Ensemble and transfer learning of soil inorganic carbon with visible near-infrared spectra

Soil inorganic carbon (SIC) dominates the soil carbon pools in semi-arid and arid areas globally. Variations in the SIC pool would substantially affect the atmospheric CO2 concentrations. The rapid and accurate measurement of SIC content using visible near-infrared (Vis-NIR) spectroscopy is of high...

Full description

Saved in:
Bibliographic Details
Main Authors: Yu Wang, Keyang Yin, Bifeng Hu, Yongsheng Hong, Songchao Chen, Jing Liu, Lili Yang, Jie Peng, Zhou Shi
Format: Article
Language:English
Published: Elsevier 2025-04-01
Series:Geoderma
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S0016706125000953
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850185440135479296
author Yu Wang
Keyang Yin
Bifeng Hu
Yongsheng Hong
Songchao Chen
Jing Liu
Lili Yang
Jie Peng
Zhou Shi
author_facet Yu Wang
Keyang Yin
Bifeng Hu
Yongsheng Hong
Songchao Chen
Jing Liu
Lili Yang
Jie Peng
Zhou Shi
author_sort Yu Wang
collection DOAJ
description Soil inorganic carbon (SIC) dominates the soil carbon pools in semi-arid and arid areas globally. Variations in the SIC pool would substantially affect the atmospheric CO2 concentrations. The rapid and accurate measurement of SIC content using visible near-infrared (Vis-NIR) spectroscopy is of high significance for the management of soil carbon pools in semi-arid and arid regions. Ensemble learning is a novel and advanced modeling approach. However, it has been applied less in soil spectroscopy, and its transfer capability has not been evaluated. Therefore, we hypothesized that the use of the ensemble technique could further SIC prediction accuracy and have a better model transfer capability. In this study, a stacking model was developed using 990 soil samples collected from the Alar Reclamation region in South Xinjiang, China. The stacking model consists of 10 base models (support vector machine (SVM), partial least squares algorithm (PLSR), multi-layer perceptron (MLP), etc.). Two strategies (hyperparameter-adjusted and −unadjusted) were used to transfer the model to other target areas including Shaya and Wensu Counties on the southern border of China. Our results demonstrate that the SIC content could be predicted accurately using the stacking models (R2p = 0.81). The stacking model outperformed all the individual models and significantly improved the prediction accuracy of SIC. The R2p of the stacking models improved by 0.05–0.21, and the root mean square error (RMSEP) reduced by 0.33–1.44 g kg−1. Additionally, the stacking models displayed superior model transfer capability. Compared with direct transfer, the stacking model with fine-tuning of the hyperparameters displayed better model stability and generalization. Moreover, the average R2p improved by over 0.09 compared with the stacking model with unadjusted hyperparameters. Overall, stacking ensemble learning is a potential method for predicting SIC with good transfer capabilities. Our results provide new tools and strategies for the accurate estimation of SIC in semi-arid and arid regions.
format Article
id doaj-art-85f2701265db4bcc88d9c4cee705dc95
institution OA Journals
issn 1872-6259
language English
publishDate 2025-04-01
publisher Elsevier
record_format Article
series Geoderma
spelling doaj-art-85f2701265db4bcc88d9c4cee705dc952025-08-20T02:16:44ZengElsevierGeoderma1872-62592025-04-0145611725710.1016/j.geoderma.2025.117257Ensemble and transfer learning of soil inorganic carbon with visible near-infrared spectraYu Wang0Keyang Yin1Bifeng Hu2Yongsheng Hong3Songchao Chen4Jing Liu5Lili Yang6Jie Peng7Zhou Shi8College of Agriculture, Tarim University, Alar 843300, China; College of Natural Resources and Environment, Northwest A&F University, Yangling, Shaanxi 712100, ChinaCollege of Agriculture, Tarim University, Alar 843300, ChinaDepartment of Land Resources Management, School of Public Administration, Jiangxi University of Finance and Economics, Nanchang 330013, ChinaState Key Laboratory of Soil and Sustainable Agriculture, Chinese Academy of Sciences, Nanjing 210008, ChinaZJU-Hangzhou Global Scientific and Technological Innovation Center, Zhejiang University, Hangzhou 311215, China; College of Resources and Environment, Zhejiang University, Hangzhou 310058, ChinaCollege of Natural Resources and Environment, Northwest A&F University, Yangling, Shaanxi 712100, ChinaCollege of Agriculture, Tarim University, Alar 