
Research Article


Estimating Cultivar Coefficients of a Spring Wheat Using GenCalc and GLUE in DSSAT 

O.M. Ibrahim,
A.A. Gaafar,
Asal M. Wali,
M.M. Tawfik
and
Marwa M. ElNahas



ABSTRACT

Background and Objective: Crop simulation models are used for simulating crop growth as affected by management and climate. Simulating the growth of a certain variety in a certain soil, climate and management needs specific parameters of that variety due to the genetic variations among varieties, which are called genetic coefficients. Methodology: Decision Support System for Agrotechnology Transfer (DSSAT) cropping system model has two programs for estimating specific parameters of a variety. Genotype coefficient calculator (GenCalc) and Generalized Likelihood Uncertainty Estimation (GLUE). An experiment was conducted during winter seasons of 2012/2013 and 2013/2014 to simulate the effect of three rates of nitrogen fertilizer (75, 100 and 125 kg N/feddan) on grain yield and its components of wheat cultivar Sakha 93 and to make a comparison between GenCalc and GLUE in their ability to assess the genetic coefficient of the cultivar. Results: Results showed that GenCalc program performed better than GLUE. The results of model validation revealed that the average of the difference between the simulated and observed parameters when using GenCalc were 4.02, 3.96 and 4.14% for biological yield, grain yield and straw yield, respectively, while they were 5.47, 8.32 and 6.12% for the same aforementioned parameters when using GLUE. The GLUE has three disadvantages, first it does not provide estimation for PHINT (Interval between subsequent leaf tip appearances), 2nd it does not provide options for keeping some coefficients fixed, while others are being calibrated like GenCalc, for example in wheat crop there are spring wheats and winter wheats, in GenCalc it can set P1V (required days for vernalization) at 0 meaning that this variety is spring, while in the same time GenCalc are calibrating the other coefficients, 3rd GLUE takes a lot of time for calibration. Conclusion: However, GLUE is more easily to use than GenCalc.





Received: February 11, 2016;
Accepted: May 12, 2016;
Published: June 15, 2016


INTRODUCTION
Crop modeling is used in many areas of agricultural researches. Crop growth models, such as grain cereal model simulate crop growth and grain yield in response to weather, soil conditions and management practices^{1}. Crop models have been validated and applied in agriculture in many study fields^{2}, such as assessing the effect of climate change on crop production^{3}, assessing cultivar performance^{4}, evaluating the adaptation of a new cultivar to a certain location^{5}, exploring the interaction between genotype and environment^{6}, prediction of crop yield^{7} and optimizing management^{8}. Specific field experiments are designed to estimate cultivar coefficients^{9}, which requires sampling of growth data at intervals during the growing season. Anothai et al.^{10} found that the minimum data needed for the estimation of cultivar coefficients in crop simulation model may be lowered to two developmental stages along with plant growth analysis data on three dates, with no need to measure Leaf Area Index (LAI). However, this reduced sampling method for growth data is still time consuming and requires many labor work. Another approach for estimation of cultivar coefficients without conducting a specific field experiments containing intensive data collection is to estimate the cultivar coefficients by using endofseason data. Genotype coefficient calculator (GenCalc) is used for the optimization of cultivar genetic coefficients. The GenCalc software had been developed to facilitate the calculation of cultivar coefficients from cultivar trial data^{11}. Another program for estimating cultivar coefficients is Generalized Likelihood Uncertainty Estimation (GLUE). The GLUE cultivar specific parameters estimation method was integrated into DSSAT using the R language.
MATERIALS AND METHODS
An experiment was conducted at Soil Salinity and Alkalinity Laboratory, Alexandria, Egypt during 2012/2013 and 2013/2014 winter seasons to simulate the effect of three levels of nitrogen fertilizer 75, 100 and 125 kg N/feddan (1 feddan = 0.42 ha) on grain yield and its components of wheat cultivar Sakha 93 using DSSAT. The design of nitrogen fertilizer levels was RCBD with four replications, where the nitrogen fertilizer levels were the first factor of an experiment including another factor (salicylic acid), only the data of nitrogen fertilizer levels under the control level of salicylic acid was used since DSSAT does not simulate the effect of salicylic acid, each experimental plot area was 1.125 m^{2} containing sandy loam soil, with organic matter content (0.63%) and calcium carbonate content (2.3%). It was classified as nonsaline soil Ece = 1.82 dS m^{–1} and soil pH (1:2.5) = 7.53. Cation Exchange Capacity (CEC) was 20 cmolc kg^{–1}, every plot contains four rows, the grains were sown in lateNovember in each year, before sowing all plots were fertilized by adding superphosphate 15.5% P_{2}O_{5} at a rate of 100 kg/feddan, potassium sulphate 48% K_{2}O at a rate of 100 kg/feddan and the nitrogen fertilizer rates were added at the rates of control, 100 and 125 kg N/feddan of ammonium sulphate 20.5% N in three doses, at sowing, at the first irrigation and at the 2nd irrigation. At the end of the experiment, number of grain per spike, number of spikes per meter square, 1000 grain weight (g), grain yield gram per plot, biological yield gram per plot, straw yield gram per plot, harvest index as the ratio between grain yield and biological yield were measured. The GenCalc program of (DSSAT version 4.5) was used to estimate the cultivar coefficients of the wheat cultivar. The GenCalc is a software used for the calculation of cultivar coefficients for use in many crop models^{12} including the CERES wheat model, which has 7 cultivar coefficients that describe growth and development of a wheat cultivar (Table 1).
Because Sakha 93 variety is a spring wheat, P1V (required days for vernalization) was set to 0. The GenCalc starts with the initial values of the cultivar coefficient. The algorithm searches in the output of the crop model file and based on the difference between simulated and observed target variables, it tends whether to increase or decrease the value of the coefficient that is being optimized. The sequences for the optimization procedures followed with GenCalc are summarized in Fig. 1. The first step was to set P1V to 0. Then, the coefficient of days to anthesis (ADAP) was adjusted to produce the lowest RMSE between the simulated and observed values of days to anthesis. The next step was adjusting the days from anthesis to maturity (MDAP) to obtain the lowest RMSE between the simulated and observed days to maturity. Then interval between subsequent leaf tip appearances (PHINT) was adjusted based on the target of leaf number on main stem. Next, standard, nonstressed mature tiller weight including grain (G3) was adjusted based on grain yield components then the standard kernel size under optimum conditions (G2) was adjusted until the simulated and observed values for final grain size provide the lowest RMSE. Then, the coefficient of kernel number per unit canopy weight at anthesis (G1) was calibrated based on grain yield components. Finally, readjust G1, G2 and G3 at the same time. The GLUE is simple to use program, just select a crop, then a cultivar from a list of cultivars included in the database for the crop in DSSAT and the treatments from experiments, in which that cultivar was grown, define at least 6000 runs for phonological parameters and another 6000 runs for growth parameters.
Table 1:  Cultivar coefficients and growth parameters of wheat CERES model 


