© H. Aboodeh et al., Published by EDP Sciences, 2024

1 Introduction

Rapeseed is considered the most important oilseed crop in the temperate and subtropical regions of the world. This plant ranks second after soybean in terms of supply of edible oil in the world. In addition to providing edible oil, rapeseed is used to provide fodder for livestock and poultry by feeding domestic animals from the leaves and grain meal of this plant (Moradi-Telavat and Siadat, 2012). The study on the evaluation of the life cycle of biological energy obtained from oilseeds such as rapeseed, shows that the development of the cultivation of this plant with the aim of producing electricity, in addition to providing human edible oil and also its use in the preparation of animal and poultry fodder, can reduce the emission of greenhouse gases in Compare with electricity generation from fossil fuels (Cocco et al., 2014). In the years leading to the first decade of the 21st century, the yield of rapeseed in some European countries has not increased and in some countries it has decreased. In addition, in arid and semi-arid regions of the world, rapeseed production faces various challenges, including water deficit and high temperatures during the flowering to ripening stage. This plant with three-carbon photosynthetic cycle is significantly affected by climate change. So that in the coming years, predicting the reaction of rapeseed yield under the influence of climate change will be a very important research topic. Different plant models such as e.g. DSSAT, Cropsyst, etc. have shown acceptable accuracy in describing the process of growth and yield formation of canola. These models describe the growth and yield formation of rapeseed based on light reception and efficiency of radiant energy consumption (Wang et al., 2022). However, in many regions, increasing yield or maintaining current yields in rapeseed fields will face a significant limitation of water resources (Fathi et al., 2010).The AquaCrop model describes the growth and yield of the crop based on the transpiration efficiency. Therefore, it may accurately describe the growth and yield of rapeseed in areas where water scarcity is considered the most important limiting factor.

Although the lack of irrigation can reduce agricultural production, it is a fundamental way to increase water consumption efficiency (Blum, 2009; Babalola and Ekinsaluna, 2016). The relationship between plant performance and water supply is based on empirical production functions (Doorenbos and Kassam (1979)), but these functions are not reliable in areas where they have not been developed. Therefore, recalibrated simulation models, such as the AquaCrop plant growth model developed by the World Food and Agriculture Organization (FAO), are increasingly used as an alternative solution for rapid assessment of plant performance in response to water limitation (Raes et al., 2012; Foster et al., 2014; Sayyahi et al., 2020). The AquaCrop model is highly accurate in simulating grain and biomass yield under conditions of full irrigation and mild water stress (Heng et al., 2009). Dirwai et al. (2021) investigated the AquaCrop plant growth model for rapeseed under different irrigation regimes and found that the calculated seed yield was respectively 1.32, 0.73 and 0.56 tons per hectare under the irrigation regimes of 100 %, 75 % and 50 %, while the simulated yield was estimated to be 1.48, 1.15 and 0.75 tons per hectare, respectively. Sayyahi et al., 2020 also used the AquaCrop model to simulate the performance and efficiency of sugar beet water consumption, with the model estimating a yield of 1.79 and 0.99 tons per hectare, respectively. The AquaCrop model is also accurate in simulating plant density, planting date, and water requirement, and has an acceptable performance (Hsiao et al., 2009). The inputs of the model are seed germination time, initial soil moisture, and depth. Irrigation water in full irrigation treatment does not have any sensitivity (Babazadeh and Mehdi, 2012). The model also performs well in simulating the growth of potato plants under different irrigation conditions in the southeast of Spain, with an index of agreement (d) and coefficient of explanation (R²) above 0.9 (Montoya et al., 2016). Although there has been limited research on rapeseed, Ebrahimi-Pak et al. (2018) calibrated and verified the AquaCrop model using data obtained from a rapeseed field in Qazvin. The results of this study were in good agreement with the actual data, indicating that the model can be used to simulate the efficiency of water consumption and yield of canola.

Researchers around the world have confirmed the validity of the AquaCrop model in simulating the performance of different plants in different environments (Huang et al., 2017; Raes et al., 2012; Zeleke et al., 2011; Stricevic et al., 2014 Ahmadi et al., 2015; Akumuga et al., 2017; Jovzi et al., 2020). Although the AquaCrop model has been used in the simulation of many plants, there have been limited experiments conducted to evaluate the model’s accuracy in simulating the performance of rapeseed plants under different irrigation management conditions. Therefore, this research aims to evaluate the accuracy of the AquaCrop model in predicting the seed yield, biomass, and water consumption efficiency of rapeseed.

