Update a biogeochemical model with process-based 1 algorithms to predict ammonia volatilization from 2 fertilized uplands and rice paddy fields 3

Accurate simulation of ammonia (NH3) volatilization from fertilized croplands is crucial to 13 enhancing fertilizer-use efficiency and alleviating environmental pollution. In this study, a 14 process-oriented model, CNMM-DNDC (Catchment Nutrient Management Model 15 DeNitrification-DeComposition), was evaluated and modified using NH3 volatilization observations 16 from 44 and 19 fertilizer application events in cultivated upland areas and paddy rice fields in China, 17 respectively. The original CNMM-DNDC model not only performed poorly in simulating NH3 18 volatilization from upland areas but also failed to simulate NH3 volatilization from paddy rice fields. In 19 the modified CNMM-DNDC model, the major modifications for simulating NH3 volatilization from 20 uplands were primarily derived from a peer-reviewed and published study. NH3 volatilization from 21 uplands was jointly regulated by the factors of wind speed, soil depth, clay fraction, soil temperature, 22 soil moisture, vegetation canopy, and rainfall-induced canopy wetting. Moreover, three principle 23 modifications were made to simulate NH3 volatilization from paddy rice fields. First, the simulation of 24 the floodwater layer and its pH were added. Second, the effect of algal growth on the diurnal 25 fluctuation of floodwater pH was introduced. Finally, the Jayaweera-Mikkelsen model was introduced 26 to simulate NH3 volatilization. The modified model showed remarkable performances in simulating the 27 cumulative NH3 volatilization of the calibrated and validated cases, with drastically significant 28 zero-intercept linear regression of slopes of 0.94 (R = 0.76, n = 40) and 0.98 (R = 0.71, n = 23), 29 https://doi.org/10.5194/bg-2021-341 Preprint. Discussion started: 17 January 2022 c © Author(s) 2022. CC BY 4.0 License.


Introduction 36
Synthetic fertilizer application, as the secondary largest contributor to ammonia (NH 3 ) emissions 37 after livestock production, accounts for approximately 30% to 50% of anthropogenic NH 3 emissions 38  Park et al., 2008). However, these models do not distinguish between the simulation modules of 50 NH 3 volatilization for uplands and rice paddy fields but rather use the same algorithm (Cannavo et al.,51 2008; Li, 2016). It is worth emphasizing that the mechanisms of NH 3 volatilization are completely 52 different between fertilized uplands and rice paddy fields due to the presence of floodwater over rice 53 paddy soils. Recent studies also indicate that estimating NH 3 emissions without considering rice 54 cultivation results in large uncertainties (Riddick et  biogeochemical models (e.g., DNDC) adopted scientific processes and algorithms applied in simulating 191 NH 3 volatilization from fertilized cultivated uplands to calculate NH 3 volatilization from rice paddy 192 fields without considering floodwater over soils (Cannavo et al., 2008;Li, 2016). Therefore, floodwater 193 over rice paddy soils was added to the modified CNMM-DNDC. To add this component, the modified 194 CNMM-DNDC adopted the Jayaweera-Mikkelsen model (i.e., J-M model), based on the two-film 195 theory of mass transfer (Jayaweera and Mikkelsen, 1990a), which is one of the most widely applied 196 process-based models for simulating NH 3 volatilization from rice paddy fields. The J-M model consists 197 of two processes ( Fig. 2): (i) the chemical processes of NH 4 + ions and aqueous NH 3 (NH 3(aq) ) 198 equilibrium in floodwater and (ii) the volatilization processes of NH 3(aq) transfer in the form of NH 3 gas 199 (NH 3(air) ) across the water-air interface to the atmosphere (Rxn1). k d (first-order, s −1 ) and k a 200 (second-order, L mol −1 s −1 ) are referred to as the dissociation and association rate constants for 201  Jayaweera and Mikkelsen (1990a). 208 In the modified CNMM-DNDC, the pH of the floodwater, which is the negative logarithm of 209 [H + ] w , is related to the initial pH of water for flooding and that of surface soil. When the floodwater 210 depth is less than 0.04 m, the pH of the floodwater is equal to the mean of the initial pH of water for 211 flooding and that of surface soil, both of which are the inputs of the modified model. Otherwise, the pH 212 of the floodwater is equal to the initial pH of the water for flooding. On the one hand, [H + ] w is regulated 213 by urea hydrolysis in floodwater, the algorithm of which was derived from that of urea hydrolysis 214 affecting soil pH in the model. On the other hand, many studies have found that a marked diurnal 215 fluctuation in floodwater pH is associated with algal photosynthesis, which was elevated with solar 216 radiation (De Datta, 1995;Fillery and Vlek, 1986). Therefore, a ratio of the daytime solar shortwave 217 radiation effect on algal photosynthesis (R slr , 0−1) was established by the authors using Eq. (4) as a 218 quadratic function of the simulation time (t, 06:00 to 21:00 with a 3-hour interval) of a day. R slr at the 219 other moments with no or extremely little solar radiation in a day was set as 0. The effect of algal 220 growth (f alg ) on floodwater pH was calculated by Eq. (5), where the adjusted coefficient (k alg , 0−1) was 221 calibrated to 0.75 or 0.6 when the floodwater depth was no more than or more than 0.04 m, respectively. 222 The floodwater pH of (t+1)th was modified by the floodwater pH of tth and f alg using Eq. 6, which was 223 set as no more than 10. factors of soil organic carbon and soil moisture. Therefore, urea hydrolysis in floodwater is only 230 determined by the floodwater temperature. To simplify the calculation, the floodwater temperature is 231 arbitrarily set equal to the temperature in the first soil layer in the modified model. Given that ABC 232 decomposition in floodwater was not involved in the original CNMM-DNDC, this study directly 233 adopted the algorithm of ABC decomposition in upland soils used in Li et al. (2019), and this process 234 was regulated by soil temperature, pH and the applied depth of fertilizer. However, ABC decomposition 235 in floodwater is different from that in upland soils; i.e., the ABC concentration is uniformly distributed 236 in the floodwater, and the effect factors (i.e., temperature, pH and depth) applied should be those of 237 floodwater rather than those of soil. Therefore, this study ignored the effect of soil depth and retained 238  (Jayaweera and Mikkelsen, 1990a). As shown in Eq. (7), k a is affected by its 245 relationship with floodwater temperature (T f , K) based on Alberty (1983): 246 11 9 6 2 a f f =3.8 10 3.4 10 7.5 10 k d is derived from the relationship with the equilibrium constant for NH 4 + /NH 3(aq) (K) and k a (Eq. 247 can be determined using the model input of wind speed measured at a height of 10 m (U 10 , m s −1 ), 260 based on Eq. (15) derived from Jayaweera and Mikkelsen (1990a The CNMM-DNDC with the above modifications is hereinafter referred to as the modified 265 CNMM-DNDC. The cases for model calibration were identified on the basis of covering as many 266 climate conditions, soil properties and management practices as possible. Therefore, for the simulation 267 of NH 3 from urea application on uplands and rice paddy fields, 26 typical cases of DBW, FQU, and QZ 268 and 10 typical cases of DY, FQP, and SZ were used for model calibration. Regarding the simulation of 269 NH 3 from the ABC application on uplands and rice paddy fields, 3 typical cases of DBW and YT and 1 270 typical DY case were conducted for model calibration. The remaining 23 independent cases were 271 provided for model validation. For the meteorological data inputs, the reported 3-hourly meteorological data from the weather 282 station at the experimental site were used. If these data were not available, then data from the adjacent 283 weather station in the China Meteorological Administration (CMA, http://www.data.cma.cn) were 284 adapted by referring to the reported average or maximum values (Table S1, Text S1). 285 The necessary inputs of surface soil properties at the individual upland sites for the modified 286  Table S4, respectively. 314

