Introduction
Rainfed lowland rice ecosystems cover 9.5 × 106 ha, equivalent to more than 80% of the total area of rice (Oryza sativa L.) in the Mekong region of Southeast Asia (Fukai and Ouk, Reference Fukai and Ouk2012). The climate in this region is tropical monsoonal, with a wet season (from June to November) followed by a prolonged dry season, and irregular rainfall both from year to year and within years. Rice is grown mostly during the wet season, but with frequent intermittent drought (Tsubo et al., Reference Tsubo, Fukai, Basnayake, Tuong, Bouman and Harnpichitvitaya2007). The area of rice cultivation in Cambodia has doubled during the last three decades, from 1.4 × 106 ha in 1987 to 3.0 × 106 ha in 2017 (FAOSTAT, 2018). The average rice yield has also increased during the same period, from 1.4 t ha−1 to 3.5 t ha−1, but the yield of rainfed lowland rice in the wet season remains lower than 3.0 t ha−1. This compares with a mean yield of at least 6.6 t ha−1 in irrigated rice cultivation in Australia, China, Japan, Korea, and the United States (FAOSTAT, 2018).
Most soils in rainfed lowlands of the Mekong region are infertile, and rice yield is limited by this poor fertility (Bell and Seng, Reference Bell, Seng, Seng, Craswell, Fukai and Fischer2004). In Cambodia, coarse-textured soils are widespread in lowland rice environments and lack sufficient N, P, and often K (Oberthur et al., Reference Oberthur, Dobermann and White2000). Following a large national soil survey, the soils were classified into 11 groups according to their nutrient management requirements. These groups are easily distinguishable by local people without requiring any equipment: Bakan, Kampong Siem, Kbar Po, Kein Svay, Koktrap, Krakor, Labansiek, Orung, Prateah Lang, Prey Khmer, and Toul Samroung (White et al., Reference White, Dobermann, Oberthur and Ros2000). This survey found that soil fertility varied greatly, with unfertilized rice yielding from 0.6 to 2.6 t ha–1. The low soil fertility results from weathering of soils with a low carbon content, as is typically seen in the Prateah Lang and Prey Khmer soil groups, which cover 25–30% and 10–12%, respectively, of the country’s rice-growing area (White et al., Reference White, Dobermann, Oberthur and Ros2000). Blair and Blair (Reference Blair and Blair2014) estimated that 87% of the country’s soils have less than 10 g kg−1 of organic carbon content. Coarse-textured soils respond poorly to nutrient management owing to their low nutrient-holding capacity (Haefele et al., Reference Haefele, Kato and Singh2016). On the other hand, loamy or clayey soils such as Toul Samroung (7–10% of the rice-growing area), Bakan (10–15%), and Krakor (15%) have the potential for high rice productivity (White et al., Reference White, Dobermann, Oberthur and Ros2000).
Although low soil fertility is a major constraint on rainfed lowland rice yields, nutrient management under fluctuating hydrological conditions is also challenging (Kato and Katsura, Reference Kato and Katsura2014). From nutrient management trials at 78 locations, Wade et al. (Reference Wade, Amarante, Olea, Harnpichitvitaya, Naklang, Wihardjaka, Sengar, Mazid, Singh and McLaren1999) concluded that the magnitude of the yield response to N was affected by the water regime. The disappearance of standing water strongly affects soil nutrient availability and fertilizer-use efficiency in rainfed lowland rice (Kato et al., Reference Kato, Tajima, Toriumi, Homma, Moritsuka, Shiraiwa, Yamagishi, Mekwatanakern, Chamarerk and Jongdee2016). The soil P dynamics in several Cambodian soils under fluctuating water regimes and the rice response to various P management regimes have been studied (Seng et al., Reference Seng, Bell, Willett and Nesbitt1999, Reference Seng, Bell and Willett2004), and the results demonstrated the importance of using a balanced NPK fertilizer for wet-season rice (Seng et al., Reference Seng, Ros, Bell, White, Hin, Fukai and Basnayake2001).