2 Materials and methods

A research study was conducted at Khuzestan University of Agricultural Sciences and Natural Resources. This area is located at a distance of 35 kilometers north of Ahvaz, in Khuzestan province, Iran. The latitude and longitude of the area are 31°36’ N and 48°53’ E respectively, with an altitude of approximately 22 meters above sea level. The table below presents the amounts of water consumed during the first and second year of the experiment, which was carried out over a period of two crop years − 2021–2022 and 2022–2023. The location of the experiment was classified as a dry region, based on Do Marten’s classification (Ghasemifar and Naserpoor, 2014).

The water consumption data from the two-year experiment are presented in Table 1.

Tables 2 and 3 provide information on the meteorological parameters during the two crop years and the soil of the research field during the experiment period.

The experiment was designed using a split factorial research design based on randomized complete blocks, containing three replications. The experiment involved testing three levels of irrigation as the main plot, which included (1) Control (no interruption of irrigation), (2) Interruption of irrigation at the beginning of flowering (phenology code 60) until the formation of 50 % of the pods (phenology code 75), and (3) Interruption of irrigation at the stage of pod formation (phenology code 99) (Zavareh and Emam, 2008), until the stage of harvest (phenology code 99). The experiment also tested three levels of plant density (800000, 1100000, and 1400000 plants per square meter) and two cultivars (Hayola 4815 and Aram), which were arranged in subplots. Hayola 4815 cultivar, an early growth period cultivar, originated from Australia, while Aram, a mid-early growth period cultivar, originated from Iran. The spring cultivars were obtained from the Karaj Seed and Plant Breeding Research Institute.

The amount of irrigation water for each experimental plot was determined by calculating the area of the plot and measuring the water using a meter. It is important to note that the irrigation interruption at each stage was carried out based on the growth stage of each cultivar.

Nitrogen was supplied from a 46 % urea fertilizer source at a rate of 200 kg per hectare. The application was done twice: first during the four-leaf stage (phenology code 19) and then after thinning operations were completed and the second stage at the beginning of budding (phenology code 51). This information is based on a study conducted by Zavareh and Emam in 2008.

The AquaCrop plant growth model was used to analyze the results. The model covers a wide range of agricultural and horticultural products and uses climate information, plant management, and soil information. Some fixed parameters are specific to the plant, which were used from the default values of the model and calibrated with plant data. The model is based on Doorenbos and Kassam (1979) equation (Relation 1) and separates ETa into evaporation from the soil surface (Es) and transpiration (Tr). It also separates the final economic performance (Y) into biological performance (B) and harvest index (HI). Separating Eta into Es and Tr means that evaporation is not considered in product production (Raes et al., 2009).

(1)

(2)

where Y_max is the maximum performance, Ya is the actual performance, ET_max is the maximum evaporation and transpiration, ETa is the actual evaporation and transpiration and Ky is the proportionality factor between the relative loss of performance and the relative reduction of evaporation and transpiration. CC is the crown cover in the plant development stage (percent), CC0 is the initial crown cover (percent), CGC is the growth factor of the crown cover (photo of the day) and t is the time (days). Therefore, the rate of plant transpiration based on crown coverage is obtained from equation (3):

(3)

where, Ks and Kc are coefficients of water stress and plant stress, respectively. The amount of grain yield is also calculated using equation (4) (Raes et al., 2009).

(4)

In the above relationship, HI₀ is the reference harvest index, Y is the grain yield, f_HI is the coefficient that adjusts the reference harvest index, and B is the biological performance obtained using the following relationship (Raes et al., 2009).

(5)

In relation K_sb, temperature stress coefficient, WP* parameter is the normalized water productivity.

The Aquacrop plant growth model does not simulate the leaf area index, but instead represents foliage growth as canopy green cover (CC). CC is the fraction of the soil surface covered by the canopy and can vary from zero (before greening) to its maximum amount (CCx), which can be close to 100 % depending on the crop type and plant density. There are different ways of calculating CC, including: a) using digital images taken within a week of canopy development; b) measuring the leaf surface index and light extinction coefficient; and c) using the ratio of the measured values of photosynthetically active radiation (PAR) above and below the canopy. In this study, the third method of using equations (Farahani et al., 2009) is used to estimate CC more accurately.

