The speed and accuracy of conservation planning could be improved if maps indicating areas where grassed waterways should be placed to reduce erosion could be easily created. For five central Kentucky fields, elevation data were obtained with real time kinematic (RTK) global positioning system (GPS) and from US Geological Survey (USGS) digital elevation models (DEMs). Terrain attributes were calculated from these datasets which were used as predictor variables for neural network and logistic regression analyses. Grassed waterway prediction models were developed with these analyses. The type of activation function, type of standardization procedure, number of neurons, number of preliminary runs, and number of hidden layers had little impact on the results of the neural network analysis. Logistic regression and neural network analyses produced similar erosion prediction maps. The type of flow direction algorithm used to calculate terrain attributes did not change prediction maps substantially. Grassed waterways could be predicted in most cases with the RTK data but only in some cases with the USGS data. This modeling approach was robust and could aid conservation planners in identifying suitable areas for waterways more efficiently if accurate elevation data can be acquired.

The Maputaland Coastal Plain is situated in north-eastern KwaZulu-Natal Province, South Africa. The Maputaland Coastal Plain and underlying aquifer are two separate but inter-linked entities. This area with high permeable cover sands, low relief and regional geology that slopes towards the Indian Ocean, hosts a variety of important wetlands in South Africa (e.g. 66% of the recorded peatlands). The wetlands overlie and in some cases also connect to the underlying regional water-table. The apparent distribution of wetlands varies in response to periods of water surplus or drought, and over the long-term has been reduced by resource (e.g. agriculture, forestry) and infrastructure (e.g. urbanisation) development. Accurate wetland mapping and delineation in this environment is problematic due to the ephemeral nature of wetlands and extensive land-use change. Furthermore the deep aeolian derived sandy soils often lacks soil wetness indicators in the soil profile. It is postulated that the aquifer is the source of water to rivers, springs, lakes and wetlands (and vice versa). However, the role of groundwater in the sustainability of hydro-ecological systems is unclear. Consequently this research attempted to determine spatial and temporal changes in the distribution of these wetlands, their susceptibility to human development, understand the landscape processes and characterise and classify the different wetland types. An underlying assumption of the hydrogeomorphic wetland classification concept in South Africa is that wetlands belonging to the same hydrogeomorphic unit share common features in terms of environmental drivers and processes. Given the above, the objectives of this thesis relating to the north-eastern corner of the Maputaland Coastal Plain are to: 1) Map the distribution of wetlands and their relation to other land-use; 2) Characterise the landscape processes shaping the dynamics of wetland type and their distribution; 3) Classify wetlands by applying hydrogeomorphic wetland classification system. This study used Landsat TM and ETM imagery acquired for 1992 and 2008 (dry) and Landsat ETM for 2000 (wet) along with ancillary data. Wetland type characteristics were described using terrain unit position in the landscape, SRTM DEM, land surveyor elevation measurements along with long-term rainfall records, in situ water-table levels with soil analysis and geology and vegetation descriptions. A conceptual model was used to account for the available data, and output from a hydrology model was used to support the interpretation of wetland distribution and function. Wetlands in the study area include permanent wetlands (swamp forests and reed/sedge wetlands), but the majority of sedge/moist grassland wetlands are temporary systems. The wetland distribution reflects the rainfall distribution and groundwater discharge in lower lying areas. The weathering of the Kosi Bay Formation is a key factor in wetland formation. Because of an increase in clay content with depth, the pore-space and hydraulic conductivity are reduced which causes water to impede on this layer. The nature of the aquifer and regional geology that slope towards the east along with extreme rainfall events in wet and dry periods are contributing drivers of wetland and open water distribution. In 2008 (a dry year) the smaller wetland extent (7%) could primarily identify “permanent” groundwater-fed wetland systems, whereas for the wet year (2000) with larger wetland extent (18%) both “temporary” and “permanent” wetlands were indicated. Comparison between both dry years (1992 and 2008) indicates an 11% decrease in wetland (sedge/moist grassland) and a 7% increase in grassland