Search

A critical issue in digital soil mapping (DSM) is the selection of data sampling method for model training. One emerging approach applies instance selection to reduce the size of the dataset by drawing only relevant samples in order to obtain a representative subset that is still large enough to preserve relevant information, but small enough to be easily handled by learning algorithms. Although there are suggestions to distribute data sampling as a function of the soil map unit (MU) boundaries location, there are still contradictions among research recommendations for locating samples either closer or more distant from soil MU boundaries. A study was conducted to evaluate instance selection methods based on spatially-explicit data collection using location in relation to soil MU boundaries as the main criterion. Decision tree analysis was performed for modeling digital soil class mapping using two different sampling schemes: a) selecting sampling points located outside buffers near soil MU boundaries, and b) selecting sampling points located within buffers near soil MU boundaries. Data was prepared for generating classification trees to include only data points located within or outside buffers with widths of 60, 120, 240, 360, 480, and 600m near MU boundaries. Instance selection methods using both spatial selection of methods was effective for reduced size of the dataset used for calibrating classification tree models, but failed to provide advantages to digital soil mapping because of potential reduction in the accuracy of classification tree models. [ABSTRACT FROM AUTHOR]

Land cover is a spatial information of great relevance for a variety of models for estimating sediment yield and to measure the potential of the landscape carbon sequestration. The classification of land cover by the supervised method requires training areas, these areas must be representative of each class of land cover. For the classification decision tree (DT) algorithm, the complexity of DT, results in different values of accuracies for thematic maps. Thus, the objective of this study was to estimate the minimum sample density in a DT model which would allow to discriminate land cover classes, evaluate the size of the generated DT model, as well as, identify the more difficult land cover class to be mapped. Satellite images from RESOURCESAT-1 as well as spectral indices were used in the study. The minimum sample density varied between 0.15 and 0.30% of the total area for each class, this sampling interval allowed better results than 80% for kappa index. The smallest grouping of observations in the same terminal leaf was 45 observations. In this study the most difficult land use classes to be mapped were forest and rice crops due to spectral similarity of shaded forests with irrigated rice crops. [ABSTRACT FROM AUTHOR]

Soil is increasingly being recognized as having an important role in ecosystems, as well as for food production and global climate regulation. For this reason, the demand for relevant and updated soil information is increasing. Soil science researchers are being demanded to produce information in different spatial resolutions with associated quality in what is being called Digital Soil Mapping (DSM). Due to an increasing number of papers related to the DSM in Brazil, it is necessary to discuss the main characteristics of those studies related to the automated mapping of soil classes, which will enable a broader perspective of the subject and guide future works and demands. The mapping of soil classes using DSM techniques is recent in the country, the first publication in this topic occurred just in 2006. Among the predictive functions the predominant is logistic regression. The soil formation factor relief was used in all studies reviewed. Quality of predictive models was evaluated employing error matrix and kappa which were the most common procedures. The consolidation of this automated approach as an auxiliary tool to the conventional soil mapping will demand training of young soil scientists to use geoinformation technologies and quantitative tools to handle aspects of soil variability. [ABSTRACT FROM AUTHOR]

The aim of this study was to apply an automated technique of soil mapping from predictive models developed at a reference area, into nearby areas of the same soil-landscape relationships. Generalized linear models were developed using nine terrain attributes derived from a digital elevation model as covariate predictors, with soil classes, obtained from a conventional soil survey, as dependent variables. The models were able to distinguish the three main forms of the local landscape. Soil classes with pedogenesis intimately tied to the predictive covariates obtained the best results. The soil maps generated, showed a reproducibility of 46.12% and an accuracy of 21.06%. [ABSTRACT FROM AUTHOR]