Search

Modeling soil detachment rates at the regional scale is important for better understanding of the processes of erosion and the development of erosion models. Soil erodibility is an important factor for predicting soil loss, but its direct measurement at the watershed scale is difficult, time-consuming and costly. This study used stepwise multiple-linear regression (MLR) and artificial neural networks (ANNs) to model Water Erosion Prediction Project (WEPP) soil erodibility parameters, including the baseline inter-rill erodibility (K ib ), baseline rill erodibility (K rb ) and critical shear stress (τ cb ) of cropland conditions in calcareous soils of northwest Iran. Simulated inter-rill and rill erosion experiments were conducted at 100 locations with three replications. K ib , K rb and τ cb and basic soil properties were measured at each location. Auxiliary variables related to soil erodibility were derived from a Landsat 7 satellite image and a 30 m × 30 m digital elevation model (DEM). MLR and ANN models were employed to predict K ib , K rb and τ cb using two groups of input variables: i) more easily measurable basic soil properties (pedo-transfer functions (PTFs)) and ii) more easily measurable basic soil properties and auxiliary data (soil spatial prediction functions (SSPFs)). The results indicated that the WEPP models performed poorly in comparison to the derived models. PTFs and SSPFs generated from ANN models provided more reliable predictions than the MLR models. ANN-based SSPF models yielded the best results (with the highest R 2 and lowest RMSE values) for predicting K ib and K rb . ANN-based PTF model performed reasonably well for predicting τ cb . These results show that information from terrain attributes and remote sensing data are potential auxiliary variables for improving prediction of soil erodibility parameters. [ABSTRACT FROM AUTHOR]

Estimation of soil organic matter (SOM) at unsampled locations is crucial in agronomical and environmental studies. In this study, the ability of geostatistical methods such as ordinary kriging (OK), simple kriging (SK) and cokriging (CK) and hybrid geostatistical methods such as regression-simple kriging (RSK)/-ordinary kriging (ROK) and artificial neural network-simple kriging (ANNSK)/-ordinary kriging (ANNOK) was evaluated to predict SOM content. To this end, a set of 100 soil samples were collected from 0 to 15 cm depth of agricultural soils in Selin plain, northwest of Iran. The organic carbon was measured using Walkley–Black method. An auxiliary variable was provided by remote sensing data (Landsat 7 ETM +). Three performance criteria including mean error (ME), root mean square error (RMSE) and coefficient of determination (R 2 ) were used to evaluate the performance of the derived models. The results showed that the ANN model that used principal components (PCs) as input variables, performed better than the multiple linear regression (MLR) model. The hybrid geostatistical methods, which include ANNOK, ANNSK, ROK and RSK provided more reliable predictions than the geostatistical methods, which include SK, OK and CK. In general, the best prediction method for the estimation of SOM spatial distribution was the ANNOK model, which had the smallest RMSE (0.271%) and the highest R 2 (0.633). It was concluded that information from Landsat ETM + imagery is potential auxiliary variables for improving spatial prediction, monitoring SOM and development of high quality SOM maps, which is the primary step in site-specific soil management. [ABSTRACT FROM AUTHOR]

The objective of this research was to determine the best model and compare performances in terms of producing land use maps from six supervised classification algorithms. As a result, different algorithms such as the minimum distance of mean (MDM), Mahalanobis distance (MD), maximum likelihood (ML), artificial neural network (ANN), spectral angle mapper (SAM), and support vector machine (SVM) were considered in three areas of Iran's dry climate. The selected study areas for dry climates were Shahreza, Taft and Zarand in Isfahan, Yazd, and Kerman Provinces, respectively. Three Landsat ETM+ images and topographical maps of 1:25,000-scale were used in the present study. In addition, training samples for each land use were constructed using GPS and extensive field surveys. The training sites were divided into two categories; one category was used for image classification and the other for classification accuracy assessment. Results show that for the dry climate areas, Maximum Likelihood and Support Vector Machine algorithms with averages of 0.9409 and 0.9315 Kappa coefficients are the best algorithms for land use mapping. The ANOVA test was performed on Kappa coefficients, and the result shows that there are significant differences at the 1% level, between the different algorithms for the dry climate zones. These results can be used for land use planning, as well as environmental and natural resources purposes in study areas. [ABSTRACT FROM AUTHOR]

• K ib , K rb and τ cb were predicted by aspatial (MLR and ANN models) and spatial methods. • Type I data developed with measured soil properties and auxiliary data. • Type II data developed using principal components. • Type I ANN models were the best aspatial model to predict K ib and K rb. • Type II ANN models were the best aspatial model to predict τ cb. • ANN-Kriging was the best spatial model to predict soil sensitivity to erosion. Soil erosion prediction and conservation planning require detailed soil data under different environmental conditions. When such data are needed at the field scale, aspatial and spatial models could be used to predict soil erosion processes. This study was conducted to develop spatial models including geostatistical models (i.e. ordinary kriging (OK) and cokriging (CK)) and hybrid geostatistical models (i.e. multiple linear regression-kriging (MLRK) and artificial neural network-kriging (ANNK)) for estimating WEPP baseline soil sensitivity to erosion parameters for calcareous agricultural soils in northwest Iran. Inter-rill and rill erosion simulation experiments were carried out at 100 locations at the field scale with 3 replications. At each location, the soil properties (organic matter, calcium carbonate equivalent, sand, silt, clay, base infiltration rate) were measured and auxiliary data obtained from attributes derived from digital elevation models (elevation, slope, stream power index, wetness index and sediment transport index) and remote sensing data (three visible bands, NIR, SWIR (5 and 7 bands), NDVI indices). The MLR and ANN models were used to estimate baseline inter-rill soil sensitivity to erosion (K ib) and rill soil sensitivity to erosion (K rb and τ cb) using two types of input data. For the first type of prediction models (type I), the measured soil properties and auxiliary data were used, whereas for the second type of models (type II), principal components (PCs) based on the soil and auxiliary data were used. In comparison to the models that were developed here, the WEPP inter-rill and rill soil sensitivity to erosion models showed a relatively poor performance. The ANN models developed here predicted K ib , K rb and τ cb parameters better than the MLR models using both types of data (type I and II). Moreover, the results indicated that the ANNK model was the most appropriate spatial or hybrid model for predicting soil sensitivity to erosion parameters. [ABSTRACT FROM AUTHOR]