Bayesian modelling of a decision support system for irrigation

• Ano de defesa: 2024
• Autor(a) principal: Ribeiro, Vitor Pinto
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Estadual Paulista (Unesp)
• Palavras-chave em Português: Bayesian networks; Probabilistic inference; Influence diagrams; Evapotranspiration; Redes bayesianas; Inferência estatística; Diagramas de influência; Evapotranspiração; Inferência (Estatística); Sistemas estocásticos
• Link de acesso: https://hdl.handle.net/11449/255844

The Sustainable Development Goals are a set of seventeen goals proposed by the United Nations as guidelines to promote the sustainable growth of signatory nations. Such objectives include ones related to food security and safe water availability. Brazil is one of the leading countries in agricultural production, which ties most of its consumptive water usage to farming and irrigation. Most farmers activate their irrigation systems before dawn to save on water, electricity and related fees and require planning on the volume of water to supply the crops with enough to ensure optimal productivity. The main component of water loss is evapotranspiration, a complex and non-linear phenomenon that is a function of climatological variables and crop and soil parameters. Many analytical methods to estimate evapotranspiration require hard-to-obtain data or are approximations with a high region specificity. With the development of new computational tools, many researchers began to employ machine learning and remote sensing to improve the assessment of hydro-climatological variables. Bayesian Network is a class of probabilistic models based on graphs and appropriate to deal with stochastic problems and under scenarios of missing data with the main drawback as devolving into an NP-hard optimisation problem. A probabilistic model based on Bayesian Networks called Inference Diagram provides the toolset to develop support decision systems over stochastic scenarios. The objectives of this project are threefold. First, to implement and validate a bio-inspired algorithm to build a Bayesian Network from data. Second, to create a Bayesian Network capable of estimating evapotranspiration from climatological data even under missing data. And third, to propose an Influence Diagram model that accommodates hydro-climatological stochastic variables and provides suggestions on irrigation scheduling. This project's results achieve the first two objectives, each resulting in a published conference paper, while the third goal requires more development time.