Fig. 1: 
Sequence for calibrating the cultivar coefficients using GenCalc 
The first step was to choose a cultivar to calibrate then choose treatments to be used in estimating the cultivar coefficients then run the program for phonological parameters first then update the cultivar file with phonological parameters produced by GLUE then rerun the program again for growth parameters and finally update the cultivar file with the growth parameters produced from the second run.
RESULTS AND DISCUSSION
Model calibration (estimating cultivar coefficients): Table 2 shows both the initial and final values of cultivar coefficients used and generated by GenCalc and GLUE. The data used for estimating cultivar coefficients was endofseason data. The results demonstrated that GenCalc has the capability to let the user set the value of P1V to 0, since the variety under study is spring wheat and has no requirements for vernalization, while GLUE has not this capability, where it deals with the cultivar as winter wheat. There is a contradiction with MaldonadoIbarra et al.^{13} who stated that spring wheats have small values of vernalization, however, a partial agreement with Lobell and OrtizMonasterio^{14} was found who stated that vernalization days of spring wheat was set to 0.5. On the other hand, GenCalc was able to adjust PHINT, while GLUE was not able to adjust this coefficient resulting in inaccurate results for that coefficient and consequently negatively affected the accuracy of the model. However, both GenCalc and GLUE were similar in their efficiency in estimating G1 and G2. The GenCalc gave reasonable estimation of G3, while GLUE underestimated that coefficients. This results for coefficient PHINT of cultivar Sakha 93 is agreed with those obtained by Fayed et al.^{15} who stated that PHINT for Sakha 93 was 120 and partially agreed in P1V coefficient, where they stated that P1V was 0.5 for Sakha 93. However, P1D, P5, G1 and G2 were not agreed with this results, this may be due to the difference in locations of the experiments. However, a contradiction with Liu and Tao^{16} was found, where they stated that GLUE was accurately assessed P5 and PHINT for Maize.
Data in Table 3 and 4 show the comparisons between the simulated values and the observed values and indicated that the cultivar coefficients that were estimated from model calibration using GenCalc were more efficient than those estimated from model calibration using GLUE. The calculated average absolute relative error value between mean values of simulated and observed grain yield (tons/feddan) was 1.44 and 1.64%, for number of spikes per meter square was 12.39 and 140.81%, for number of grains per spike was 18.65 and 60.78% when using GenCalc and GLUE, respectively, which revealed that GenCalc performed better than GLUE in estimating cultivar coefficients more efficient in predicting the previous agronomic parameters.
Table 2:  Initial and final values of genetic coefficients used in the study for cultivar Sakha 93  
Table 3: 
Simulated and observed values of the agronomic parameters (model calibration) using GenCalc and GLUE 