Since there is growing evidence that the response of crop yield to management depends on the local growth environment, agricultural guidelines have become increasingly site-specific (Carberry et al., Reference Carberry, Liang, Twomlow, Holzworth, Dimes, McClelland, Huth, Chen, Hochman and Keating2013). A well-known example is the development of site-specific nutrient management for tropical rice (Sharma et al., Reference Sharma, Panneerselvam, Castillo, Manohar, Raj, Ravi and Buresh2019). The recommendations for rice were based on a large dataset from nutrient omission trials in the target regions and accounted for the relationship between nutrient supply and crop demand in each field (Dobermann and Fairhurst, Reference Dobermann and Fairhurst2000). Wongboon et al. (Reference Wongboon, Sansen, Lertna, Saleeto, Srisomphan, Chamarerk, Haefele and Kato2014) found a clear yield improvement under site-specific nutrient management compared with blanket fertilizer recommendations (i.e., one recommendation applied to a large region) for nutrient management in rainfed lowland rice in Thailand. Although inorganic fertilizer is generally applied at low rates in Cambodia, its use in rice cultivation has become popular. A survey of 1223 Cambodian rice farmers during the 1990s revealed that 82% applied fertilizer (Ouk et al., Reference Ouk, Men, Nesbitt, Fukai and Basnayake2001). Although potash fertilizer was not popular, most farmers used N and phosphate fertilizers; the rates were 15–20 kg N ha−1, 14–20 kg P2O5 ha−1, and 0–3 kg K2O ha−1 (Mutert and Fairhurst, Reference Mutert and Fairhurst2002; Ouk et al., Reference Ouk, Men, Nesbitt, Fukai and Basnayake2001).
Together with the development of national soil classification systems in Cambodia based on a database of country-wide soil tests (Oberthur et al., Reference Oberthur, Dobermann and White2000), White et al. (Reference White, Dobermann, Oberthur and Ros2000) proposed a user-friendly guideline for soil-specific nutrient management, which was subsequently adopted by a governmental agency, the Cambodian Agricultural Research and Development Institute (Seng et al., Reference Seng, Ros, Bell, White, Hin, Fukai and Basnayake2001). However, the overall fertilizer recommendations for each soil group provided poor net returns (gross economic return minus total production cost) due to low inputs of fertilizers that resulted in low yield (Blair and Blair, Reference Blair and Blair2014). The total fertilizer consumption in Cambodia has increased from 7873 t N during the 2002–2005 period to 63,784 t during the 2012–2015 period, from 12,512 t P2O5 in 2002–2005 to 17,112 t in 2012–2015, and from 1033 t K2O in 2002–2005 to 3926 t in 2012–2015 (FAOSTAT, 2018). Although much research has emphasized the importance of the timing and rate of N application in rainfed lowland rice (Haefele et al., Reference Haefele, Kato and Singh2016; Homma et al., Reference Homma, Horie, Shiraiwa and Supapoj2007), our knowledge of the optimal rates of P and K application on different soil types in the Mekong region is still limited (Seng et al., Reference Seng, Ros, Bell, White, Hin, Fukai and Basnayake2001).
As more Cambodian rice farmers are beginning to apply more fertilizers to their fields, there is an urgent need to adjust the existing nutrient management guidelines for wet-season rice (White et al., Reference White, Dobermann, Oberthur and Ros2000) to maximize the return from fertilizer investments. Therefore, we designed a study to evaluate rice productivity and the economic feasibility of various nutrient management treatments with different rates of N, P2O5, and K2O based on current farmer practice to upgrade soil-specific fertilizer recommendations. We studied rice in four major Cambodian rice-growing provinces for 2 years on the three most common soil types.
Materials and Methods
Setup of field experiments
Field experiments were conducted at government research stations or in farmer fields in four Cambodian provinces (Battambang, Kampong Thom, Pursat, and Siem Reap) during the wet seasons (June to November) of 2016 and 2017 (Table 1). The monthly mean temperature in Cambodia ranges from 26 to 31 °C throughout the year. The Toul Samroung (high-fertility clay), Prateah Lang (moderate-fertility clay), and Prey Khmer (coarse sand) soils are classified as Endoaqualfs, Plinthustalfs, and Psamments, respectively, in the USDA classification (White et al., Reference White, Dobermann, Oberthur and Ros2000). Wet-season rice is mostly grown on these soil types (Seng et al. Reference Seng, Ros, Bell, White, Hin, Fukai and Basnayake2001). At each site, we selected terraced fields at mid-slope positions along the toposequence, which occupy almost 40% of the country’s total rice area (Ouk et al., Reference Ouk, Men, Nesbitt, Fukai and Basnayake2001).
