Gaosong Shi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Research Interests
Gaosong Shi has a total of 25 co-authors. Here are the top co-authors by number of shared publications:
- Wei Shangguan
School of Atmospheric Sciences, Sun Yat-sen University, Zhuhai, China
7 shared publications - Qingliang Li
School of Artificial Intelligence, Changchun Normal University, Changchun 130032, China
5 shared publications Chunyan Wang
4 shared publicationsLu Li
4 shared publications- Xingjie Lu
School of Atmospheric Sciences, Sun Yat-Sen University, Zhuhai 519082, China
4 shared publications Ye Zhang
Sun Yat-sen University
4 shared publications
Latest Publications
- Open Access
- 7 May 2026
- Published by Elsevier in SSRN Electronic Journal
Explaining global soil organic carbon mapping discrepancies with explainable AI: A comparison of soil type linkage and digital soil mapping
Soil property maps are essential for characterizing soil spatial heterogeneity and are critical for supporting land surface modelling, agricultural management, and climate change research. However, substantial discrepancies among existing global soil property datasets resulting from methodological differences, particularly between the conventional soil type-linkage (STL) approach and modern digital soil mapping (DSM). In this study, we use an explainable artificial intelligence (XAI) framework that integrates extreme gradient boosting, Shapley additive explanations, and partial dependence plots to investigate the consistency and environmental sensitivity of soil organic carbon content (SOC) predictions with environmental factors. We compare two representative global maps: the Global Soil Dataset for Earth System Modeling (GSDE), based on STL, and SoilGrids 2.0 (SG), generated using the DSM framework. Focusing on topsoil (0-5cm) and subsoil (30-60cm) SOC in Southwest China, we assessed spatial differences between datasets and used soil profile observations as a reference to evaluate the consistency of SOC-environment relationships, serving as an external benchmark. The results indicate that while both datasets captured the general pattern of higher SOC in mountainous regions and lower SOC in the Sichuan Basin, their predictions diverged substantially in humid, high-elevation areas and across sharp topographic transitions. Benchmarking revealed that SoilGrids aligned more closely with observed values in both mean and variability, whereas GSDE systematically underestimated central tendencies and reduced variance. Interpretability analyses identified temperature and precipitation as the dominant predictors of SOC, with clear threshold responses, and highlighted important interaction effects among climatic and topographic factors. These findings clarify structural sources of disagreement between the two mapping paradigms and provide guidance on dataset selection, regional bias assessment, and the development of next-generation global soil property products. Importantly, the XAI framework provides new, process-oriented insights into how environmental controls and their interactions are differentially represented across global soil datasets.
- 7 May 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessExplaining global soil organic carbon mapping discrepancies with explainable AI: A comparison of soil type linkage and digital soil mapping
Soil property maps are essential for characterizing soil spatial heterogeneity and are critical for supporting land surface modelling, agricultural management, and climate change research. However, substantial discrepancies among existing global soil property datasets resulting from methodological differences, particularly between the conventional soil type-linkage (STL) approach and modern digital soil mapping (DSM). In this study, we use an explainable artificial intelligence (XAI) framework that integrates extreme gradient boosting, Shapley additive explanations, and partial dependence plots to investigate the consistency and environmental sensitivity of soil organic carbon content (SOC) predictions with environmental factors. We compare two representative global maps: the Global Soil Dataset for Earth System Modeling (GSDE), based on STL, and SoilGrids 2.0 (SG), generated using the DSM framework. Focusing on topsoil (0-5cm) and subsoil (30-60cm) SOC in Southwest China, we assessed spatial differences between datasets and used soil profile observations as a reference to evaluate the consistency of SOC-environment relationships, serving as an external benchmark. The results indicate that while both datasets captured the general pattern of higher SOC in mountainous regions and lower SOC in the Sichuan Basin, their predictions diverged substantially in humid, high-elevation areas and across sharp topographic transitions. Benchmarking revealed that SoilGrids aligned more closely with observed values in both mean and variability, whereas GSDE systematically underestimated central tendencies and reduced variance. Interpretability analyses identified temperature and precipitation as the dominant predictors of SOC, with clear threshold responses, and highlighted important interaction effects among climatic and topographic factors. These findings clarify structural sources of disagreement between the two mapping paradigms and provide guidance on dataset selection, regional bias assessment, and the development of next-generation global soil property products. Importantly, the XAI framework provides new, process-oriented insights into how environmental controls and their interactions are differentially represented across global soil datasets.