Table 4: 
Average of the difference (%) between simulated and observed values of the agronomic parameters (model calibration) using GenCalc and GLUE 

Ns: Non significant, **Significant 
Table 5:  Simulated and observed values of the agronomic parameters (model validation) using GenCalc and GLUE 

However, the calculated average absolute relative error value between mean values of simulated and observed 1000 grain weight was 5.66 and 3.25%, for biological yield (tons/feddan) was 7.16 and 4.74%, for straw yield (tons/feddan) was 10.55 and 7.3% when using GenCalc and GLUE, respectively. An agreement was found with He et al.^{17} who found that there was a relative error between 0.23 and 11.05% between simulated and observed values of grain yield in winter wheat. There were a significant difference between GenCalc and GLUE in calibration of number of spikes per meter square and number of grains per spike, while there were no significant differences in calibrating the other agronomic parameters. The results from the present study are in general agreement with those obtained by Pal et al.^{18} who stated that GenCalc was efficient in estimating cultivar coefficients to predict grain yield, biological yield and straw yield of wheat.
Model validation: The results of model validation in Table 5 and 6 reveal that the difference between the simulated values and the observed values indicated that the cultivar coefficients that were estimated from model calibration using GenCalc and were used for model validation were more efficient than those derived from model calibration using GLUE in predicting crop yield parameters.

Fig. 2: 
Comparison between GenCalc and GLUE in predicting crop yield parameters 
Table 6:  Average of the difference (%) between simulated and observed values (model validation) of the agronomic parameters using GenCalc and GLUE 

Ns: Non significant, **Significant 
There were no significant differences between GenCalc and GLUE in predicting all the studied crop yield parameters except for number of grains per spike and number of spike per meter square as shown in Fig. 2. The absolute relative error between simulated and observed grain yield (tons/feddan) was 3.96 and 8.32% and for biological yield (tons/feddan) was 4.02 and 5.47% and for straw yield (tons/feddan) was 4.14 and 6.12%, for number of spikes per meter square was 5.57 and 141.79%, for number of grain/spike was 7.52 and 56.5% when using GenCalc and GLUE, respectively, which means that GenCalc performed better than GLUE in calibrating cultivar coefficients more efficient in predicting the crop yield parameters. A similar grain yield was obtained by Asal et al.^{19}. The calculated average absolute relative error value between mean values of simulated and observed 1000 grain weight was 1.51 and 2.99%, while it was 2.7 and 5.4% for harvest index and was 10.14 and 15.4% for number of grain per meter square when using GenCalc and GLUE, respectively. The GenCalc accurately predicting number of grains per spike and number of spikes per meter square, while GLUE under estimated number of grains per spike and over estimated number of spikes per meter square. The results from the present study are in partial agreement with those obtained by Fayed et al.^{15}.
CONCLUSION
The DSSAT was able to simulate the effect of different rates of nitrogen fertilizer on wheat grain yield and its components. The two programs (GenCalc and GLUE) included in DSSAT for estimating cultivar coefficients were used to assess the coefficients. The results from the present study revealed that GenCalc was more accurate than GLUE in estimating cultivar coefficients of spring wheat, however, GLUE was more easily to use than GenCalc. There was a significant difference between GenCalc and GLUE in both number of grains per spike and number of spikes per meter square in both model calibration and model validation where GenCalc accurately predicting those parameters, while GLUE under estimated number of grains per spike and over estimated number of spikes per meter square.

