Digital soil mapping supervised and unsupervised classification methods were evaluated to aide soil survey of unmapped areas in the western United States. Supervised classification of landscape into mountains and basins preceded unsupervised classification of data chosen by iterative data reduction. Principal component data reduction, ISODATA classification, Linear combination of principal components, Zonal averaging of linear combination by ISODATA class, Segmentation of the image into polygons, and Attribution of polygons by majority ISODATA class (PILZSA process) comprised steps isolating unique soil-landscape units. Input data included ASTER satellite imagery and USGS 30-m elevation layers for environmental proxy variables representing soil forming factors. Results indicate that PILZSA captured general soil patterns when compared to an existing soil survey while also detecting fluvial soils sourced from different lithologies and unique mountain areas not delineated by the original survey. PILZSA demonstrates potential for soil pre-mapping, and sampling design efforts for soil survey and survey updates.

Monitoring and identifying the state of rangelands on a landscape scale can be a time consuming process. In this thesis, remote sensing imagery has been used to show how the process of classifying different ecological sites and states can be done on a per pixel basis for a large landscape. Twenty-seven years' worth of remotely sensed imagery was collected, atmospherically corrected, and radiometrically normalized. Several vegetation indices were extracted from the imagery along with derivatives from a digital elevation model. Dominant vegetation components from five major ecological sites in Rich County, Utah, were chosen for study. The vegetation components were Aspen, Douglas-fir, Utah juniper, mountain big sagebrush, and Wyoming big sagebrush. Training sites were extracted from within map units with a majority of one of the five ecological sites. A Random Forests decision tree model was developed using an attribute table populated with spectral biophysical variables derived from the training sites. The overall out-of-bag accuracy for the Random Forests model was 97.2%. The model was then applied to the predictor spectral and biophysical variables to spatially map the five major vegetation components for all of Rich County. Each vegetation class had greater than 90% accuracies except for Utah juniper at 81%. This process is further explained in chapter 2. As a follow-on effort, we attempted to classify vegetation ecological states within a single ecological site (Wyoming big sagebrush). This was done using field data collected by previous studies as training data for all five ecological states documented for our chosen ecological site. A Maximum Likelihood classifier was applied to four years of Landsat 5 Thematic Mapper imagery to map each ecological state to pixels coincident to the map units correlated to the Wyoming big sagebrush ecological site. We used the Mahalanobis distance metric as an indicator of pixel membership to the Wyoming big sagebrush ecological site. Overall classification accuracy for the different ecological states was 64.7% for pixels with low Mahalanobis distance and less than 25% for higher distances.

Submitted by Jorge Silva (jorgelmsilva@ufrrj.br) on 2017-08-22T18:59:33Z No. of bitstreams: 1 2013 - Andr? Luis Oliveira Villela.pdf: 14328753 bytes, checksum: ce4f856fddd576111ae58d83bad8de61 (MD5) / Made available in DSpace on 2017-08-22T18:59:34Z (GMT). No. of bitstreams: 1 2013 - Andr? Luis Oliveira Villela.pdf: 14328753 bytes, checksum: ce4f856fddd576111ae58d83bad8de61 (MD5) Previous issue date: 2013-08-29 / CAPES / PETROBRAS / The Brazilian territory region covered by the Amazon rainforest, due to its continental dimensions and difficulty of access and various interests in extractive activities, has great demand for information to provide support for the occupation, exploitation, and systematic recuperation thus keeping environmental safeguards. The regional soil information available is scarce and in scales inconsistent with the current demands, and the investments in new areas of research in the region is still insufficient. With technological developments, especially in the area of informatics that enables the storage and analysis of large banks of pedological data, the soil mapping techniques improved considerably. Pedometric techniques have been used to store and to explore large databases, thus enabling the improvement of existing soil databases and allowing manufacture of new products in larger scale and mapped areas, with low investment required. The hypothesis of this study is that the technique of reference area may allow the systematic digital mapping of soils from the Solim?es Formation, in the Amazon State. The general objective was to develop and compare methods for mapping soils in the Oil Province Uruc? (AM), using relief covariates. A conventional pedological survey of an area of 8.000 hectares, at