Evaluation of AquaCrop model for corn simulation under different management of nitrogen fertilizer in Karaj

Introduction
The use of plant models is one of the methods that help researchers to understand the response of plants to different agricultural managements without conducting numerous experiments that require spending a lot of time and money. The use of plant models for simulating the reaction of plants to water deficit has a relatively long history and has gone through four stages: infancy, adolescence, youth and maturity. These models were introduced to researchers about fifty years ago by developing computer programs and taking into account the biochemical and biophysical mechanisms of solar energy available for the production of chemical energy and plant biomass. Then, from the late 60s and with the advent of computers, the adolescent stage of plant modeling began. In the youth period of plant modeling, some beliefs of the previous years were lost and detailing and validation of models were considered. Among the first measures taken to provide an acceptable model can be the results of research (1998) by Gerik et al. Cited. The results of these researchers' studies led to the presentation of a model called SORKAM, which was able to simulate the dynamic growth of sorghum (Sorghum bicolor). The maturation period of plant modeling started in the 90s and has continued until now. During this period, more comprehensive software for the simulation of crop plants was developed by research centers around the world, including WOFOST, SWAP and MARS and modeling the plant was widely used in different countries.
The AquaCrop plant model, which was developed by the Food and Agriculture Organization since 2009, is one of the user-friendly, flexible and widely used models that is widely used by researchers due to the closeness of the simulation results to real conditions. The AquaCrop model is used to simulate crops under various stresses, including fertilizer stress. In the developed version of the model, a semi-quantitative method is used to simulate nitrogen fertilizer stress. However, this model has not been evaluated to simulate the effect of different amounts of nitrogen fertilizer under different application methods. For this reason, this issue is addressed in the present study.
 
Materials and Methods
To achieve the research objective, the data was collected from the research project carried out in the 500-hectare research farm of the Seed and Plant Breeding Research Institute, located at 50.58° East longitude and 35.56° N latitude and 1312 m altitude. In this study, nitrogen fertilizer at three levels (N1: 100, N2: 80, and N3: 60) and application time in two methods (T1: three equal usage including 4-6 leaf stage, 10 leaf stage and reproduction stage, and T2: Four equal usage including 4-6 leaf stage, 10 leaf stage, reproduction stage, and inoculation stage) were considered. Then, all treatments were compared with the control, which included traditional fertilization applications in the region. The input data of the AquaCrop model includes four groups of climatic, plant, soil and farm management data. Climatic data includes maximum and minimum daily temperature, rainfall, reference plant evapotranspiration (ET0) and annual average CO2 concentration. Soil data includes saturated hydraulic conductivity, soil texture and volumetric soil moisture at the points of crop capacity and permanent wilting. Farm management data also includes farm management and fertility and irrigation. To evaluate the plant model, calibration was done first. For this purpose, the AquaCrop model was evaluated based on the conditions without fertilizer stress and using the data collected from the farm in the first year. Then, in order to calibrate this model under fertilizer stress conditions, it was necessary to determine the reduction coefficients of coverage development, maximum coverage, average reduction and normalized water productivity reduction percentage.
 
Results and Discussion
The difference between the simulated and observational yield for the control was about 5.7%. The AquaCrop model had an overestimation error to simulate control yield. The mean difference between the simulated and observed yield for T1 was about 4.4%. The highest and lowest yield differences were observed for N3T1 (5.9%) and N1T1 (1.8%) treatments, respectively. The average difference between the simulated and observed yield in T2 was about 5.7%, which is equal to T1. The highest and lowest yield differences in T2 treatment were obtained in N3T2 (4%) and N1T2 (1%), respectively. Therefore, with increasing fertilizer stress, the difference between simulated and observational yield increased. The results showed that the AquaCrop model had an overestimation error (MBE <0) to simulate corn grain yield and an underestimation error (0<MBE) to simulate water productivity. The error of the AquaCrop model for yield simulation was about 0.36 t ha^-1 in the T1 method and about 0.24 t ha^-1in the T2 method. According to NRMSE statistics values, the accuracy of this model for simulating yield in both fertilization methods was in the excellent category (NRMSE <0.1). The error of the AquaCrop model for simulating water productivity in the T1 method was about 0.29 kg.m^-3 and in the T2 method was about 0.30 kg.m^-3. However, the accuracy of this crop model to simulate water productivity in both fertilization methods was in the middle category (NRMSE <0.3). According to all the results, the accuracy of this crop model to simulate the yield was better than water productivity and its use for similar conditions is recommended.
 
Conclusion
The results showed that the accuracy and efficiency of the AquaCrop model for simulating corn yield were acceptable in both calibration and validation stages. No difference was observed in the accuracy of the AquaCrop model for simulating corn yield under both fertilization methods. The error of this plant model to simulate water productivity increased slightly but its efficiency was acceptable. As the fertilizer stress increased, the accuracy of the AquaCrop model decreased. The reason was the increase in the error of this plant model to simulate the development of vegetation under fertilizer stress conditions. Based on all the results, since the fertilizer division methods have not been simulated with this plant model so far, it is possible to rely on the accuracy of the output of this plant model in the mentioned conditions, and its use for similar conditions is recommended.