Document Type : Research Paper
Authors
1
Former M.Sc. student, , Department of Soil Science, Faculty of Agriculture, Shahid Chamran University of Ahvaz, Khuzestan, Iran
2
Professor, Department of Soil Science, Faculty of Agriculture, Shahid Chamran University of Ahvaz, Khuzestan, Iran
3
Professor, Department of Soil Science, Faculty of Agriculture, Shahid Chamran University of Ahvaz, Khuzestan, Iran,
4
Lecturer, Department of Soil Science, Faculty of Agriculture, Shahid Chamran University of Ahvaz, Khuzestan, Iran
Abstract
Introduction: Previous studies indicate that many soils worldwide lack zinc or contain it in a form that is unavailable to plants. In Iran, over 56% of agricultural soils have DTPA-extractable zinc levels below 0.75 mg/kg, while only 31% exceed 1 mg/kg. Therefore, it is crucial to understand the factors influencing zinc concentration and prioritize their significance in the distribution of this essential element in soil. Employing an effective model to comprehend the logical connections within the results is essential. Artificial neural networks, known for their ability to model complex and non-linear relationships between soil variables, have been widely used in numerous studies over the past few decades. Khuzestan province, located in the southwest of Iran, possesses high potential for producing a variety of agricultural and horticultural products due to its water resources and fertile soils. With an output of 17.5 million tons of agricultural and horticultural products, Khuzestan plays a crucial role in ensuring the country's food security. Considering the diverse topography and land uses, this study aims to estimate the available zinc content in the surface soils of a selected area in northern Khuzestan province using artificial neural networks and to identify the key factors controlling its distribution across the study area.
Materials and Methods: Most of the study area is within the catchment area of the Karun River, which forms the backbone of the hydrographic network in Khuzestan province. The eastern sectors are located in the catchment area of the Zohre-Jarhari Rivers. The region has a hot and dry climate, with average rainfall ranging from 240 mm in the center to 590 mm in the eastern parts. Various geological formations, such as Asmari, Gachsaran, and Mishan, are present in this area. The soil moisture regimes include aridic, ustic, and xeric (in the eastern parts), while the temperature regimes are classified as thermic and hyperthermic. The soils in the study area are primarily classified as Aridisol, Inceptisol, and Entisol. Two hundred and three (203) surface (0-10 cm) composite samples were randomly collected from an area of 27,452 square kilometers in northern Khuzestan province. Sampling took place under sunny conditions during winter. The soil samples were air-dried and then passed through a 2 mm sieve. The pH and electrical conductivity of the soil were measured in pastes and saturated extracts, respectively. Soil texture was determined using the pipette method; organic matter was assessed by the wet oxidation method with potassium dichromate; and calcium carbonate equivalent was measured through back titration. To measure the available zinc content in the soils, 10 grams of each soil sample were weighed, and 20 ml of a 0.005 M DTPA solution was added. The suspension was shaken for 2 hours at 120 rpm. Subsequently, the solid and solution phases were separated using Whatman 42 filter paper. Finally, the zinc content of the extracts was measured using a GBS Variant AA atomic absorption device. In the dataset, zinc concentrations of less than 3 mg/kg, between 3 and 6 mg/kg, and greater than 6 mg/kg were categorized as low, medium, and high classes of available zinc content, respectively. A Multilayer Perceptron (MLP) neural network was employed to model available zinc content alongside other soil variables in the SPSS v26 environment, with the optimal structure of the neural network determined through a trial-and-error approach. The normality of the data was assessed using the Kolmogorov-Smirnov test within SPSS v26 software.
Results and Discussion: The results indicated that the average content of available soil zinc in the study area is 2.76 mg/kg. The average calcium carbonate equivalent of the soils is 42.25%, suggesting that the soils can be classified as calcareous. Additionally, with an average pH of 7.65, the soils in the region exhibit alkaline reactions. The average soil organic carbon content is 12.6 g/kg. The findings also suggest that the studied soils primarily belong to the loam family. Among the examined soil properties, the highest coefficient of variation is associated with soil salinity (131.9%), while the lowest is related to soil pH (4.2%). The results of the Kolmogorov-Smirnov test revealed that only the silt content of the soils exhibited a normal distribution pattern; the other variables did not pass the test of normality. When various configurations of multilayer perceptron artificial neural networks were evaluated, it was found that the MLPSO7 model, with a stable arrangement of 3 output classes, 5 hidden layer neurons, and 7 input layers, using the sigmoid function for both input and output activation, performed best in predicting available zinc values in soil samples. The results also indicate that organic carbon, calcium carbonate equivalent, and electrical conductivity were the most influential factors in developing the model. The analysis of the estimation matrix for available zinc classes shows that the MLPSO7 model accurately estimated the class with low concentrations of available Zn more effectively than the other two classes. Furthermore, the soil variables used to model the available zinc contents in the study area were able to accurately estimate these contents (97.4% for the training dataset, 97.1% for the test dataset, and an overall accuracy of 90.8%). In contrast, the class with high concentrations of Zn (>6 mg/kg) produced weaker estimates, which can be attributed to the small number of samples examined in this category.
Conclusion: It can be concluded that the estimates derived from the optimal neural network model demonstrate good flexibility and efficiency in justifying the available Zn contents in the study area. For future large-scale studies, greater attention must be given to the significant roles of organic matter, the abundance of calcareous compounds, and the salt content of the soils in determining available Zn levels.
Keywords
Subjects