Table 1. Characteristics of the study sites in rainfed lowland areas of Cambodia
CEC, cation-exchange capacity. Experiments were conducted at the same sites in Kampong Thom, Pursat, and Siem Reap in both years.
* Monthly total rainfall (mean ± S.E.) during the wet season (from June to November).
† Soil type classification according to White et al. (Reference White, Dobermann, Oberthur and Ros2000).
‡ Based on the Bray-2 method.
We examined nine (2016) and seven (2017) nutrient management regimes at each site (Tables 2 and 3) and grew a popular high-quality rice cultivar (‘Phka Rumduol’) under rainfed lowland conditions at all locations. The fertilizer application rates in the farmer management regime were based on a recent survey of 1200 rice farmers in 2013 (Mr. Sourn Sophoan, Cambodia Agricultural Value Chain Program, pers. comm.; 35–25–5 kg ha−1 of N–P2O5–K2O). Fertilizer application in the soil-specific management regime followed the recommendations of the Cambodian Agricultural Research and Development Institute, as described above. We also tested several treatments based on the addition of 20–30 kg ha−1 of N (as top-dressing), 15 kg ha−1 of P2O5, or 20 kg ha−1 of K2O to the fertilizer rate under current farmer management (Table 2) or soil-specific management (Table 3). Supplemental Table S1 provides details of the fertilizer amounts in each treatment during the study period. Amounts of urea, diammonium phosphate, and potassium chloride were determined to provide these fertilizer rates in each treatment. Treatments were arranged in a randomized complete block design with four replicates at each site. Each plot covered 15.0 m2 (3.0 m × 5.0 m). The transplanting date ranged from 23 August to 5 September in 2016 and from 14 July to 10 August in 2017, depending on the site. After the creation of bunds (20 cm high) around each plot and soil puddling, two or three 3-week-old seedlings were transplanted into each hill (at a hill spacing of 20 cm × 20 cm). Weeds, insects, and diseases were carefully controlled following the standard protocol in each province.
Table 2. Nutrient management treatments in the trials of rainfed lowland rice in Cambodia in 2016
* Days after transplanting.
† Panicle initiation.
‡ The application regime was based on standard farmer practices in the wet season (Mr. Sourn Sophoan, Cambodia Agricultural Value Chain Program, pers. comm.).
§ The dominant soil in Battambang and Kampong Thom provinces.
¶ The dominant soil in Pursat Province.
** The dominant soil in Siem Reap Province.
Table 3. Nutrient management treatments in the trials of rainfed lowland rice in Cambodia in 2017
* Days after transplanting.
† Panicle initiation.
‡ The dominant soil in Battambang and Kampong Thom provinces.
§ The dominant soil in Pursat Province.
¶ The dominant soil in Siem Reap Province.
Measurements
Soil properties were determined from composite samples taken at 0–20 cm (topsoil) and at 20–40 cm (subsoil) at five randomly chosen positions at each site before the experiment. We measured the following parameters using the standard protocols in Dobermann and Fairhurst (Reference Dobermann and Fairhurst2000): pH (1:1 v/v soil–H2O mixture), total soil C, total soil N, cation-exchange capacity (CEC), soil texture, plant-available K (extracted by ammonium acetate), and plant-available P (Bray-2). Daily rainfall was recorded at the meteorological station in each province. Surface hydrological conditions were recorded weekly as a score by the method of Ohno et al. (Reference Ohno, Banayo, Bueno, Kashiwagi, Nakashima, Iwama, Corales, Garcia and Kato2018): 0, flooded soil (standing water in the field); 1, saturated soil (no standing water but soggy); 2, moist soil (unsaturated but not dry); or 3, dry soil (based on soil color and the presence of cracks).
The date of 50% heading in each plot was recorded. At maturity, plant height (in five hills) and panicle number (in 10 hills) were measured in each plot. In addition, five representative medium-sized panicles were collected, and the numbers of ripened and empty grains were counted to determine the number of spikelets per panicle and the filled grain percentage. The paddies were manually harvested at maturity from a 5-m2 area in each plot and manually threshed. Grain moisture content was determined immediately after threshing using a Riceter grain moisture meter (Kett Electric Laboratory, Tokyo, Japan), and grain yields were converted to values at a 14% moisture content.
