Digital soil mapping: strategy for data pre-processing

Abstract

SUMMARYThe region of greatest variability on soil maps is along the edge of their polygons, causing disagreement among pedologists about the appropriate description of soil classes at these locations. The objective of this work was to propose a strategy for data pre-processing applied to digital soil mapping (DSM). Soil polygons on a training map were shrunk by 100 and 160 m. This strategy prevented the use of covariates located near the edge of the soil classes for the Decision Tree (DT) models. Three DT models derived from eight predictive covariates, related to relief and organism factors sampled on the original polygons of a soil map and on polygons shrunk by 100 and 160 m were used to predict soil classes. The DT model derived from observations 160 m away from the edge of the polygons on the original map is less complex and has a better predictive performance.

Mapping soils in two watersheds using legacy data and extrapolation for similar surrounding areas

“…The maps generated by the PM methods resulted in lower accuracies, probably because the polygons represent extrapolation of information from the sampled points to non-sampled areas during the creation of the soil map and are allowed to contain inclusions of other soil classes that differ from the dominant in the polygons (IGBE, 2015), which may have hindered the modeling of soil classes by decision trees and consequent less accuracy on the maps generated from this procedure of extraction of information, although, homogeneity of soils within polygons on maps is expected to increase as the scale of the soil maps increase. Although the predicted maps did not present very high accuracies, they were consistent with other studies that used decision trees as a support for soil mapping (Caten, Dalmolin and Ruiz 2012;Giasson et al, 2006;Giasson et al, 2011).…”

“…Thus, the points used for the buffers are mostly located in transition regions between soil classes, which may have affected the modeling in the procedure of extracting information. In addition, an adequate contribution of predictor variables for these transitional locations in modeling the decision trees may be diminished because of the less precise delineation of polygons at the creation of the soil map, as stated by Caten, Dalmolin and Ruiz (2012), which may also have contributed to the lower accuracy of this method as a basis for extracting information in VCW. Legros (2005) compared the delineation of mapping units made by different pedologists, finding out that the polygons present common areas, but the boundaries tend to be more divergent since they are transitional zones between soil classes.…”

“…Decision trees can also help in the choice of terrain attributes that can constitute the best predictors for soil classes and properties. This data mining technique was used by Caten, Dalmolin and Ruiz (2012), Scull, Franklin and Chadwick (2005), Zhou, Zhang and Wang (2004), helping understand the distribution of soils on the landscape.…”

“…Thus, those places may adequately represent environmental conditions of a soil class to occur and, for this reason, to study them may be an alternative to improve the efficiency of decision trees to model and predict soil classes and properties. Caten, Dalmolin and Ruiz (2012) and Giasson et al (2015) used semidetailed soil maps to obtain information for training decision trees to map soil classes, avoiding samples at different distances from map unit boundaries. However, the use of a detailed soil map as source of data, the comparison between information collected avoiding polygon boundaries with information obtained within varying distances from sampling points, where profile descriptions were performed, culminating with the use of the best generated models to extrapolate information to map soils of areas with similar environmental conditions, assessing their accuracy, may constitute alternatives to improve existing soil maps.…”

Using relief subclasses for soil map elaboration with support of geoprocessing tools

“…Segundo Tencaten et al (2012), um mapa de solos pode ser deduzido de informações pedológicas já existentes. Contudo, nosso país possui um banco de dados deficiente neste sentido, sendo que apenas uma porção reduzida do território nacional apresenta levantamento em escalas maiores que 1:25.000 (Chagas et al, 2011).…”