REFERENCES 
1: Miao, Y., D.J. Mulla, W.D. Batchelor, J.O. Paz, P.C. Robert and M. Wiebers, 2006. Evaluating management zone optimal nitrogen rates with a crop growth model. Agron. J., 98: 545553. CrossRef  Direct Link 
2: Tsuji, G.Y., G. Hoogenboom and P.K. Thornton, 1998. Understanding Options for Agricultural Production. Kluwer Academic Publishers, Dordrecht, The Netherlands
3: Mall, R.K., M. Lal, V.S. Bhatia, L.S. Rathore and R. Singh, 2004. Mitigating climate change impact on soybean productivity in India: A simulation study. Agric. Forest Meteorol., 121: 113125. CrossRef  Direct Link 
4: Banterng, P., A. Patanothai, K. Pannangpetch, S. Jogloy and G. Hoogenboom, 2006. Yield stability evaluation of peanut lines: A comparison of an experimental versus a simulation approach. Field Crops Res., 96: 168175. CrossRef  Direct Link 
5: Chapman, S.C., G.L. Hammer, D.W. Podlich and M. Cooper, 2002. Linking BioPhysical and Genetic Models to Integrate Physiology, Molecular Biology and Plant Breeding. In: Quantitative Genetics, Genomics and Plant Breeding, Kang, M.S. (Ed.). CABI Publishing, New York, pp: 167188
6: Phakamas, N., A. Patanothai, K. Pannangpetch, S. Jogloy and G. Hoogenboom, 2008. Dynamic patterns of components of genotype x environment interaction for pod yield of peanut over multiple years: A simulation approach. Field Crops Res., 106: 921. CrossRef  Direct Link 
7: Soler, C.M.T., P.C. Sentelhas and G. Hoogenboom, 2007. Application of the CSMCERESMaize model for planting date evaluation and yield forecasting for maize grown offseason in a subtropical environment. Eur. J. Agron., 27: 165177. CrossRef  Direct Link 
8: Paz, J.O., C.W. Fraisse, L.U. Hatch, A.G.Y. Garcia and L.C. Guerra et al., 2007. Development of an ENSObased irrigation decision support tool for peanut production in the southeastern US. Comput. Electr. Agric., 55: 2835. CrossRef  Direct Link 
9: Suriharn, B., A. Patanothai, K. Pannangpetch, S. Jogloy and G. Hoogenboom, 2007. Determination of cultivar coefficients of peanut lines for breeding applications of the CSMCROPGROPeanut model. Crop Sci., 47: 607619. CrossRef  Direct Link 
10: Anothai, J., A. Patanothai, K. Pannangpetch, S. Jogloy, K.J. Boote and G. Hoogenboom, 2008. Reduction in data collection for determination of cultivar coefficients for breeding applications. Agric. Syst., 96: 195206. CrossRef  Direct Link 
11: Hunt, L.A. and S. Pararajasingham, 1994. GenCalc. In: DSSAT Version 3, 34, Tsuji, G.Y., G. Uehara and S. Balas (Eds.). University of Hawaii, Honolulu, Hawaii, pp: 201234
12: Hunt, L.A., S. Pararajasingham, J.W. Jones, G. Hoogenboom, D.T. Imamura and R.M. Ogoshi, 1993. GENCALC: Software to facilitate the use of crop models for analyzing field experiments. Agron. J., 85: 10901094. CrossRef  Direct Link 
13: MaldonadoIbarra, I., G.R. Rodriguez and D. CastilloRosales, 2015. Determination of genetic coefficients of three spring wheat varieties under a Mediterranean environment applying the DSSAT model. Chilean J. Agric. Res., 75: 418424. CrossRef  Direct Link 
14: Lobell, D.B. and J.I. OrtizMonasterio, 2007. Impacts of day versus night temperatures on spring wheat yields: A comparison of empirical and CERES model predictions in three locations. Agron. J., 99: 469477. CrossRef  Direct Link 
15: Fayed, T.B., E.I. ElSarag, M.K. Hassanein and A. Magdy, 2015. Evaluation and prediction of some wheat cultivars productivity in relation to different sowing dates under North Sinai region conditions. Ann. Agric. Sci., 60: 1120. CrossRef  Direct Link 
16: Liu, Y. and F. Tao, 2013. Probabilistic change of wheat productivity and water use in China for global mean temperature changes of 1°, 2° and 3°C. J. Applied Meteorol. Climatol., 52: 114129. CrossRef  Direct Link 
17: He, J., M.D. Dukes, J.W. Jones, W.D. Graham and J. Judge, 2009. Applying GLUE for estimating CERESMaize genetic and soil parameters for sweet corn production. Trans. ASABE., 52: 19071921. CrossRef  Direct Link 
18: Pal, R.K., K.S. Rawat, J. Singh and N.S. Murty, 2015. Evaluation of CSMCERESwheat in simulating wheat yield and its attributes with different sowing environments in Tarai region of Uttarakhand. J. Applied Nat. Sci., 7: 404409. Direct Link 
19: Asal, M., A. Elham, O.M. Ibrahim and E.G. Ghalab, 2015. Can humic acid replace part of the applied mineral fertilizers? A study on two wheat cultivars grown under calcareous soil conditions. Int. J. Chem. Tech. Res., 8: 2026. Direct Link 



