I detta examensarbetesprojekt genomfördes en flygkartering över deponiområdet Stomsjö i Värnamo kommun, mha. en drönare, eller den i detta sammanhang mer använda benämningen UAV (Unmanned Aerial Vehicle). Värnamo kommuns tekniska avdelning var i behov av beräkning av massor vid deponin, ett område på ca 15 hektar samt modellering av densamma. Den låga kostnaden för inköp av UAV och programvara motiverade kommunen att driva egen verksamhet jämfört med att köpa tjänsterna från konsulter. Projektets syfte är att utveckla en effektiv arbetsmetod för kartering i 2D och 3D med UAV samt att dokumentera arbetsgång och hur den geografiska informationen ska kvalitetssäkras och testas. Flygningen är den första och denna rapport ska utgöra ett underlag för kommande flygningar inom kommunen. Insamling av geografisk information utfördes med quadrokoptern DJI Phantom 4 från fyra olika flyghöjder 50, 75, 100 och 120 meter. Fyra 3D-modeller, ortofoton och digitala höjdmodeller (DEM) har tagits fram i programvaran Agisoft. Sammanlagt mättes 6 markstöd in över området samt en kontrollruta (5x5 punkter) på en hårdgjord asfaltsyta. Utifrån kontrollrutan gjordes en jämförelse mellan inmätta GPS-punkters höjdvärden mot rastervärden från respektive höjdmodell. Vid samtliga flygningar erhålls en upplösning (GSD) på mindre än 3 cm/pix i ortofoto. Upplösningen för samtliga höjdmodeller var mindre än 6 cm/pix. Lantmäteriet har under år 2015 genomfört flygfotografering på 2 500 m höjd över området. En jämförelse mellan Lantmäteriets höjddata mot höjddata genererad från UAV-flygfotograferingen gjordes genom en slumpmässig spridning av punkter på hårdgjorda ytor. Resultaten visar att 100 meters flygningen bäst överensstämmer mot Lantmäteriets data. Volym- och areaberäkning gjordes för den södra deponin. Det avgränsade områdets areal är ca 34 300 m2 och volymen 290 000 m3. / In this project the area Stomsjö in Värnamo municipality was mapped using a Unmanned Aerial Vehicle (UAV). Since 1972 Stomsjö landfill is a part of the municipality. The municipality´s technical department needed a calculation and modelling of mass in a landfill, comprising an area of 15hectares. The purpose of the project is to develop an effective mapping method in 2D and 3D with UAV data, and to document the process to ensure geographical information quality. The flight performed in the study constitutes a basis for further upcoming flights within the municipality. The acquisition of geographical data was made at four different altitudes 50, 75, 100 and 120 meters using a DJI Phantom 4 quadcopter. Four 3D models, orthophotos and Digital Elevation Models (DEMs) were created with the software Agisoft PhotoScan. A total of 6 Ground Control Points (GCP) and a control surface on asphalt (5x5 points) were used for evaluation of the models accuracy. A comparison between measured GPS points and raster values from each flight were made on a control surface. The resolution for each generated orthophoto was less than 3 cm/pix. The resolution of the DEMs was less than 6 cm/pix. Lantmäteriet (The Swedish Mapping, Cadastral and Land Registration Authority) conducted aerial photograph acquisition at 2 500 m altitude over the area in 2015. A comparison between altitude data from Lantmäteriet and altitude data from UAV was made through random points generation. The acquisition at 100 meters altitude showed the lowest deviation forms the data derived by Lantmäteriet. Volume and area measurements were performed at the southern part of the landfill. The selected area is about 34 300 m2 in size and the volume amounts to 290 000 m3.

