For model calibration, five popular upland rice genotypes namely, Dawk Pa-yawm, Mai Tahk, Bow Leb Nahng, Dawk Kha 50 and Dawk Kahm were grown on 8^th July 2018 to calibrate the CSM-CERES-Rice model by using a temperature tolerant cultivar (Basmati 385). The experiment was conducted as a split plot using Complete Block Design (CRD) as main plots with three replications. The main plots had five upland rice genotypes and sub plots had three different temperature treatments (28˚C, 30˚C and 32˚C). Temperature treatments were set up by using the data logger (UA–003–09 HOBO Pendant, for Temp/Light) from starting of the experiment. Each plot had 5 rows, with 10 plants each. Standard agronomic practices were followed for treatment of fertilizers, weed and insect control. The crop management data (i.e., phenological data) required for the simulation of the crop model include planting date, (50% and 100%) germination date, flowering and maturity, tillers number, panicles number, leaf area index, grain yield, biomass and 1000 grain weight were recorded harvest.

CERES-Rice present in DSSAT v4.7which is an advanced physiologically based model, was used to calibrate and evaluate the crop simulation model. Genetic co-efficients for the five upland rice genotypes were used to calibrate CERES-Rice model. Soil analysis was carried out before started of the experiments to analyze soil fertility and to carryout proper fertilizer management. Weather parameters including maximum and minimum temperatures, rainfall with air intensity were recorded from Kho Hong Agro meteorological station. Daily solar radiation (MJm⁻² day⁻¹) was calculated by using weatherman tools in DSSAT v4.7. Dawk Kahm genotype was used as border crop to avoid the varietal errors. To run the model the following five input files were created:

For model evaluation, selected five upland rice genotypes namely Dawk Pa-yawm, Mai Tahk, Bow Leb Nahng, Dawk Kha 50 and Dawk Kahm and three temperature treatments were laid out in a Completely Randomized Design (CRD) under split plot design structure with three replications in the 2017/2018 and 2018/2019 growing seasons at the research area of Plant Science Department, faculty of Natural Resources, Prince of Songkla University Hat Yai campus, Thailand. Soil samples were collected fifteen days before sowing for important chemical and physical characterization (Table 3). We must do two seasons research experimental work for model evaluation. Here planting was done on 24^th

July for the 2017/2018 and 2018/2019 seasons. Different fertilizers whenever necessary. Other management practices were carried out accordingly. The number of days to anthesis and physiological maturity, and grain filling data were collected. Moreover, grain yield and total plant biomass was measured at physiological maturity.

The analysis of variance (ANOVA) to evaluate cultivars growth and development and the effect of different temperatures was done. Mean separation was done by Least Square Difference (LSD) for split plot design to see the varietal differences. Test of significance between the 2017/2018 and 2018/2019 experiments and simulated and measured quantities was performed by R/agricolae program [14]. This research focused to simulate the effect of different temperatures on yield performed for phenology, grain yield, tops weight, leaf area index, harvest index, and tillers number.

Calibration is the process of adjusting some model parameters to local environmental conditions and obtains genetic coefficients for new cultivar used in modeling study [15]. According to [16] model calibration was done by the Root Mean Square Error (RMSE).

RMSE = ∑ i = 1 n ( S i − O i ) 2 n (1)

Model performance was evaluated by comparing the simulated versus observed values from upland rice experiment under rainfed conditions where an agreement index or d stat index [16] was used. The RMSEn gives the level of error associated with each evaluation between the observed and simulated values.

d = 1 − ∑ i = 1 n ( S i − O i ) 2 ∑ i = 1 n ( | S ′ i | + | O ′ i | ) 2 (2)

RMSEn = ( RMSE × 100 ) / O ¯ (3)

where S_i = simulated value, O_i = observed value, n = number of observations, O ¯ = overall mean of observed values, S ′ i = S i − O ¯ and O ′ i = O i − O ¯ .

From Figure 1 and Figure 2 it was showed that in both 2017 and 2018 years, weather conditions had no visible variations. The maximum and minimum

temperatures ranged between 25˚C to 38˚C and 18˚C to 22˚C. Highest maximum temperature (35˚C) and highest minimum temperature (26˚C) shown at ambient temperature 28˚C and 32˚C for 2018 Solar radiation ranged from 10.2 to 20 MJ∙m⁻²∙d⁻¹, rainfall was reported between 20 - 980 mm and the highest rainfall was shown in the month of November to mid-December. After that January month was second most rainfall month near about 850 mm rainfall, respectively. Extreme high temperature beyond the average temperature hampered during experiment might have negative influence on upland rice yield, crop growth and development.

Different soil layers of the study site with surroundings contain sandy clay loam texture and were highly acidic. Special management e.g., irrigation due to higher clay contents and drainage system should be adopted for heavy rainfall. As the soil was acidic, phosphorus fixation was required to improve soil pH, phosphorus fertilizers were added.

