Soil and phenological characteristics were collected at an experimental plot conducted by the Cambodian Agricultural Research and Development Institute (CARDI), located at E 11°28'34.4", N 104°48'36.5, in 2010, 2011 and 2013. Two aromatic milled rice cultivars were analyzed: (1) short duration, early maturity, a short non-photosensitive rice, cv. Sen Pidao and (2) medium duration, medium maturity, a taller photosensitive wet season rice, cv. Phka Rumduol (Kamoshita et al., 2016). The two varieties (Phka Rumduol in 1999 and Sen Pidao in 2002) were released by the CARDI to improve food security (Trak, 2009).

The future climate data for this study were collected from CORDEX-East Asia. We selected the HadGEM3-RA model, a regional climate model, to investigate the effects of climate change on rice yields in Cambodia, because the model provides the longest period. More detailed information on these models is given on the CORDEX-East Asia webpage (https://cordex-ea.climate.go.kr/). The projected data were based on two IPCC greenhouse gas concentration scenarios (RCP4.5 and RCP8.5) for this study. The yearly average changes in the projections for total daily radiation and maximum and minimum temperature are listed in Table 1. For both emission scenarios, the projected radiation and temperatures were higher than the baseline. The increase in the minimum temperature was generally greater than the increase in the maximum temperature. Compared with baseline weather conditions, the mean daily maximum temperature was projected to increase by 1.9℃ in the 2040s and 2.5℃ in the 2080s under the RCP4.5 scenario and by 2.1℃ and 4.8℃ for the respective decades under the RCP8.5 scenario.

The baseline years were the years 1991 to 2000 and the simulation years were 2041∼2050 (the 2040s), 2051∼2060 (the 2050s) and 2081∼2090 (the 2080s). The CO₂ concentrations for all the periods used for this study were 362, (499 and 571 for RCP4.5 and RCP8.5, respectively), (499 and 571 for RCP 4.5 and RCP8.5, respectively), and (532 and 801 for RCP4.5 and RCP8.5, respectively) µmol mol^-1 for the baseline, the 2040s, the 2050s, and the 2080s, respectively. However, the question on howthese CO₂ concentration scan impact on rice productivity is not addressed in this study. It is recommended that this question be addressed in a further study. Daily weather data included total solar radiation, maximum and minimum temperature, and precipitation. Cultivar coefficients should be calibrated to achieve the observed yield and biomass under no-stress growing conditions (Boote, 1999). In this study, the experiments conducted in 2011 were selected for the calibration of the cultivar coefficients, while those in 2010 (Sen Pidao and Phka Rumduol) and 2013 (Phka Rumduol only) were selected for the validation of the CERES-Rice model. Genotype coefficients were determined by the “trial and error” method to compare the best match. The estimates of genotype coefficients are summarized in Table 2. The genotype coefficients P20, G1, G2, G3, and G4 for both cultivars were similar to each other, while P1, P2R, and P5 for both cultivars are quite different to each other. However, a further study is recommended on the calibration using longer period of the experiments. The Gen-Calc program in DSSAT was employed to estimate cultivar coefficients (Hunt et al., 1993). The simulated yields of both cultivars were nearly same as the observed yields in the calibration year, while 1.2 to 3.1% of rice yields were higher for the simulated than observed values in the validation years. More detailed information on the calibration and validation of the CERES-Rice model can be found in Chun et al. (2016).

Simulations of the effect of adjusting the planting date under the RCP4.5 and RCP8.5 future climate change scenarios were conducted using the CERES-Rice model to investigate a suitable agronomic option for adaptation. Cambodia’s accommodated fertilizer rate was 50 N kg ha^-1 (CARDI, 2012) which is much lower than the other countries in Southeast Asia. According to the FAO (2010), farmers in Vietnam applied 221 kg of fertilizer on average and 108 kg on average in Thailand. For this study, considering the possible increase in fertilizer usage in the future, we simulated rice growth and yields with six fertilizer rates (N:P:K, kg ha^-1): N0 (0:0:0), N33 (33:23:30), N50 (50:23:30), N100 (100:23:30), N150 (150:23:30), and N200 (200:23:30). In this study, we focused on adaption strategies in accordance with the changing climate, considering 6 different fertilization rates. A wide range of adaptive actions can be implemented to reduce or overcome the negative effects of climate change on rice yields. For this study, the adaption measures that were investigated in this study were a combination of changing of planting dates, along with fertilization. Farmers can strategically shift the time of planting to control the plant growth-rate. To determine the optimum planting dates for the greatest yields, the planting dates were shifted from the actual planting date in 2011 (day-of-year, DOY 195) to DOY147 and to DOY 245, a shift of 48 days before and 50 days after, respectively, at an interval of 1 day. Changes in rice yields were then compared with the rice yield in the baseline. Since the effects of planting density on changes in rice yields were not investigated, a further study on the planting density is recommended.