The economic returns from applying the fertilizer regimes (the ‘gross return above fertilizer cost’) were determined by subtracting the total cost of the fertilizer inputs from the income received from the harvested paddy rice. The treatment-specific costs for fertilizers were calculated by multiplying the amount of each fertilizer by the fertilizer price. The income for each treatment in each plot was calculated by multiplying the plot’s grain yield by the grain price. We used the farm gate price of dry paddy rice (for ‘Phka Rumduol’) in this calculation, which was 0.32 USD kg−1. The price of fertilizers was 0.44 USD kg−1 for urea, 0.57 USD kg−1 for diammonium phosphate, and 0.58 USD kg−1 for potassium chloride. These prices represented the mean local market prices in 2015 and 2016.
Statistical analysis
We performed analysis of variance (ANOVA) using the generalized linear model in STAR v. 2.0.1 software, an open-access program implemented in the R package (https://bbi.irri.org). Nutrient management and location were regarded as fixed effects, replicate was treated as a random effect, and the effects of nutrient management, location, and the nutrient management × location interaction were assessed. When the ANOVA result was significant at p < 0.05, we compared pairs of values using Fisher’s least-significant-difference test.
Results
Growth conditions at the experimental sites
Soil characteristics at the experimental sites differed greatly among locations and soil layers (Table 1). The values of clay content, pH (lowest acidity), plant-available K, and CEC were the highest in the Toul Samroung soil. The Prey Khmer showed the opposite pattern, with the Prateah Lang soil showing intermediate values. There was a significant positive relationship (p < 0.01) between total C and total N, but significant negative relationships between the sand and clay contents, between the sand content and CEC, and between the sand content and available K (Figure 1).
Figure 1. Relationships between (a) the total C and total N contents, (b) between the sand and clay contents, (c) between the sand content and cation-exchange capacity (CEC), and (d) between the sand content and available K for the soils at the experimental sites in Cambodia.
All environments had more than 120 mm per month of rainfall in both years, and the rainfall was relatively evenly distributed during the growing season (Table 1). There was standing water in the field for more than two-thirds of the experimental period at all sites, and intermittent drought was not serious at any site in either year; the surface hydrology score was < 2.0 (i.e., we observed no change in soil color and no soil cracks).
Grain yield and economic profit
Grain yield ranged from 0.9 to 4.8 t ha−1 in 2016 and from 1.0 to 5.2 t ha−1 in 2017, depending on the location and nutrient management (Figure 2). There was a significant location × management interaction for yield in both years (p < 0.01). In the zero-input treatment, yield was highest at Kampong Thom, followed by Battambang, in both years; Toul Samroung soil covered both sites. The yield in the unfertilized plots was significantly positively correlated with mean soil C content (r = 0.98*, n = 4) and soil N content (r = 0.98*, n = 4) to a depth of 40 cm in 2016, and negatively correlated with the sand content (r = −0.98*, n = 4) in 2017. The yield response to 35 kg ha−1 of N input was significant at only one site (Pursat) on the Prateah Lang soil. In 2016, the maximum yield gain that resulted from nutrient management, i.e., an increase of 3.2 t ha–1 on the Prateah Lang soil, 1.2–1.5 t ha–1 on the Toul Samroung soil, and 0.9 t ha–1 on the Prey Khmer soil (Figure 2), was significantly correlated with the yield response to N-only input (r = 0.97*, n = 4).
Figure 2. Grain yield in the different nutrient management treatments in rainfed lowland rice in Cambodia in (a) 2016 and (b) 2017. Bars labeled with different letters differed significantly at each location in a given year. Nutrient regimes (Tables 2 and 3): ZI, zero input; NI, N input; FM, farmer management; FMN, farmer management + N; FMK, farmer management + K; FMNP, farmer management + NP; FMNK, farmer management + NK; SM, soil-specific management; SMN, soil-specific management + N; SMP, soil-specific management + P; SMK, soil-specific management + K; SMNK, soil-specific management + NK; SMNP, soil-specific management + NP.