Made available in DSpace on 2014-06-11T19:31:33Z (GMT). No. of bitstreams: 0 Previous issue date: 2008-12-02Bitstream added on 2014-06-13T20:02:19Z : No. of bitstreams: 1 celinski_tm_dr_botfca.pdf: 1773028 bytes, checksum: 4e269402cffb336eabab0615c60d49d5 (MD5) / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) / Este trabalho teve por objetivo a discriminação de classes de cobertura do solo em imagens de sensoriamento remoto do satélite CBERS-2 por meio do Classificador Árvore de Decisão. O estudo incluiu a avaliação de combinações de atributos da imagem para melhor discriminação entre classes e a verificação da acurácia da metodologia proposta comparativamente ao Classificador Máxima Verossimilhança (MAXVER). A área de estudo está localizada na região dos Campos Gerais, no Estado do Paraná, que apresenta diversidade quanto aos tipos de vegetação: culturas de inverno e de verão, áreas de reflorestamento, mata natural e pastagens. Foi utilizado um conjunto de dezesseis (16) atributos a partir das imagens, composto por: bandas do sensor CCD (1, 2, 3, 4), índices de vegetação (CTVI, DVI, GEMI, NDVI, SR, SAVI, TVI), componentes de mistura (solo, sombra, vegetação) e os dois primeiros componentes principais. A acurácia da classificação foi avaliada por meio da matriz de erros de classificação e do coeficiente kappa. A coleta de amostras de verdade terrestre foi realizada utilizando-se um aparelho GPS de navegação para o processo de georreferenciamento, para serem usadas na fase de treinamento dos classificadores e também na verificação da acurácia. O processamento das imagens e a geração dos mapas temáticos foram realizados por meio do Sistema de Informações Geográficas SPRING, sendo as rotinas desenvolvidas na linguagem de programação LEGAL. Para a geração do Classificador Árvore de Decisão foi utilizada a ferramenta See5. Na definição das classes, buscou-se um alto nível discriminatório a fim de permitir a separação dos diferentes tipos de culturas presentes na região nas épocas de inverno e de verão. A classificação por árvore de decisão apresentou uma acurácia total de 94,5% e coeficiente kappa igual a 0,9389, para a cena 157/128; para... / This work aimed to discriminate classes of land cover in remote sensing images of the satellite CBERS-2, using the Decision Tree Classifier. The study includes the evaluation of combinations of attributes of the image to a better discrimination between classes and the verification of the accuracy of the proposed methodology, comparatively to the Maximum Likelihood Classifier (MLC). The geographical area used is situated in the region of the “Campos Gerais”, in the Paraná State, which presents diversities concerning the different kinds of vegetations: summer and winter crops, reforestation areas, natural forests and pastures. It was used a set of sixteen (16) attributes from images, composed by bands of the sensor CCD (1, 2, 3, 4), vegetation indices (CTVI, DVI, GEMI, NDVI, SR, SAVI, TVI), mixture components (soil, shadow, vegetation) and the two first principal components. The accuracy of the classifications was evaluated using the classification error matrix and the kappa coefficient. The collect of the samples of ground truth was performed using a navigation device GPS to the georeference process to be used in the training stage of the classifiers and in the verification of the accuracy, as well. The processing of the images and the generation of the thematic maps were made using the Geographic Information System SPRING, and the routines were developed in the programming language LEGAL. The generation of the Decision Tree Classifier was made using the tool See5. A high discriminatory level was aimed during the definition of the classes in order to allow the separation of the different kinds of winter and summer crops. The classification accuracy by decision tree was 94.5% and kappa coefficient was 0.9389 to the scene 157/128; to the scene 158/127, it presented the values 88% and 0.8667, respectively. Results showed that the performance of the Decision Tree Classifier was better... (Complete abstract click electronic access below)