(6)

In these equations, PAR_above and PAR_below are the photosynthetic active radiation above and below the canopy, respectively.

2.1 Evaluation of plant growth model

In order to evaluate and measure the validity of the model in estimating grain yield, biomass yield and water efficiency from root mean square (RMSE), root mean square normalized error (NRMSE), mean skew error (MBE), efficiency Model (EF), explanation coefficient (R²) and relative error index (RE) were used:

(7)

(8)

(9)

(10)

(11)

(12)

(13)

In the equations presented above, Pi represents the simulated value, while Oi represents the measured value. P is the average of the simulated values, O is the average of the measured values, the average of the observed variable values Ō and n is the number of data. The RMSE statistic is always positive and the closer it is to zero, the more suitable it is. Values less than 0.1 for the NRMSE statistic indicate high accuracy of the model. Moreover, the mentioned statistic values in the range of 10–20 %, 20–30 %, and more than 30 % indicate good, medium, and poor accuracy, respectively.

The MBE statistic’s positive value indicates that the aquacrop model has estimated the value of the desired parameter more than the actual value, while the negative values indicate that the model has obtained a smaller number in the estimation of the desired parameter. The EF statistic indicates the correctness of the data fitting and varies from negative infinite values in the worst-case scenario to a value of one when the data is fully fitted. R²’s value changes from zero to one, and the closer it is to one, the better the fit of the data. A positive value of RE indicates an overestimation of the model, while its negative values indicate an underestimation of the model.

The “compatibility (d)” index, introduced by Milmot, is a measure of how closely the simulated data matches the observational data (Willmott, 1982). This index ranges from zero to one, with a value closer to one indicating a better model and more reliable simulated values. Ideally, this index should be equal to one. In practice, values greater than 0.5 are generally considered acceptable for simulations (Moriasi et al., 2007).

3 Results and discussion

Calibration means estimating model parameters in such a way that the difference between the measured and simulated values estimated by the model is minimized.

The first year’s results (2021–2022) were used to recalibrate the model and the second year’s results (2022–2023) were used for Aquacrop model validation.

The AquaCrop model was implemented by entering measured plant data and other management factors into the model. The model already had some default plant growth factors, but based on the results obtained from the first year of the experiment, some of these factors required recalibration, which were presented in Tables 4 and 5.

Figures 1 and 2 illustrate the calibration and verification stages of the AquaCrop model for grain yield characteristics of experimental treatments. In the calibration stage, the highest and lowest yields of rapeseed were observed in two treatments. The first treatment was without interrupting irrigation (control) at a density of 1100,000 plants per hectare, using the Hayola 4815 cultivar. The second treatment was at the interruption of irrigation in the beginning of flowering until the formation of 50 % of the pods, using the Aram cultivar at the same density.

During the verification stage of the study, it was observed that the highest simulated and measured value was obtained when irrigation was not interrupted (control) at a density of 1400,000 plants per hectare and with the Hyola 4815 cultivar. On the other hand, the lowest value was observed when irrigation was interrupted during the beginning of flowering until the formation of 50 % of the pods, at a density of 1400,000 plants per hectare and with the Aram cultivar. Based on the results, it can be concluded that the interruption of irrigation during the flowering period has a significant impact on reducing the yield of canola. This conclusion is supported by the findings of Ghasemian (2020) and Zareei and Rezaeizad (2019).

The assessment and measurement of the reliability of the model used to estimate the grain yield of rapeseed cultivars have been carried out and the parameters used are listed in Table 6. During the calibration and evaluation stages of the model, the normalized error index (NRMSE) was used to measure the seed yield, and it was observed to be in the range of 10 % to 20 % and less than 10 %. A value of less than 10 % indicates that the modeling of rapeseed yield has been carried out in an appropriate and excellent manner. In Sandhu and Irmak’s (2019) research on the simulation of corn seed yield, the model statistics during calibration and validation were observed, with NRMSE values of 7.7–12.1 and EF values of 0.8–0.7, respectively.

The model’s efficiency index (EF) for the experimental treatments examined during both the calibration and evaluation stages was close to one, indicating that the model estimates grain yield with high accuracy. The index value (MBE) was obtained for calibration and evaluation of experimental, positive treatments. It shows that the AquaCrop model tends to overestimate the yield of rapeseed.

