INTRODUCTION

In Cambodia, rice-based farming systems have held a dominant position (Mak, 2001) with rice supplying 60% of the subsistence need of over 80% of the farmers (Yu and Fan, 2011). Rice is grown during the wet season, mostly in rainfed lowland in Cambodia. The yield of wetseason rice has increased by more than two-fold from 1.19 t/ha in 1980 to 3.09 t/ha in 2018 (MAFF, 2019) owing to the introduction of higher-yielding varieties and the improvement of soil management techniques (Ouk, 2015). However, the current rice yield is still constrained partly by the water stress in rainfed areas and partly by sandy soils with low fertility (Fukai and Ouk, 2012). In particular, the recorded rice yield loss due to water stress has ranged from 12% to 46% under irregular rainfall (Ouk et al., 2006).

The system of rice intensification (SRI) was first established by Fr. Henri de Laulanie in Madagascar in the early 1980s, and has since proved its efficacy in increasing rice yields (SRI International Network and Resources Center). The SRI primarily targets at increasing the rice harvest for smallholder farmers using considerably less water and seeds without relying on external inputs such as chemical fertilizers. The SRI practices include planting (1) very young seedlings, i.e. from 8 to 12 days old (at the 2 leafstage); (2) one plant per hill to avoid root competition, instead of 3-4 plants together; (3) in wider spaces with a square grid pattern (more than 25 × 25 cm) to encourage greater root and canopy growth; along with (4) water management to keep the rice fields intermittently flooded and drained rather than continuously flooded during the vegetative growth period; (5) weeding frequently; and (6) refraining from chemical fertilizers in favor of farmyard manure (Uphoff, 1999;Stoop et al., 2002;Uphoff, 2003;Stoop and Kassam, 2005;Thakur et al., 2015;Thakur et al., 2016).

The main benefits reported under the SRI include water saving (Nugroho et al., 2018;Kavishe et al., 2021) and increased rice yield (Ceesay et al., 2006;Kabir and Uphoff, 2007;Sinha and Talati, 2007;Stoop et al., 2009;Adusumilli and Laxmi, 2011;Styger et al., 2011;Ndiiri et al., 2013;Islam et al., 2014;Uphoff, 2014;Hidayati and Triadiati, 2016;Thakur et al., 2016;Lee and Kobayashi, 2018).

In Cambodia, the SRI was first introduced to 20 farmers across 18 villages by an NGO called the Cambodian Center for Study and Development in Agriculture (CEDAC) in 2000 (Yang, 2002). Previous studies in Cambodia have shown that the SRI increased rice yield and reduced the use of chemical fertilizer and seeds in rainfed lowland fields (Anthofer, 2004;Ly et al., 2012;Ches and Yamaji, 2016;Ly et al., 2016;Mishra et al., 2016;Lee and Kobayashi, 2017). None of these studies on the SRI in Cambodia has, however, investigated how the yield increase delivered by the SRI was attained in interaction with the water constraint and other determinants of the rainfed rice production. Indeed, Lee and Kobayashi (2017) reported that in the rainfed lowland rice in Southern Cambodia, the rice yield was increased by the SRI only in the fields with a water supply specifically at the beginning of the season. The farmers recognized the dependence of the SRI-induced yield increase on the water availability, among other factors, in their decision to adopt the SRI (Lee and Kobayashi, 2018). It must be noted that the finding in the preceding study was based on interview surveys, and that their conclusions require substantiation by direct observations in the field.

Here we closely observed farmers’ practices and compared the rice plant growth and yield of the SRI practices with those of the non-SRI practices in rainfed fields of southern Cambodia. At the same time, we monitored climatic and soil regimes in the fields, and analyzed soil properties to identify the relationship between the SRIinduced changes, if any, in rice growth and yield and the climatic and edaphic constraints in the rainfed environment.

In this study, we aimed at understanding, across a threeyear study, the mechanisms whereby the SRI affected rice growth and yield for rainfed lowland fields in Southern Cambodia.

STUDY AREA AND METHODS OF STUDY

Study site

We undertook an on-farm experiment in the rainfed paddies of Popel commune (11° 04' 67" N, 104° 40' 79" E) in the Tram Kak district in Takeo Province during the wet seasons in 2012, 2013, and 2015 (Fig. 1). Rice is the main crop in Takeo province, which is one of the granary zones with high rice yields in Cambodia. Tram Kak is among the districts where the SRI has been rapidly and widely adopted by farmers. A detailed description of the study site is provided in Lee and Kobayashi (2017, 2018).