- Open Access
Reducing location error of legacy soil profiles leads to improvement in digital soil mapping
- Open Access
- 18 January 2024
- Published by Wiley
Projections of Soil Organic Carbon in China: The Role of Carbon Fluxes Revealed by Explainable Artificial Intelligence
The impact of carbon fluxes on soil organic carbon (SOC) remains underexplored. We employed machine learning to model SOC dynamics. Our findings project an increase in China’s SOC through to the year 2100 across various Shared Socioeconomic Pathways. Sensitivity analyses have identified carbon fluxes as the main drivers for this projected rise, followed by climate and land use. Further examination using an explainable artificial intelligence method, Shapley Additive Explanations, has uncovered both spatial and temporal variations in how gross primary production (GPP) influences SOC levels. Notably, GPP’s contribution on SOC is initially negative at low levels, turning positive once a threshold of approximately 3 gC m-2d-1 is surpassed. Beyond a GPP of about 7 gC m-2d-1, its positive contribution to SOC plateaus. Critical zones for soil carbon sequestration are located around 400 mm annual precipitation line.
- 18 January 2024
- Published by Wiley
Open AccessProjections of Soil Organic Carbon in China: The Role of Carbon Fluxes Revealed by Explainable Artificial Intelligence
The impact of carbon fluxes on soil organic carbon (SOC) remains underexplored. We employed machine learning to model SOC dynamics. Our findings project an increase in China’s SOC through to the year 2100 across various Shared Socioeconomic Pathways. Sensitivity analyses have identified carbon fluxes as the main drivers for this projected rise, followed by climate and land use. Further examination using an explainable artificial intelligence method, Shapley Additive Explanations, has uncovered both spatial and temporal variations in how gross primary production (GPP) influences SOC levels. Notably, GPP’s contribution on SOC is initially negative at low levels, turning positive once a threshold of approximately 3 gC m-2d-1 is surpassed. Beyond a GPP of about 7 gC m-2d-1, its positive contribution to SOC plateaus. Critical zones for soil carbon sequestration are located around 400 mm annual precipitation line.
- Top Cited PapersOpen Access
- 30 November 2022
- Published by Copernicus GmbH in Earth System Science Data
- Vol. 14 (12) , 5267-5286
A 1 km daily soil moisture dataset over China using in situ measurement and machine learning
High-quality gridded soil moisture products are essential for many Earth system science applications, while the recent reanalysis and remote sensing soil moisture data are often available at coarse resolution and remote sensing data are only for the surface soil. Here, we present a 1 km resolution long-term dataset of soil moisture derived through machine learning trained by the in situ measurements of 1789 stations over China, named SMCI1.0 (Soil Moisture of China by in situ data, version 1.0). Random forest is used as a robust machine learning approach to predict soil moisture using ERA5-Land time series, leaf area index, land cover type, topography and soil properties as predictors. SMCI1.0 provides 10-layer soil moisture with 10 cm intervals up to 100 cm deep at daily resolution over the period 2000–2020. Using in situ soil moisture as the benchmark, two independent experiments were conducted to evaluate the estimation accuracy of SMCI1.0: year-to-year (ubRMSE ranges from 0.041 to 0.052 and R ranges from 0.883 to 0.919) and station-to-station experiments (ubRMSE ranges from 0.045 to 0.051 and R ranges from 0.866 to 0.893). SMCI1.0 generally has advantages over other gridded soil moisture products, including ERA5-Land, SMAP-L4, and SoMo.ml. However, the high errors of soil moisture are often located in the North China Monsoon Region. Overall, the highly accurate estimations of both the year-to-year and station-to-station experiments ensure the applicability of SMCI1.0 to study the spatial–temporal patterns. As SMCI1.0 is based on in situ data, it can be a useful complement to existing model-based and satellite-based soil moisture datasets for various hydrological, meteorological, and ecological analyses and models. The DOI link for the dataset is http://dx.doi.org/10.11888/Terre.tpdc.272415 (Shangguan et al., 2022).
- 30 November 2022
- Published by Copernicus GmbH in Earth System Science Data
- Vol. 14 (12) , 5267-5286
Top Cited PapersOpen AccessA 1 km daily soil moisture dataset over China using in situ measurement and machine learning
High-quality gridded soil moisture products are essential for many Earth system science applications, while the recent reanalysis and remote sensing soil moisture data are often available at coarse resolution and remote sensing data are only for the surface soil. Here, we present a 1 km resolution long-term dataset of soil moisture derived through machine learning trained by the in situ measurements of 1789 stations over China, named SMCI1.0 (Soil Moisture of China by in situ data, version 1.0). Random forest is used as a robust machine learning approach to predict soil moisture using ERA5-Land time series, leaf area index, land cover type, topography and soil properties as predictors. SMCI1.0 provides 10-layer soil moisture with 10 cm intervals up to 100 cm deep at daily resolution over the period 2000–2020. Using in situ soil moisture as the benchmark, two independent experiments were conducted to evaluate the estimation accuracy of SMCI1.0: year-to-year (ubRMSE ranges from 0.041 to 0.052 and R ranges from 0.883 to 0.919) and station-to-station experiments (ubRMSE ranges from 0.045 to 0.051 and R ranges from 0.866 to 0.893). SMCI1.0 generally has advantages over other gridded soil moisture products, including ERA5-Land, SMAP-L4, and SoMo.ml. However, the high errors of soil moisture are often located in the North China Monsoon Region. Overall, the highly accurate estimations of both the year-to-year and station-to-station experiments ensure the applicability of SMCI1.0 to study the spatial–temporal patterns. As SMCI1.0 is based on in situ data, it can be a useful complement to existing model-based and satellite-based soil moisture datasets for various hydrological, meteorological, and ecological analyses and models. The DOI link for the dataset is http://dx.doi.org/10.11888/Terre.tpdc.272415 (Shangguan et al., 2022).
