Assessment of spatial variability of phosphorous and potassium in Sistan Plain using geo-statistical and artificial intelligence methods

Mir, H. (2014) Assessment of spatial variability of phosphorous and potassium in Sistan Plain using geo-statistical and artificial intelligence methods. Masters thesis, University of Zabol.

[img]
Preview
Text
Assesment of spatial variability of.pdf

Download (139kB) | Preview

Abstract

The current study was an attempt to delve into the spatial variability of phosphorus and potassium using artificial intelligence and geo-statistics methods in Sistan plain, southeast of Iran. Firstly, 300 soil samples were taken from 0-30 cm depth with 1.5×1.5 km distances. For this, networks with mentioned distances were considered on topographic maps of the region and the coordinates were determined based on UTM. Then, coordinates exported to GPS and the sampling procedures performed. Having been transferred to laboratory, samples were dried, passed through a 2-mm sieve, and their physicochemical properties were measured. The results of geo-statistics method suggested that simple Co-Kriging with circular model is the best model predicting phosphorus and potassium. Resulted maps of geo-statistics illustrated that the highest amounts of available phosphorus and potassium were located in the north and northwest of the plain and their values were decreased moving across west to east. The lowest values were found on the southeastern part of the region which might be due to low amounts of organic matter and coarse texture soils. The resulted structures for predicting available phosphorus and potassium are made by artificial neural network, the best model including 10 nodes in input and 1 in output layer. The hidden layer composed of 15 and 13 neurons for phosphorus and potassium, respectively. Moreover, optimum iteration was 1000 and the most efficient transfer function was Tansig. Comparing the best geo-statistics and perceptron neural networks method suggested that having been allocated less RMSE and MAE values, the latter model which used all soil properties as input data, enjoys higher accuracy for predicting available phosphorus and potassium.

Item Type: Thesis (Masters)
Uncontrolled Keywords: Sistan plain, Geo-statistics, neural network, Available phosphorus, Available potassium
Subjects: S Agriculture > S Agriculture (General)
T Technology > TA Engineering (General). Civil engineering (General)
Depositing User: admin admin1 admin2
Date Deposited: 18 Apr 2016 07:13
Last Modified: 18 Apr 2016 07:21
URI: http://eprints.uoz.ac.ir/id/eprint/366

Actions (login required)

View Item View Item