Neural networks in climate spatialization and their application in the agricultural zoning of climate risk for sunflower in different sowing dates

• Autor(a) principal: Aparecido, Lucas Eduardo de Oliveira
• Data de Publicação: 2019
• Outros Autores: de Moraes, José Reinaldo da Silva Cabral [UNESP], Rolim, Glauco de Souza [UNESP], Martorano, Lucieta Guerreiro, de Meneses, Kamila Cunha [UNESP], Valeriano, Taynara Tuany Borges [UNESP]
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://dx.doi.org/10.1080/03650340.2019.1566715; http://hdl.handle.net/11449/190057
• topic: climate modeling; crop zoning; Helianthus annus; multi-layer perceptron network

Sunflower is a species that is sensitive to local climate conditions. However, studies that use artificial neural networks (ANNs) to evaluate this influence and create tools such as agricultural zoning of climate risk (ZARC) have not been conducted for this species. Due to the importance of sunflower as a human food source and for biodiesel production, and also the necessity of conducting research to evaluate the suitability of this oleaginous species under different climatic conditions. Thus, we seek to construct a ZARC for sunflower in Brazil simulating sowing on different dates and using meteorological elements spatialized by ANNs. Climate data were used: air temperature (T), rainfall (P), relative air humidity (UR), solar radiation (MJ_m−2_d−1) and wind velocity (U2). Climatic regions considered suitable for the cultivation of sunflower had average annual values for T between 20 and 28°C, P between 500 and 1.500 mm per cycle, and soil water deficit (DEF) below 140 mm per cycle. A neural network is an efficient tool that can be used in spatialization of climate variables quickly and accurately. Sunflower sowing in the spring and summer are the ones that provide the largest suitable areas in southeastern Brazil, with 58.13 and 64.36% of suitable areas, respectively.