- Open Access
- 8 June 2022
- Published by Copernicus GmbH
Supplementary material to "A 1-km daily soil moisture dataset of China based on in-situ measurement using machine learning"
- 8 June 2022
- Published by Copernicus GmbH
Open AccessSupplementary material to "A 1-km daily soil moisture dataset of China based on in-situ measurement using machine learning"
- Open Access
- 8 June 2022
- Published by Copernicus GmbH in EGUsphere
- Vol. 2022, 1-29
A 1-km daily soil moisture dataset of China based on in-situ measurement using machine learning
High quality gridded soil moisture products are essential for many Earth system science applications, and they are usually available from remote sensing or model simulations with coarse resolution. Here we present a 1 km resolution long-term dataset of soil moisture derived through machine learning trained with in-situ measurements of 1,789 stations, named as SMCI1.0. Random Forest is used to predict soil moisture using ERA5-land time series, leaf area index, land cover type, topography and soil properties as covariates. SMCI1.0 provides 10-layer soil moisture with 10 cm intervals up to 100 cm deep at daily resolution over the period 2010–2020. Using in-situ soil moisture as the benchmark, two independent experiments are conducted to investigate the estimation accuracy of the SMCI1.0: year-to-year experiment (ubRMSE ranges from 0.041–0.052 and R ranges from 0.883–0.919) and station-to-station experiment (ubRMSE ranges from 0.045–0.051 and R ranges from 0.866–0.893). SMCI1.0 generally has advantages over other gridded soil moisture products, including ERA5-Land, SMAP-L4 and SoMo.ml. However, the high errors of soil moisture often located in North China Monsoon Region. Overall, the highly accurate estimations of both the year-to-year and station-to-station experiments ensure the applicability of SMCI1.0 to studies on the spatial-temporal patterns. As SMCI1.0 is based on in-situ data, it can be useful complements of existing model-based and satellite-based datasets for various hydrological, meteorological, and ecological analyses and modeling. SMCI1.0 can be accessed at http://dx.doi.org/10.11888/Terre.tpdc.272415 (Shangguan et al., 2022).
- 8 June 2022
- Published by Copernicus GmbH in EGUsphere
- Vol. 2022, 1-29
Open AccessA 1-km daily soil moisture dataset of China based on in-situ measurement using machine learning
High quality gridded soil moisture products are essential for many Earth system science applications, and they are usually available from remote sensing or model simulations with coarse resolution. Here we present a 1 km resolution long-term dataset of soil moisture derived through machine learning trained with in-situ measurements of 1,789 stations, named as SMCI1.0. Random Forest is used to predict soil moisture using ERA5-land time series, leaf area index, land cover type, topography and soil properties as covariates. SMCI1.0 provides 10-layer soil moisture with 10 cm intervals up to 100 cm deep at daily resolution over the period 2010–2020. Using in-situ soil moisture as the benchmark, two independent experiments are conducted to investigate the estimation accuracy of the SMCI1.0: year-to-year experiment (ubRMSE ranges from 0.041–0.052 and R ranges from 0.883–0.919) and station-to-station experiment (ubRMSE ranges from 0.045–0.051 and R ranges from 0.866–0.893). SMCI1.0 generally has advantages over other gridded soil moisture products, including ERA5-Land, SMAP-L4 and SoMo.ml. However, the high errors of soil moisture often located in North China Monsoon Region. Overall, the highly accurate estimations of both the year-to-year and station-to-station experiments ensure the applicability of SMCI1.0 to studies on the spatial-temporal patterns. As SMCI1.0 is based on in-situ data, it can be useful complements of existing model-based and satellite-based datasets for various hydrological, meteorological, and ecological analyses and modeling. SMCI1.0 can be accessed at http://dx.doi.org/10.11888/Terre.tpdc.272415 (Shangguan et al., 2022).