Although ANOVA results (Table 4) for yield contributing traits using least significant differences (LSD) test (P < 0.05 and P < 0.01) revealed that some traits showed highly significant differences with temperatures and genotypes. Grain yield, tops weight, harvest index, flowering date, and maturity date was significantly affected by temperatures. Non-significant difference for tillers number, and leaf area index occurred possibly due to the optimum input of temperatures at the early stage. Whereas all the phenological traits had shown highly significant

variation among different genotypes. Grain yield, flowering date and maturity date had interaction effects between temperature and genotype.

Table 5 showed that both grain yield at harvesting period and tops weight at physiological maturity varied significantly (P < 0.05) during 2017/2018 and 2018/2019 seasons, except harvest index at 50% flowering did not fluctuate. So, there was significant variation among genotypes in the 2018 and 2019 seasons. Also, there was inter seasonal variation in the tested variables for tops weight and grain yield. The similarity in the parameters for the 2018 and 2019 growing seasons suggested that at the growing condition period rainfed condition, nutrients, fertilizers, and other managements were affected the yield contributing traits. Though, sowing date for both seasons was the same 24^th July, and weather elements were comparatively similar in both years, finally the similarity could not be expected.

Experiment for estimation cultivar coefficient was used for model calibration. Data of observed and simulated days to anthesis, physiological maturity, grain yield, by product and tops weight were collected for all the genotypes (Table 6(a) and Table 6(b)) indicating that cultivar specific parameters within the model were reasonably adjusted. Similarly, there was a good relationship among observed and simulated variables (d state values were respectively 0.97, 0.98, 1.0,

1.0 and 0.99 for all the yield attributes. Dawk Pa-yawm showed high RMSE with respect to both number of anthesis days and physiological maturity than other genotypes. Mai Tahk had higher RMSE with respect to grain yield and tops weight also indicated that was high yielding genotypes and the yield is associated with the growth duration since it took longer than others to attain anthesis and physiological maturity. Besides, Bow Leb Nahng had lower RMSE with anthesis days, maturity days, grain yield, by product and tops weight indicated lower yielding genotype.

Table 7 showed that Dawk Kahm genotype required few thermal units to complete juvenile stage (P1) compared to Mai Tahk genotype. This allowed Mai Tahk required extra time to accumulate photosynthesis before flowering and hence higher yielding genotype compared to other cultivars. Dawk Kahm was formerly bred for temperature tolerance condition indicated that few temperature units of short duration were just required to attain end of juvenile stage. Dawk Kha 50 required few temperature units from anthesis to physiological maturity (P5), contrasting Mai Tahk with highest thermal time requirement (G4). Spikelet number coefficient (G1) and single grain weight (G2) also high in Mai Tahk cultivar and lower in Dawk kahm cultivar. This resembled to differences in grain size between the two cultivars. The rate of grain development was high in Mai Tahk compared to others since this could be a temperature tolerant cultivar for rainfed environment with upland condition. Temperature tolerant coefficient (G5) was for the cultivars varied from 0.78˚C-1.09˚C for Dawk kahm to Mai Tahk genotypes.

The cultivar specific parameters found from experiments were used to evaluate CSM-CERES-Rice for simulating different temperature treatments under rainfed upland condition. The model simulated well the average number of anthesis to maturity days with high degree of agreement as indicated by the agreement index (d-statistics) (Table 8). This is an indication that the model calibration and resulting cultivar specific parameters were reasonably estimated. Generally, there were significant different (p < 0.05) between observed and simulated data at all temperature treatments and in all variables. Especially simulated yield decreased as temperature increased in both model simulation and experimental observations. This suggests that the CERES-Rice model is sensitive to climatic variables such as temperature. Furthermore t-test showed significant difference (P < 0.05) between simulated and observed yields at all temperature treatments (Table 9).

For model evaluation, a comparison was made for the five upland rice genotypes between simulated and observed grain yield at 28˚C, 30˚C and 32˚C temperature treatments (Table 7). In this study all the genotypes had shown a close agreement between observed and simulated values of grain yield data. Results showed that Mai Tahk and Bow Leb Nahng were the best cultivars due to

increasing simulated grain yield and best performed with the moderate temperature at 30˚C (Mai Tahk 5500 kg∙ha⁻¹ and Bow Leb Nahng 5000 kg∙ha⁻¹). Whereas at 32˚C temperature treatment, a large variation for all the genotypes decreased in yield with the maximum temperature increased. Highest reduction in simulated yield was recorded for Dawk Kha 50 cultivar (3650 kg∙ha⁻¹ at 28˚C, 4220 kg∙ha⁻¹ at 30˚C and 3055 kg∙ha⁻¹ at 32˚C) compared to others while minimum temperature increased at high temperature (30˚C) stage.