Estimación de humedad del suelo mediante regresiones lineales múltiples en Llano Brenes
Date
Authors
Palominos Rizzo, María Teresa
Villatoro Sánchez, Mario
Alvarado Hernández, Alfredo
Cortés Granados, Víctor
Paguada Pérez, Darwin
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Introducción. La humedad del suelo es una variable muy importante en el suministro de agua para la agricultura
y es su principal recurso. Sin embargo, su medición en campo suele presentar limitaciones, por lo que su predicción
es necesaria para diversas actividades de planificación agrícola e investigación. Objetivo. Predecir la humedad diaria
del suelo a escala de cultivo, a partir de información meteorológica mediante modelos de regresión lineal múltiple.
Materiales y métodos. El estudio se desarrolló en Llano Brenes, Alajuela, Costa Rica. Se instalaron sensores de
reflectometría de dominio temporal (TDR) y registraron información cada veinte minutos de humedad de suelo desde
noviembre 2018 a diciembre 2019. El suelo se clasificó a nivel taxonómico como Lithic Ustorthents, en una finca
con cultivo de café en producción. Se tomaron muestras de suelo no disturbadas para la calibración de los TDR y se
realizó un análisis de estabilidad temporal. El primer modelo (RLM1) fue una regresión lineal múltiple con variables
meteorológicas, en el segundo modelo (RLM2) además de las variables meteorológicas, se separó la precipitación en
subperíodos, los cuales se introdujeron como variables “dummy”, mientras que el tercer modelo (PCA) consistió en un
análisis de componentes principales y un modelo de regresión lineal. Resultados. Los modelos RLM2 (R2 = 0,838) y
PCA (R2 = 0,823) presentaron un mejor desempeño en comparación con el modelo RLM1 (R2 = 0,540). Sin embargo,
el modelo RLM2 se consideró más útil, debido a su simplicidad y a que presentó los mejores indicadores de bondad
de ajuste. Conclusión. Los modelos de regresión lineal con variables meteorológicas permitieron estimar la humedad
del suelo, debido a que esta tendió a seguir los patrones estacionales y las variaciones de la precipitación, tal como se
observó en el RLM2 con la separación de subperíodos
Introduction. Soil moisture is a very important variable in the water supply for agriculture and it is its main resource. However, its field measurement usually has limitations, so its prediction is necessary for various agricultural planning and research activities. Objective. To predict daily soil moisture at the crop scale from meteorological information through multiple linear regression models. Materials and methods. The study was carried out in Llano Brenes, Alajuela, Costa Rica. Time domain reflectometry (TDR) sensors were installed and soil moisture information was recorded every twenty minutes from November 2018 to December 2019. The soil was classified at the taxonomic level as Lithic Ustorthents, in a farm with coffee cultivation in production. Undisturbed soil samples were taken for TDR calibration and a temporal stability analysis was performed. The first model (RLM1) was a multiple linear regression with meteorological variables, in the second model (RLM2) in addition to the meteorological variables, the precipitation was separated into sub-periods which were introduced as dummy variables, while the third model (PCA) consisted of a main component analysis and a linear regression model. Results. The RLM2 (R2 = 0.838) and PCA (R2 = 0.823) models performed better than the RLM1 model (R2 = 0.540). However, the RLM2 model was considered more useful due to its simplicity and the fact that it presented the best goodness-of-fit indicators. Conclusion. The linear regression models with meteorological variables allowed estimating soil moisture, because it tends to follow seasonal patterns and variations in precipitation, as observed in the RLM2 with the separation of sub-periods.
Introduction. Soil moisture is a very important variable in the water supply for agriculture and it is its main resource. However, its field measurement usually has limitations, so its prediction is necessary for various agricultural planning and research activities. Objective. To predict daily soil moisture at the crop scale from meteorological information through multiple linear regression models. Materials and methods. The study was carried out in Llano Brenes, Alajuela, Costa Rica. Time domain reflectometry (TDR) sensors were installed and soil moisture information was recorded every twenty minutes from November 2018 to December 2019. The soil was classified at the taxonomic level as Lithic Ustorthents, in a farm with coffee cultivation in production. Undisturbed soil samples were taken for TDR calibration and a temporal stability analysis was performed. The first model (RLM1) was a multiple linear regression with meteorological variables, in the second model (RLM2) in addition to the meteorological variables, the precipitation was separated into sub-periods which were introduced as dummy variables, while the third model (PCA) consisted of a main component analysis and a linear regression model. Results. The RLM2 (R2 = 0.838) and PCA (R2 = 0.823) models performed better than the RLM1 model (R2 = 0.540). However, the RLM2 model was considered more useful due to its simplicity and the fact that it presented the best goodness-of-fit indicators. Conclusion. The linear regression models with meteorological variables allowed estimating soil moisture, because it tends to follow seasonal patterns and variations in precipitation, as observed in the RLM2 with the separation of sub-periods.
Description
Keywords
Contenido de agua en el suelo, Coffea, Zona tropical, Soil water content, Tropical zone