According to the research findings, it was observed that the maximum relative error of underestimation occurred during the calibration phase. This happened in the stage of irrigation interruption, specifically during the beginning of flowering until the formation of 50 % of the pods, and at a density of 1,400,000 plants per hectare for the Aram cultivar. Additionally, the cultivar Hayola 4815 with a density of 1,100,000 plants per hectare showed the highest relative error of underestimation at the stage of irrigation interruption during the formation of pods until harvest. The details can be found in Table 7.

During the validation phase of the research, it was observed that the highest amount of relative error overestimation occurred in both stages of interruption of irrigation: the beginning of flowering until the formation of 50 % of the pods, and until harvest at a density of 1,400,000 plants per hectare. Furthermore, the research findings suggest that the percentage of relative error increased with the increase in plant density in most of the investigated treatments.

Overall, the AquaCrop model showed underestimation in the calibration stage and overestimation in the evaluation stage. The average relative error values for the calibration and validation stages were −0.29 and 10.51, respectively (Tab. 7).

It is observed that overestimation of model during validation the stage of interruption of irrigation in the beginning of flowering until the formation 50 % of the pods may be due to weather conditions during the experiment (Tab. 7). A similar finding was reported by Ebrahimi Ebrahimipak et al. (2018) where rapeseed yield was lower than the simulated value under water stress conditions. In another study by Zelki et al. (2011) on rapeseed calibration and evaluation, the yield error was obtained with the values of 4.7 and −2.1, respectively. Furthermore, in wheat plant simulation using the AquaCrop model, the accuracy of grain yield simulation was observed to be more accurate than biomass characteristics, as reported by Zhang et al. (2013).

The study aimed to measure the biomass values of rapeseed cultivars under different irrigation interruption treatments. To measure and predict the biomass, the AquaCrop model was employed during the calibration and validation stages. The treatment with a density of 1400000 plants per hectare and Hayola 4815 cultivar showed the highest amount of biomass during the calibration phase (as shown in Fig. 3). In the verification stage, the highest measured amount of biomass was observed in the control treatment with the same density of plants per hectare and Hayola 4815 cultivar. The highest simulated biomass was recorded in the control treatment and the density of 1100000 plants per hectare and Aram cultivar (as shown in Fig. 4). Hayola 4815 and Aram cultivars showed the highest amount of biomass in both stages, respectively (as shown in Fig. 4).

At a density of 800,000 plants per hectare, the lowest biomass value was observed during the stage interruption of irrigation from the beginning of flowering until the formation of 50 % of the pods, and also during the simulated stage interruption of irrigation from the pod formation stage until harvest. These stages showed the lowest biomass values for the Hayola 4815 cultivar, as indicated in Figure 4.

According to the data presented in Table 6, the normalized root mean square error (NRMSE) for the biomass simulation was 19.99 % during the calibration stage and 15.36 % during the validation stage. This indicates that the biomass simulation has been done accurately, as the statistics fall within the medium range of 20 % to 30 %.

A study conducted by Ziaii et al. (2015) compared the performance of AquaCrop and CERES-Maize models in corn simulation. It was found that the normalized error rate for the AquaCrop model ranged from 20 % to 40 %, while the CERES-Maize model had a normalized error rate ranging from 20 % to 80 %. Similarly, in the research conducted by Sandhu and Irmak (2019), the AquaCrop model overestimated the final biomass in 2009 while the NRMSE value was 31 %. However, in 2010, the NRMSE value was 5.3 %.

The relative error (RE) values for biomass treatments are presented in the Table 8, showing that the stages of irrigation interruption had the lowest estimation error. Paredes et al. (2014) demonstrated the high ability of the AquaCrop model to accurately estimate the biomass and yield of corn in low irrigation conditions when simulating the effects of water stress on the performance of corn plants.

The treatment with the highest measured and simulated amount was observed when irrigation was interrupted during the stage of pod formation until harvest. This treatment had a density of 800000 plants per hectare and used the Hayola 4815 cultivar. The control treatment, which had a density of 1400000 plants per hectare and used the Hayola 4815 cultivar, also showed high results (see Fig. 6).

