Estimación del nivel de estrés hídrico en frutales mediante técnicas machine learning para aplicación en sistemas de riego inteligentes

  1. Juan D. González-Teruel 1
  2. Victor Blanco 2
  3. Pedro José Blaya-Ros 1
  4. Rafael Domingo Miguel 1
  5. Fulgencio Soto Vallés 1
  6. R. Torres Sánchez 1
  1. 1 Universidad Politécnica de Cartagena
    info

    Universidad Politécnica de Cartagena

    Cartagena, España

    ROR https://ror.org/02k5kx966

  2. 2 Washington State University
    info

    Washington State University

    Pullman, Estados Unidos

    ROR https://ror.org/05dk0ce17

Libro:
XLII Jornadas de Automática: libro de actas, Castellón, 1 a 3 de septiembre de 2021

Editorial: Universitat Jaume I ; Servizo de Publicacións ; Universidade da Coruña ; Comité Español de Automática

ISBN: 978-84-9749-804-3

Año de publicación: 2021

Páginas: 477-484

Congreso: Jornadas de Automática (42. 2021. Castellón)

Tipo: Aportación congreso

DIALNET GOOGLE SCHOLAR HANDLE: https://hdl.handle.net/2183/28246

Objetivos de desarrollo sostenible

Resumen

Water is a limited resource in arid and semi-arid regions. This is the case of the Mediterranean area, where its demographic and climatic conditions make it particularly prone to farming, demanding a major percentage of water resources. Deficit irrigation strategies have proved to be successful, but it is essential to control crop water stress. The measurement of crop water stress is currently associated with midday stem water potential, which is very time-consuming. At an agricultural perspective, it would be interesting to define qualitative levels of crop water stress and to be able to estimate them from variables whose measurement can be automated, so that intelligent irrigation systems can be implemented based on the water needs of the crop. In this work we present a preliminary study to obtain a model capable of predicting five levels of crop water stress from time data of water potential and volumetric water content in the soil and different agro-climatic variables. Multiple Machine Learning algorithms have been evaluated, obtaining a maximum estimation accuracy of 72.4%.

Fuente de los datos: Dialnet