Land suitability modeling by parametric-based neural networks and fuzzy methods for soybean production in a semi-arid region

• Published in: Modeling earth systems and environment Vol. 2; no. 2; pp. 1 - 11
• Main Authors: Bagherzadeh, Ali, Ghadiri, Elnaz, Souhani Darban, Ali Reza, Gholizadeh, Amin
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01-06-2016; Springer Nature B.V
• Subjects: Area; Arid regions; Arid zones; Chemistry and Earth Sciences; Computer Science; Crop production; Cultivation; Earth and Environmental Science; Earth Sciences; Earth System Sciences; Ecosystems; Environment; Evaluation; Fertility; Fuzzy logic; Geographical information systems; Interpolation; Land use; Limiting factors; Math. Appl. in Environmental Science; Mathematical Applications in the Physical Sciences; Modelling; Neural networks; Original Article; Performance indices; Physicochemical processes; Physicochemical properties; Physics; Production methods; Regression analysis; Regression coefficients; Semi arid areas; Semiarid lands; Soil; Soil fertility; Soil properties; Soil surveys; Soybeans; Statistics for Engineering; Surveying; Tables; Fuzzy logic; Land evaluation; Yield; Soybean; Neural networks; Suitability
• ISSN: 2363-6203; 2363-6211
• DOI: 10.1007/s40808-016-0152-4

Land evaluation is the process of predicting land use potential on the basis of its attributes. In the present study, the qualitative land suitability evaluation using parametric learning neural networks and fuzzy models was investigated for irrigated soybean production based on FAO land evaluation frameworks (FAO
1976
,
1983
,
1985
) and the proposed methods by Sys et al. (
1991c
) and Hwang and Yoon (
1981
) in Neyshabour plain, Northeast of Iran. Some 41 land units were studied at the study area by a precise soil survey and their morphological and physicochemical properties. The Climatic and land qualities/characteristics for soybean crop were determined using the tables of soil and crop requirements developed by Sys et al. (
1993
). An interpolation function in GIS was used to map values to scores in terms of land qualities/characteristics for the land utilization type. Our results indicated that the most limiting factor for soybean cultivation in the study area was soil fertility properties. The values of land indexes by neural networks model ranged from 29.77 in some parts in west and middle to 57.45 in the north west and east parts of the study area, which categorized the plain from marginally suitable (S3) to moderately suitable (S2) classes. The land index values by fuzzy model varied between 16.10 and 47.80 which classified from marginally not suitable (N1) to marginally suitable (S3) classes. The coefficient of determination between the neural network land index values and the corresponding fuzzy values revealed a high correlation (
R
2
 = 0.966) between two models. The exponential regression coefficient (
R
2
) between the land indexes of neural networks and fuzzy models with the observed soybean yield in the study area varied between 0.610 and 0.514 respectively, which revealed higher performance of neural networks in predicting land suitability for irrigated soybean production in the study area.