distribution over time. Some areas that appear to be grassland in the dry years were actually temporary wetland, based on the larger wetland extent (16%) in 2000. The 2008 Landsat TM dataset classification for the entire Maputaland Coastal Plain gave an overall 80% mapping accuracy. Landscape settings identified on this coastal aquifer dominated by dune formations consist of 3 types: plain (upland and lowland), slope and valley floor. Although the wetland character is related to regional and local hydrogeology as well as climate affecting the temporal and spatial variability of the wetlands this research confirms that the patterns and wetland form and function are predominantly shaped by the hydrogeomorphic setting and not the rainfall distribution. The following wetland types were identified: permanent wetlands such as peat swamp forests, peat reed and sedge fens; temporary wetland systems such as perched depressions, and sedge/moist grasslands. The Hydrogeomorphic wetland classification system was applied using a semi-automated method that was 81% accurate. The following hydrogeomorphic units could be identified: one floodplain, i.e., Siyadla River Floodplain, channelled valley-bottoms, unchannelled valley-bottoms, depressions on modal slope values <1%, seepage wetlands on modal slope values 1-2%. However, evaluation of the hydrogeomorphic classification application results suggests that the “flat” hydrogeomorphic class be revised. It did not fit meaningfully on the upland plain area. This research finding concludes wetland function does depend on landscape setting and wetland function is not truly captured by the hydrogeomorphic type classification. Not all depression on the coastal plain function the same way and three types of depressions occurs and function differently, i.e., perched depression with no link to the regional water-table vs. depressions that are linked with the regional water-table on plain, slope and valley floor landscape settings. Overall, this research study made a useful contribution in characterising and classifying wetland type and distribution for a high priority wetland conservation area in South Africa. Applying similar methods to the broader Maputaland Coastal Plain will particularly benefit from the research findings. The importance of using imagery acquired in wet and dry periods as well as summer and winter for a more comprehensive wetland inventory of the study area, is stressed. To manage the effects of climate variability and development pressure, informed land-use planning and rehabilitation strategies are required based on landscape analysis and interpretation.

Soil hydraulic properties, such as saturated hydraulic conductivity (Ks), have high spatial variation, but little is known about how to vary a few measurements of Ks over an area to model hydrology in a watershed with complex topography and multiple land uses. Variations in soil structure, macropores (especially in soil that shrink and swell), land use, and soil development can cause large variations in Ks within one soil type. Characterizing the impacts of soil properties that might vary systematically with land use and terrain attributes on Ks rates would provide insight on how management and human activity affect local and regional hydrology. The overall objective of this research was to develop a strategy for using published infiltration and Ks measurements by the Natural Resources Conservation Service for watershed hydrology applications in a Vertisol, and to extend this knowledge toward developing recommendations for future infiltration measurements. To achieve this goal, soil infiltration measurements were collected across three catenas of Houston Black and Heiden clays (fine, smectitic, thermic Udic Haplusterts) under three land uses (improved pasture, native prairie, and conventional tillage row crop). Measurement locations were selected to account for variation in terrain attributes. Overall, Ks values were not significantly different across different landscape positions; however, in fields under similar land uses, Ks values were found to be lower in the footslope positions and higher in the backslope positions. The pedotransfer function, ROSETTA, provided estimates of 64 percent of the overall variability in Ks while also providing accurate estimates of the mean of Ks when particle size distribution and bulk density are used as inputs in the model. Through the use of multiple regression analysis, soil antecedent water content, bulk density, clay content, and soil organic carbon along with two indicator variables for the catenas were highly correlated (r2 = 0.59) with Ks. The indicator variables explained 17 percent of the variation in Ks that could not be explained by measured soil properties. It is recommended that when NRCS measures Ks on benchmark soils, especially high clay soils, that they collect particle size distribution, bulk density, organic carbon, and antecedent water content data.

