Comparative Analysis of Carbon Density Simulation Methods in Grassland Ecosystems: A Case Study from Gansu Province, China
An accurate assessment of grassland carbon stocks is essential for understanding their role in China’s terrestrial carbon cycle. At regional scales, combining remote sensing technology with carbon density has become a common approach. However, substantial variability among remote sensing inversion m...
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-01-01
|
| Series: | Remote Sensing |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2072-4292/17/1/172 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1841548977582899200 |
|---|---|
| author | Luyao Wu Jiaqiang Du Xinying Liu Lijuan Li Xiaoqian Zhu Xiya Chen Yue Gong Yushuo Li |
| author_facet | Luyao Wu Jiaqiang Du Xinying Liu Lijuan Li Xiaoqian Zhu Xiya Chen Yue Gong Yushuo Li |
| author_sort | Luyao Wu |
| collection | DOAJ |
| description | An accurate assessment of grassland carbon stocks is essential for understanding their role in China’s terrestrial carbon cycle. At regional scales, combining remote sensing technology with carbon density has become a common approach. However, substantial variability among remote sensing inversion models, particularly in theoretical foundations, variable selection, and algorithmic implementation, introduces significant uncertainty into estimating grassland carbon density. This study focuses on the grassland ecosystems in Gansu Province, China, employing both an overall approach (without distinguishing between grassland types) and a stratified approach, classifying the grassland into seven distinct types: alpine meadow steppe, temperate steppe, lowland meadow, alpine meadow, mountain meadow, shrubby grassland, and temperate desert. Using remote sensing, topography, climate, and 490 measured sample data points, this study employs five representative inversion models from three model categories: parametric (single-factor model and stepwise multivariate linear regression), spatial (geographically weighted regression (GWR) and multi-scale geographically weighted regression (MGWR)), and non-parametric (random forest (RF)). Inversion models were constructed for four components of the grassland ecosystem: aboveground (AGBC) and belowground biomass carbon density (BGBC), dead organic matter carbon density (DOMC), and soil organic carbon density (SOC). The applicability of each model was then systematically compared and analyzed. The main conclusions are as follows: (1) The overall estimation results demonstrate that the GWR model is the optimal choice for inverting AGBC, DOMC, and SOC, with coefficients of determination (<i>R</i><sup>2</sup>) of 0.67, 0.60, and 0.92, respectively. In contrast, the MGWR model is best suited for BGBC, with an <i>R</i><sup>2</sup> value of 0.73. (2) The stratified estimation results suggest that the optimal inversion models for AGBC and BGBC are predominantly the MGWR and RF models selected through the recursive feature elimination algorithm. For DOMC, the optimal model is a spatial model, while SOC is most accurately estimated using the GWR and RF models selected via the Boruta algorithm. (3) When comparing the inversion results of the optimal overall and stratified approaches, the stratified estimations of AGBC, BGBC, and DOMC (<i>R</i><sup>2</sup> = 0.80, 0.78, and 0.73, respectively) outperformed those of the overall approach. In contrast, the SOC estimates followed an opposite trend, with the overall approach yielding a higher <i>R</i><sup>2</sup> value of 0.92. (4) Generally, variable selection significantly enhanced model accuracy, with spatial and non-parametric models demonstrating superior precision and stability in estimating the four carbon density components of grassland. These findings provide methodological guidance for converting sample point carbon density data into regional-scale estimates of grassland carbon storage. |
| format | Article |
| id | doaj-art-2961d7c195e5441b9dbe9ee584e7aa96 |
| institution | Kabale University |
| issn | 2072-4292 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Remote Sensing |
| spelling | doaj-art-2961d7c195e5441b9dbe9ee584e7aa962025-01-10T13:20:28ZengMDPI AGRemote Sensing2072-42922025-01-0117117210.3390/rs17010172Comparative Analysis of Carbon Density Simulation Methods in Grassland Ecosystems: A Case Study from Gansu Province, ChinaLuyao Wu0Jiaqiang Du1Xinying Liu2Lijuan Li3Xiaoqian Zhu4Xiya Chen5Yue Gong6Yushuo Li7State Key Laboratory of Environmental Criteria and Risk Assessment, Chinese Research Academy of Environmental Sciences, Beijing 100012, ChinaState Key Laboratory of Environmental Criteria and Risk Assessment, Chinese Research Academy of Environmental Sciences, Beijing 100012, ChinaCollege of Forestry, Central South University of Forestry and Technology, Changsha 410004, ChinaState Key Laboratory of Environmental Criteria and Risk Assessment, Chinese Research Academy of Environmental Sciences, Beijing 100012, ChinaState Key Laboratory of Environmental Criteria and Risk Assessment, Chinese Research Academy of Environmental Sciences, Beijing 100012, ChinaState Key Laboratory of Environmental Criteria and Risk Assessment, Chinese Research Academy of Environmental Sciences, Beijing 100012, ChinaState Key Laboratory of Environmental Criteria and Risk Assessment, Chinese Research Academy of Environmental Sciences, Beijing 100012, ChinaState Key Laboratory of Environmental