Mehrabi gohari, E., Taghizadeh Mehrjardi, R. (2014). Accuracy assessments of artificial neural network for prediction of some soil parameters in FASHAND zone. , 27(2), 3-11. doi: 10.22092/wmej.2014.106242
Elham Mehrabi gohari; Ruhollah Taghizadeh Mehrjardi. "Accuracy assessments of artificial neural network for prediction of some soil parameters in FASHAND zone". , 27, 2, 2014, 3-11. doi: 10.22092/wmej.2014.106242
Mehrabi gohari, E., Taghizadeh Mehrjardi, R. (2014). 'Accuracy assessments of artificial neural network for prediction of some soil parameters in FASHAND zone', , 27(2), pp. 3-11. doi: 10.22092/wmej.2014.106242
Mehrabi gohari, E., Taghizadeh Mehrjardi, R. Accuracy assessments of artificial neural network for prediction of some soil parameters in FASHAND zone. , 2014; 27(2): 3-11. doi: 10.22092/wmej.2014.106242

Accuracy assessments of artificial neural network for prediction of some soil parameters in FASHAND zone

پژوهش های آبخیزداری
Article 1, Volume 27, Issue 2 - Serial Number 103, July 2014, Pages 3-11    PDF (503.57 K)
Document Type: Research
DOI: 10.22092/wmej.2014.106242
Authors
Elham Mehrabi gohari* 1; Ruhollah Taghizadeh Mehrjardi2
1Academic member, Facualty of Agriculture,payam noor university
2Assistance Professor of Soil Science, Faculty of Agriculture, Natural Resources and Desert Studies, Ardakan University, Yazd, Iran
Abstract
Direct measurement of cation exchange capacity and bulk density parameters is hard and costly; ‎henceforth indirect methods of pedotransfer functions for predicting soil parameters are developed ‎based on easily obtained parameters. In this regard, for estimating the cation exchange capacity and ‎bulk density 63 samples were taken from 15 profiles in FASHAND zone in TEHRAN. Using ‎pedotransfer function and having cation exchange capacity and bulk density as output variables, ‎variablesthe normality of the input data were controlled by MINITAB software based on ‎KOLOMOROF-SMIRNOV normality test. For developing the pedotransfer functions multiple ‎regression and artificial neural network methods were enaged. Determining test and training data‏ ‏and then finding the best type of neural network, a hidden layer with 2 to 10 neurons was selected ‎for an optimum simulation in terms of RMSA, ME and AARE indices.  Input data to the network ‎for predicting cation exchange capacity involved clay, silt, sand and organic material in percent ‎while for predicting bulk density‏ ‏clay, silt, sand, lime, saturation moisture and organic material in ‎percent were selected. Using multiple regrsion techniques and some experimental equations like ‎Breeuwsma and teammates (1986), Matrique and Bell (1991) andvanKeulen (1995) regression ‎equations were developed. Because of nonlinear relationship between input and output variables, ‎neural networks had better performance  in comparison with  the regression models  which for ‎cation exchange capacity and bulk density gave coefficient of determination as  0.89 and 0.74, ‎respectively.   Regarding pedotransfer functions employed for estimating cation exchange capacity ‎Matrique et al. (1991) models with a coefficient determination of 0.24 gave the same response ‎which in comparison with the Breeuwsma (1986) models of a determination coefficient of 0.06 is ‎more acceptable.‎
Keywords
Pedotransfer function; Cation Exchange Capacity; Bulk density
References
Statistics
Article View: 1,339
PDF Download: 1,742


counter
Journal management system. designed by sinaweb