The yield variations of both cultivars at six different fertilizer rates for the baseline years (1991∼2000) are shown in Fig. 1. Generally, rice yields increased with more fertilizer but reached a plateau if more than 100 kg N ha^-1 of fertilizer was applied. This result implies that 100 kg N ha^-1 of fertilizer rate can be economically better and is in good agreement with that of Chun et al. (2016). They reported that 100 kg N ha^-1 of fertilizer rate would be cost-effective in Cambodia through a benefit-cost ratio analysis. For Sen Pidao, the rice yield was 3,264 kg ha^-1 without any fertilizer (N0), increasing to 5,334 kg ha^-1 with applied nitrogen at 33 kg N ha^-1 (N33), and reaching the yield of 6,533 kg ha^-1 with applied nitrogen at 100 kg N ha^-1 (N100). For the Phka Rumduol cultivar, rice yield increased from 3,713 kg ha^-1 without fertilizer application (N0) to 7,725 kg ha^-1 with nitrogen fertilizer at 100 kg N ha^-1 (N100).

Fig. 1. 
Average simulated rice yields of Sen Pidao and Phka Rumduol at 6 different fertilizer rates for the baseline years (1991∼2000). The vertical bars represent±SE (standard errors). The fertilizer rates (N:P:K, kg ha^-1) are: N0 for 0:0:0, N33 for 33:23:30, N50 for 50:23:30, N100 for 100:23:30, N150 for 150:23:30, N200 for 200:23:30.

To assess the impacts of climate change on rice yields, we simulated the rice yields for two cultivars with the projected future climate variables, i.e. a combination of temperature, precipitation, and solar radiation for each year from 2041∼2050, 2051∼2060 and 2081∼2090 (Fig. 2 and Fig. 3). For the Sen Pidao cultivar under the RCP4.5 scenario, yields decreased under the effects of climate change (Fig. 2a). The yields were projected to decrease by 15∼27%, and 21∼34% in the 2040s and 2080s, respectively. For the Phka Rmuduol cultivar under the RCP4.5 scenario, the yields decreased much less than for Sen Pidao, starting with a 7% decrease in the 2020s at fertilizer F2 to a 21% decrease in the 2080s at fertilizer N100 (Fig. 2b).

Fig. 2. 
The projected rice yield change (%) relative to the simulated yield for the baseline years (1991∼2000) for Sen Pidao (a) and Phka Rumduol (b) at 6 different fertilizer rates under the RCP4.5 scenario for the 2040s, 2050s and 2080s. The fertilizer rates (N:P:K, kg ha^-1) are: N0 for 0:0:0, N33 for 33:23:30, N50 for 50:23:30, N100 for 100:23:30, N150 for 150:23:30, N200 for 200:23:30.

Under the RCP8.5 scenario, rice yields for Sen Pidao decreased with climate change for all combinations of variables (Fig. 3a). The simulated declines in yields were up to 59% in the 2080s (N33). The yields in the 2080s were generally much lower than the baseline. For the Phka Romduol cultivar, the relative changes in yields in the 2040s and 2050s were much less than for Sen Pidao, while the changes in yields in the 2080s were much larger than in other years (Fig. 3b). These results can be explained by considering that temperature was the largest contribution to decreases in rice yields in Cambodia (Li et al., 2017). Li et al. (2017) reported that the contribution of temperature to rice yield changes account for approximately 64.7%. Chun et al. (2016) reported that approximately 4.5℃ of temperature was projected to increase in the 2080s. The ADB (2009) projected that temperature would increase by 4.8 ℃ by 2100 for the Southeast Asia region. In some cases, yields were projected to slightly increase in the 2040s at N0 and N33 fertilizer levels.