843300, ChinaCollege of Agriculture, Tarim University, Alar 843300, China; Key Laboratory of Genetic Improvement and Efficient Production for Specialty Crops in Arid Southern Xinjiang of Xinjiang Corps, Alar 843300, China; Corresponding author.College of Resources and Environment, Zhejiang University, Hangzhou 310058, ChinaSoil inorganic carbon (SIC) dominates the soil carbon pools in semi-arid and arid areas globally. Variations in the SIC pool would substantially affect the atmospheric CO2 concentrations. The rapid and accurate measurement of SIC content using visible near-infrared (Vis-NIR) spectroscopy is of high significance for the management of soil carbon pools in semi-arid and arid regions. Ensemble learning is a novel and advanced modeling approach. However, it has been applied less in soil spectroscopy, and its transfer capability has not been evaluated. Therefore, we hypothesized that the use of the ensemble technique could further SIC prediction accuracy and have a better model transfer capability. In this study, a stacking model was developed using 990 soil samples collected from the Alar Reclamation region in South Xinjiang, China. The stacking model consists of 10 base models (support vector machine (SVM), partial least squares algorithm (PLSR), multi-layer perceptron (MLP), etc.). Two strategies (hyperparameter-adjusted and −unadjusted) were used to transfer the model to other target areas including Shaya and Wensu Counties on the southern border of China. Our results demonstrate that the SIC content could be predicted accurately using the stacking models (R2p = 0.81). The stacking model outperformed all the individual models and significantly improved the prediction accuracy of SIC. The R2p of the stacking models improved by 0.05–0.21, and the root mean square error (RMSEP) reduced by 0.33–1.44 g kg−1. Additionally, the stacking models displayed superior model transfer capability. Compared with direct transfer, the stacking model with fine-tuning of the hyperparameters displayed better model stability and generalization. Moreover, the average R2p improved by over 0.09 compared with the stacking model with unadjusted hyperparameters. Overall, stacking ensemble learning is a potential method for predicting SIC with good transfer capabilities. Our results provide new tools and strategies for the accurate estimation of SIC in semi-arid and arid regions.http://www.sciencedirect.com/science/article/pii/S0016706125000953Soil inorganic carbonVis-NIR spectroscopyStacking modelModel transferMachine learning
spellingShingle Yu Wang
Keyang Yin
Bifeng Hu
Yongsheng Hong
Songchao Chen
Jing Liu
Lili Yang
Jie Peng
Zhou Shi
Ensemble and transfer learning of soil inorganic carbon with visible near-infrared spectra
Geoderma
Soil inorganic carbon
Vis-NIR spectroscopy
Stacking model
Model transfer
Machine learning
title Ensemble and transfer learning of soil inorganic carbon with visible near-infrared spectra
title_full Ensemble and transfer learning of soil inorganic carbon with visible near-infrared spectra
title_fullStr Ensemble and transfer learning of soil inorganic carbon with visible near-infrared spectra
title_full_unstemmed Ensemble and transfer learning of soil inorganic carbon with visible near-infrared spectra
title_short Ensemble and transfer learning of soil inorganic carbon with visible near-infrared spectra
title_sort ensemble and transfer learning of soil inorganic carbon with visible near infrared spectra
topic Soil inorganic carbon
Vis-NIR spectroscopy
Stacking model
Model transfer
Machine learning
url http://www.sciencedirect.com/science/article/pii/S0016706125000953
work_keys_str_mv AT yuwang ensembleandtransferlearningofsoilinorganiccarbonwithvisiblenearinfraredspectra
AT keyangyin ensembleandtransferlearningofsoilinorganiccarbonwithvisiblenearinfraredspectra
AT bifenghu ensembleandtransferlearningofsoilinorganiccarbonwithvisiblenearinfraredspectra
AT yongshenghong ensembleandtransferlearningofsoilinorganiccarbonwithvisiblenearinfraredspectra
AT songchaochen ensembleandtransferlearningofsoilinorganiccarbonwithvisiblenearinfraredspectra
AT jingliu ensembleandtransferlearningofsoilinorganiccarbonwithvisiblenearinfraredspectra
AT liliyang ensembleandtransferlearningofsoilinorganiccarbonwithvisiblenearinfraredspectra
AT jiepeng ensembleandtransferlearningofsoilinorganiccarbonwithvisiblenearinfraredspectra
AT zhoushi ensembleandtransferlearningofsoilinorganiccarbonwithvisiblenearinfraredspectra