Digital Soil Mapping Using Artificial Neural Network on Upper Terrace of Aras River in ParsAbad-e-Moghan

Majid Safarifar | Ruhollah Taghizadeh-Mehrjardi | Hossein Torabi Golsefidi

---

URL :   http://research.shahed.ac.ir/WSR/WebPages/Report/PaperView.aspx?PaperID=159228
Date :  2021/10/20
Publish in :    هفدهمين کنگره علوم خاک ايران و چهارمين همايش ملي مديريت اب در مزرعه

Link :  https://civilica.com/doc/1312617/
Keywords :Khoda-afarin, Soil profile, Kappa index, Digital elevation map (DEM)

Abstract :

---

Digital soil mapping can predict the relationship between environmental auxiliary data and soil characteristics, classes or soil properties in areas of the regions that have not been sampled; consequently, by less profile digging and less laboratory analysis, it will save time and money. On the other hand, continuity and gradual change of soil is also shown by this method. This research was aimed at providing a digital map of the soil classes of the region by artificial neural network model on the level of sub-greatgroups of soil and identifying the effective parameters in predicting soil classes on the 12500 hectares of the second development unit of Khoda-afarin area. In this study, data from an existing soil bank were used, including physico-chemical properties of 497 soil profiles, which were dug in a regular grid arrangement of 500 to 500 square meters. The auxiliary parameters used in this study were Landsat 8 satellite data, digital elevation map, and land parameters extracted from it and geological map. The region data were divided into two sections: training (75 of the data) and test (25 of the data). The results showed that overall accuracy (72) and Kappa index (0.65) were for neural network model in the education section. While in the test section, the overall accuracy of the artificial neural network was 57.26. Also, the results showed that five features, including of elevation, geological map, clay index, base level channel network, and valley depth were the most important parameters for predicting soil classes. In general, the results showed that in the studied area, the efficiency and accuracy of artificial neural network was suitable in spatial modeling and prediction of soil sub-greatgroups. Therefore, it is recommended to use artificial neural network method in similar areas in order to provide a digital soil mapping

https://civilica.com/doc/1312617/