Orientador: CéliaRegina Lopes Zimback / Banca: Zacarias Xavier de Barros / Banca: Marco Antonio M.Biaggioni / Banca: Marcelo Giovaneti Canteri / Banca: Ivo Mario Mathias / Resumo: Este trabalho teve por objetivo a discriminação de classes de cobertura do solo em imagens de sensoriamento remoto do satélite CBERS-2 por meio do Classificador Árvore de Decisão. O estudo incluiu a avaliação de combinações de atributos da imagem para melhor discriminação entre classes e a verificação da acurácia da metodologia proposta comparativamente ao Classificador Máxima Verossimilhança (MAXVER). A área de estudo está localizada na região dos Campos Gerais, no Estado do Paraná, que apresenta diversidade quanto aos tipos de vegetação: culturas de inverno e de verão, áreas de reflorestamento, mata natural e pastagens. Foi utilizado um conjunto de dezesseis (16) atributos a partir das imagens, composto por: bandas do sensor CCD (1, 2, 3, 4), índices de vegetação (CTVI, DVI, GEMI, NDVI, SR, SAVI, TVI), componentes de mistura (solo, sombra, vegetação) e os dois primeiros componentes principais. A acurácia da classificação foi avaliada por meio da matriz de erros de classificação e do coeficiente kappa. A coleta de amostras de verdade terrestre foi realizada utilizando-se um aparelho GPS de navegação para o processo de georreferenciamento, para serem usadas na fase de treinamento dos classificadores e também na verificação da acurácia. O processamento das imagens e a geração dos mapas temáticos foram realizados por meio do Sistema de Informações Geográficas SPRING, sendo as rotinas desenvolvidas na linguagem de programação LEGAL. Para a geração do Classificador Árvore de Decisão foi utilizada a ferramenta See5. Na definição das classes, buscou-se um alto nível discriminatório a fim de permitir a separação dos diferentes tipos de culturas presentes na região nas épocas de inverno e de verão. A classificação por árvore de decisão apresentou uma acurácia total de 94,5% e coeficiente kappa igual a 0,9389, para a cena 157/128; para... (Resumo completo, clicar acesso eletrônico abaixo) / Abstract: This work aimed to discriminate classes of land cover in remote sensing images of the satellite CBERS-2, using the Decision Tree Classifier. The study includes the evaluation of combinations of attributes of the image to a better discrimination between classes and the verification of the accuracy of the proposed methodology, comparatively to the Maximum Likelihood Classifier (MLC). The geographical area used is situated in the region of the "Campos Gerais", in the Paraná State, which presents diversities concerning the different kinds of vegetations: summer and winter crops, reforestation areas, natural forests and pastures. It was used a set of sixteen (16) attributes from images, composed by bands of the sensor CCD (1, 2, 3, 4), vegetation indices (CTVI, DVI, GEMI, NDVI, SR, SAVI, TVI), mixture components (soil, shadow, vegetation) and the two first principal components. The accuracy of the classifications was evaluated using the classification error matrix and the kappa coefficient. The collect of the samples of ground truth was performed using a navigation device GPS to the georeference process to be used in the training stage of the classifiers and in the verification of the accuracy, as well. The processing of the images and the generation of the thematic maps were made using the Geographic Information System SPRING, and the routines were developed in the programming language LEGAL. The generation of the Decision Tree Classifier was made using the tool See5. A high discriminatory level was aimed during the definition of the classes in order to allow the separation of the different kinds of winter and summer crops. The classification accuracy by decision tree was 94.5% and kappa coefficient was 0.9389 to the scene 157/128; to the scene 158/127, it presented the values 88% and 0.8667, respectively. Results showed that the performance of the Decision Tree Classifier was better... (Complete abstract click electronic access below) / Doutor

<p>Soil residual herbicides are a vital component of the best management practices (BMPs), to provide early-season weed control in most cropping systems. The availability of a biologically effective dose of a soil residual herbicide in the soil solution is dependent on several soil parameters including soil texture, organic matter (OM), and pH.  Soil residual herbicides are currently applied as a uniform application rate over an individual field; yet soil properties can vary spatially within agricultural fields. Therefore, areas of the field are being over- and under-applied when using a uniform application rate. By integrating variable-rate (VR) technology with soil residual herbicides, the correct rate could be applied based on the intra-field soil variability. However, the extent of spatial soil variability within a field and the impact on herbicide application rates has not been well-characterized to inform whether soil residual herbicide applications should move towards variable rate applications. Therefore, the objectives of this research were to 1) determine the extent of intra-field variability of soil texture and organic matter in ten commercial Indiana fields, 2) quantify the reliability of five different combinations of spatial soil data sources, 3) determine the impact of soil sample intensity on map development and the classification accuracy for VR applications of soil residual herbicides, 4) quantify the impact of VR herbicide application on the total amount and spatial accuracy of herbicide applied according to product labels, and 5) determine if the intensive spatial characterization of soil properties is related to weed seedbank abundance and species richness to improve predictive weed management using soil residual herbicides.