Propiedades bioquímicas del suelo en agroecosistemas de la zona centro sur estimadas por espectroscopía vis-nir
Loading...
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad de Concepción
Abstract
El uso de la espectroscopia del infrarrojo cercano visible (Vis-NIR) presenta un enfoque no destructivo, rápido, fiable y rentable para predecir las propiedades físicas, químicas y bioquímicas del suelo. En este estudio, se estimaron propiedades bioquímicas como la respiración basal del suelo, la biomasa microbiana activa (hidrólisis de diacetato de fluoresceína), la actividad βglucosidasa y la materia orgánica particulada-C (POM), variables clave en el ciclo del carbono. Se muestrearon 70 sitios en la zona centro-sur de Chile, específicamente en las regiones del Maule, Biobío, Ñuble y Araucanía (35°58′ S 72°58′ O) en cultivos tradicionales, pastizales y frutales a una profundidad de 0- 30 cm. Las muestras fueron analizadas en laboratorio y escaneadas usando un espectrofotómetro UV-Vis-NIR cubriendo un rango de 175 a 3300 nm y una resolución de 1 nm. Se utilizaron modelos multivariantes que incluían PLS, un método de regresión lineal, y Random Forest (RF), basado en el aprendizaje automático. Se evaluaron siete métodos de preprocesamiento, y la corrección ortogonal de la señal (OSC) resultó ser la mejor técnica de preprocesamiento en los modelos PLS. Los modelos PLS, con preprocesamiento OSC, y corrección multiplicativa de la dispersión (MSC), mostraron una alta precisión en rasgos como la respiración basal y la β-glucosidasa (R2validación: 0,99-0,98; RMSECV: 0,05-0,101, respectivamente). La FDA requirió transformaciones logarítmicas, mientras que la POM-C fue sensible a las correcciones espectrales (R2validación: 0,95; RMSECV: 1,0486). Random Forest (RF) presentó resultados heterogéneos: La POM-C se predijo bien (RPD = 2,08; R2validación = 0,77, RMSEP:1,128), pero la biomasa y la β-glucosidasa tuvieron rendimientos bajos. Basándose en estos resultados, la espectroscopia del suelo muestra potencial como herramienta para la estimación de los atributos bioquímicos del suelo.
The use of visible near-infrared spectroscopy (Vis-NIR) presents a nondestructive, fast, reliable, and cost-effective approach to predict soil physical, chemical and biochemical properties. In this study, biochemical properties such as basal soil respiration, active microbial biomass (Fluorescein diacetate hydrolysis), β-glucosidase activity and particulate organic matter-C (POM), key variables in the carbon cycle, were estimated. Seventy sites were sampled in south-central Chile, specifically in the regions of Maule, Biobío, Ñuble and Araucanía (35°58′ S 72°58′ W) in traditional crops, pastures, and fruit trees at a depth of 0-30 cm. The samples were analyzed in a laboratory and scanned using an UV-Vis-NIR spectrophotometer covering a range from 175 to 3300 nm and a resolution of 1 nm. Multivariate models including PLS, a linear regression method, and Random Forest (RF), which is based on machine learning, were used. Seven preprocessing methods were evaluated, and the orthogonal signal correction (OSC) emerged as the best preprocessing technique in PLS modelling. PLS models, with preprocessing OSC, and multiplicative scatter correction (MSC), showed high accuracy on traits such as basal respiration and β-glucosidase (R2validation: 0.99-0.98; RMSECV: 0.05-0.101, respectively). The FDA required logarithmic transformations, while POM was sensitive to spectral corrections (R2validation: 0.95; RMSECV: 1.0486). Random Forest (RF) presented heterogeneous results: POM was well predicted (RPD = 2.08; R2validation = 0.77, RMSEP:1.128), but biomass and β-glucosidase had low performances. Based on these results, soil spectroscopy shows potential as a tool for the estimation of soil biochemical attributes.
The use of visible near-infrared spectroscopy (Vis-NIR) presents a nondestructive, fast, reliable, and cost-effective approach to predict soil physical, chemical and biochemical properties. In this study, biochemical properties such as basal soil respiration, active microbial biomass (Fluorescein diacetate hydrolysis), β-glucosidase activity and particulate organic matter-C (POM), key variables in the carbon cycle, were estimated. Seventy sites were sampled in south-central Chile, specifically in the regions of Maule, Biobío, Ñuble and Araucanía (35°58′ S 72°58′ W) in traditional crops, pastures, and fruit trees at a depth of 0-30 cm. The samples were analyzed in a laboratory and scanned using an UV-Vis-NIR spectrophotometer covering a range from 175 to 3300 nm and a resolution of 1 nm. Multivariate models including PLS, a linear regression method, and Random Forest (RF), which is based on machine learning, were used. Seven preprocessing methods were evaluated, and the orthogonal signal correction (OSC) emerged as the best preprocessing technique in PLS modelling. PLS models, with preprocessing OSC, and multiplicative scatter correction (MSC), showed high accuracy on traits such as basal respiration and β-glucosidase (R2validation: 0.99-0.98; RMSECV: 0.05-0.101, respectively). The FDA required logarithmic transformations, while POM was sensitive to spectral corrections (R2validation: 0.95; RMSECV: 1.0486). Random Forest (RF) presented heterogeneous results: POM was well predicted (RPD = 2.08; R2validation = 0.77, RMSEP:1.128), but biomass and β-glucosidase had low performances. Based on these results, soil spectroscopy shows potential as a tool for the estimation of soil biochemical attributes.
Description
Tesis presentada para optar al grado de Magíster en Ciencias Agronómicas
Keywords
Propiedades del suelo - Chile, Espectroscopia, Bioquímica de suelos