Sensitivity analysis 315
Sensitivity analysis was adopted to investigate the regulating factors in the modified 316 CNMM-DNDC that simulates NH 3 volatilization following fertilizer application. Meteorological 317 variables (i.e., 3-hourly averages of T air and W; 3-hourly totals of P and R during measurement periods 318 of NH 3 volatilization), soil properties (i.e., soil clay fraction, pH, SOC content and BD), and field 319 management practices (i.e., water management (irrigation water amount or depth of floodwater) and 320 nitrogen fertilization type, dose and depth) were involved in this sensitivity test. U37 in QZ and P4 in 321 CS were chosen as the baseline cases to assess the model's behavior in simulating NH 3 volatilization 322 from uplands and rice paddy fields, respectively. One reason for this selection was that U37 and P4 323 were geographically located near the center of the region for upland and rice paddy cases, respectively. 324 Another reason was that the selected cases implement general Chinese management practices. The 325 authors altered only one item at a time by maintaining the others constant. The model input items of the 326 3-hourly average of W, 3-hourly totals of P and R during the measurement periods of NH 3 327 volatilization, as well as the soil clay fraction, SOC content, nitrogen fertilization dose, and depth of 328    Table  369 2), with only 16% (seven of forty-four) of cases suffering from an |RMB| larger than 100%. 370