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / The requirement for quantitative soil information has increased as a consequence of the global scenarios. The digital soil mapping (DSM) seeks to produce consistent data with the current needs through the generation of functional soil maps in multi-scales. The aim of this study was to evaluate the altimetry quality and limitations of digital elevation model (DEM) in order to assist in choosing the most suitable DEM to derive terrain attributes (TA) to develop spatial prediction functions to be applied to digital mapping of soil organic carbon (SOC) of farm scale in the Plateau of the state of Rio Grande do Sul (RS). The study was conducted on a 937 ha area located in the municipality of Giruá, RS, Brazil. They collected 243 sampling points in the 0-5 cm layer and an accomplished planialtimetric survey (PS) considered in this study the truth of the ground of altitude values. DEM evaluated were: DEM-PS (generated from the PS), DEM-LETTER (generated by interpolating the level of topographic map curves), DEM-ASTER, DEM- SRTM and DEM-TOPODATA. The DEM were evaluated for precision altimetry through statistical test value of the square root of the mean square error (RMSE) and application of the Brazilian Cartographic Standard for defining the scale of each DEM based on the accuracy of the altitude. TA derived from each DEM were faced with the AT derived from the DEM-PS. The results showed that the DEM-PS presented the best quality of elevation data (RMSE = 1.93 m), followed by DEM-SRTM (RMSE = 5.95 m), DEM- (RMSE = 8.28 m), DEM-TOPODATA (RMSE = 9.78 m) and DEM-ASTER (RMSE = 15.57 m). The size of the area and gently rolling relief were the main factors that influenced the results. The DEM-PS is suited in scale 1: 10,000 Class D, while DEM-LETTER and DEM-SRTM are suited in scale 1: 50,000 class B, the DEM-TOPODATA the scale 1: 50,000 class D and the DEM-ASTER scale 1: 100,000 Class B. With regard to TA, the DEM-SRTM and DEM-TOPODATA present results closer to the DEM-PS and smaller RMSE values for each TA assessed. The prediction function constructed from the DEM-PS derived from the TA and vegetation index Landasat-7 obtained images explained only 16% of the variance in SOC area. The small spatial resolution of DEM-PS and images associated with the use only of these predictors were the main factors that influenced the results. Based on these results, the DEM-SRTM and DEM-TOPODATA can be used in DSM semi-detailed soil classes. In the case of the SOC DSM suggest the use of these DEM associated with field control points to verify the precision altimetry and the inclusion of variables related to soil management practices. / A demanda por informações quantitativas de solos em nível detalhado de bacias hidrográficas vêm aumentado em decorrência dos cenários globais. O mapeamento digital de solos (MDS) visa gerar dados compatíveis com essas necessidades por meio da geração de mapas funcionais de solos em multi-escalas. O objetivo desse trabalho foi avaliar a qualidade altimétrica e limitações de MDE com a finalidade de auxiliar na escolha do MDE mais adequados para derivar atributos do terreno (AT) para desenvolver funções de predição espacial para serem aplicadas ao mapeamento digital do carbono orgânico do solo em escala de propriedade rural no Planalto do Estado do Rio Grande do Sul (RS). O estudo foi realizado em uma área de 937 ha localizada no município de Giruá, RS, Brasil. Foram coletados 243 pontos amostrais na camada de 0-5 cm e realizado um levantamento planialtimétrico (LP) considerado neste estudo a verdade do terreno dos valores de altitude. Os MDE avaliados foram: MDE-LP (gerado a partir do LP), MDE-CARTA (gerado pela interpolação das curvas de nível da carta topográfica), MDE-ASTER, MDE-SRTM e MDE-TOPODATA. Os MDE foram avaliados quanto à precisão altimétrica por meio de teste estatísticos, valor da raiz quadrada do erro médio quadrático (RMSE) e aplicação da Norma Brasileira de Cartografia para definição da escala de cada MDE com base na precisão da altitude. Os AT derivados de cada MDE foram confrontados com os AT derivados do MDE-LP. Os resultados mostraram que o MDE-LP apresentou a melhor qualidade dos dados de altitude (RMSE = 1,93 m), seguido dos MDE-SRTM (RMSE = 5,95 m), MDE-CARTA (RMSE = 8,28 m), MDE-TOPODATA (RMSE = 9,78 m) e MDE-ASTER (RMSE = 15,57 m). O tamanho da área e relevo suave ondulado foram os principais fatores que influenciaram nos resultados. O MDE-LP se adequou na escala 1:10.000 classe D, enquanto os MDE-CARTA e MDE-SRTM se adequaram na escala 1:50.000 classe B, o MDE-TOPODATA a escala 1:50.000 classe D e o MDE-ASTER escala 1:100.000 classe B. Em relação aos AT, os MDE- SRTM e MDE-TOPODATA apresentam resultados mais próximos do MDE-LP e os menores valores de RMSE para cada AT avaliado. A função de predição construída a partir dos AT derivados do MDE-LP e índice de vegetação obtido de imagens Landasat-7 explicou apenas 16% da variância do COS na área. A resolução espacial pequena do MDE-LP e das imagens associado ao uso apenas dessas variáveis preditoras foram os principais fatores que influenciaram nos resultados. Com base nesses resultados, os MDE- SRTM e MDE-TOPODATA podem ser utilizados no MDS semidetalhado de classes de solos. No caso do MDS do COS sugere-se o uso desses MDE associado com pontos de controle de campo para verificação da precisão altimétrica e a inclusão de variáveis relacionadas a práticas de manejo do solo.

Research on digital soil mapping (DSM) in Brazil is subject to the existence of legacy soil data obtained through conventional methods for extrapolation of soil-landscape relations. In areas with no available soil maps, it is necessary to develop methodologies for the acquisition of these data in a scale compatible with the needs of the users. In this context, the main objective of this study was to develop a methodology, through techniques of DSM, to predict soil classes at semidetailed level, in a region of gently undulating relief delimited by a topographic map in a scale of 1:50,000. Techniques for relief modeling were used to elaborate and evaluate the quality of the digital elevation models used in the area covered by the topographic map and in the reference area (RA). The RA technique was used to establish soil-landscape relations in the DSM. A basemap in a scale of 1:5,000 was created to support the implementation of soil mapping by conventional methods for the RA. Decision Tree (DT) technique was used to build the prediction model based on the soil map and eight terrain attributes of the RA. Two DSM strategies were tested in order to obtain the data to create the classification rules (DSM 1 and DSM 2). Each strategy employed the eight terrain attributes as predictor variables: elevation (ele), distance to channel network (dis), slope (dec), aspect (asp), topographic wetness index (twi), profile curvature (per), plan curvature (pla), and landform classes (lan). The previous selection of digital elevation models to extract the terrain attributes aggregated quality to the use of the predictor variables that participated in the production of the model. The use of RA in areas with limitation of data proved to be an efficient strategy to improve the understanding of soil-landscape relations for prediction of occurrence of soil classes through the DSM method. The comparison between field data and the digital soil map resulted in a global accuracy (GA) of 66.15% and Kappa of 0.35 for the DSM 1, and GA of 65.58% and Kappa of 0.27 for the DSM 2. The approach of soil survey through the conventional method in the RA proved appropriate, since it contributed to the knowledge of predominant soil categories, as well as reduced the number of field observations in the area covered by the topographic map. / As pesquisas em mapeamento digital de solos (MDS) realizadas no Brasil são dependentes da existência de dados legados de solos obtidos por métodos convencionais para extrapolação das relações solo-paisagem. Em áreas onde não há disponibilidade de mapas de solos, torna-se necessário desenvolver metodologias para aquisição desses dados em escala compatível com a necessidade dos usuários. Nesse contexto, o objetivo geral desta pesquisa foi desenvolver uma metodologia, por intermédio de técnicas de MDS para predizer classes de solos em nível semidetalhado, numa região de relevo suave ondulado, tendo como limite uma carta topográfica na escala 1:50.000. Foram utilizadas técnicas de modelagem do relevo para a elaboração e a avaliação da qualidade dos