Na pecuária extensiva de corte é importante saber a quantidade de alimento disponível na pastagem para que se possa estimar a quantidade ideal de animais que devem ocupar aquela área. Na Empresa Brasileira de Pesquisa Agropecuária - Pecuária Sul, a massa de forragem disponível em determinadas áreas experimentais é estimada por um método direto, onde são usadas gaiolas para isolar determinada área da pastagem e, a cada mês, a vegetação dessa área é cortada, seca e depois pesada. Dessa forma, obtém-se a massa seca da amostra coletada que é utilizada para a estimar a taxa de acúmulo de pasto para o mês posterior. Este trabalho propõe um modelo computacional baseado em Redes Neurais Artificiais Long Short-Term Memory com o intuito de melhorar a estimativa de disponibilidade de pasto feita atualmente na Empresa Brasileira de Pesquisa Agropecuária - Pecuária Sul, com base em dados amostrais da disponibilidade de pasto, coletados por meio do método direto, e meteorológicos. A metodologia é baseada em pesquisa exploratória e experimental interdisciplinar, e seu desenvolvimento, realizado na forma de estudo de caso, apontou alguns problemas no processo de amostragem de disponibilidade de pasto que vinha sendo realizado. Buscando contornar os problemas identificados, foi definida uma arquitetura onde os dados amostrais de disponibilidade de pasto e meteorológicos, são persistidos em um banco de dados geográfico, após um processo de Extract, Transform, and Load. Depois de carregados no banco de dados geográfico, os dados passam por um pré-processamento que os prepara para o processo de treinamento da Redes Neural Artificial Long Short-Term Memory desenvolvida. Os resultados mostram que a metodologia proposta é capaz de estimar com significativa acurácia a disponibilidade de pasto instantânea e do mês posterior, uma vez que o erro médio verificado é de mesma magnitude do erro associado ao método de amostragem utilizado.

Nanopesticides (Npes) carry the potential of increased efficacy while reducing application rates, hence increasing agricultural productivity in a more sustainable way. However, given its novelty, the environmental risk assessment of these advanced materials is mostly absent. In the present study we investigated the ecotoxicity of a commercial insecticide, with reported nanofeatures, Karate Zeon®, and compared it to its active substance lambda-cyhalothrin. It is hypothesised that the use of the nanopesticide Karate Zeon® poses lower risk to enchytraeids than its active substance. The standard non-target soil invertebrate Enchytraeus crypticus was used, and exposure was done in LUFA 2.2 soil in 4 tests (endpoints: days): avoidance test [avoidance behaviour: 2 days], OECD standard reproduction test [survival, reproduction plus adults' size: 28 days] and its extension [total number organisms: 56 days], and Full Life Cycle (FLC) test [hatching and juveniles' size: 13 days; survival, reproduction and adults' size: 46 days]. Results showed that enchytraeids did not avoid Karate Zeon® nor its active substance lambda-cyhalothrin, which could be due to neurotoxicity. There was no indication of increased toxicity with prolonged exposure (46, 56d) compared to the standard (28d) for neither of the materials, being overall equally toxic in terms of hatching, survival, and reproduction. The FLCt results indicated that the juvenile stage was the most sensitive, resulting in higher toxicity for the adult animals when exposed from the cocoon stage. Although toxicity was similar between Karate Zeon and lambda-cyhalothrin, different patterns of uptake and elimination cannot be excluded. The benefits of using Karate Zeon will rely on reduced application rates.

