Users of soil survey products are mostly interested in understanding how soil properties vary in space and time. The aim of digital soil mapping (DSM) is to represent the spatial variability of soil properties quantitatively to support decision-making. The goal of this study is to evaluate DSM techniques (Regression Trees - RT and Multiple Linear Regressions - MLR) and the ability of these tools to predict mineral fraction content under a wide variability of landscapes. The study site was the entire Guapi-Macacu watershed (1,250.78 km²) in the state of Rio de Janeiro in the Southeast region of Brazil. Terrain attributes and remote sensing data (with 30 m of spatial resolution) were used to represent landscape co-variables selected as an input in predictive models in order to develop the explanatory variables. The selection of sampling sites was based on the Latin Hypercube algorithm. A representative set of one hundred points with feasible field access was chosen. Different input databases were tested for prediction of mineral fraction content (harmonized and original data). The Spline algorithm was used to harmonize data according to the GlobalSoil. Net consortium standards. The results showed better performance from the RT models, using input from an average of six covariates; the simplest MLR model used twice as many input variables, creating more complex models without gaining precision. Furthermore, better R² values were obtained using RT models, irrespective of harmonization of soil data. The harmonized dataset from the 0.00-0.05 and 0.05-0.15 m layers, in general, presented better results for the clay and silt, with R² values of 0.52 (0.00-0.05 m) and 0.69 (0.05-0.15 m), respectively. Prediction of sand content showed better results when the original depth data was used as an input, although all regression tree models had R² values greater than 0.52. The RT models provided a better statistical index than MLR for all predicted properties; however, the variance between models suggests similarity of performance. Regarding harmonization of soil data, both input databases (harmonized or not) can be used to predict soil properties, since the variance of model performance was low and generalization of the soil maps showed similar trends. The products obtained from the digital soil mapping approach make it possible to integrate the factor of uncertainties, providing easier interpretation for soil management and land use decisions.

O objetivo deste trabalho foi o desenvolvimento de modelos com diferentes conjuntos de dados, para estimar a densidade de solos de regiões tropicais montanhosas, a partir de atributos de solos comumente encontrados nas análises de perfis de solos descritos nos levantamentos regionais. O conjunto total de dados compõe-se de 163 amostras e foi dividido em seis grupamentos, dos quais três com 73 amostras, com o máximo de 32 covariáveis, e três com 163 amostras, com o máximo de 18 covariáveis. Testaram-se modelos de regressão linear múltipla (RLM) e randomForest (RF). A menor incerteza entre os modelos foi alcançada pelo RLM2, com R² de 0,56, 13 covariáveis e 73 amostras. Nos grupamentos com 163 amostras, os melhores modelos foram os RF, com R² médio de 0,48. A raiz quadrada da média do erro ao quadrado variou entre 0,09 e 0,14. As covariáveis mais importantes no modelo RF foram: carbono orgânico, hidrogênio, areia fina e grossa, saturação por bases e capacidade de troca catiônica. Pelo método "stepwise regression", as variáveis mais importantes foram: a relação silte/argila; areia grossa e fina; carbono orgânico; saturação por bases; e potássio.

Planning sustainable use of land resources requires reliable information about spatial distribution of soil physical and chemical properties related to environmental processes and ecosystemic functions. In this context, cation exchange capacity (CEC) is a fundamental soil quality indicator; however, it takes money and time to obtain this data. Although many studies have been conducted to spatially quantify soil properties on various scales and in different environments, not much is known about interactions between soil properties and environmental covariates in the Brazilian semiarid region. The goal of this study was to evaluate the efficiency of random forest and cokriging models applied to predict CEC in the Brazilian semiarid region. The covariates used to predict CEC consist of images from Landsat 5 TM and a legacy soil map (scale 1:10,000). The sample set comprises 499 samples from the topsoil layer (0.00-0.20 m), where 375 samples were used in training processes and 124 as validation samples. The cokriging model (R² = 0.57 and RMSE = 7.22 cmol_c kg^-1) performed better in predicting CEC than the random forest model (R² = 0.47 and RMSE = 7.89 cmol_c kg^-1). The approach used showed potential for estimating CEC content in the Brazilian semiarid region by using covariates obtained from orbital remote sensing and the legacy soil map.