modelos digitais de elevação utilizados na área de abrangência da carta topográfica e na área de referência (AR). A técnica de AR foi empregada para estabelecer as relações solo-paisagem no MDS. Foi construída uma base cartográfica, na escala 1:5.000, que serviu de apoio para executar o mapeamento de solos com técnicas convencionais para a AR. A técnica de árvore de decisão (AD) foi utilizada para construção do modelo preditivo com base no mapa de solos e em oito atributos de terreno da AR. Duas estratégias de MDS foram testadas com o objetivo de obter os dados para gerar as regras de classificação (MDS 1 e MDS 2), sendo que cada estratégia empregou os oito atributos de terreno como variáveis preditoras: elevação (ele), distância da hidrografia (dis), declividade (dec), orientação de vertente (asp), índice de umidade topográfica (twi), curvatura em perfil (per), curvatura planar (pla) e classes de geoformas (lan). A seleção prévia de modelos digitais de elevação para extrair os atributos de terreno agregou qualidade no uso das variáveis preditoras que participaram da construção do modelo. O uso da AR em locais com limitação de dados mostrou-se uma estratégia eficiente para aprimorar o conhecimento referente às relações solo-paisagem com vistas à predição de ocorrência de classes de solos geradas pelo método de MDS. A comparação entre a verdade de campo e o mapa digital de solos resultou numa exatidão global (EG) de 66,15% e Kappa de 0,35 para o MDS 1 e uma EG de 65,58% e Kappa de 0,27 para o MDS 2. A abordagem de levantamento de solos pelo método convencional na AR demonstrou ser apropriada, visto que contribuiu para o conhecimento dos tipos de solos predominantes, assim como reduziu a quantidade de observações de campo na área de abrangência da carta topográfica.

Soil erosion by water is a main form of land degradation worldwide. The problem has been addressed, among others, in the United Nations Sustainability Goals. However, for mitigation of erosion consequences and adequate management of affected areas, reliable information on the magnitude and spatial patterns of erosion is needed. Although such need is often addressed by erosion modelling, precise erosion monitoring is necessary for the calibration and validation of erosion models and to study erosion patterns in landscapes. Conventional methods for quantification of rill erosion are based on labour-intensive field measurements. In contrast, remote sensing techniques promise fast, non-invasive, systematic and larger-scale surveying. Thus, the main objective of this study was to develop and evaluate automated and transferable methodologies for mapping the spatial extent of erosion rills from a single acquisition of remote sensing data. Data collected by an uncrewed aerial vehicle was used to deliver a highly detailed digital elevation model (DEM) of the analysed area. Rills were classified by two methods with different settings. One approach was based on a series of decision rules applied on DEM-derived geomorphological terrain attributes. The second approach utilized the random forest machine learning algorithm. The methods were tested on three agricultural fields representing different erosion patterns and vegetation covers. Our study showed that the proposed methods can ensure recognition of rills with accuracies between 80 and 90% depending on rill characteristics. In some cases, however, the methods were sensitive to very small rill incisions and to similar geometry of rills to other features. Additionally, their performance was influenced by the vegetation structure and cover. Besides these challenges, the introduced approach was capable of mapping rills fully automatically at the field scale and can, therefore, support a fast and flexible assessment of erosion magnitudes.

This research work concentrate on developing digital soil maps to support field based plant phenotyping research. We have developed soil organic matter content (OM), cation exchange capacity (CEC), natural soil drainage class, and tile drainage line maps using topographic indices and aerial imagery. Various prediction models (universal kriging, cubist, random forest, C5.0, artificial neural network, and multinomial logistic regression) were used to estimate the soil properties of interest.