Fig. 3. 
The projected rice yield change (%) relative to the simulated yield for the baseline years (1991∼2000) for Sen Pidao (a) and Phka Rumduol (b) at 6 different fertilizer rates under the RCP8.5 scenario for the 2040s, 2050s and 2080s. The fertilizer rates (N:P:K, kg ha^-1) are: N0 for 0:0:0, N33 for 33:23:30, N50 for 50:23:30, N100 for 100:23:30, N150 for 150:23:30, N200 for 200:23:30.

Under the RCP4.5 scenario, yields decreased as the planting date was moved to earlier in the year and increased for a planting date somewhat later, depending on the fertilization rate, cultivar and simulated year (Fig. 4). For example, in the 2040s at fertilizer N0, the highest yields for Sen Pidao were achieved by planting on DOY 229, resulting in a 12.7% increase relative to the baseline years (1991∼2000), while at fertilizer N200, the highest yields were achieved by planting on DOY 242, with a 41.4% increase (Fig. 4a). However, for Phka Rumduol, the highest yields in the 2040s at fertilizer N0 were achieved by planting on DOY 198, which is an increase of only 3.9%; the highest yield for fertilizer N200 was at DOY 228, which represents a yield increase of 13.3% (Fig. 4d).

Fig. 4. 
Changes in simulated mean rice yields (%) for the Sen Pidao (a∼c) and Phka Rumduol (d∼f) cultivars under the RCP4.5 climate scenario at 6 different fertilizer rates with shifting planting date (a shift from the actual planting date in 2011 of 48 days before and 50 days after, respectively, at an interval of 1 day). (a) and (d): the 2040s, (b) and (e): the 2050s, and (c) and (f): the 2080s. The fertilizer rates (N:P:K, kg ha^-1) are: N0 for 0:0:0, N33 for 33:23:30, N50 for 50:23:30, N100 for 100:23:30, N150 for 150:23:30, N200 for 200:23:30.

Under the RCP8.5 scenario, yields increase as the planting date is shifted to later in the season (Fig. 5). Yields for Sen Pidao under the RCP8.5 scenario varied with planting dates and had more clear peaks except for the 2080S than under the RCP4.5 scenario. At the highest fertilization levels, yields were the more sensitive to the planting date. In the 2040s, yields increased by 15% when the planting date was shifted to DOY 214 at fertilizer N0, and increased to 40.9% when the planting date was shifted to DOY 239 at fertilizer N200 (Fig. 5a). In the 2080s (Fig. 5c), the trends of changes in rice yields were different from the other two (i.e., the 2040s and 2050s). Changes in rice yields showed a tendency to increase as a result of shifting the planting date. For Phka Rumduol, the variations in yields were less sensitive to shifting the planting date compared with Sen Pidao (Fig. 5a to c). In the 2040s, the highest yields for Phka Rumduol only increased by 1.3% at fertilizer N0 and by 8.4% at fertilizer N200 (Fig. 5d).

Fig. 5. 
Changes in simulated mean rice yields (%) for the Sen Pidao (a∼c) and Phka Rumduol (d∼f) cultivars under the RCP8.5 climate scenario at 6 different fertilizer rates rates with shifting planting date (a shift from the actual planting date in 2011 of 48 days before and 50 days after, respectively, at an interval of 1 day). (a) and (d): the 2040s, (b) and (e): the 2050s, and (c) and (f): the 2080s. The fertilizer rates (N:P:K, kg ha^-1) are: N0 for 0:0:0, N33 for 33:23:30, N50 for 50:23:30, N100 for 100:23:30, N150 for 150:23:30, N200 for 200:23:30.

Many studies have shown that adjusting the planting dates for rice cultivation will be helpful for improving rice yields under the effects of climate change (e.g., Azdawiyah et al., 2015; Chun et al., 2016; Dharmarathna et al., 2014; Li et al., 2017). By changing the planting date, plant development stages are shifted, which suggests that rice plants may grow under a more appropriate temperature range and/or sufficient solar radiation for grain-filling; therefore, shifting the rice planting date following climate change can allow for rice plants to be subjected to more optimal conditions during the plant development stages.

