Improving nitrogen management via a regional management plan for Chinese rice production

In China, rice is grown primarily in four major agroecological regions and eight agroecological subregions: Northeast China (NE1, NE2), the Yangtze River Valley (YRV: Upper Yangtze River, UYR; Middle Yangtze River, MYR; Lower Yangtze River, LYR), South China (SC1, SC2), and Southwest China (SW; figure 1). The total area of rice production in the eight subregions is approximately 29 million ha (table S1), which represents 96% of the total rice production in China. Detailed information for location, weather and rice management on each of these subregions are provided in text S1.

We used a multistage random sampling technique to select representative farmers for a face-to-face, questionnaire-based household survey conducted in 2008 and 2009. Surveys were carried out in all major rice production regions throughout the agroecological regions of China. In total, 6611 farmers from 55 counties in 21 rice-producing provinces were surveyed to collect information about N fertilizer application rates and rice yields (table 1). In each province, two to three counties were randomly selected, two townships were randomly selected from each county, two to four villages were randomly selected as sample villages from each township, and 20–25 farmers from the villages were randomly surveyed.

In total, 1177 on-farm N fertilizer experiments were conducted in 173 counties of 21 provinces from 2005 to 2010 in the NE (n = 136), YRV (n = 543), SC (n = 334), and SW (n = 164) regions (table 2). All of the counties included in the farmer surveys were also part of these 173 counties.

All experimental fields received the same treatments without replication: no N fertilizer (N0), medium N application rate (MN), 50% MN, and 150% MN. The MN rate was determined by local agricultural extension employees according to their experience and target yield (1.1 times the average yield of the past 5 years). All experiments received 97–270 kg N ha⁻¹. The granular urea was applied 40% at sowing or transplanting, 25% at the tillering stage, and the remainder at the panicle differentiation stage. All experimental fields also received 30–180 kg P₂O₅ ha⁻¹ and 24–165 kg K₂O ha⁻¹ (table S2), which were applied by broadcasting before sowing or transplanting rice. No manure was used.

Individual plots were about 50 m² (5 m × 10 m). The management of all 1177 on-farm experiments, including sowing or transplanting rice, and control of diseases, insect pests and weeds, was done by local farmers according to their experience. Local farmers determined the timing of planting or transplanting, the variety of rice, and the time of harvest. At the time of harvest, a 2.5 m × 4 m section was harvested from each experimental plot to measure yield. The grain yield was adjusted to a moisture content of 14%.

A RONR was calculated for each subregion through several steps. First, a quadratic regression model was used to assess the grain yield response to N application rate for the 1177 on-farm N response trials using PROC NLIN of SAS (version 9.1.3; SAS Institute Inc. 2006), showing that yield significantly (P < 0.05) responded to the N rate (Wallach and Loisel 1994). Next, from the response curve equation for each experiment, we calculated the following variables for each 1 kg N increment from 0 to 250 kg N ha⁻¹: yield increase (amount above the yield in the N0 treatment), the gross return in Chinese yuan for the yield increase (rice grain yield times grain price), the N fertilizer cost (N application rate times fertilizer price), and the net return on N application (gross return in Chinese yuan minus the cost of N fertilizer). Finally, for each incremental N rate, we averaged the net return across all N response models for each subregion and rotation type. The N application rate with the greatest average net return on investment was defined as the RONR for rice in each subregion (Sawyer et al 2006). A range of N rates around the RONR rate can be defined to be within the economically profitable N rate range (PNRR) when it results in a net return to N within ±14.87 yuan ha⁻¹ (US$1.00 acre⁻¹; Sawyer et al 2006) can be defined as the smallest and largest profitable N rate (PNRR low, PNRR high), respectively. N application rates outside this range were defined as under- or overapplication in each subregion.

The partial factor productivity (PFP) was defined the ratio of crop yield per unit of applied N fertilizer (Dobermann 2005). The partial N balance was defined as N application minus above-ground N uptake (Zhang et al 2013). The details of these two parameters were shown in text S2.

Total GHG emissions during the entire life cycle of rice production, including CO₂, methane (CH₄), and nitrous oxide (N₂O), comprise three components: GHG emissions during and following N, P, and K fertilizer application, production, and transportation; GHG emissions from pesticide and herbicide production (delivered to the gate); and diesel fuel consumption during sowing, harvesting, and tilling (Forster et al 2007).