O objetivo deste trabalho foi avaliar a influência da resolução espacial do modelo digital de elevação e da eficiência de modelos Random Forest sobre a predição dos teores de areia, argila e carbono orgânico, com uso de número reduzido de amostras. O trabalho foi realizado em área de Cerrado com diversidade litológica, no Estado do Mato Grosso do Sul, tendo-se utilizado atributos morfométricos, dados do sensor TM Landsat 5 e litologia como covariáveis preditoras. Dados da camada superficial (0,0-0,2 m) de 175 perfis de solos (0,009 perfis km^-2) e de 26 covariáveis preditoras foram utilizados com resolução espacial de 30 (conjunto 1) e 90 m (conjunto 2). A análise realizada pelo Random Forest mostrou que as covariáveis de nível de base do canal de drenagem, da elevação e da litologia foram as mais importantes para explicar a variabilidade. A validação dos modelos apresentou similaridade entre os conjuntos quanto à predição de areia, argila e carbono orgânico, o que explica os seguintes valores de variabilidade espacial, respectivamente: 44, 40 e 33%, para a resolução de 30 m; e de 45, 46 e 33%, para a resolução de 90 m. A resolução espacial das covariáveis preditoras tem pouca influência sobre a predição dos atributos, e a abordagem por Random Forest apresenta potencial de utilização para estimar atributos do solo.

There is a relation of vegetation physiognomies with soil and geological conditions that can be represented spatially with the support of remote sensing data. The goal of this research was to map vegetation physiognomies in a mountainous area by using Sentinel-2 Multispectral Instrument (MSI) data and morphometrical covariates through data mining techniques. The research was based on red-edge (RE) bands, and indices, to classify phytophysiognomies at two taxonomic levels. The input data was pixel sampled based on field sample sites. Data mining procedures comprised covariate selection and supervised classification through the Random Forest model. Results showed the potential of bands 3, 5, and 6 to map phytophysiognomies for both seasons, as well as Green Chlorophyll (CLg) and SAVI indices. NDVI indices were important, particularly those calculated with bands 6, 7, 8, and 8A, which were placed at the RE position. The model performance showed reasonable success to Kappa index 0.72 and 0.56 for the first and fifth taxonomic level, respectively. The model presented confusion between Broadleaved dwarf-forest, Parkland Savanna, and Bushy grassland. Savanna formations occurred variably in the area while Bushy grasslands strictly occur in certain landscape positions. Broadleaved forests presented the best performance (first taxonomic level), and among its variation (fifth level) the model could precisely capture the pattern for those on deep soils from gneiss parent material. The approach was thus useful to capture intrinsic soil-plant relationships and its relation with remote sensing data, showing potential to map phytophysiognomies in two distinct taxonomic levels in poorly accessible areas.

Quantitative soil–landscape models offer a method for conducting soil surveys that use statistical tools to predict natural patterns in the occurrence of particular map units across a landscape. The aim of the present study was to predict soil units in a watershed with wide variation in landscape conditions. The approach relied on a modelling of soil-forming factors in order to understand the variability of the landscape components in the region. Models were generated for landscape attributes related to pedogenesis, specifically elevation, slope, curvature, compound topographic index, Euclidean distance from stream networks, landforms map, clay minerals index, iron oxide index and normalised difference vegetation index, along with an existing geology map. The soil classification was adapted from the World Reference Base System for Soil Resources, and the predominant soil taxonomic orders observed were Ferrasols, Acrisols, Gleysols, Cambisols, Fluvisols and Regosols. The algorithms used to predict the soil units were based on decision tree (DT) and random forest (RF) methods. The criteria used to evaluate the models’ performance were statistical indices, coherence between predicted units and the legacy map, as well as accuracy checks based on control samples. The best performing model was found to be the RF algorithm, with resulting statistical indices considered excellent (overall = 0.966, kappa = 0.962). The accuracy of the map as determined by control points was 67.89%, with a kappa value of 61.39%.