Document Type : Research Paper

Authors

• Alireza Dahmardeh ¹
• Ali Shahriari ²
• Mohammad reza Pahlavan Rad ³
• Asma Shabani ⁴
• MARYAM GHOEBANI ⁴

¹ Graduated MSc Student, Soil Science Department, Faculty of Water and Soil, University of Zabol, Zabol, Iran

² Assistant Professor of Soil Science Department, Faculty of Water and Soil, University of Zabol, Zabol, Iran

³ Assistant Professor, Soil and Water Research Department, Golestan Agricultural and Natural Resources Research and Education Center, AREEO, Iran

⁴ Academic Staff, Soil Science Department, Faculty of Water and Soil, University of Zabol, Zabol, Iran

Abstract

Introduction Crop yield modeling is an important part of ecological modeling because it makes possible plant production prediction and increase understanding of how it works. In other words, plant and crop growth simulation and yield modeling are mathematical expressions of plant growth stages and processes under the influence of environmental and managerial factors. Wheat is one of the key crops grown worldwide and is a source of nourishment for millions of people around the world. Therefore, studying this strategic crop is very importance. On the other hand, more than 70% of wheat and 84% of barley in Sistan and Baluchestan province were produced in Sistan plain and wheat has the highest area under cultivation among different crops, in this arid region. So, the aim of this study was modeling wheat yield with some soil characteristics and determination of the most important soil factors affecting wheat yield in the Sistan plain.
Materials and Methods This research was done in the educational and research farm of University of Zabol. Topsoil (0-30 cm) sampling of 100 soil sample was done randomly. Clay, silt, sand abundances and soil texture class, soil pH, electrical conductivity, apparent electrical conductivity of soil, organic carbon, phosphorus, potassium and nitrogen were measured by conventional methods. Wheat plant samples were taken from a one m2 plot and the grain weight, 1000-grain weight and total weight were measured. Performance modeling was performed by three methods of multi-linear regression (MLR), multi-layer perceptron (MLP) and support vector machines (SVMs) by two kernels types linear(SVM-L) and radial basic function (SVM-RBF). It should be noted, before modeling, 80% of the data were selected for modeling (or training) and 20% for testing (or validation) of the models. These data (training and validation) were the same for all models. Coefficient of determination (R2) and the root mean square error (RMSE) were the criteria for comparing the models. Sensitivity analysis was used to determine the most important soil factors affecting wheat yield.
Results and Discussion The results of soil properties analyses showed that the soil of this area is non-saline and alkaline soil, has a medium to coarse soil texture and the soil fertility conditions are poor to moderate. The results of comparing the models showed that the highest R2 and the lowest RMSE in estimating all three wheat yield indices were related to the MLP method (grain weight with R2= 0.61, 1000-grain weight with R2= 0.64 and total yield with R2= 0.76). After MLP, with less difference, the SVMs method with two kernels types of linear (grain weight with R2= 0.54, 1000-grain weight with R2= 0.44 and total yield with R2= 0.65) and radial basic function (grain weight with R2= 0.48, 1000-grain weight with R2= 0.58 and total yield with R2= 0.67) showed the better modeling and finally the MLR (grain weight with R2= 0.20, 1000-grain weight with R2= 0.27 and total yield with R2= 0.40) showed the lowest accuracy in modeling the yield components of wheat. The results of sensitivity analysis of wheat yield components showed that total soil nitrogen, clay, silt and soil organic matter had the highest on wheat yield components (grain weight: nitrogen, clay and organic matter; 1000-grain weight: nitrogen, silt and clay; and total yield: clay, organic matter and nitrogen) and soil pH had the least effect on it, maybe because of its low variation.
Conclusion Due to harsh environmental conditions in the arid regions, the study of crops yield is very important for the optimal management of facilities and resources. Investigating the application of several wheat yield modeling methods using some soil characteristics in the arid region of Sistan showed that the perceptron neural network (MLP) performed better in predicting the yield components of wheat than other models. Also, some chemical and physical properties of soil that affect the soil fertility and water storage conditions in the soil (soil nitrogen, organic matter, clay and silt contents), were the most affecting factors on the yield of wheat in this arid region. It is important to note that attention to other soil properties as well as climatic parameters and studies and monitoring wheat yield for several years can can lead to more accurate modeling of this strategic crop and thus optimal farm management.

Keywords

• Wheat yield prediction
• multi-linear regression (MLR)
• multi-layer perceptron (MLP)
• support vector machines (SVMs)
• sensitivity Analysis
• Arid regions

Main Subjects

• Precision Agriculture