$$\begin{eqnarray}&&{\rm{Total}}\;{\rm{GHG}}={{\rm{GHG}}}_{{\rm{mt}}}+{\rm{Total}}\;{{\rm{N}}}_{2}{\rm{O}}\times \mathrm{44/28}\\ &&\quad \quad \quad \quad \quad \times \;298+{{\rm{GHG}}}_{{\rm{other}}}+{{\rm{GHG}}}_{{\rm{CH}}4},\end{eqnarray} \tag{ 1 }$$

$$\begin{eqnarray}&&{\rm{Total}}\;{{\rm{N}}}_{2}{\rm{O}}\\ &&={\rm{Direct}}\;{{\rm{N}}}_{2}{\rm{O}}+{\rm{ammonia}}\left({{\rm{NH}}}_{3}\right){\rm{volatilization}}\\ &&\times 1\%+\left({\mathrm{NO}}_{3}^{-}\right.\left.{\rm{leaching}}+{\rm{N}}{\rm{runoff}}\right)\times 0.75\%,\end{eqnarray} \tag{ 2 }$$

$$\begin{eqnarray}&&{\rm{Direct}}\;{{\rm{N}}}_{2}{\rm{O}}\;{\rm{emission}}\;\left({\rm{kg}}\;{\rm{N}}\;{{\rm{ha}}}^{-1}\right)\\ &&\quad \;=\;0.37\;\mathrm{exp}\left(0.0061\times {\rm{N}}\;{\rm{rate}}\right),\end{eqnarray} \tag{ 3 }$$

$$\begin{eqnarray}&&{\rm{Ammonia}}\;{\rm{volatilization}}\;\left({\rm{kg}}\;{\rm{N}}\;{{\rm{ha}}}^{-1}\right)\\ &&\quad =0.16\times {\rm{N}}\;{\rm{rate}}+2.97,\end{eqnarray} \tag{ 4 }$$

$$\begin{eqnarray}&&{\rm{N}}\;{\rm{leaching}}\;{\rm{and}}\;{\rm{N}}\;{\rm{runoff}}\left({\rm{kg}}\;{\rm{N}}\;{{\rm{ha}}}^{-1}\right)\\ &&\quad =2.221\;\mathrm{exp}\left(0.0048\times {\rm{N}}\;{\rm{rate}}\right),\end{eqnarray} \tag{ 5 }$$

where GHG_mt is the GHG emission originating from mining and manufacturing and N fertilizer transportation (table S3). Total N₂O (kg N ha⁻¹) included direct and indirect N₂O emissions from ammonia volatilization, N leaching and N runoff. The N losses from direct N₂O, ammonia volatilization and N leaching and N runoff were calculated based on an empirical N loss model (Cui et al 2014). N rate is the N fertilizer application rate (kg N ha⁻¹).

GHG_other represents GHG emission of P and K fertilizer production and transportation, pesticide and herbicide production, and transportation, diesel fuel consumption (table S3). GHG_CH4 was calculated with 1625 kg CO₂ eq ha⁻¹ for single rice cropping system in northeast China, 5225 kg CO₂ eq ha⁻¹ for rice-upland crop rotation system in south China, and 6125 and 8075 kg CO₂ eq ha⁻¹ for early rice and late rice in the double rice cropping system, respectively (Feng et al 2013). The other details of this calculation of GHG emission were shown in text S4.

Across all 6611 farms, the average rice grain yield, weighted by the area of rice production in each subregion, was 7.14 Mg ha⁻¹ and the N application rate averaged 210 kg N ha⁻¹ (table 1). Across all farms, the N application rate ranged from 30 to 744 kg N ha⁻¹, with a coefficient of variation (CV) of 50%, and grain yield ranged from 2.5 to 12.8 Mg ha⁻¹. No significant correlation was observed between the N rate and crop grain yield, indicating overuse and misuse N fertilization as being common occurrences in Chinese rice production (figure S1). As a result, large variation in the partial N balance existed across the farms, which ranged from −116 to 588 kg N ha⁻¹, with a CV of 125% (table 1).

Large variation in N fertilizer application and grain yield was observed across the eight agroecological subregions (table 1). N application rates ranged from 131 kg ha⁻¹ (NE1) to 316 kg ha⁻¹ (LYR) and rice grain yields ranged from 6.47 Mg ha⁻¹ (SC2-ER) to 8.07 Mg ha⁻¹ (NE2). Across all eight subregions, the PFP_N averaged 37 kg kg⁻¹ (table 1) and ranged from 25 kg kg⁻¹ (LYR) to 60 kg kg⁻¹ (NE1). The PNB averaged 82 kg N ha⁻¹ and ranged from 7 kg N ha⁻¹ (NE1) to 173 kg N ha⁻¹ (LYR) (table 1). The estimated GHG emission intensity, weighted by the area of rice production in each subregion, averaged 1390 kg CO₂ eq Mg⁻¹ grain (table 1) and ranged from 820 kg CO₂ eq Mg⁻¹ grain (NE2) to 1932 kg CO₂ eq Mg⁻¹ grain (SC2-LR; table 1). This large variation in the intensity of GHG emissions was due to high N fertilizer use (244 kg N ha⁻¹), low grain yield (6.47 Mg ha⁻¹), and high emission factors for soil CH₄ emission (8026 kg CO₂ eq ha⁻¹) in SC2-LR, and low N fertilizer use (163 kg N ha⁻¹), high grain yield (8.07 Mg ha⁻¹), and low emission factors for soil CH₄ emission (1634 kg CO₂ eq ha⁻¹) in NE2.

