Comparison of Machine Learning and Geostatistical Methods on Mapping Soil Organic Carbon Density in Regional Croplands and Visualizing Its Location‐Specific Dominators via Interpretable Model

Land Degradation and Development, Год журнала: 2025, Номер unknown

Опубликована: Март 17, 2025

ABSTRACT High‐precision soil organic carbon density (SOCD) map is significant for understanding ecosystem cycles and estimating storage. However, the current mapping methods are difficult to balance accuracy interpretability, which brings great challenges of SOCD. In present research, a total 6223 samples were collected, along with data pertaining 30 environmental covariates, from agricultural land located in Poyang Lake Plain Jiangxi Province, southern China. Furthermore, ordinary kriging (OK), geographically weighted regression (GWR), random forest (RF), empirical Bayesian (EBK), three hybrid models (RF‐OK, RF‐EBK, RF‐GWR), constructed. These used SOCD (soil density) study region high resolution m. After that, shapley additive explanations (SHAP) quantify global contribution spatially identify dominant factors that influence variation. The outcomes suggested compared single geostatistics model model, RF method emerged as most effective predictive showcasing superior performance (coefficient determination ( R 2 ) = 0.44, root mean squared error (RMSE) 0.61 kg m −2 , Lin's concordance coefficient (LCCC) 0.58). Using SHAP, we found properties contributed prediction (81.67%). At pixel level, nitrogen dominated 50.33% farmland, followed by parent material (8.11%), available silicon (8.00%), annual precipitation (5.71%), remaining variables accounted less than 5.50%. summary, our offered valuable enlightenment toward achieving between interpretability digital mapping, deepened spatial variation farmland

Язык: Английский

---

A Novel Framework for Improving Soil Organic Carbon Mapping Accuracy by Mining Temporal Features of Time-Series Sentinel-1 Data

Land, Год журнала: 2025, Номер 14(4), С. 677 - 677

Опубликована: Март 23, 2025

Digital soil organic carbon (SOC) mapping is used for ecological protection and addressing global climate change. Sentinel-1 (S-1) microwave radar remote sensing data offer critical insights into SOC dynamics through tracking variations in moisture vegetation characteristics. Despite extensive studies using S-1 mapping, most focus on either single or multi-date periods without achieving satisfactory results. Few have investigated the potential of time-series high-accuracy mapping. This study utilized from 2017 to 2021 analyze temporal correlation between southern Xinjiang, China. The primary objective was determine optimal monitoring period SOC. Within this period, feature subsets were extracted variable selection algorithms. performance partial least squares regression, random forest, convolutional neural network–long short-term memory (CNN-LSTM) models evaluated a 10-fold cross-validation approach. findings revealed following: (1) exhibited both interannual monthly variations, with July October. volume reduced by 73.27% relative initial dataset when determined. (2) Introducing significantly improved CNN-LSTM model (R2 = 0.80, RPD 2.24, RMSE 1.11 g kg⁻1). Compared single-date 0.23) 0.33) data, R2 increased 0.57 0.47, respectively. (3) newly developed vertical–horizontal maximum mean annual cumulative indices made significant contribution (17.93%) Therefore, integrating selection, deep learning offers enhancing accuracy digital

Язык: Английский

---

Mapping Soil Organic Carbon by Integrating Time-Series Sentinel-2 Data, Environmental Covariates and Multiple Ensemble Models

Sensors, Год журнала: 2025, Номер 25(7), С. 2184 - 2184

Опубликована: Март 30, 2025

Despite extensive use of Sentinel-2 (S-2) data for mapping soil organic carbon (SOC), how to fully mine the potential time-series S-2 still remains unclear. To fill this gap, study introduced an innovative approach mining data. Using 200 top samples as example, we revealed temporal variation patterns in correlation between SOC and subsequently identified optimal monitoring time window SOC. The integration environmental covariates with multiple ensemble models enabled precise arid region southern Xinjiang, China (6109 km2). Our results indicated following: (a) exhibited both interannual monthly variations, while July August is SOC; (b) adding properties texture information could greatly improve accuracy prediction models. Soil contribute 8.85% 61.78% best model, respectively; (c) among different models, stacking model outperformed weight averaging sample terms performance. Therefore, our proved that spectral from window, integrated has a high accurate mapping.

Язык: Английский

---

Assessment of land degradation in different land uses by modeling soil salinity and soil erodibility coupled Vis-NIR spectroscopy and machine learning model

Infrared Physics & Technology, Год журнала: 2025, Номер unknown, С. 105835 - 105835

Опубликована: Март 1, 2025

Язык: Английский

---