A China dataset of soil properties for land surface modelling (version 2, CSDLv2)

Earth system science data, Journal Year: 2025, Volume and Issue: 17(2), P. 517 - 543

Published: Feb. 7, 2025

Abstract. Accurate and high-resolution spatial soil information is crucial for efficient sustainable land use, management, conservation. Since the establishment of digital mapping (DSM) GlobalSoilMap working group, significant advances have been made in terms availability quality globally. However, accurately predicting variation over large complex areas with limited samples remains a challenge, especially China, which has diverse landscapes. To address this we utilised 11 209 representative multi-source legacy profiles (including Second National Soil Survey World Information Service, First regional databases) soil-forming environment characterisation. Using advanced ensemble machine learning high-performance parallel-computing strategy, developed comprehensive maps 23 physical chemical properties at six standard depth layers from 0 to 2 m China 90 resolution (China dataset surface modelling version 2, CSDLv2). Data-splitting independent-sample validation strategies were employed evaluate accuracy predicted maps' quality. The results showed that significantly more accurate detailed compared traditional type linkage methods (i.e. CSDLv1, first dataset), SoilGrids 2.0, HWSD 2.0 products, effectively representing across China. prediction all intervals ranged good moderate, median model efficiency coefficients most ranging 0.29 0.70 during data-splitting 0.25 0.84 validation. wide range between 5 % lower 95 upper limits may indicate substantial room improvement current predictions. relative importance environmental covariates predictions varied property depth, indicating complexity interactions among multiple factors formation processes. As used study mainly originate conducted 1970s 1980s, they could provide new perspectives on changes, together existing based 2010s. findings make important contributions project can also be Earth system better represent role hydrological biogeochemical cycles This freely available https://www.scidb.cn/s/ZZJzAz (last access: 17 November 2024​​​​​​​) or https://doi.org/10.11888/Terre.tpdc.301235 (Shi Shangguan, 2024).

Language: Английский

---

Using Constrained K-Means Clustering for Soil Texture Mapping with Limited Soil Samples

Agronomy, Journal Year: 2025, Volume and Issue: 15(5), P. 1220 - 1220

Published: May 17, 2025

Soil texture is one of the most important physical properties soil and plays a crucial role in determining its suitability for crop cultivation. Currently, supervised classification machine learning methods are commonly used digital mapping. However, these may not yield optimal predictive performance due to limited number samples. Therefore, we propose using Constrained K-Means Clustering combine small labeled samples with large amount unlabeled data, thereby achieving improved prediction In this study, focused on typical hilly region northern Jurong City, Jiangsu Province, China, as our mapping model. GF-2 remote sensing imagery ALOS elevation model (DEM), along their derived variables, were employed environmental variables. Clustering, choice distance method key parameter. Here, four different (euclidean, maximum, manhattan, canberra) compared results those random forest (RF) multilayer perceptron (MLP) models. Notably, euclidean within achieved highest overall accuracy (OA), Kappa coefficient, Macro F1 Score, values 0.77, 0.68, 0.75, respectively. These higher than obtained by RF MLP models 0.12, 0.18, 0.26, This indicates that demonstrates strong Moreover, land use (LU), multi-resolution ridge top flatness index (MRRTF), topographic position (TPI), plan curvature (PlC) emerged variables predicting texture. Overall, proves be an effective approach, offering novel perspective

Language: Английский

---

Reply on RC1

Published: Oct. 23, 2024

Abstract. Accurate and high-resolution spatial soil information is crucial for efficient sustainable land use, management, conservation. Since the establishment of digital mapping (DSM) GlobalSoilMap working group, significant advances have been made in globally. However, accurately predicting variation over large complex areas with limited samples remains a challenge, especially China, which has diverse landscapes. To address this we utilized 11,209 representative multi-source legacy profiles (including Second National Soil Survey World Information Service, First regional databases) soil-forming environment characterization. Using advanced Quantile Regression Forest algorithms high-performance parallel computing strategy, developed comprehensive maps 23 physical, chemical fertility properties at six standard depth layers from 0 to 2 meters China 90 m resolution (China dataset surface modeling version 2, CSDLv2). Data-splitting independent validation strategies were employed evaluate accuracy predicted quality. The results showed that significantly more accurate detailed compared traditional type linkage methods (i.e., CSDLv1, first dataset), SoilGrids 2.0, HWSD 2.0 products, effectively representing across China. prediction most 0–5 cm interval ranged good moderate, Model Efficiency Coefficients ranging 0.75 0.32 during data-splitting 0.88 0.25 sample validation. wide range between 5 % lower 95 upper limits may indicate substantial room improvement current predictions. relative importance environmental covariates predictions varied depth, indicating complexity interactions among multiple factors formation processes. As used study mainly originate 1970s 1980s, they could provide new perspectives changes together existing based on 2010s profiles. findings make important contributions project can also be Earth system better represent role hydrological biogeochemical cycles This freely available accessed https://doi.org/10.11888/Terre.tpdc.301235 (Shi et al, 2024).

Language: Английский

---