The main soil type in the study area is the Prateah Lang group (Plintic Acrisol), which covers about 58% of the district and is predominantly a lowland rice soil. The soil has sandy topsoil less than 40 cm deep over subsoil with a loamy or clay texture. The soil texture is coarse with 5% clay, 22% silt, and 73% sand (Seng et al., 2001;Hin et al., 2005). The effective rooting depth is often restricted by a firm to extremely hard plow pan occurring within the top 15-20 cm (White et al., 1997).

Preliminary survey

During the harvest season in 2011 prior to the field experiment, we conducted an interview survey covering 29 households in total. We interviewed the farmers using an open questionnaire that covered farm size, crop management, nutrient management, weed and pest management practices, and SRI practices, if any.

The farmers referred to farmyard manure (FYM) in the local unit of “lotte”, which represents a cart-full of farmyard manure. The farmyard manure of one lotte was multiplied by a factor of 200 kg lotte^-1 to quantify the manure applications on a mass basis.

On-farm experiment and crop management

Out of the 25 households interviewed for the on-farm experiment, we selected eight. With some households being studied in more than one field, we conducted the onfarm experiment in a total of 15 fields during wet season of the three years of 2012 (9 fields), 2013 (8 fields), and 2015 (13 fields) (Table A1 in Appendix). SRI was applied in 8 out of the 15 fields. We did not conduct the experiment in 2014, when the rainfall was so low that the farmers gave up to plant rice in many fields (Lee and Kobayashi, 2017, 2018). The fields were characterized by their agronomic practice, topography, and availability of supplementary water supply (Table A1). In some fields, the farmers had supplementary water sources such as ponds and pumpingup of river water in addition to the rainwater. SRI was applied in the fields across the availability of supplementary water and topography of the field, which we considered the major determinants of rice growth and yield at the site of experiment (Table A1 in Appendix).

All agronomic managements on water level, weeds, and fertilizers, except the choice of rice variety was determined by the farmers according to their own practices. For research purposes, we urged the farmers to plant a popular modern variety, Phka Rumdul, in all fields for three years. Phka Rumdul is a photoperiod-sensitive medium-maturity variety, and flowers mostly in October with potential yields of 3.5–5.5 t ha⁻¹. Two fields (SRI-2-L and SRI-2-U) were nevertheless planted to a late variety in 2012, and the other two fields (SRI-3 and Non-SRI-2) were directly sown in 2015 (Table A1 in Appendix).

The farmers sowed the seeds between 10 May and 26 July (2012), between 05 July and 15 July (2013), and between 28 June and 01 September (2015). They transplanted seedlings at 20 days - 40 days after sowing except for direct seeding in the two fields (SRI-3 and Non-SRI-2) in 2015 (Table A1 in Appendix). We recorded the date when plants reached 50% flowering and that of physiological maturity.

Measurements of climatic variables

We recorded weather variables including rainfall amount, minimum and maximum temperatures, relative humidity, and solar radiation with an automated measurement system (EM 50, Decagon Devices, Inc., USA) during the growing period in a farmer’s field (SRI-2 field) located in Cham Pol commune.

Nutrient, soil and plant measurements

In some households we analyzed farm-yard manure for nitrogen and phosphoric acid contents. At the time of rice harvest in 2015, we collected a composite sample of top soils from five random sub-samples taken from the plow layer of 25 cm depth in 13 fields, which included all the fields for the on-site experiment in 2015 (Table A1 in Appendix) with some additional fields. We airdried and sieved (< 2 mm particle diameter) all soil samples, and analyzed them for soil properties: soil total nitrogen and organic carbon (Rayment and Higginson, 1992), and soil texture by a hydrometer method (Gee and Bauder, 1986).

At physiological maturity, we determined panicle number, number of spikelets per panicle, 1,000-grain weight, number of filled grains, and grain yield from the triplicates of a 1 m × 1 m area in each field. We measured grain moisture content immediately after threshing in each field at harvest (Grain moisture meter, G-won Hitec, Co., Ltd, Korea) and used it to express the grain yield at 14% moisture.