Nanopesticides (Npes) offer improved efficacy compared to their conventional forms while reducing the usage/application rates, hence being more sustainable options. However, there is still a knowledge gap on the Npes environmental impacts. To support the safety of nano-enabled pesticides, the present study aimed at assessing the toxicity of the commercial Npe NUCOP-M and the active substance copper oxychloride, using the ecotoxicological soil model Enchytraeus crypticus and LUFA 2.2 soil. Bioassays were performed to assess various endpoints within short-to longer-term exposures: avoidance behaviour (2 d), hatching (13 d), survival, reproduction and organisms' size (based on the standard OECD test (28 d), the OECD extension (56 d), and the Full Life Cycle test - FLCt (46 d)). Based on the standard OECD test and its extension, NUCOP-M had a similar level of toxicity as copper oxychloride without indications of increase in toxicity over time (28 versus 56 d). The shorter-term exposures (2 and 13 d) showed higher toxicity for copper oxychloride. The exposure from cocoon stage (FLCt) seemed to provide an adaptative advantage (reduced toxicity) to NUCOP-M. The differences might be related to a slower release of Cu2+ ions from NUCOP-M, which seems to account for the toxicity at longer-term. Based on the recommended application doses (ca. 1.72 mg NUCOP-M kg-1, i.e. 0.62 mg Cu kg-1 in the topsoil) there is no unacceptable risk of NUCOP-M on the enchytraeid population.

The work described in this dissertation focuses on the methods for analyzing MS and ML that are used in PF. The purpose of the work is to investigate these methods on their practical application to a specific set of data. In the course of the work, the following tasks were completed: the current state of affairs in the field of PF was investigated, the theoretical foundations of the methods of MS and ML were investigated, which were subjected to practical tests on a specific set of data. Conclusions were drawn about the advantages and disadvantages of these methods. A selection of works of scientists engaged in research on the introduction of a specific set of nutrients into the soil was also investigated. The most important contributions to this work are the practical application of various methods of analysis, as well as the design of a DST designed to help farmers integrate PF into their pilot training farms.

Megatrends, such as rising population and the associated increased demand for food, as well as limited resources, result in a higher need for a more efficient agriculture. Nowadays, agri-culture is moving into the Digital Age. Information technology and electronics are increasingly shaping agricultural practices. Digital applications help in many areas, for example fertilisa-tion, crop protection and, also, animal husbandry. Agricultural machinery is equipped with intelligent technologies for precision farming. Automated work processes have arrived in the field and on the farm. Nevertheless, the digital transformation in agriculture is far from com-plete. Over the next decade, 5G network as well as Big Data will play a critical role in further driving digital transformation in German agriculture, enabling higher agricultural efficiency. These two digital phenomena will allow farmers to be more productive and informed in real-time about their plants and animals. This thesis examines how agriculture is taking advantage of digital technologies and highlights the main use cases of 5G and Big Data in arable and livestock farming. The thesis also exam-ines the extent to which the fifth generation of mobile networks and Big Data will enhance German agriculture and discusses benefits along with potential bottlenecks associated with these digital enablers.

Um dos passos fundamentais no processamento de imagens para um sistema de visão artificial é a segmentação dos objetos de interesse na cena, e um dos métodos mais utilizados é a limiarização, em especial quando o objetivo é agrupar os pixels em duas classes. Neste método, o valor do limiar determina o número de pixels atribuídos a cada classe, além de influenciar a dimensão e a forma dos objetos nas imagens segmentadas. A utilização de métodos automáticos para definição do limiar, não só evitaria a influência de operadores mas, também, tornaria mais rápida a escolha dos limiares no campo, onde a variação da iluminação influencia os valores dos pixels. Este trabalho objetivou implementar e avaliar dois métodos automáticos de limiarização para identificação de plantas de milho atacadas pela lagarta do cartucho. Foram utilizadas imagens de plantas atacadas e não-atacadas, em três épocas, correspondendo a diferentes dias após a infestação. As plantas foram reunidas em três grupos de 10, sendo as imagens de cada grupo obtidas sob uma intensidade luminosa diferente. As imagens processadas com o índice do excesso de verde normalizado foram limiarizadas, automaticamente, e comparadas com a limiarização manual das mesmas imagens. Os resultados obtidos pelos dois métodos automáticos de limiarização foram satisfatórios, apresentando média acima de 99% de exatidão global, evidenciando-se, portanto, que ambos os métodos têm potencial para serem utilizados em um sistema de identificação de plantas de milho atacadas pela lagarta do cartucho.

