Comparison of near and mid-infrared reflectance spectroscopy for the estimation of soil organic carbon fractions in Madagascar agricultural soils

Ramifehiarivo Nandrianina, Barthès Bernard, Cambou Aurélie, Chapuis-Lardy Lydie, Chevallier Tiphaine, Albrecht Alain, Razafimbelo Tantely. 2023. Comparison of near and mid-infrared reflectance spectroscopy for the estimation of soil organic carbon fractions in Madagascar agricultural soils. Geoderma Regional, 33:e00638, 13 p.

Résumé : Assessing the different pools of soil organic carbon (SOC) improves our understanding of how and at what rate the different forms of carbon (C) are being formed or lost in soils. Physical fractionation of soil organic matter (SOM) has often been used to separate and quantify SOC pools, but this approach is very tedious and can rarely be performed on large sample sets. Infrared spectroscopy has proven useful for time- and cost-effective quan- tification of total SOC, which prompted us to study its ability to characterise the distribution of SOC in physical fractions. This study aimed to compare the potential of near- and mid-infrared reflectance spectroscopy (NIRS and MIRS, respectively) to predict the distribution of SOC in particle-size and particle-density fractions, using spectra of unfractionated soils. A set of 134 sieved (< 2 mm) soil samples originating from seven sites in con- trasting pedoclimatic regions of Madagascar was studied. For each sample, five SOM fractions were separated: the particulate organic matter fraction (POM) and the particle-size fractions >200 μm, 50–200-μm, 20–50-μm and < 20 μm. The mass (g fraction 100 g 1 soil), SOC concentration (gC kg 1 fraction) and SOC amount (gC fraction kg 1 soil) were determined for each fraction in the laboratory. The NIR and MIR spectra were acquired on finely ground (< 0.2 mm) aliquots of unfractionated soil. Then, spectra were used for the prediction of each variable (i.e., the mass, SOC concentration and amount of each fraction, and SOC content of unfractionated soil), which was achieved with locally weighted partial least squares regression (LW-PLSR) on NIRS and MIRS data separately. For both spectral ranges, the same samples were used for calibration (n = 109, selected for spectral representativeness) and validation (n = 25). Models based on NIRS and MIRS yielded excellent predictions for SOC content in unfractionated soil (R2 = 0.98 and ratio of performance to interquartile range RPIQ ≥13) and accurate predictions (R2 ≥ 0.75 and RPIQ ≥2) for the mass, SOC concentration and SOC amount in most fractions. The predictions of SOC concentrations and SOC amounts were better in the fractions <200 μm (R2 ≥ 0.85 and RPIQ >3), especially in the fraction <20 μm (R2 > 0.9 and RPIQ >4.5). In most cases, NIRS slightly outperformed MIRS. This result contradicts most previous studies performed on soils from temperate regions but confirms those performed on soils from tropical regions. Infrared spectroscopy allowed accurate prediction of SOC distribution in particle-size fractions, which paves the way to high-throughput characterization of SOM.

Mots-clés Agrovoc : spectroscopie infrarouge, matière organique du sol, spectroscopie, carbone, agroécosystème

Mots-clés géographiques Agrovoc : Madagascar

Mots-clés libres : Soil fractionation, Locally weighted PLSR, Infrared reflectance spectroscopy, Soil organic matter, Tropical soils, Ferralsols, Arenosols, Fluvisols

Agences de financement hors UE : Fonds Français pour l'Environnement Mondial, Fondation BNP Paribas

Auteurs et affiliations

• Ramifehiarivo Nandrianina, Université d'Antananarivo (MDG) - auteur correspondant
• Barthès Bernard, IRD (FRA)
• Cambou Aurélie, IRD (FRA)
• Chapuis-Lardy Lydie, IRD (FRA)
• Chevallier Tiphaine, IRD (FRA)
• Albrecht Alain, CIRAD-PERSYST-UPR AIDA (REU)
• Razafimbelo Tantely, Université d'Antananarivo (MDG)

Source : Cirad-Agritrop (https://agritrop.cirad.fr/610781/)

Métriques

[ Page générée et mise en cache le 2024-12-18 ]