We calculated the fraction of filled grains by dividing the total number of filled grains by the total number of spikelets. We divided the grain and straw as sub-samples into culms, sheaths, and leaf for plant dry weight, after ovendrying at 75°C for over 72 h to a constant weight, which we measured at the Cambodian Agricultural Research and Development Institute, and measured the plant nitrogen content by the micro-Kjeldahl method (Kjeltec auto 1030, analyzer, Foss, Inc., USA). We calculated the amount of nitrogen accumulation in rice plants by multiplying the dry-matter weight by the nitrogen content (Tanaka et al., 2012).

Data analysis

We analyzed results of preliminary interview survey for difference between the fields under SRI and conventional non-SRI practices. We also analyzed the effects of SRI management on the crop performance in the on-site experiment using the mixed linear model, where we assigned the field to random effect variable, and other factors, e.g. the SRI practice and year, to fixed effect variables. Values measured in triplicate per field were averaged so as to make the field as the unit of the statistical analysis. We also analyzed the relationships among the crop performance variables across the years and cultivation method. For statistical analyses of the onsite experiment, we used only the records of the transplanted standard variety excluding the data of the late variety and direct seeding (Table A1 in Appendix). We used the JMP-Pro 16.0 software (SAS Institute, USA) for the statistical analyses.

RESULTS

Comparison between SRI and non-SRI practices in crop management as found in the field interview survey

The average seedling age at transplanting was 27 days in the SRI and 35 days in the non-SRI fields, with a statistically significant difference (P = 0.019) (Table 1). The transplanting density was lower in the SRI than in the non- SRI (P = 0.005), and the number of seedlings per hill was also lower in the SRI than in the non-SRI (P < 0.001). The average number of times of plowing was higher in the SRI (2.8) than in the non-SRI (2.4) (P = 0.013).

Farmers applied farmyard manure in the rice fields at significantly higher mean rate (4.8 t ha^-1) in the SRI fields than that (2.7 t ha^-1) in the non-SRI fields (P = 0.018) (Table 1). All the farmers mentioned their use of inorganic fertilizers in their fields. In study area, the farmers, without knowing the recommended fertilization rate, applied fertilizers according to their own experiences. They applied urea (46% N) and di-ammonium phosphate (N: P: K = 18: 46: 0) 2-3 times (basal application and topdressing) during the growth period at average nitrogen rates of 2.7 kg N ha^-1 in the SRI and 33 kg N ha^-1 in the non-SRI fields, which differed statistically (P = 0.045) (Table 1). They performed weeding 1-3 times by hand but did not apply pesticides. The SRI and the non-SRI did not differ with respect to the weed or pest management, however.

Climate and crop management during the on-farm experiment

Monthly summary of rainfall, humidity, temperatures, and solar radiation are given for the rice growing periods over the three years of the experiment (Table 2). The total rice season rainfall was greater in 2015 (1,040 mm) than in 2012 (421 mm) and 2013 (638 mm). Nevertheless, 2012 is remembered as a bumper year and 2015 as a drought year among the farmers. This is because the rainfall in the transplanting period from July to August was much lower in 2015 than that in 2012 and 2013. We recorded the average monthly minimum and maximum temperatures as 23.2 °C-26.0 °C and 30.1 °C- 33.9 °C, respectively, during the growing season for the three years. Average monthly solar radiation over the growth period was 18 MJ m^-2 day^-1, 15 MJ m^-2 day^-1, and 18 MJ m^-2 day^-1 in 2012, 2013, and 2015, respectively.

The farmers’ crop management and the crop phenology depended on annual rainfall. Seed sowing was done more than 25 days earlier in the SRI than in the non-SRI fields in 2012 a bumper year with sufficient rainfall whereas the difference was not significant for other years leading to the significant interaction between year and cultivation method (Table 3). The average number of seedlings per hill and transplanting density were lower in the SRI than those in the non-SRI fields, with a highly significant difference in the number of plants per hill (Table 3). Heading and harvest were significantly delayed in 2015 the drought year than in the other years (Table 3). Effects of supplementary water or topography was not significant on either crop management or crop phenology, and, hence, omitted from the statistical model.

Nutrient management and soil properties in the onfarm experiment

The farmers applied more FYM in SRI than Non-SRI fields for 2012 and 2013, but the difference was negligible in 2015, the drought year (Table 4). In contrast, the SRI fields received much less applications of inorganic fertilizers than the Non-SRI fields (Table 4). Total nitrogen and phosphorus inputs were also significantly lower in the SRI fields than those in non-SRI fields (Table 4).