Nutrient management regime and the nutrients that significantly increased yield relative to farmer management (2016) or soil-specific management (2016 and 2017) differed between locations. In 2016, additional input of N to farmer management or to soil-specific management did not increase yield at any site relative to farmer management alone (Figure 2a). At Kampong Thom, farmer management + K, farmer management + NK, soil-specific management, and soil-specific management + N gave the highest yields. At Battambang, all treatments except zero input and N-only input had similar yields. Thus, the minimum fertilizer inputs required to achieve the highest yield on the Toul Samroung soil were 35–25–0 kg ha−1 of N–P2O5–K2O. At Pursat, farmer management + NK provided the highest yield, which was significantly higher than soil-specific management + N, farmer management + K, and farmer management + NP; the minimum fertilizer inputs required to achieve the highest yield on the Prateah Lang soil were 65–25–25 kg ha−1 of N–P2O5–K2O. At Siem Reap, farmer management + NK provided the highest yield; the minimum fertilizer inputs required for the highest yield on the Prey Khmer soil were 65–25–25 kg ha−1 of N–P2O5–K2O.
In 2017, adding inputs of K, N and K, or N and P to soil-specific management gave the highest yield at Kampong Thom, whereas adding only P did not increase yield (Figure 2b). At Battambang, adding N and K inputs or N and P inputs to soil-specific management provided the highest yields. At Pursat, adding N and K to soil-specific management provided the highest yield, followed by adding N and P to soil-specific management. At Siem Reap, adding N and P to soil-specific management significantly increased yield.
Combining the results for the 2 years with the fertilizer inputs in each treatment, rice yield was maximized by adding at least 63–38–0 kg ha−1 or 63–23–20 kg ha−1 of N–P2O5–K2O to the Toul Samroung soil, 57–23–35 kg ha−1 or 65–25–25 kg ha−1 of N–P2O5–K2O to the Prateah Lang soil, and 65–25–25 kg ha−1 of N–P2O5–K2O to the Prey Khmer soil.
The economic profit from the fertilizer inputs followed similar trends to those for grain yield (Figure 3). There was a significant location × management interaction for the gross return above fertilizer cost in both years (p < 0.01). In other words, the nutrient management treatment that gave the maximum benefit depended on the location. The maximum additional benefit from improved nutrient management relative to the currently recommended soil-specific management ranged from 9 USD ha–1 (Pursat in 2016) to 487 USD ha–1 (Pursat in 2017). In 2016, farmer management + K, farmer management + NK, soil-specific management, and soil-specific management + N gave the highest profits at Kampong Thom, whereas all treatments except zero input and N-only input provided similar profits at Battambang (Figure 3a). At Pursat, farmer management + NK and soil-specific management gave the highest profits. At Siem Reap, there was no significant increase in the economic benefit from any fertilizer input in 2016. In 2017, soil-specific management + the addition of K, NK, or NP gave the highest profit at Kampong Thom, whereas additional N and K inputs or adding N and P inputs to soil-specific management gave the highest profit at Battambang (Figure 3b). At Pursat, adding N, K, NK, or NP to soil-specific management gave the highest profits. At Siem Reap, adding nutrients to soil-specific management did not increase profits in 2017.
Figure 3. Gross return above fertilizer cost (GRAFC) in the different nutrient management treatments in rainfed lowland rice in Cambodia in (a) 2016 and (b) 2017. Bars labeled with different letters differed significantly at each location in a given year. Nutrient regimes (Tables 2 and 3): ZI, zero input; NI, N input; FM, farmer management; FMN, farmer management + N; FMK, farmer management + K; FMNP, Farmer management + NP; FMNK, farmer management + NK; SM, soil-specific management; SMN, soil-specific management + N; SMP, soil-specific management + P; SMK, soil-specific management + K; SMNK, soil-specific management + NK; SMNP, soil-specific management + NP.
Combining the results for the 2 years reveals that profits were maximized by applying 63–38–0 kg ha−1 or 63–23–20 kg ha−1 of N–P2O5–K2O on the Toul Samroung soil and by applying 57–23–35 kg ha−1 of N–P2O5–K2O on the Prateah Lang soil relative to soil-specific management, whereas applying soil-specific management (22–9.2–24 kg ha−1 of N–P2O5–K2O) was enough to maximize the profit on the Prey Khmer soil.