Ammonia volatilization from rice paddy fields 371
The observed CAVs in all cases of rice paddy fields (2 and 17 cases for ABC and urea applications, 372 respectively) during the measurement periods totaled 5.9-39.8 kg N ha −1 (mean: 18.1 kg N ha −1 , Table  373 2), with fertilizer application doses of 40.5-162.2 kg N ha −1 (mean: 81.4 kg N ha −1 ). Given the lack of 374 the capacity to simulate the water-flooded layer over rice paddy fields, the original CNMM-DNDC 375 could not simulate NH 3 volatilizations from rice paddy fields. The corresponding CAVs simulated by 376 the modified CNMM-DNDC totaled 3.4-39.1 kg N ha −1 (mean: 16.2 kg N ha −1 , Table 2 floodwater temperature (Fig. 5a-b), although the simulated floodwater temperatures of several certain 406 days for P9 were lower than the observations. The modified CNMM-DNDC, which introduced the 407 effect of algal growth on floodwater pH, generally simulated the observed daily elevated floodwater pH 408 resulting from algal photosynthetic activity (Fig. 3a-b and Fig. 5c-e). The simulation of calibrated (P1, 409 P9 and P10, Fig. 3a and Fig. 5c-d) and validated cases (P2 and P19, Fig. 3b and Fig. 5e) Fig. 3c-d, Fig. 4a-b and Fig. 5f-h. Compared to the observed floodwater NH 4 + 414 concentrations of the ABC cases, the model simulation underestimated the peak concentration on the 415 first day after ABC application for the P1 case but captured the peak concentration of the P2 case (Fig.  416 3c-d). The modified CNMM-DNDC generally captured the observed temporal pattern in the daily 417 NH 4 + concentrations during the observation periods following urea application, although discrepancies 418 existed in the magnitudes of some cases; e.g., the model overestimated the floodwater NH 4 + 419 concentration in the P7 and P8 cases ( Fig. 4a-b) and underestimated that in the P6 (Fig. 4b) and P19 420 cases (Fig. 5h) (Table 4). 424

Regulating factors of the modified model in simulating ammonia volatilization 425
The sensitivity analysis indicated that NH 3 volatilization from upland soils was primarily 426 regulated by field management practices (Fig. 7a). The changes in N dose, the different N types and the 427 implementation of irrigation had considerable effects on NH 3 volatilization from upland soils. In 428 addition, a fertilization depth of 15 cm resulted in a -23% change in NH 3 volatilization, and the 429 increase in irrigation amount had an inhibitory effect on NH 3 volatilization. Moreover, in comparison to 430 other soil properties, the changes in soil SOC had a greater influence (-19% to 16%) on NH 3 431 volatilization. Among all considered meteorological variables, NH 3 volatilization from upland soils 432 appeared to be the most sensitive response to changes in air temperature (Fig. 7a). However, NH 3 433 volatilization from rice paddy soils was sensitive to changes in fertilization and floodwater 434 management, which increased with N dosage and decreased with the depth of fertilizer application and 435 that of floodwater (Fig. 7b). For all soil variables considered in the sensitivity analysis, only the 436 changes in soil pH had a great influence on NH 3 volatilization from rice paddy soils. In addition, NH 3 437 volatilization from rice paddy soils decreased with solar radiation. Vlek, 1986; Mikkelsen et al., 1978). The addition of a suitable photosynthetic inhibitor also controlled 499 the pH of the floodwater, implying that the increase in pH was caused by algal growth (Bowmer and 500 Muirhead, 1987). Therefore, more observational data on the effect of algal growth on floodwater pH and 501 subsequent NH 3 volatilization are needed to improve the simulation of the modified model on NH 3 502 volatilization from rice paddy fields. 503 Many studies have found that the depth of surface floodwater has a substantial influence on NH 3 504 volatilization (Fillery et al., 1984;Freney et al., 1988;Hayashi et al., 2006). The sensitivity analysis of 505 this study also indicated that NH 3 volatilization from rice paddy fields was sensitive to changes in the 506 depth of surface floodwater (Fig. 7b). Jayaweera and Mikkelsen (1990b)

Differences between ammonia volatilization from upland and rice paddy fields 529
According to the above results, the regulatory factors affecting NH 3 volatilization from rice paddy 530 fields were demonstrated to be different from those from cultivated uplands. NH 3 volatilization from 531 cultivated uplands was primarily influenced by the regulatory factors of soil properties and field 532 management practices. However, given the existence of floodwater over rice paddy field soils, NH 3 533 volatilization from rice paddy fields was additionally affected by flooding management strategies, such 534 as floodwater pH and depth. Therefore, the mechanisms and algorithms applied in simulating NH 3 535 volatilization from uplands are not appropriate for simulating NH 3 volatilization from rice paddy fields. 536 In the modified CNMM-DNDC, NH 3 volatilization following nitrogen fertilizer application in 537 cultivated upland soils was based on first-order kinetics. However, the modified CNMM-DNDC 538 adopted the J-M model, which was based on the two-film theory of mass transfer, to calculate NH 3 539 volatilization following nitrogen fertilizer application in rice paddy field soils. The results suggest that 540 the application of two different mechanisms according to the distinguished properties of cultivated 541 uplands and rice paddy fields to simulate NH 3 volatilization is necessary for process-based 542 biogeochemical models, such as the CNMM-DNDC used in this study. 543   and Yingtan (YTA). c O and S are the cumulative NH 3 volatilization (kg N ha −1 ) observed and simulated by the modified CNMM-DNDC, respectively; RMB is the relative model bias (%) of the modified model, each of which was determined as the relative difference between the simulated and observed values. d The depth of floodwater table (cm). For the cases "*" and " # ", the exact depth of the floodwater 850