Modelo de estimativa para a previsão meteorológica para fins agrícolas utilizando machine learning
Carregando...
Arquivos
Data
2022-07-29
Autores
Vieira, Igor Cristian de Oliveira
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
A previsão das condições climáticas auxilia nas tomadas de decisões devido a sua grande influência nos cultivos, possibilitando estratégias produtivas mais eficientes e diminuição de danos, principalmente nas fases mais críticas da cultura do milho que determinam seu potencial produtivo. Foi desenvolvido um modelo de regressão linear múltipla (RLM) para previsão decendial dos elementos meteorológicos com pelo menos 2 meses de antecedência para 15 localidades que são destaque na produção de milho no Brasil. Foi utilizado um conjunto de dados diários de temperatura média, mínima e máxima do ar, velocidade do vento, umidade relativa e radiação global fornecidos pelo sistema NASA/POWER e dados de precipitação obtidos do banco de dados da Agência Nacional de Águas (ANA) (2003 a 2019), organizados em decêndios (DEC) em função do ciclo médio do milho e agrupados em dois tipos de clima (Am e Aw). As previsões utilizando 14 DEC nos dois tipos climáticos apresentaram, em média, elevada acurácia para todos os elementos, com destaque para a velocidade do vento que com 2 meses de antecedência atingiu R²ajust > 0,9 e RMSE e ES < 0,9 m/s. A chuva, mesmo com maior variabilidade no clima Am, obteve um R²ajust > 0,62 e com 12 DEC R²ajust > 0,55 e em ambos RMSE e ES < 30 mm, enquanto no clima Aw com 14 DEC R²ajust > 0,70 e com 12 DEC R²ajust > 0,60 e em ambos RMSE e ES < 50 mm. Conclui-se que a RLM é uma ferramenta eficaz para a previsão de elementos climáticos em escala decendial.
Climatic conditions forecast helps in decision making due to its great influence on crops, enabling more efficient production strategies and damages reduction, especially in the most critical corn phenological phases when the productive potential is determined. A multiple linear regression model (MLR) was developed for ten-day forecasting of meteorological elements at least 2 months in advance for 15 locations with emphasis on corn production in Brazil. A set of daily data of average, minimum and maximum air temperature, wind speed, relative humidity and global radiation was provided by the NASA/POWER system and precipitation data was obtained from the National Water Agency (ANA) database (2003 to 2019). The data set was organized into ten days period (DEC) according to the average maize cycle and grouped into two climate classification (Am and Aw). Forecasts with 14 DEC in bothweather classification obtained, on average, high accuracy for all elements, with emphasis on the wind speed which, 2 months in advance, reached R²adjust > 0.9, RMSE and ES < 0.9 m/ s. For rain, even with higher variability, in Am had an R²adjust > 0.62 and with 12 DEC R²adjust > 0.55 and in both RMSE and ES < 30 mm and in the climate Aw with 14 DEC R²adjust > 0.70 and with 12 DEC R²adjust > 0.60 and in both RMSE and ES < 50 mm. In this way, MLR can serve as an effective tool in predicting climate elements on a ten-day scale.
Climatic conditions forecast helps in decision making due to its great influence on crops, enabling more efficient production strategies and damages reduction, especially in the most critical corn phenological phases when the productive potential is determined. A multiple linear regression model (MLR) was developed for ten-day forecasting of meteorological elements at least 2 months in advance for 15 locations with emphasis on corn production in Brazil. A set of daily data of average, minimum and maximum air temperature, wind speed, relative humidity and global radiation was provided by the NASA/POWER system and precipitation data was obtained from the National Water Agency (ANA) database (2003 to 2019). The data set was organized into ten days period (DEC) according to the average maize cycle and grouped into two climate classification (Am and Aw). Forecasts with 14 DEC in bothweather classification obtained, on average, high accuracy for all elements, with emphasis on the wind speed which, 2 months in advance, reached R²adjust > 0.9, RMSE and ES < 0.9 m/ s. For rain, even with higher variability, in Am had an R²adjust > 0.62 and with 12 DEC R²adjust > 0.55 and in both RMSE and ES < 30 mm and in the climate Aw with 14 DEC R²adjust > 0.70 and with 12 DEC R²adjust > 0.60 and in both RMSE and ES < 50 mm. In this way, MLR can serve as an effective tool in predicting climate elements on a ten-day scale.
Descrição
Palavras-chave
Agrometeorologia, Meteorologia, Big data, Previsão tempo, Clima, Temperatura atmosférica