Este trabalho teve como objetivo avaliar o uso de índices espectrais, retirados de imagens digitais, para discriminar diferentes doses de N no feijoeiro. O trabalho, conduzido em vasos de 8 dm³, teve cinco tratamentos (0; 50; 100; 150 e 200 kg de N ha-1), com dez repetições. As imagens foram adquiridas aos 30; 40 e 50 dias após a emergência. Foram desenvolvidas funções discriminantes quadráticas, tendo como vetores de entrada as médias dos "pixels" de diferentes combinações dos quatro índices espectrais testados. Três diferentes tamanhos de blocos de imagem foram testados 9 x 9; 20 x 20 e 40 x 40 "pixels". Os melhores resultados foram alcançados pelos blocos de 9 x 9 e 20 x 20 "pixels", apresentando classificação 94; 96 e 96% superior à classificação ao acaso para os blocos 9 x 9 "pixels" e 92; 94 e 94% para os blocos 20x20 "pixels" aos 30; 40 e 50 dias após a emergência, respectivamente.

ABSTRACT: Knowing the spatial variability of sugarcane biomass in the early stages of development may help growers in their management decision-making. Proximal canopy sensing is a promising technology that can identify this variability but is limited to quantifying plant-specific parameters. In this study, we evaluated whether biometric variables integrated with canopy reflectance data can assist in the generation of models for early-stage sugarcane biomass prediction. To substantiate this assertion, four sugarcane-producing fields were measured with an active crop canopy sensor and 30 sampling plots were selected for manually quantifying chlorophyll content, plant height, stalk number and aboveground biomass. We determined that Random Forest and Multiple Linear Regression models are similarly able to predict biomass, and that associating biometric variables such as number of stalks and plant height with reflectance data can assist model performance, depending on the attributes selected. This indicates that, when estimating biomass in the early stages, sugarcane growers can carry out site-specific management in order to increase yield and reduce the use of inputs.

This study aimed to develop classifiers based on different combinations of spectral bands and vegetation indices from original, segmented and reflectance images in order to determine the levels of leaf nitrogen and chlorophyll in the bean, and to define the best time and best variables. A remote-sensing system was used, consisting of a helium balloon and two small-format digital cameras. Besides the individual spectral bands, four vegetation indices were tested: simple ratio, normalized difference, normalized difference in the green band, and modified-chlorophyll absorption. The classifiers proved to be efficient in determining levels of leaf nitrogen and chlorophyll. The best time for determining leaf N content was at 13 DAE (stage V4). The best classifiers for that time used as input variables two indices from segmented reflectance images, one index related to the canopy structure and the other related to chlorophyll, with a Kappa ranging from 0.26 to 0.31. The best time to discriminate leaf chlorophyll content was 21 DAE (stage V4). The best classifier used as input variables two original images, one in the red band and one in the blue with a Kappa of 0.47.

ABSTRACT The aim of this study was to evaluate the spatial dependence between chemical properties of the soil and yield components in the soybean using precision farming techniques. Samples of the soil and plants were taken from georeferenced points to determine the chemical properties of the soil and the yield components. The results were submitted to Pearson correlation analysis, descriptive statistics and geostatistics. The coefficient of variation showed a wide range of distribution for the chemical attributes of the soil, with the highest indices being found for the levels of available phosphorus (102%) and potassium (72.65%). Soil pH and organic matter showed a coefficient of variation of 5.96 and 15.93% respectively. Semivariogram analysis of the yield components (productivity, 1,000-seed weight and number of seeds) and the chemical properties of the soil (organic matter, pH, phosphorus, potassium, calcium, magnesium, boron, manganese and zinc) fitted the spherical model with moderate spatial dependence, with values ranging from 200 to 700 m. Spatial distribution by means of map interpolation was efficient in evaluating spatial variability, allowing the identification and quantification of regions of low and high productivity in the production area, together with the distribution of soil attributes and their respective levels of availability to the soybean plants.