the detail level, was executed to be used as a reference area (RA), in the augmentation of the map using digital soil mapping (MDS) techniques for an area of 73.000 hectares after downscaling and grouping of the legend. The numerical modeling of the terrain (MDT) was used (11 covariates derived from MDT) for further application of this soil formation factor as a predictor of the map units, in discriminant functions (DF), and in an expert system based on a tree model classification (AC). Four MDS models were developed, where two were trained using the studied region conceptual model of the pedologist, and the other two were trained with models based on a statistical analysis of the reference area information. The techniques were effective for predicting the mapping units (MU) in the study region, with overall accuracy (EG) ranging from 74.62 % to 88.81 %, and the kappa index was between 0.68 and 0.85. The MDS based in the expert system and AC showed significantly better results in terms of the kappa index, general EG, and the EG for 3 of the 4 mapping units in the area. Although the FD had not the highest accuracy levels, they showed a great potential for use in MDS, especially for preliminary mapping for the pedological survey of new regions, using knowledge of AR neighboring areas. The limitations were observed in the use of FD for mapping unities with small territorial expression, and it is recommended to increase the number of training observations in a way inversely proportional to the frequency of observation of these MUs. The major contribution of this work to scientific community was the establishment of bases and techniques of MDS, using AR and the soil relief relationship, for systematic mapping of new soils form Solim?es Formation. / A regi?o do territ?rio brasileiro coberta por floresta tropical amaz?nica, por suas dimens?es continentais e dificuldade de acesso e interesses diversos em atividades extrativistas, apresenta forte demanda por informa??es gerais que possam servir como subs?dio para a ocupa??o, explora??o e recupera??o ordenada e ambientalmente equilibrada. As informa??es pedol?gicas dispon?veis sobre a ?rea s?o escassas e em escalas incompat?veis com as demandas atuais, e os investimentos em novas frentes de pesquisa na regi?o ainda s?o insuficientes. Com a evolu??o tecnol?gica, sobretudo na ?rea da inform?tica que possibilita o armazenamento e an?lises de extensos bancos de dados pedol?gicos, as t?cnicas de mapeamento pedol?gico v?m se aperfei?oando consideravelmente. T?cnicas de pedometria t?m sido utilizadas para armazenar e explorar grandes bancos de dados e t?m possibilitado o aperfei?oamento das bases pedol?gicas existentes e permitido a confec??o de novos produtos em escala e ?reas mapeadas maiores, com menores investimentos exigidos. A hip?tese deste trabalho ? de que a t?cnica de ?rea de refer?ncia permite o mapeamento digital sistem?tico dos solos da regi?o da forma??o Solim?es, e o objetivo geral foi desenvolver e comparar m?todos de mapeamento de solos da Forma??o Solim?es, na Prov?ncia Petrol?fera de Urucu, AM, utilizando covari?veis do relevo. Foi executado um levantamento pedol?gico convencional de uma regi?o com 8.000 ha, em n?vel de detalhe para ser utilizado como ?rea de refer?ncia (AR) para a amplia??o do mapa, com t?cnicas de mapeamento digital de solos (MDS) para uma ?rea de 73.000 ha com redu??o de escala e agrupamento de legenda. Foi ent?o elaborada modelagem num?rica do terreno (MDT) (11 covari?veis derivadas do MDT) para posterior utiliza??o deste fator de forma??o do solo, como preditor das unidades de mapeamento, em fun??es discriminantes (FD) e um sistema especialista baseado em modelo de ?rvores de classifica??o (AC). Foram desenvolvidas 4 cartas MDS, sendo duas treinadas por modelos baseados no modelo conceitual do ped?logo sobre a regi?o em estudo, e duas treinadas por modelos baseados em an?lise estat?stica de informa??es sobre a ?rea de referencia. As t?cnicas mostraram-se eficientes para predi??o de unidades de mapeamento (UM) na regi?o de estudo, com exatid?o global (EG) variando entre 74,62% a 88,81% e ?ndice kappa entre 0,68 e 0,85. O MDS baseado em sistema especialista e AC apresentou resultados sensivelmente melhores em termos de ?ndice kappa, EG geral e EG para 3 das 4 UM da ?rea. Embora as FD n?o tenham apresentado os maiores ?ndices de acur?cia, estas tem grande potencial de uso em MDS, sobretudo para a confec??o de mapas preliminares para o levantamento pedol?gico de novas regi?es, utilizando-se do conhecimento de AR de ?reas vizinhas. Foram observadas limita??es no emprego de FD para o mapeamento de UM?s com pequena express?o territorial, sendo recomend?vel o aumento do n?mero de observa??es de treinamento inversamente proporcional ? frequ?ncia de observa??o destas UM?s. A maior contribui??o deste trabalho para a comunidade cient?fica foi o estabelecimento de bases e t?cnicas de MDS, utilizando AR e rela??o solo-relevo para o mapeamento sistem?tico de novas ?reas da forma??o Solim?es.