The topsoil analysis demonstrated that soil organic carbon or total soil nitrogen contents were not significantly different between SRI and non-SRI soils at harvest in 2015 (Table A2 in Appendix). All fields had predominantly sandy texture (69% sand on average in SRI and 76% sand on average in non-SRI) with low clay contents (Table A2 in Appendix).

Yield components, plant biomass, grain yield, and N uptake in the on-farm experiment

Across the three years, the SRI significantly increased grain yield (P = 0.043) and number of spikelets per land area (P = 0.054) (Table 5). SRI also increased nitrogen amount in rice plants at harvest with a weak statistical significance (P = 0.071; Table 5). Topography significantly altered the crop performance: the grain yield, number of spikelets and N accumulated in plant were less in the higher fields than those in the middle and the lower fields (Table 5 and Figure A1 in Appendix). Plant dry biomass and N amount at harvest also differed significantly between years being best in 2012 followed by 2013, and poorest in the drought year of 2015 (Table 5).

The effects of supplementary water availability on crop performance were not significant, and, hence, not included in the analysis of variance. Nitrogen in plant at harvest was closely related with the grain yield (Fig. 2A) and the number of spikelets per m² (Fig. 2B) across the SRI and non- SRI fields as well as the topography and years.

Rice plants in the SRI fields had significantly more panicles per hill (P = 0. 027) and spikelets per panicle (P = 0.051) than those in the non-SRI fields (Table 6). Among the 3 years, the number of spikelets per panicle was more variable on relative basis than the other yield components (Table A3 in Appendix), although the difference between years was not significant (P = 0.109; Table 6). Thousand grain weight was not affected by SRI but significantly differed between the 3 years, although the extent of its interannual variability was small (Table A3). The fraction of filled spikelet in total spikelet was not affected by any of the variables (Tables 6 and A3). The effect of topography was not significant on any of the yield components, and, hence, was not included in the statistical analysis.

DISCUSSION

In rainfed lowland rice farming, the crop performance is subjected to variability in diverse management practices, soil fertility, water regimes, as well as unpredictable and variable rainfall patterns (Inthavong et al., 2014;Haefele et al., 2016). The efficacy of SRI practices in enhancing the crop performance under rainfed conditions needs to be discussed against these determinants and uncertainties.

Effects of lower planting density in the SRI fields in the on-farm experiment

In the SRI practice, early seed sowing and planting of younger seedlings would allow the rice plants a longer period of vegetative growth for greater biomass and nutrient accumulation. In this study area, however, the large variability of the rainfall early in the rice season (Table 2) often prevents the farmers from adopting the SRI practice at the early stages. In 2012, for example, the good rainfall in July enabled the earlier seed sowing and earlier planting and heading in SRI fields (Table 3). In 2015, in contrast, the early season drought forced some SRI farmers to make the seed sowing for the second round, which delayed the seed sowing significantly in comparison to the other two years (Table 3).

While the seedling preparation is subjected to the natural fluctuations of early rainfalls, the farmers can control the population density of the seedlings as evidenced by the significant difference between the SRI and non-SRI fields in the number of plants per hill and the number of hills per land (Table 3). Compared with the seedlings planted at a high density, seedlings planted in a wider spacing would have greater tiller production for stronger individual plants (San-oh et al., 2004;Horie et al., 2005;Pasuquin et al., 2008;Ray and Barik, 2015). The greater number of spikelets per panicle in SRI fields in Table 6 conforms to the findings in a previous report (Thakur et al., 2016). While the effect of SRI on this yield component is only weakly significant (P = 0.051), its increase should have increased the grain yield (Table 5) since the other yield components showed no significant effects of SRI (Table 6).

It is also noteworthy that, despite the reduction in planting density (Table 3), the number of panicles per land area was not reduced in the SRI fields due to the increased number of panicles per hill (Table 6). Several previous studies reported similar findings (Uphoff, 2003;Thakur et al., 2016;Nugroho et al., 2018), where the SRI resulted in prolific tillering and associated root development, and increased panicle development that more than compensated for the reduced plant population on an area basis.