Crop growth traits associated with grain yield
We found statistically significant effects of location, management, and the location × management interaction on all agronomic traits in both years, except for the interaction on spikelets per panicle in 2017 (Table 4). In general, days to heading decreased with increasing fertilizer input. Plant height was lowest at Siem Reap in both years, but it increased with increasing fertilizer inputs. The effects of N inputs, P inputs, or both on panicle number were significant in both years; the highest number was obtained under farmer management + NP in 2016 and under soil-specific management + NP in 2017. Panicle number, therefore, responded significantly and positively to N and P inputs. Spikelets per panicle were highest under farmer management + NK or NP in 2016 and under soil-specific management + N, NP, or NK in 2017. The filled grain percentage also increased significantly with increasing fertilizer inputs in both years.
Table 4. Agronomic characteristics of rainfed lowland rice under the different nutrient management treatments in Cambodia
* Days from sowing.
Significance: * and ** indicate a significant effect at p < 0.05 and 0.01, respectively. ns, not significant.
Values of a parameter in a given year labeled with different letters differ significantly.
Nutrient regimes (Tables 2 and 3): ZI, zero input; NI, N input; FM, farmer management; FMN, farmer management + N; FMK, farmer management + K; FMNP, farmer management + NP; FMNK, farmer management + NK; SM, soil-specific management; SMN, soil-specific management + N; SMP, soil-specific management + P; SMK, soil-specific management + K; SMNK, soil-specific management + NK; SMNP, soil-specific management + NP.
The agronomic traits that were most strongly correlated with grain yield differed among locations (Table 5). Panicle number and spikelets per panicle were significantly positively correlated with yield in both years at Battambang, whereas no traits were significantly associated with yield in either year at Kampong Thom. Plant height, spikelets per panicle, and the filled grain percentage at Pursat and the plant height and panicle number at Siem Reap were significantly positively correlated with yield in both years.
Table 5. Correlation coefficients (Pearson’s r) for the relationships between grain yield and agronomic characteristics under different nutrient management treatments during the wet season in Cambodia
Significance: * and ** indicate p < 0.05 and 0.01, respectively.
ns, not significant.
Discussion
Most rice farmers in rainfed lowlands of the Mekong region grow old, tall cultivars such as ‘Phka Rumduol’ in Cambodia and ‘KDML105’ in Thailand that respond less well to soil nutrients than high-yielding cultivars in irrigated lowlands (Fukai and Ouk, Reference Fukai and Ouk2012). Notwithstanding, nutrient limitation of rice growth was obvious. Yield of unfertilized rice ranged from 0.9 to 3.3 t ha–1 in 2016 and from 1.0 to 3.0 t ha–1 in 2017 (Figure 2), which were correlated with the total soil N and C in 2016 and with the sand content in 2017. The sand content was negatively correlated with the clay content, CEC, and available K (Figure 1); that is, a high sand content decreased the soil’s ability to retain nutrients. Our finding that rice yield in the unfertilized plots strongly reflected the inherent soil fertility or nutrient-holding capacity agrees with previous studies (Homma et al., Reference Homma, Horie, Shiraiwa, Supapoj, Matsumoto and Kabaki2003; Sharma et al., Reference Sharma, Panneerselvam, Castillo, Manohar, Raj, Ravi and Buresh2019). As White et al. (Reference White, Dobermann, Oberthur and Ros2000) suggested, the Toul Samroung soil proved to be the most fertile of the three soil groups used in this study, probably because of its high clay and CEC.
The response of yield to the different fertilizer inputs also varied among the soil groups. Although the yield increase caused by adding 35 kg N ha−1 explained only 11–60% of the maximum yield gain resulted from nutrient management (Figure 2a; see the Result section), the correlation between the yield increase caused by the N-only input and that caused by the balanced inputs of NPK indicated that rice yield poorly responds to integrated nutrient management on the soil which merely provides a slight yield increase by N-only input. This also supports the previous statement that N is the primary element in nutrient management for Cambodian rice (Seng et al., Reference Seng, Ros, Bell, White, Hin, Fukai and Basnayake2001). The reason why the Prateah Lang soil gave the greatest yield response to nutrient management remains unknown. Although the soil had high sand content, low CEC, and low carbon content leading to a low yield in the zero-input plots, the degree of weathering was not serious compared with the Prey Khmer soil. Presumably, the soil could still temporarily hold the externally applied nutrients.
