission was observed towards the maturity of the crop after water was drained from the field. Babu et al. [13] tested the DNDC model in wide regions of India. They discussed that the influence of the rhizosphere on the ecological drivers is not yet incorporated in the model, so the model simulates flooded anoxic soils with suppressed rates of nitrification, leading to zero N 2 O emissions in continuously flooded rice fields. The average rates of observed and simulated N 2 O fluxes in CF were 6.3 and 8.1 g N ha −1 d −1 in the summer and 8.2 and 7.4 g N ha −1 d −1 in the monsoon season, respectively ( Table 3). The RMSE values for the simulated N 2 O fluxes in the CF were 18.7 and 24.3 g N ha −1 d −1 in the summer and monsoon season, respectively.
Although the seasonal variability of N 2 O fluxes was simulated reasonably under AWD, the model overestimated N 2 O emissions after the additional nitrogen fertilization in all rice-growing seasons (Figure 4). When the soil is well aerated under AWD irrigation, the oxidation, i.e., nitrification, of available nitrogen dominates and NO is the most common gas emitted from the soil instead of N 2 O [29], and therefore the observed emission peaks after additional fertilization were lower compared with the simulated one. Moreover, frequent aeration under AWD significantly increased soil redox conditions up to +485, which might be overestimated by the model. Under actual field conditions, although an increase in soil redox potential was observed after introducing the drying period in AWD, the soil was still saturated, and therefore the soil redox potential did not reach positive values ( Figure 2). As a result, the model overestimated soil N 2 O emissions compared to observed ones. The average rates of observed and simulated N 2 O fluxes were 15.3 and 39.1 g N ha −1 d −1 in the summer and 9.7 and 26.8 g N ha −1 d −1 in the monsoon season, respectively (Table 2). High RMSE values of 60.7 and 59.6 g N ha −1 d −1 in the summer and monsoon season, respectively, stated that the model poorly predicted N 2 O emissions under AWD irrigation.
In previous applications of the DNDC-Rice model to tropical rice soil in The Philippines [21], the simulated and observed N 2 O emissions from the AWD pots were higher than those from the CF pots, but the DNDC-Rice model could not predict the timing and magnitude of the high N 2 O pulses which created a higher RMSE for AWD irrigation (124 g N ha −1 d −1 ) than for CF (2.23 g N ha −1 d −1 ). In this study, the DNDC-Rice model predicted high magnitude N 2 O peaks after additional nitrogen fertilization in AWD treatment in all rice-growing seasons. This might be due to overestimation of soil nitrification under frequent soil aeration in AWD-related high soil redox values (Figure 4), since N 2 O production in paddy rice soils was mainly regulated by nitrification [21]. flooded conditions, the consistently low soil Eh (Figure 2) resulted in complete denitrification, and consequently reduced N2O emission [3]. Ussiri and Lal [28] discussed that prolonged flooding promotes the development of strong anaerobic conditions in soils, reducing any N2O produced in the paddy fields to N2. The increase in N2O emissions from AWD treatments under N fertilization was due to the abundant N supply and the suitable soil moisture conditions due to successive moist and dry periods during the rice-growing season.

Cumulative Emissions and Total Global Warming Potential
The average observed and simulated cumulative CH 4 emissions in CF were 73.7 and 69.0 kg C ha −1 , respectively, with a relative variation of −6.5% during the summer season and 131.0 and 135.6 kg C ha −1 with the variation of 4.2% during the monsoon season ( Table 4). The simulated emissions for CF were 6.4% lower in the summer season and 3.5% higher in the monsoon season than the corresponding observed values.
The average observed and simulated cumulative CH 4 emissions in AWD were 38.5 and 35.2 kg C ha −1 , respectively, with the variation of −9.3% during the summer season and 99.4 and 87.3 kg C ha −1 with the variation of −12.7% during the monsoon season ( Table 4). The simulated emissions for AWD were 8.6% and 12.2% lower than the observed ones in the summer and monsoon, respectively. Overall, the DNDC-Rice model reasonably estimated the total CH 4 emission in CF and reproduced the suppressive effect of AWD treatment on CH 4 emission well (Figure 5a). Katayanagi et al. [21] tested the DNDC-Rice model by using the data from The Philippines under CF and AWD conditions. They observed that the simulated emissions for CF and AWD were 9.8% lower and 0.76% higher, respectively, than the observed values. In this study, low variations between the observed and simulation values for CF and AWD indicated that the DNDC-Rice model simulated CH 4 emission well. Thus, the model can be used for the estimation of CH 4 emissions under both water management conditions in the double-cropping paddy rice system in the tropical regions. Previous studies also demonstrated the advantage of using DNDC-Rice for estimating the general effect of midseason drainage or intermittent drainage on CH 4 reduction instead of conducting the corresponding long-term field experiments [6,30].