Effects of topography in the on-farm experiment

In the region of this study, rice growth and yield were very strongly influenced by the topography, and were much poorer in the higher fields than in the middle and the lower fields (Figure A1). The effects of the topography could be accounted for by the soil moisture content in the fields, as reported for rainfed lowland rice in Northeast Thailand (Homma et al., 2003) and Southern Laos (Tsubo et al., 2006). In lower fields, the water is more easily retained, since the water table is often close to the land surface and the main rooting zone. In contrast, the water table in the higher fields is often below the main rooting zone, and the only water available to the seedlings with later planting of early maturing variety is retained above the hard pan (Miyagawa and Kuroda, 1988;Homma et al., 2003;Fukai and Ouk, 2012;Haefele et al., 2016). Rice yields in rainfed fields thus reflect the constraints of variable water conditions at different parts of the sloped topography (Tsubo et al., 2006).

Effects of SRI practices on nutrient budget in the on-farm experiment

In rainfed lowland paddies, rice productivity is strongly correlated with soil organic carbon content (SOC) (Homma et al., 2003;Haefele and Konboon, 2009), which indicates the critical importance of soil-born nutrient supply. For SRI in Madagascar, rice yields were reported to have been increased by the high N-supplying ability of the soil from accumulated SOC (Tsujimoto et al., 2009). A more recent study in Cambodia also indicated that the application of FYM was significantly correlated with rice yield (6 t ha^-1 SRI vs 3 t ha^-1 non-SRI) (Ly et al., 2016). Although we did not find significant difference in soil carbon content between the soils in SRI and non-SRI fields (Table A2), SRI fields had a higher input of farmyard manure than that of the non-SRI as we observed in the preliminary interview (Table 1) and the on-farm experiment (Table 4).

While the calculated total nitrogen input was less in SRI than non-SRI fields due to the very small inputs of chemical fertilizer input in the SRI fields, the nutrient availability to the rice plants may not have been less in SRI. The predominantly sandy soil should have only poorly retained the nutrients applied as inorganic fertilizers. The lower planting density along with the smaller number of plants per hill (Table 3) should have in effect made a larger amount of soil-borne nutrients available for the plant growth at the early stages. The very low rate of inorganic fertilizers would have also been conducive for natural nitrogen fixation beyond the early stages.

The effect of plant nitrogen content on spikelet differentiation reported in a previous study (Kobayasi and Horie, 1994) also suggests greater nitrogen availability that increased the number of spikelets per panicle (Table 6) and thereby grain yield (Table 5). Close relationships as established between the grain yield and the number of spikelets and plant N accumulation (Yoshida, 1981) were observed in the on-farm experiment in this study also (Fig. 2). A similar result was reported in an SRI–non-SRI comparison study in Madagascar (Barison and Uphoff, 2011).

It must be noted that, in the SRI fields in 2012 and 2013, the rice plants accumulated about 7 g m^-2 of nitrogen (Table 5), and that the total nitrogen input from the farmyard manure and a minor amount of inorganic fertilizers was only 30 kg ha^-1 or 3 g m^-2 (Table 4). Closing the nitrogen budget would require quantification of the effects of SRI practices, e.g. the lower planting density and higher amount of organic matter application along with the reduced or no application of inorganic fertilizers, on the efficient use of nitrogen within the local agricultural ecosystems. It is important to note that farmers in the rainfed lowlands of the Cambodian production system are already operating with limited resources in a vulnerable production environment. Thus, the implementation of the SRI based on agroecosystem principles could lead to more stable and sustainable production systems with less use of external inputs and addition of locally available organic matter via the enhanced soil nutrient supply to the rice.

CONCLUSION

The on-farm experiment showed that reduced planting density should increase plant N accumulation at harvest and thereby grain yield via the increased number of spikelets per panicle. The reduced planting density in combination with the increased nitrogen supply in the form of farmyard manure would facilitate the efficient use of the nutrient with lower leachate losses from the sandy soils. The availability of manure does not appear to be a major constraint at present, but it could become a limiting factor if the SRI is adopted at much larger scales in

적 요

벼 집약 시스템(System of Rice Intensification, SRI)은 보 다 지속 가능한 벼 재배 전환 방법으로 제안되었다. 그러나 천수답 지역에서의 다양한 환경적 제약으로 인해 SRI 수행을 채택하기 어려운 경우가 많다.