Across all 1177 on-farm experiments, the national rice grain yield under the N0 treatment, weighted by rice area in each subregion, was 5.24 Mg ha⁻¹ (NE2; table 2). The MN rate, as determined by local agronomists, averaged 166 kg ha⁻¹ and ranged from 102 kg ha⁻¹ (NE1) to 226 kg ha⁻¹ (LYR). The corresponding grain yield under the MN rate averaged 7.44 Mg ha⁻¹ and ranged from 6.75 Mg ha⁻¹ (SC2-LR) to 9.17 Mg ha⁻¹ (NE2; table 2). Adding more N beyond MN (+50%) did not increase yields from the MN levels, but reduced PFP_N; treatments below MN (−50%) produced 7% lower yield, 6.94 Mg ha⁻¹, compared to MN (table 2).

The RONR, weighted by the area of rice production in each subregion, averaged 167 kg N ha⁻¹ and ranged from 114 (NE1) to 224 kg N ha⁻¹ (LYR) (figure 2 and table 3). Grain yield averaged 7.67 Mg ha⁻¹ and ranged from 6.69 Mg ha⁻¹ (SC2-ER) to 8.97 Mg ha⁻¹ (LYR; table 3). The estimated N balance, weighted by rice area in each subregion, was 23 kg N ha⁻¹ (table S4). The estimated GHG emission intensity averaged 1236 kg CO₂ eq Mg⁻¹ grain (table 3), including contributions of 84, 182, and 970 kg CO₂ eq Mg⁻¹ grain from N fertilizer application on farmland, N fertilizer production, and transportation, respectively (data not shown). The estimated GHG emissions intensity ranged from 735 kg CO₂ eq Mg⁻¹ grain (NE2) to 1709 kg CO₂ eq Mg⁻¹ grain (SC2-LR).

Zoom In Zoom Out Reset image size

The lowest and largest profitable N rate, weighted by the area of rice production in each subregion, was 158 and 176 kg N ha⁻¹, yielding 7.65 and 7.68 Mg ha⁻¹, the estimated GHG emission intensity averaged 1225 and 1249 kg CO₂ eq Mg⁻¹ grain (table 3). Compared to RONR, adding more N (PNRR high) did not increase yields, but increased GHG emission intensity; adding less N (PNRR low) did not reduce yields, but reduced GHG emission intensity (table 3).

If all famers were to use a uniform N rate with average of 210 kg N ha⁻¹, the estimated GHG emission intensity among 6611 farm fields was reduced by 6%, from 1390 to 1305 kg CO₂ eq ha⁻¹ (table S5). Further, a regional N management approach would reduce the N fertilizer application rate by 20.5% (from 210 to 167 kg N ha⁻¹), compared to current practices. Grain yield would increase by 7.4% (7.14 versus 7.67 Mg ha⁻¹) and estimated GHG emission intensity would decrease by 11%, from 1390 to 1236 kg CO₂ eq Mg⁻¹ grain. As a consequence of increased grain yield, reduced N rates and GHG emission, the regional N management significantly increased net economic gains by 10.2% from 2669 to 2941 $ ha⁻¹ compared to current farmer's practice (table S6).

In this study, we define N overuse as the amount of N application greater than the largest profitable N rate (118–234 kg N ha⁻¹) and N deficiency as N application lower than the smallest profitable N rate (108–214 kg N ha⁻¹) in each subregion. Across all farms, 3734 sites (56%) had N overuse and 2153 sites (33%) had N underuse (data not shown). Across the eight agroecological subregions, all except NE2 showed a high proportion of N overuse (>50% sites), ranging from 50% (UYR) to 74% (LYR; data not shown). NE2 had a high proportion of N underuse at 56% of the sites. The proportion of sites with N underuse in other subregions ranged from 19% (LYR) to 41% (UYR and SW; data not shown).

Of the eight agroecological subregions, NE2, NW1, UYR, MYR-Sr, LYR, and SC1-Sr, showed the highest increase in yield potential (>0.53 Mg ha⁻¹ mean yield increase; figures 3(a) and S2(a)), ranging from 0.71 to 1.37 Mg ha⁻¹ and accounting for 44% of the total area in rice production. The subregions with the highest potential for N reduction (>43 kg N ha⁻¹ mean N reduction; figures 3(b) and S2(b)) were LYR, SC1-ER, SC1-LR, SC1-Sr, SC2-ER, and SC2-LR, ranging from 44.3 to 92.3 kg N ha⁻¹ and accounting for 39% of the total area in rice production. Reduced N rates in other subregions ranged from 7.8 to 40.4 kg N ha⁻¹. Subregions with a high potential to decrease GHG emission intensity (>154 kg CO₂ eq Mg⁻¹ grain; figures 3(c) and S2(c)) included UYR, MYR-Sr, LYR, SC1-Sr, and SC2-LR, ranging from 177 kg CO₂ eq Mg⁻¹ grain to 362 kg CO₂ eq Mg⁻¹ grain and accounting for 43% of the total area in rice production. Reduced estimated GHG emissions in other regions ranged from 42 kg CO₂ eq Mg⁻¹ grain to 136 kg CO₂ eq Mg⁻¹ grain.

Zoom In Zoom Out Reset image size