The minimum underestimation error of calibration was observed for the treatment was interruption of irrigation in the stage of formation of pods until harvest, with a density of 1400000 plants per hectare and the Hayola 4815 cultivar (Tab. 9). The reason for the reduction in model error in this treatment is the application of stress in the final periods of plant growth, which resulted in a significant decrease in plant water productivity, even in real conditions.

After verifying the water consumption efficiency, it was observed that the control treatment with a density of 800000 plants per hectare and the Hayola 4815 cultivar had the minimum relative error index of underestimation (as per Tab. 9). Additionally, most of the examined treatments showed higher efficiency of water consumption than the measured value, indicated by the positive MBE values in Table 6.

During the validation phase, all experimental treatments except for the control treatment, the Hayola 4815 cultivar at a density of 1400000 plants per hectare, and the Aram cultivar at 1100000 plants per hectare had lower water consumption efficiency than the measured value. This is confirmed by the negative MBE values. The measurements in Table 6 support this finding.

The normalized squared error (NRMSE) for water efficiency simulation was 6.22 % during the calibration stage, and 14.15 % during the validation stage (Tab. 6). This indicates that the model had appropriate accuracy in simulating water efficiency. The coefficient of explanation (R²) ranged from 0.90 to 0.98, while the efficiency coefficient (EF) ranged from 0.95 to 0.96 in the calibration and validation stages, respectively. These results further confirm the AquaCrop model’s ability to simulate water consumption efficiency. Overall, Table 6 shows that there was a difference between the actual and simulated values of grain yield, biomass, and water use efficiency in the investigated treatments.

According to a study conducted by Kumar et al. 2014, the AquaCrop model performed better at simulating grain yield than it did at measuring water consumption efficiency and biomass. In another study by Zeleke et al. 2011, the RMSE statistic for the simulation of rapeseed yield with the Aquacrop model was between 1.75 and 2.58 tons per hectare. This research produced more accurate results for seed yield and water use efficiency, which aligns with the findings of Ebrahimi Pak et al. 2018 for rapeseed simulation.

Fig. 1 Simulated and measured values of rapeseed grain yield (calibration stage).

Fig. 2 Simulated and measured values of rapeseed grain yield (validation stage).

Fig. 3 Simulated and measured values of the biomass of rapeseed cultivars (calibration stage).

Fig. 4 Simulated and measured values of rapeseed biomass (validation stage).

Fig. 5 Simulated and measured values of rapeseed water consumption efficiency (calibration stage).

Fig. 6 Simulated and measured values of rapeseed water consumption efficiency (validation stage).

4 Conclusion

The AquaCrop plant growth model was used in a recent research study to predict seed yield, biomass and water efficiency under different levels of irrigation interruption, plant density, and rapeseed cultivars. The R² statistic index showed that the model accurately estimated grain yield and water efficiency with higher accuracy than biomass characteristics. During the calibration stages, the normalized error value (NRMSE) of grain yield prediction, water efficiency efficiency, and rapeseed biomass were obtained, which showed that the model has good accuracy in predicting these parameters. However, the AquaCrop model showed overestimation and underestimation during the calibration and validation of biomas and waterproductivity respectively, while for grain yield, underestimation and overestimation were obtained. The relative error increased with an increase in plant density. Changes in irrigation management and different cultivars and densities resulted in changes in measured parameters during two crop years. Based on the results, the powerful AquaCrop model can be used to optimize water consumption and irrigation management for rapeseed.

Acknowledgments

Authors thanks grammarly AI for its help to improve overall English writing of manuscript.

Also, we would like to thank the vice-chancellor of education and graduate education and the vice-chancellor of research and technology of Agricultural Sciences and Natural Resources University of Khuzestan for supporting the implementation of this research.

Conflicts of interest

The authors declare that they have no conflicts of interest with regard to this article.

References

All Tables

All Figures

	Fig. 1 Simulated and measured values of rapeseed grain yield (calibration stage).
In the text

	Fig. 2 Simulated and measured values of rapeseed grain yield (validation stage).
In the text

	Fig. 3 Simulated and measured values of the biomass of rapeseed cultivars (calibration stage).
In the text

	Fig. 4 Simulated and measured values of rapeseed biomass (validation stage).
In the text

	Fig. 5 Simulated and measured values of rapeseed water consumption efficiency (calibration stage).
In the text

	Fig. 6 Simulated and measured values of rapeseed water consumption efficiency (validation stage).
In the text