The averaged observed and simulated cumulative N 2 O emissions in CF were 0.72 and 0.79 kg N ha −1 , respectively, with the relative variation of 27.3% during the summer and 0.59 and 0.42 kg N ha −1 with the relative variation of −35.3% during the monsoon season ( Table 4). The simulated emissions for CF were 9.7% higher in the summer and 28.8% lower than the observed value in the monsoon season.
The average observed and simulated cumulative N 2 O emissions in AWD were 1.1 and 3.8 kg N ha −1 , respectively, with the variation of 267.4% during the summer season and 1.18 and 2.82 kg N ha −1 with the variance of 233.8% during the monsoon season ( Table 4). The simulated emissions for N 2 O were 245.5% and 139.0% higher in the summer and monsoon season, respectively, than the observed values. The result showed that a negative or positive effect of CF and AWD irrigations on N 2 O emissions observed in the measurement was not adequately reproduced by the model (Figure 4). This result was also supported by the correlation analysis (Figure 5b). Katayanagi et al. [21] observed that the simulated N 2 O emissions for CF and AWD were 87% and 29% lower, respectively, than the observed values. High range of estimation error value in this study (−35.9% to +514.1%) was comparable to the error values that ranged from −220% to +28.6% [13] and from −66% to +265% [19] and it was hypothesized that these errors were caused by inaccurate estimation of nitrogen release rates from fertilizers, including coated urea. The averaged observed and simulated cumulative N2O emissions in CF were 0.72 and 0.79 kg N ha −1 , respectively, with the relative variation of 27.3% during the summer and 0.59 and 0.42 kg N ha −1 with the relative variation of −35.3% during the monsoon season ( Table 4). The simulated emissions for CF were 9.7% higher in the summer and 28.8% lower than the observed value in the monsoon season.
The average observed and simulated cumulative N2O emissions in AWD were 1.1 and 3.8 kg N ha −1 , respectively, with the variation of 267.4% during the summer season and 1.18 and 2.82 kg N ha −1 with the variance of 233.8% during the monsoon season ( Table 4). The simulated emissions for N2O were 245.5% and 139.0% higher in the summer and monsoon season, respectively, than the observed values. The result showed that a negative or positive effect of CF and AWD irrigations on N2O emissions observed in the measurement was not adequately reproduced by the model (Figure 4). This result was also supported by the correlation analysis (Figure 5b). Katayanagi et al. [21] observed that the simulated N2O emissions for CF and AWD were 87% and 29% lower, respectively, than the observed values. High range of estimation error value in this study (−35.9% to +514.1%) was comparable to the error values that ranged from −220% to +28.6% [13] and from −66% to +265% [19] and it was hypothesized that these errors were caused by inaccurate estimation of nitrogen release rates from fertilizers, including coated urea.
The average observed and simulated total GWP in CF were 3678 and 3498 kg CO2 eq ha −1 in the summer season and 6210 and 6342 kg CO2 eq ha −1 in the monsoon season, respectively ( Table 4). The simulated emissions for CF were 4.9% lower in the summer season and 2.1% higher than the observed values in the monsoon season. The average observed and simulated total GWP in AWD were 2260 of N 2 O to the GWP compared with that of CH 4 , it is less important to modify the model to account for N 2 O emission from paddy rice fields for estimation of total GWP.

Conclusions
This study is the first attempt for field validation of the DNDC-Rice model by using the observed CH 4 and N 2 O emissions data from double-cropping paddy rice under continuous flooding and water-saving irrigation in Tamil Nadu, India. The model predicted cumulative CH 4 emissions and total GWP for CF and AWD treatments for all rice-growing seasons well. However, there were some discrepancies between observed and simulated daily CH 4 fluxes at the beginning of the growing season, indicating that the model was less successful in predicting seasonal pattern of emissions during the rice-growing season. Due to high fluctuation in the soil Eh value during the drying period of AWD irrigation, the model needs to be improved for calculation of soil Eh in response to soil aeration, though soil Eh does not directly influence CH 4 emissions in simulation by this model. Moreover, further modification of the nitrification and denitrification rates under AWD irrigation will be needed for reasonable prediction of N 2 O emissions from double-cropping paddy rice under frequent soil aeration in tropical rice production.