Suprimento hídrico e índices de vegetação para estimativa de produtividade de milho com machine learning

• Autor(a) principal: Avozani, Amanda
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Manancial - Repositório Digital da UFSM
• dARK ID: ark:/26339/001300000vbjp
• Texto Completo: http://repositorio.ufsm.br/handle/1/29498
• topic: Agricultura de precisão; Ciência de dados; Aeronaves remotamente pilotadas; Precision agriculture; Machine learning; Remotely piloted aircraft; CNPQ::CIENCIAS AGRARIAS::AGRONOMIA

The development and productivity of maize crops depend crucially on water availability, making this variable extremely important for achieving high levels of productivity. In this regard, the study aimed to investigate the influence of water supply on crop development and productivity, utilizing factorial multivariate analysis (FMA) and artificial neural networks (ANNs). The study was conducted in two environments, one irrigated and one non-irrigated. Irrigation was carried out using a central pivot that applied a cumulative water depth of 36.25 mm throughout the growing cycle. Four microstations were installed in each environment, equipped with soil moisture sensors and measurements of precipitation (rainfall and irrigation). During the phenological stages, remotely piloted aircraft flights and multispectral sensors were conducted to generate vegetation indices. Data analysis showed that irrigation significantly altered the productive system, even with the application of just over 10% of the recommended irrigation. In the relationships between variables in the non-irrigated environment, the influence of a severe water deficit from mid-November/21 to mid-January/22 was observed, reflected by stress-related vegetation indices (PSRI), low productivity, and dependence on precipitation. In the irrigated environment, the addition of 36 mm through three irrigations during the critical period caused significant changes. The plants exhibited greater vegetative vigor and physiological activity, resulting in higher productivity of 7.81 t ha−¹ , a 46.35% increase compared to the non-irrigated environment (4.19 t ha−¹ ). ANNs were used to estimate maize productivity, and their estimates were influenced by variables such as soil water content measurement by sensors and the PSRI vegetation index. The ANNs presented specific models for each environment in the maize production system, with a (6-4-1) architecture consisting of 6 neurons, with a focus on the participation of soil sensor variables in the 10cm and 30cm layers and the PSRI vegetation index in the input layers. It was concluded that irrigation significantly altered the maize production system, and the FMA analysis detected the influence of irrigation on the analyzed variables and productivity