Critical role of vegetation and human activity indicators in the prediction of shallow groundwater quality distribution in Jianghan Plain with LightGBM algorithm and SHAP analysis

Groundwater serves as an indispensable resource for freshwater, but its quality has experienced a notable decline over recent decades. Spatial prediction of groundwater quality (GWQ) can effectively assist managers in groundwater remediation, management, and risk control. Based on the traditional in...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Chemosphere (Oxford) Ročník 376; s. 144278
Hlavní autoři: Xiong, Hanxiang, Wang, Jinghan, Yang, Chi, Li, Shuyi, Li, Xiaobo, Xiong, Ruihan, Wang, Yuzhou, Ma, Chuanming
Médium: Journal Article
Jazyk:angličtina
Vydáno: England Elsevier Ltd 01.05.2025
Témata:
Algorithms
Boosting Machine Learning Algorithms
China
Environmental Monitoring - methods
Groundwater - chemistry
Groundwater management
Groundwater quality
Human Activities
Human activity
Humans
LightGBM
Urbanization
Vegetation
Water Pollutants, Chemical - analysis
Water Quality
China
Groundwater management
Groundwater quality
Human activity
LightGBM
Vegetation
ISSN:0045-6535, 1879-1298, 1879-1298
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Groundwater serves as an indispensable resource for freshwater, but its quality has experienced a notable decline over recent decades. Spatial prediction of groundwater quality (GWQ) can effectively assist managers in groundwater remediation, management, and risk control. Based on the traditional intrinsic groundwater vulnerability (IGV) model (DRASTIC) and three vegetation (V) indicators (NDVI, EVI, and kNDVI) and four human activity (H) indicators (land use, GDP, urbanization index, and nighttime light), we constructed four models for GWQ spatial prediction in the Jianghan Plain (JHP), namely DRASTI, DRASTIH, DRASTIV, and DRASTIVH, excluding the conductivity (C) indicator due to its uniformly low values. LightGBM algorithm, Tree-structured Parzen Estimator (TPE) optimization method, and SHapley Additive exPlanations (SHAP) analysis are used for model setting, calibration, and interpretation, respectively. The results show that nitrogen-related GWQ parameters have higher weights, and the model performs exceptionally well when considering all the indicators (accuracy = 0.840, precision = 0.824, recall = 0.832, F1 score = 0.828, AUROC = 0.914). Notably, the introduced indicators (NDVI, EVI, kNDVI, nighttime light, GDP, and urbanization index) rank as the top six in terms of importance, while traditional DRASTI and land use indicators show lower significance. Based on SHAP analysis, poor GWQ primarily occurs in areas with either extremely high or extremely low GDP and urbanization index values, and human activities are the primary cause of poor GWQ in JHP, potentially involving urbanization, industrial and agricultural activities, as well as fertilizer usage. Finally, the methodological framework proposed in this study is encouraged to be applied to diverse regions, such as plains, karst areas, mountainous regions, and coastal areas, to support effective future groundwater management. [Display omitted] •The spatial prediction of groundwater's entropy based WQI was showcased.•Four models were compared based on traditional DRASTIC framework.•LightGBM algorithm and SHAP analysis were employed.•Vegetation and human activity indicators have high importance.
Bibliografie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:0045-6535
1879-1298
1879-1298
DOI:10.1016/j.chemosphere.2025.144278