Assessing the utility of Munsell soil color in building and evaluating spectral models for soil clay content prediction
2024
Dharumarajan, S. | Gomez, Cécile | Lalitha, M. | Vasundhara, R. | Hegde, R. | Patil, N.G. | Indian Council of Agricultural Research (ICAR) | Laboratoire d'étude des Interactions Sol - Agrosystème - Hydrosystème (UMR LISAH) ; Institut de Recherche pour le Développement (IRD)-AgroParisTech-Institut National de Recherche pour l’Agriculture, l’Alimentation et l’Environnement (INRAE)-Institut Agro Montpellier ; Institut national d'enseignement supérieur pour l'agriculture, l'alimentation et l'environnement (Institut Agro)-Institut national d'enseignement supérieur pour l'agriculture, l'alimentation et l'environnement (Institut Agro) | Indo-French Cell for Water Sciences = Cellule Franco Indienne de Recherche en Science de l’Eau (IFCWS = CEFIRSE) ; Indian Institute of Science [Bangalore] (IISc Bangalore) | The authors thank the Karnataka Watershed Development Department and the World Bank for funding the Sujala III and REWARD project. The authors thank the ATCHA project, ANR-16-CE03-0006, and the PNTS 2023-08 774 "TankSed" project for supporting the work. | ANR-16-CE03-0006,ATCHA,Accompagner l'adaptation de l'agriculture irriguée au changement climatique(2016)
International audience
显示更多 [+] 显示较少 [-]英语. The present study examined how the use of soil color can help build and evaluate clay content prediction models from laboratory visible and near infrared spectroscopic data. This study was based on a regional database containing 449 soil samples collected over Karnataka state in India, which has been divided into red soils (240 samples) and black soils (209 samples) based on their Munsell soil color. Partial least squares regression models were calibrated and validated from both the regional datasets and subsets stratified as red and black soils. In addition, a random forest model was used to classify the validation soil samples into black and red classes to evaluate models’ performance. First, while the clay content predicted by the regression model built from regional data was evaluated as correct at regional scale ( R 2 val of 0.75), this model was evaluated as more accurate over black ( R 2 val of 0.8) than red ( R 2 val of 0.63) soil samples. Second, the regression models built from subsets stratified per soil color provided different performances than the regression model built from the regional data, both at the regional scale and soil color scale. In conclusion, this study demonstrated that (1) predictions are highly dependent on calibration data, (2) the interpretation of prediction performances relies heavily on validation data, and (3) pedological knowledge, such as soil color, can be effectively employed as an encouraging covariate in both the construction and evaluation of regression models.
显示更多 [+] 显示较少 [-]