The greater yield response to K fertilizer than to P fertilizer under N-fertilized conditions on the Prateah Lang soil, and the poorest response to fertilizer applications on the Prey Khmer soil, confirmed previous research (Seng et al., Reference Seng, Ros, Bell, White, Hin, Fukai and Basnayake2001; White et al., Reference White, Dobermann, Oberthur and Ros2000). Interestingly, the yield response to K input combined with N input was also significant on the Toul Samroung soil, except at Battambang in 2016 (Figure 2). Previous studies suggested a low requirement for supplemental K on the Toul Samroung soil, since the soil is already relatively rich in plant-available K (Table 1; Blair and Blair, Reference Blair and Blair2014; Oberthur et al., Reference Oberthur, Dobermann and White2000; White et al., Reference White, Dobermann, Oberthur and Ros2000). It is likely that long-term intensive use of N and P fertilizers on this fertile soil leads to increased K uptake by plants, thereby gradually depleting the available K in many fields (Seng et al., Reference Seng, Ros, Bell, White, Hin, Fukai and Basnayake2001). This would lead to a significant K deficit when adequate N and P inputs were provided. The soil nutrient status determined by basic soil analysis such as that in the present study may not always be associated with the rice yield response to additional nutrient inputs (Banayo et al., Reference Banayo, Haefele, Desamero and Kato2018a; Sharma et al., Reference Sharma, Panneerselvam, Castillo, Manohar, Raj, Ravi and Buresh2019) because simple estimates of plant-available nutrient contents cannot always predict their relationship with plant nutrient uptakes (Dobermann and Fairhurst, Reference Dobermann and Fairhurst2000). Our results support those in previous studies of soil-specific nutrient management for Cambodian rice, but clearly suggest that there is an opportunity for upgrade.
Yield improvements caused by nutrient management can be attributed to physiological improvements that increase crop growth. That is, the different yield responses to nutrient management at different locations will be mediated by the response of critical growth mechanisms to location-specific factors. Panicle number increased significantly with the application of N and P, whereas spikelets per panicle responded more strongly to N (Table 4), as was suggested in previous studies (Dobermann and Fairhurst, Reference Dobermann and Fairhurst2000; Kato and Katsura, Reference Kato and Katsura2010). The increased filled grain percentage caused by fertilizer application (Table 4) is likely to have been supported by increased carbohydrate accumulation during the post-anthesis stage to support grain growth (Liu et al., Reference Liu, Won, Banayo, Nie, Peng and Kato2019). In the absence of any specific growth constraints, sink size (the number of spikelets m−2) is a major yield constraint for rainfed lowland rice (Banayo et al., Reference Banayo, Haefele, Desamero and Kato2018b). We saw this effect at Battambang, where panicle number and panicle size were strongly and significantly correlated with yield in both years (Table 5). However, on the same soil type at Kampong Thom, panicle number was significantly correlated with the yield response to nutrient management only in 2016. This is probably due to the fact that these soils were more fertile than those at the other sites. Although this soil is generally fertile (Oberthur et al., Reference Oberthur, Dobermann and White2000), our results suggest that nutrient management on the Toul Samroung soil should aim to increase spikelet production (i.e., to increase the sink size). The filled grain percentage was lowest at Pursat in both years and was significantly associated with yield at this site (Table 5). Presumably, the positive yield response to K inputs on the Prateah Lang soil discussed above is mediated by the improved grain growth and filled grain percentage. A K top-dressing during the reproductive stage is sometimes suggested, as the improved plant K status promotes grain filling in rice (Dobermann and Fairhurst, Reference Dobermann and Fairhurst2000). On the other hand, crop growth during the vegetative stage seems critical for yield on the most infertile soil (Prey Khmer), where improved nutrient management was associated with increases in plant height and panicle number but not with growth during the reproductive stage.
