Manizheh Razavi Hosain Abad; Alireza Amirian Chekan; Mohammad Faraji; Jamal Mosavian
Abstract
Soil aggregate stability and its spatial distribution can be considered as a good indicator for assessing the results of measures conducted for mitigation soil erosion. This study was conducted in two adjacent sites in Chahmari region, Kuzestan province. At one site afforestation and contour furrowing ...
Read More
Soil aggregate stability and its spatial distribution can be considered as a good indicator for assessing the results of measures conducted for mitigation soil erosion. This study was conducted in two adjacent sites in Chahmari region, Kuzestan province. At one site afforestation and contour furrowing were conducted to control soil erosion and the adjacent site with no controlling measures was considered as control. A total of 150 soil samples were collected from the surface layer (0-5 cm) of two sites and mean weight diameter of aggregates (MWD) were measured using dry and wet sieving (MWDd and MWDw, respectively). Based on digital soil mapping (DSM) approach and to map MWD spatially, several environmental covariates were derived from a Landsat 8 image and a digital elevation model (DEM). Two machine learning algorithms including artificial neural networks (ANN) and regression trees (RT) were used to predict MWD with covariates as inputs. Results indicated a significant difference between MWDd in two sites, but no significant difference was found between MWDw. Correlation analysis revealed no correlation between MWDw and all terrain attributes derived from the DEM, but significant correlations were obtained between MWDd and some terrain attributes. Most covariates derived from Landsat images had significant correlation with both MWDw and MWDd. ANN and TR had relatively high and almost the same accuracy in predicting MWDw, but in predicting MWDd, ANN was superior to RT. In general, the findings showed good performance of DSM techniques in predicting and spatial mapping of MWD.
Asghar Rahmani; Fereydoon Sarmadian; Sayed Roholla Mousavi; Seyyed Erfan Khamoshi
Abstract
Conventional soil mapping is related to High density sampling, affected by scale and expert knowledge So using of new data mining methods in digital soil properties mapping was the main aim of this study for resolving conventional soil survey problems. In this research, 62 surface soil samples based ...
Read More
Conventional soil mapping is related to High density sampling, affected by scale and expert knowledge So using of new data mining methods in digital soil properties mapping was the main aim of this study for resolving conventional soil survey problems. In this research, 62 surface soil samples based on regular grid and expert knowledge opinion were selected after that soil organic carbon(SOC), clay content and CaCO3 were determined in some part of Dryland Kuhin region with area of 372 ha. Data sets were divided to two 80%(calibration) and 20%(validation), respectively. From digital elevation model with 10-meter spatial resolution were derived 19 geomorphometric attribute in SAGA GIS software. Three geomorphometric covariate included TPI, TRI, DEM and landform map unit were chosen PCA and expert knowledge. RStudio and SoLIM Solution software were used for random forest (RF) and fuzzy logic modelling, respectively. The RF modelling results show that for SOC, clay and CaCO3 based on determination coefficient (R2) had 0.63,0.75,0.63 and RMSE 0.17,7.5,5.77 percentage and for SoLIM method revealed that R2 0.47,0.42,0.42 and RMSE 0.2,8.08,4.68 percentage, respectively. Generally, the RF model with creating nonlinear relationship among soil properties and environmental covariate can predicted digital map with appropriate precision for management and sustainable land utilization
Mehrnaz Neyestani; Fereydoon Sarmadian; Azam Jafari; Ali Keshavarzi
Abstract
In digital soil mapping, soil characteristic and classes could be extracted truly by numerical and quantitative modelling. Hence, derived rules could be fitted to similar regions for achieving ruled relations on areas without soil information which is called as extrapolation. In the present study, achieving ...
Read More
In digital soil mapping, soil characteristic and classes could be extracted truly by numerical and quantitative modelling. Hence, derived rules could be fitted to similar regions for achieving ruled relations on areas without soil information which is called as extrapolation. In the present study, achieving digital soil class map of an area without adequate soil information by Random forest was tested by extrapolation at great group level. The results show overall accuracy 88% and kappa 0.77 of donor area which is able to fit over its similar region. Results of extrapolation show overall accuracy 81% and kappa 0.61 of recipient area which could show logical concordance to produce soil class map of recipient area without applying related soil samples of this area in short time and low cost. Since, extrapolation could be as an efficient way to predict soil classes of unknown areas.
