ePrints@IIScePrints@IISc Home | About | Browse | Latest Additions | Advanced Search | Contact | Help

Soil order knowledge as a driver in soil properties estimation from Vis-NIR spectral data – Case study from northern Karnataka (India

Dharumarajan, S and Gomez, C and Lalitha, M and Kalaiselvi, B and Vasundhara, R and Hegde, R (2022) Soil order knowledge as a driver in soil properties estimation from Vis-NIR spectral data – Case study from northern Karnataka (India. In: Geoderma Regional, 32 .

[img] PDF
Geo_Reg_32_2022.pdf - Published Version
Restricted to Registered users only

Download (3MB) | Request a copy
Official URL: https://doi.org/10.1016/j.geodrs.2022.e00596


Visible and near-infrared (Vis-NIR, 350–2500 nm) laboratory spectroscopy has been proven to provide soil properties estimations, such as clay or organic carbon (OC). However, the performances of such estimations may be dependent on pedological and spectral similarities between calibration and validation datasets. The objective of this study was to analyze how the soil order knowledge can be used to increase regression models performance for soil properties estimation. For this purpose, Random Forest regression models were calibrated and validated from both regional database (called regional models) and subsets stratified by soil order from the regional database (called soil-order models). The regional database contained 482 soil samples belonging to four soil orders (Alfisols, Vertisols, Inceptisols and Entisols) and associated with Vis-NIR laboratory spectra and six soil properties: OC, sand, silt, clay, cation exchange capacity (CEC) and pH. First, regional models provided i) high accuracy of some soil properties estimations when considering the regional strategy in the validation step (e.g., R2 val of 0.74, 0.76 and 0.74 for clay, CEC and sand, respectively) but ii) modest accuracy of these same soil properties when considering subsets stratified by soil order from the regional database in validation step (e.g., R2 val of 0.48, 0.58 and 0.38 over Vertisol for clay, CEC and sand, respectively). So the estimation accuracy appreciation is highly depending on the validation database as there is a risk of over-appreciated prediction accuracies at the soil-order scale when figures of merit are based on a regional validation dataset. Second, this work highlighted that the benefit of a soil-order model compared to a regional model for calibration depends on both soil property and soil order. So no recommendations for choosing between both models for calibration may be given. Finally, while Vis-NIR laboratory spectroscopy is becoming a popular way to estimate soil physico- chemical properties worldwide, this work highlights that this technique may be used discreetly depending on the targeted scale and targeted soil type

Item Type: Journal Article
Publication: Geoderma Regional
Publisher: Elsevier B.V.
Additional Information: The copyright for this article belongs to Elsevier B.V
Keywords: Prediction accuracy; Random forest; Regional model; Soil variability; Soil-order model; Visible near-infrared
Department/Centre: Division of Biological Sciences > Centre for Ecological Sciences
Division of Mechanical Sciences > Centre for Earth Sciences
Autonomous Societies / Centres > Karnataka State Council for Science and Technology
Date Deposited: 13 Jan 2023 09:42
Last Modified: 13 Jan 2023 09:42
URI: https://eprints.iisc.ac.in/id/eprint/79122

Actions (login required)

View Item View Item