**1. Introduction**

The challenges that the agricultural sector must deal with are multidimensional and large. On the one hand, the increase in production is intended to cover the nutritional needs of the rapidly growing population. On the other hand, limiting the use of water, fertilizers, and pesticides to protect the sustainability of agroecosystems while protecting the natural environment from problems, such as nutrient losses with nitrogen losses, is often used as a typical example. In recent years, the difficulties created by these challenges have been aggravated by the projected climate change [1–4]. Scientists apply simulation models to examine all the aforementioned challenges [4]. Simulation models are an approach to represent quantitative knowledge about the system of interest and how the different components of that system interact. Agroecosystem models can help agronomists to understand crop growth, predict crop yields, and assess managemen<sup>t</sup> for better water and nutrients used. Climate data, soil, and information about the managemen<sup>t</sup> of the agroecosystem are used to inform these models. Such agroecosystem tools can normally simulate many periods, locations, managemen<sup>t</sup> styles, and scenarios and can provide useful information to agricultural science and farming, exploring the changing aspects between the atmosphere, plants, soil, and water, assisting in crop agronomy, pest management, plant breeding, natural resources management, and evaluating the effect of climate change [5]. In this article, we present the activities that are currently carried out for an ongoing project where the agricultural policy environmental extender (APEX) model is applied in a rural region in Central Greece to assess crop production and water and nitrogen losses under current and future weather conditions.

The APEX model has been implemented for the aquifer of the Karla Basin. APEX was developed to help evaluate different land managemen<sup>t</sup> strategies regarding their environmental impact, erosion, cost, and possible water supplies. APEX simulates the nitrogen and the water process, the crop yield, at the field, farm, or watershed levels, subdividing the simulated area into several units with homogeneous soil, weather, land use, and topography commonly defined subareas [6,7]. The Karla watershed is an area with intense agricultural activity [8]. Figure 1 presents the land uses and crop classification for the Karla aquifer, as displayed within the ArcAPEX interface. After the delineation process, ArcAPEX separated the study area into 34 homogeneous subareas. The model was set to simulate 46 years in total, with the first 10 years used as a spin-up period and not considered in the calibration process. For the calibration, cumulative monthly data from 1961 to 2009 were used for the potential evapotranspiration (PET), while the crop yield of the main crops was calibrated considering the period 1980–2015.

**Figure 1.** The study area.

Two statistical criteria were used to evaluate the results obtained for PET. The Nash– Sutcliffe model efficiency (Ef) in Equation (1) and the coefficient of determination (R2) in Equation (2) indicate how well the model describes adaptation in the observed and estimated data:

$$\text{Eff} = 1 - \frac{\sum\_{t=1}^{T} \left(\mathbf{Y}\_{\text{m}}^{t} - \mathbf{Y}\_{\text{o}}^{t}\right)^{2}}{\sum\_{t=1}^{T} \left(\mathbf{Y}\_{\text{o}}^{t} - \overline{\mathbf{Y}\_{\text{o}}}\right)^{2}} \tag{1}$$

where Y O is the mean observed value, Ym is the estimated value by the model, and Yo is the observed at time t. Ef ranges from 1 (best result) to minus infinite.

$$\mathbf{R}^2 = \left[ \frac{\sum(\mathbf{x} - \overline{\mathbf{x}}) - (\mathbf{y} - \overline{\mathbf{y}})}{\sqrt{\sum(\mathbf{x} - \overline{\mathbf{x}})^2 \sum(\mathbf{y} - \overline{\mathbf{y}})^2}} \right] \tag{2}$$

where x and y are the observed and the estimated values by the model, x and y are the mean observed and estimated values by the model, respectively. R<sup>2</sup> ranges from 1 (best result) to 0 (worst result).

### **3. Results**

The APEX model was initially calibrated considering the PET. During the calibration process, four methods for the PET estimation were examined using the Hargreaves approach resulting as the best method. The results for Ef and R<sup>2</sup> are presented in Table 1. Figure 2 shows the scatter plot where the observed and simulated values of PET are compared. As reported in Table 1, the model was able to provide a good estimate of PET, resulting in an Ef value of 0.85 and R<sup>2</sup> of 0.90.

**Table 1.** Statistical Criteria Results.


**Figure 2.** Comparison of the observed and simulated PET.

The work continued with the calibration of yields of the main crops grown in the study area (cotton, maize, and winter wheat). It is worth noting that the calibration of crop yields was based on the average crop yield of winter wheat, cotton, and maize provided by the Greek Ministry of Rural Development and Food [9]. Due to the fact that APEX reports the yield as dry weight, the reported yield data has been adjusted for moisture content. We considered a moisture content between 6.5% to 8% for cotton [10] and 14% for maize and wheat. After adjusting the average observed crop yield for the moisture content, the target crop yield for calibration was 2.6–3.2 Mg ha−<sup>1</sup> for cotton, 8.6–17.2 Mg ha−<sup>1</sup> for maize, and 2.0–3.0 Mg ha−<sup>1</sup> for wheat. Having only one average reported yield available, it was not possible to conduct a statistical assessment of the performance in simulating crop yield.Figures 3–5 show the simulated crop yield for all the simulated years after the calibration process for wheat, cotton, and maize, respectively. The values reported are the average of

the yield simulated by the APEX model in all the areas where each crop is cultivated within the watershed.

**Figure 3.** Simulated wheat yield; minimum and maximum average reported yield.

**Figure 4.** Simulated cotton yield; minimum and maximum average reported yield.

**Figure 5.** Simulated maize yield; minimum and maximum average reported yield.

After the calibration process, the model was able to provide good results in simulating crop yield. The average simulated yield for wheat was 2.6 Mg ha−1, which was in the range

of the average reported yield. In some years, the yield was overestimated, probably due to an overestimation of the crop-available water that, in turn, produced no water stress and a very high crop yield. We will continue to analyze this aspect to improve the quality of the results for this crop. The average simulated yield for cotton was 2.75 Mg ha−1, which is within the range of the average reported yield with some years where the simulated yield is below the minimum or above the maximum average reported yield. Results for maize were better, with an average simulated yield of 13.1 Mg ha−<sup>1</sup> and yielded within the reported range for all examined years.

To calibrate crop yields, parameters that regulate the simulation of soil water content (soil water lower limits and soil evaporation) and the effect of water stress and high temperature on the harvest index were adjusted. Further, the harvest index for maize was revised to consider the higher harvest index of the new maize hybrids, which were obtained thanks to plant breeding and genetic improvement.

The calibration process will be continued considering the runoff and nitrate leaching. In the final step, the APEX model will be used to study the impacts of climate change scenarios on the agroecosystems of the Karla watershed.