</p> <p><br></p> <p>Commercial soil data was generated by intensively collecting 60 soil samples in a stratified random sampling pattern in 10 agricultural fields across Indiana. Analysis of this data from commercial fields confirmed inherent field variability that would benefit from multiple management zones according to the labeled rate structures of pendimethalin, s-metolachlor, and metribuzin. Therefore, further research was conducted to determine an accurate and reliable method to delineate the fields into management zones for variable-rate residual herbicide applications based on the spatial soil variability and herbicide labels. </p> <p><br></p> <p>A modified Monte Carlo cross-validation method was used to determine the best source of spatial soil data and sampling intensity for delineating management zones for variable-rate applications of pendimethalin, s- metolachlor, and metribuzin. These sources of spatial soil data included: Soil Survey Geographic database (SSURGO) data, intensive soil samples, electrical resistivity sensors, and implement mounted optical reflectance sensors using VNIR reflectance spectroscopy. The mean management zone classification accuracy for maps developed from soil samples with and without electrical conductivity was similar for 75% of all maps developed across each field, herbicide, and sampling intensity. The method of using soil sampling data combined with electrical conductivity (SSEC) maps was most frequently the top performing source of spatial soil data. The most reliable sampling intensity was one sample per hectare which resulted in lower root mean squared error (RMSE) OM values, higher management zone classification accuracy, and more reliable predictions for the number of management zones within each field. </p> <p><br></p> <p>Using VR maps developed from SSEC with one sample per hectare sampling intensity, additional research was conducted to compare the amount of herbicide and field area that was over-or under-applied with a uniform application rate compared to a VR application for 10 corn and soybean residual herbicides. Although research from our previous study documented that spatial soil variability was extensive enough to require two or more management zones for all fields, the same labeled herbicide dose defined for multiple soil conditions led to 20% of all maps not requiring a variable rate application (VRA). Additionally, no difference was shown in the total amount applied of herbicide in an individual field between a variable and uniform application rate for all herbicides. Nonetheless, nearly half of all VR maps had 10% or more of the field area misapplied with a uniform application rate and justifies further research to determine if the proper placement of residual herbicide adds value through increased weed control in the field areas being under-applied. </p> <p><br></p> <p>Similar to soil residual herbicides, weed seedbank abundance and species richness were impacted by the variable soil conditions present within the field area. The seedbanks favor the establishment in areas of the field that promote vigorous germination, growth, and reproduction next to the competing crop. Therefore, soil sampling and weed seedbank greenhouse grow-outs were conducted in four fields to gain a better understanding in the relationship between the spatial soil and weed seedbank variability. All weed seedbank characteristics were shown to be spatially aggregated. Even though no individual or combination of soil parameters consistently explained the variability of weed seedbank abundance, species richness, or individual weed species across all four fields. However, clay content was the most persistent soil parameter to negatively impact (lower seedbank values) the soil weed seedbank.</p> <p><br></p> <p>Further field studies should be conducted across multiple sites to determine if variable-rate residual herbicide applications aid farmers by reducing the risk of crop injury in over-applied field areas and increased weed control in the areas being under-applied.  These studies should also access whether earlier emergence and/or greater weed densities occur in field areas receiving sublethal herbicide doses compared to areas receiving the optimal application rate. Additional research should investigate the utility of VR residual herbicide applications when tank-mixing multiple products during an application. Particularly, when the soil parameters used for selecting the herbicide rate are not defined the same across herbicide labels </p>