Criteria and Risk Assessment, Chinese Research Academy of Environmental Sciences, Beijing 100012, ChinaAn accurate assessment of grassland carbon stocks is essential for understanding their role in China’s terrestrial carbon cycle. At regional scales, combining remote sensing technology with carbon density has become a common approach. However, substantial variability among remote sensing inversion models, particularly in theoretical foundations, variable selection, and algorithmic implementation, introduces significant uncertainty into estimating grassland carbon density. This study focuses on the grassland ecosystems in Gansu Province, China, employing both an overall approach (without distinguishing between grassland types) and a stratified approach, classifying the grassland into seven distinct types: alpine meadow steppe, temperate steppe, lowland meadow, alpine meadow, mountain meadow, shrubby grassland, and temperate desert. Using remote sensing, topography, climate, and 490 measured sample data points, this study employs five representative inversion models from three model categories: parametric (single-factor model and stepwise multivariate linear regression), spatial (geographically weighted regression (GWR) and multi-scale geographically weighted regression (MGWR)), and non-parametric (random forest (RF)). Inversion models were constructed for four components of the grassland ecosystem: aboveground (AGBC) and belowground biomass carbon density (BGBC), dead organic matter carbon density (DOMC), and soil organic carbon density (SOC). The applicability of each model was then systematically compared and analyzed. The main conclusions are as follows: (1) The overall estimation results demonstrate that the GWR model is the optimal choice for inverting AGBC, DOMC, and SOC, with coefficients of determination (<i>R</i><sup>2</sup>) of 0.67, 0.60, and 0.92, respectively. In contrast, the MGWR model is best suited for BGBC, with an <i>R</i><sup>2</sup> value of 0.73. (2) The stratified estimation results suggest that the optimal inversion models for AGBC and BGBC are predominantly the MGWR and RF models selected through the recursive feature elimination algorithm. For DOMC, the optimal model is a spatial model, while SOC is most accurately estimated using the GWR and RF models selected via the Boruta algorithm. (3) When comparing the inversion results of the optimal overall and stratified approaches, the stratified estimations of AGBC, BGBC, and DOMC (<i>R</i><sup>2</sup> = 0.80, 0.78, and 0.73, respectively) outperformed those of the overall approach. In contrast, the SOC estimates followed an opposite trend, with the overall approach yielding a higher <i>R</i><sup>2</sup> value of 0.92. (4) Generally, variable selection significantly enhanced model accuracy, with spatial and non-parametric models demonstrating superior precision and stability in estimating the four carbon density components of grassland. These findings provide methodological guidance for converting sample point carbon density data into regional-scale estimates of grassland carbon storage.https://www.mdpi.com/2072-4292/17/1/172grassland ecosystemcarbon densityvariable selectionmodel comparison |
| spellingShingle | Luyao Wu Jiaqiang Du Xinying Liu Lijuan Li Xiaoqian Zhu Xiya Chen Yue Gong Yushuo Li Comparative Analysis of Carbon Density Simulation Methods in Grassland Ecosystems: A Case Study from Gansu Province, China Remote Sensing grassland ecosystem carbon density variable selection model comparison |
| title | Comparative Analysis of Carbon Density Simulation Methods in Grassland Ecosystems: A Case Study from Gansu Province, China |
| title_full | Comparative Analysis of Carbon Density Simulation Methods in Grassland Ecosystems: A Case Study from Gansu Province, China |
| title_fullStr | Comparative Analysis of Carbon Density Simulation Methods in Grassland Ecosystems: A Case Study from Gansu Province, China |
| title_full_unstemmed | Comparative Analysis of Carbon Density Simulation Methods in Grassland Ecosystems: A Case Study from Gansu Province, China |
| title_short | Comparative Analysis of Carbon Density Simulation Methods in Grassland Ecosystems: A Case Study from Gansu Province, China |
| title_sort | comparative analysis of carbon density simulation methods in grassland ecosystems a case study from gansu province china |
| topic | grassland ecosystem carbon density variable selection model comparison |
| url | https://www.mdpi.com/2072-4292/17/1/172 |
| work_keys_str_mv | AT luyaowu comparativeanalysisofcarbondensitysimulationmethodsingrasslandecosystemsacasestudyfromgansuprovincechina AT jiaqiangdu comparativeanalysisofcarbondensitysimulationmethodsingrasslandecosystemsacasestudyfromgansuprovincechina AT xinyingliu comparativeanalysisofcarbondensitysimulationmethodsingrasslandecosystemsacasestudyfromgansuprovincechina AT lijuanli comparativeanalysisofcarbondensitysimulationmethodsingrasslandecosystemsacasestudyfromgansuprovincechina AT xiaoqianzhu comparativeanalysisofcarbondensitysimulationmethodsingrasslandecosystemsacasestudyfromgansuprovincechina AT xiyachen comparativeanalysisofcarbondensitysimulationmethodsingrasslandecosystemsacasestudyfromgansuprovincechina AT yuegong comparativeanalysisofcarbondensitysimulationmethodsingrasslandecosystemsacasestudyfromgansuprovincechina AT yushuoli comparativeanalysisofcarbondensitysimulationmethodsingrasslandecosystemsacasestudyfromgansuprovincechina |