We investigated the effects of the combinations of nitrogen fertilizer level and planting date on rice yields under climate change. We assumed that the managements in the baseline years (i.e., the recommended nitrogen fertilizer, 50 kg N ha^-1 and the planting date, DOY 195) were applied in the future periods. The simulated rice yields under the two emission scenarios at 6 different fertilizer rates with shifting planting date (a shift from the actual planting date in 2011 of 48 days before and 50 days after, respectively, at an interval of 1 day) were then compared with those under climate change at N50 (50 kg N ha^-1, there commended nitrogen fertilizer application rate in Cambodia) (Seng et al., 2001) and the planting date in the base line years (i.e., DOY 195). Changes in rice yields (%) are presented in Fig. 6 and Fig. 7 under RCP4.5 and RCP8.5, respectively.

Fig. 6. 
Changes in simulated mean rice yields (%) relative to yields with no shifting the planning date and N50 for the Sen Pidao (a∼c) and Phka Rumduol (d∼f) cultivars under the RCP4.5 climate scenario at 6 different fertilizer rates with shifting planting date (a shift from the actual planting date in 2011 of 48 days before and 50 days after, respectively, at an interval of 1 day). (a) and (d): the 2040s, (b) and (e): the 2050s, and (c) and (f): the 2080s. The fertilizer rates (N:P:K, kg ha^-1) are: N0 for 0:0:0, N33 for 33:23:30, N50 for 50:23:30, N100 for 100:23:30, N150 for 150:23:30, N200 for 200:23:30.

Fig. 7. 
Changes in simulated mean rice yields (%) relative to yields with no shifting the planning date and N50 for the Sen Pidao (a∼c) and Phka Rumduol (d∼f) cultivars under the RCP8.5 climate scenario at 6 different fertilizer rates with shifting planting date (a shift from the actual planting date in 2011 of 48 days before and 50 days after, respectively, at an interval of 1 day). (a) and (d): the 2040s, (b) and (e): the 2050s, and (c) and (f): the 2080s. The fertilizer rates (N:P:K, kg ha^-1) are: N0 for 0:0:0, N33 for 33:23:30, N50 for 50:23:30, N100 for 100:23:30, N150 for 150:23:30, N200 for 200:23:30.

For the Sen Pidao under RCP 4.5 at N100 (the cost-effective fertilizer rate, 100 kg N ha^-1), the highest increases were simulated when rice plants would be planted 30 to 45 days after the baseline planting date. However, for the Phka Rumduol cultivar under RCP 4.5 at N100, the highest increases were simulated when short shifts (7 days before and after) of planting date were made than for the Sen Pidao. For the Sen Pidao under RCP 4.5 at N100, the highest increase in rice yields (58.1%) was simulated on DOY 230 (i.e., 35 days after the baseline planting date) in the 2040s, while the highest increase (85.4%) was simulated on DOY 240 in the 2080s (Fig. 6a and c). For the Phka Rumduol cultivar under RCP 4.5 at N100, rice yields in the 2040s were simulated to increase by 30.2% when rice plants would be planted 7 days after the baseline planting date, while the highest increase in the 2050s was simulated when rice plants would be planted 7 days before the baseline planting date. Similar results were found in those under RCP 8.5 at N100 (Fig. 7). The highest increases for San Pidao were simulated when longer than a month shifts (36 to 48 days) of planting date were made (Fig. 7a to c). In contrast, for Phka Rumduol, about a week shifts of planting date were found to highest increase rice yields in the 2040s and 2050s. These results indicate that forward shifts of planting date from the baseline plating date for the Cambodian rice cultivars in the future can be more benefitted than backward shifts. These results are in good agreement with those of Li et al. (2017) and Dharmarathna et al. (2014). They reported that the CERES-Rice model simulated to increase rice yields in Sri Lanka under both A2 and B2 scenarios with advancing the rice planting date by 1 month. Jalota et al. (2013) reported that several climate variables including temperature, CO₂, rainfall, solar radiation, relative humidity, etc. compositionally contribute to the determination of the crop yield. Therefore, substantial investigation and analysis might be required to explain the yield response to climate variability. A further study on this is recommended.