Preenchimento de falhas em série de dados meteorológicos e evolução temporal de índices bioclimáticos na bacia hidrográfica do Alto Paranapanema

Carregando...
Imagem de Miniatura

Data

2024-04-19

Orientador

Saad, João Carlos Cury

Coorientador

Pós-graduação

Engenharia Agrícola - FCA 33004064038P7

Curso de graduação

Título da Revista

ISSN da Revista

Título de Volume

Editor

Universidade Estadual Paulista (Unesp)

Tipo

Tese de doutorado

Direito de acesso

Acesso abertoAcesso Aberto

Resumo

Resumo (português)

Devido a inconsistência dos padrões climáticos tem-se observado alterações nos regimes hídricos e, por este motivo, são esperados nas próximas décadas impactos ambientais e socioeconômicos consideráveis. Neste sentido, os índices bioclimáticos vêm sendo empregados em diversos países do mundo por permitirem a interpretação conjunta das variáveis climáticas, a exemplo da relação existente entre a temperatura e a evapotranspiração de referência. A evolução temporal dos índices tem permitido identificar padrões climáticos além da avaliação de tendências dentro das séries avaliadas. Para isso, é necessário o registro de dados climáticos, porém, é frequente a ocorrência de lacunas nos dados devido a inúmeros fatores, resultando na demanda por métodos de preenchimento de falhas ao longo do tempo. O presente trabalho teve como objetivo avaliar o desempenho de técnicas de preenchimento de falhas, bem como a evolução e a tendência temporal de índices bioclimáticos na Bacia Hidrográfica do Alto Paranapanema. Os dados utilizados foram obtidos em escala diária junto ao Instituto Nacional de Meteorologia (INMET) compreendendo o período de 2007 a 2020. As estações automáticas estão localizadas nos municípios de Bauru, Barra Bonita, Avaré e Itapeva, no estado de São Paulo. Após a tabulação, os dados foram convertidos para a escala mensal e separados aleatoriamente em dois grupos, calibração (70%) e validação (30%). Para o preenchimento das falhas foram analisados métodos estatísticos e métodos baseados em machine learning (Suport Vector Machine, Random Forest, Bagged Tree e Gradient Boosting Machine). Após o preenchimento das falhas, foram calculados os índices bioclimáticos: Martonne Aridity Index (IDM), UNEP Aridity Index (UNEP AI), Self-calibrating Palmer Drought Severity Index (scPDSI) e Normalized Ecosystem Drought Index (NEDI). Os resultados apontam que em cenários com tendências de aumento da temperatura, a Regressão Linear Multipla (RLM) tende a subestimação. No empregado do Suporte Vector Machine (SVM) para precipitação, foi observado que o método contempla o padrão de distribuição mesmo em períodos com maior número de falhas. Como observado, a temperatura média tem se mantido constante ao longo dos anos, já a precipitação total e efetiva, gradativamente vêm apresentando redução. Desta maneira, o IDM tende a aproximar-se cada vez mais do IDMe. Mediante os resultados é possível concluir que para elementos com menor variância como temperatura, pressão atmosférica e evapotranspiração de referência (ETo) os métodos estatísticos são capazes de predizer com elevado grau de assertividade. Em contra partida, para elementos com alta variância recomenda-se o uso de técnicas de machine learning devido a capacidade dos modelos em compreender distribuições complexas, como regimes pluviométricos e de radiação global. Em relação ao comportamento dos índices bioclimáticos é possível afirmar que as maiores mudanças ocorreram no periodo de menor disponibilidade hídrica. Entre os índices avaliados é possível afirmar que existe uma tendência de redução ao longo da série analisada, principalmente nas médias anuais. Em relação aos meses do ano, os índices tentem a apresentar maior redução nos primeiros três meses do ano (período chuvoso).

Resumo (inglês)

Due to the inconsistency of climate patterns, changes in water regimes have been observed, which is the reason why considerable environmental and socioeconomic impacts are expected in the coming decades. In this sense, bioclimatic indices have been used in several countries around the world because they allow the joint interpretation of climatic variables, such as the relationship between temperature and reference evapotranspiration. The temporal evolution of the indices has allowed the identification of climate patterns in addition to the assessment of trends within the series evaluated. To achieve this, it is necessary to record climate data, however, gaps in the data frequently occur due to numerous factors, resulting in the demand for methods to fill gaps over time. The present work aimed to evaluate the performance of gap filling techniques, evolution, and temporal trend of bioclimatic indices in the upper Paranapanema river basin. The data used was obtained on a daily scale from the National Institute of Meteorology (INMET) covering the period from 2007 to 2020. The stations were automatic, located in the municipalities of Bauru, Barra Bonita, Avaré, and Itapeva, state of São Paulo. After tabulation, the data were converted to a monthly scale and randomly separated into two groups, calibration (70%) and validation (30%). To fill in the gaps, statistical methods and methods based on machine learning (Support Vector Machine, Random Forest, Bagged Tree and Gradient Boosting Machine) were analyzed. Following the filling of the gaps, the bioclimatic indices were calculated: Martonne Aridity Index (IDM), UNEP Aridity Index (UNEP AI), Selfcalibrating Palmer Drought Severity Index (scPDSI), and Normalized Ecosystem Drought Index (NEDI). The results indicate that in scenarios with increasing temperature trends, Multiple Linear Regression (RLM) tends to underestimation. When using the Support Vector Machine (SVM) for precipitation, it was observed that the method covers the distribution pattern even in periods with a greater number of failures. As noted, the average temperature has remained constant over the years, while total and effective precipitation has gradually been decreasing. In this way, IDM tends to get closer and closer to IDMe. Based on the results, it is possible to conclude that for elements with lower variance such as temperature, atmospheric pressure, and reference evapotranspiration (ETo), statistical methods are capable of predicting with a high degree of accuracy. On the other hand, for elements with high variance, it is recommended to use machine learning techniques due to the models' ability to understand complex distributions such as rainfall and global radiation regimes. Regarding the behavior of bioclimatic indices, it is possible to state that the biggest changes occurred in the period of lower water availability. Among the indices evaluated, it is possible to state that there is a tendency to reduce the series analyzed, mainly in the annual averages. Concerning the months of the year, the indices tend to show a greater reduction in the first three months of the year (rainy period).

Descrição

Idioma

Português

Como citar

SILVA, T. M. M. Preenchimento de falhas em série de dados meteorológicos e evolução temporal de índices bioclimáticos na bacia hidrográfica do Alto Paranapanema. 2024. Tese (Doutorado em Irrigação e Drenagem) - Faculdade de Ciências Agronômicas, Universidade Estadual Paulista, Botucatu, 2024.

Itens relacionados