Inorganic fertilizer is the major component of the rice production cost in the tropics (Dobermann and Fairhurst, Reference Dobermann and Fairhurst2000). For example, rice farmers in rainfed lowlands of the Philippines invest 21–29% of the total production cost in fertilizer (Banayo et al., Reference Banayo, Haefele, Desamero and Kato2018a). Although the fertilizer application rates in Cambodia are lower than in neighboring countries (Mutert and Fairhurst, Reference Mutert and Fairhurst2002), the rates will increase as rice-based systems change to focus on economic goals rather than on subsistence farming (Fukai and Ouk, Reference Fukai and Ouk2012). The gross return above fertilizer cost was lowest in unfertilized plots at all four sites (Figure 3), indicating that the investment in fertilizer benefits rainfed lowland rice in Cambodia. Our results indicate that an additional 20 kg ha−1 of N, in combination with an additional 15 kg ha−1 of P2O5 or 20 kg ha−1 of K2O, compared with the recommended rates (43.5–23–0 kg ha−1 of N–P2O5–K2O for Toul Samroung and 36.6–23–15 kg ha−1 of N–P2O5–K2O for Prateah Lang) consistently provided the greater benefit (at least 60 USD ha–1) to rice farmers in Cambodia, in the absence of serious drought. Although greater fertilizer inputs did not reduce the benefit for farmers, the moderate nutrient inputs in the current recommendation for the Prey Khmer soil (i.e., 22–9.2–24 kg ha−1 of N–P2O5–K2O) do not have to be changed. To confirm the present results, on-farm validation trials at more locations will be needed in future research. It would be also worthwhile to explore the possibility of utilizing cheaper locally available nutrient sources, such as rock phosphate (White et al., Reference White, Nesbitt, Ros, Seng and Lor1999).
Nutrient management guidelines based on the existing user-friendly soil classification, in a region where adequate facilities for soil analysis and databases that summarize the results of nutrient omission trials are not available, would be easily adopted by the target smallholder farmers (White et al., Reference White, Dobermann, Oberthur and Ros2000). Nonetheless, there is some room for further improvement. One of the unique characteristics of rainfed lowland ecosystems is the high spatial variability in water and nutrient availability, which is largely determined by the toposequence; this is defined by sloping land separated into a series of rice terraces (Tsubo et al., Reference Tsubo, Fukai, Basnayake, Tuong, Bouman and Harnpichitvitaya2007). This suggests that the present guidelines should be amended to provide different fertilizer prescriptions for different positions in the toposequence. On coarse-textured soils such as Prey Khmer, the soil fertility, and particularly the clay content and CEC, increases down the toposequence from the higher fields to the lower fields (Homma et al., Reference Homma, Horie, Shiraiwa, Supapoj, Matsumoto and Kabaki2003). A previous study in Thailand successfully incorporated information on field positions within the toposequence, classified by local farmers (i.e., upper, medium, lower, and bottom), into the nutrient management guidelines for rainfed lowland rice (Haefele and Konboon, Reference Haefele and Konboon2009).
Conclusion
Field experiments on three soil types in Cambodia for 2 years demonstrated that there is no one fertilizer application regime for wet-season rice that can be recommended for the whole country. Fertilizer recommendations based on the soil types in Cambodia’s national soil classification systems are farmer-friendly, but our results suggest the need to adjust the recommended N–P2O5–K2O rates to account for differences in local conditions. Rice on the Prey Khmer soil, which is very sandy, responded poorly to nutrient management. The current soil-specific fertilizer recommendation proved to be one of the most effective practices in terms of increasing grain yield and the economic benefits for farmers. However, on more fertile soils with a higher clay content and CEC (the Toul Samroung and Prateah Lang soils), adding 20 kg ha−1 of N in combination with an additional 15 kg ha−1 of P2O5 or 20 kg ha−1 of K2O to the moderate rates specified in the current recommendation consistently increased rice yield and the economic return. Although P and K application rates for wet-season rice in Cambodia are still low, our results demonstrate the importance of these nutrients for improving the country’s rice production.
Acknowledgements
We thank Messrs. Sokheng Keo, Leng Layhuot, Akhara Ouk, and Seyla Sem (International Rice Research Institute) for technical assistance. The study was conducted under the Transnational PhD Program of Nagoya University’s Asian Satellite Campuses Institute and the Southeast Asian Regional Center for Graduate Study and Research in Agriculture (for K.K.). We are grateful for the financial support of the United States Agency for International Development (USAID), the International Fund for Agricultural Development (IFAD) for the Consortium for Unfavorable Rice Environments (CURE), the Consultative Group on International Agricultural Research for the Research Program on Rice Agri-Food Systems, and a Grant-in-Aid for Scientific Research from the Japan Society for the Promotion of Science (JSPS) (no. 18KK0169).
Supplementary materials
For supplementary material for this article, please visit https://doi.org/10.1017/S0014479719000309
