نوع مقاله : مقاله کامل علمی پژوهشی
نویسندگان
1 علوم خاک،دانشکده کشاورزی،دانشگاه شهرکرد،شهرکرد،ایران
2 استاد گروه علوم خاک،دانشکده کشاورزی،دانشگاه شهرکرد
3 استادیار گروه تولیدات گیاهی، دانشکده کشاورزی، دانشگاه تربت حیدریه.
4 استاد گروه علوم و مهندسی خاک،دانشکده کشاورزی، دانشگاه رفسنجان
5 دانشیار گروه علوم خاک،دانشکده علوم کشاورزی، دانشگاه گیلان
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Background and Objectives
Saffron is one of the most important economic plants and holds significant export value in the country. To determine the relationship between saffron yield and its influence by climatic, soil, and management features, and due to the complexity of the relationships between these variables; the present study aims to investigate the capability of the hybrid Genetic Algorithm-Artificial Neural Network (GA-ANN) model to identify an appropriate subset of features affecting the saffron yield and to determine the importance of each of these features.
Materials and Methods
This research was conducted in 100 farms (3 to 5 years old) under saffron cultivation from various provinces, including Khorasan (Razavi, North, and South), Isfahan, Chaharmahal and Bakhtiari, Gilan, Fars, and Alborz. From each farm, a composite soil sample was collected from a depth of zero to thirty centimeters and transferred to the laboratory. The physical and chemical properties of the soil were measured based on standard methods. Climatic characteristics of each region were collected from the nearest meteorological stations of the studied areas. Additionally, for each farm, a questionnaire was prepared to gather management information and determine the yield. To select an appropriate subset of features influencing saffron yield, the hybrid GA-ANN model was used. The sensitivity of the yield to changes in the soil, climate, and management features selected by the GA-ANN hybrid model was determined using the Statsoft method.
Results
Selection of effective features on yield using Genetic Algorithm-Artificial Neural Network (GA-ANN) Among all possible combinations, the combination with 22 features that had the least error and the least number of inputs was considered the best combination. The features selected by the algorithm included: the average minimum temperature of the growth cycle, the average absolute minimum temperature of the growth cycle, the maximum relative humidity of the growth cycle, the average relative humidity of the growth cycle, the average daily sunshine hours of the growth cycle, the average temperature of the growth cycle, the temperature at the first irrigation, the average maximum relative humidity of the growth cycle, soil calcium carbonate equivalent, soil cation exchange capacity, total soil nitrogen, soil iron, zinc, and manganese, planting type, cultivation system, weed control method, irrigation system, average number of irrigation rounds per year, planting depth, average bulb weight, and planting density. The importance of the selected variables in relation to the yield was determined using the Statsoft method. All features identified as influential factors on yield in feature selection were also among the important and effective features on yield in sensitivity analysis. The most important climatic features affecting yield were, in order, the average maximum relative humidity, average temperature, minimum temperature, absolute minimum temperature during the growth cycle, and temperature at the first irrigation. Among the soil features, in order, manganese, zinc, iron, cation exchange capacity, total nitrogen, and soil calcium carbonate equivalent were the influential variables on saffron yield. In terms of management, correct irrigation management, bulb weight and density, proper planting depth, and proper weed management had a significant impact on saffron yield.
Conclusion
The results showed that using the hybrid artificial neural network algorithm to select and identify an appropriate subset of features affecting saffron yield among soil, climate, and management variables in the study areas, while saving time and cost, can yield more satisfactory results from pattern recognition and data processing processes. These results can have significant applications in sustainable land management and in determining the more suitable locations for saffron cultivation.
کلیدواژهها [English]
)