Zhang et al. (2025) Dynamic monitoring of ecological security patterns in arid zone oases: a remote sensing-based ecological index evolution analysis

Identification

• Journal: Scientific Reports
• Year: 2025
• Date: 2025-12-10
• Authors: Longlong Zhang, Jiaqi Zhai, Fan He, Xing Li, Tao Wang
• DOI: 10.1038/s41598-025-29774-w

Research Groups

State Key Laboratory of Water Cycle and Water Security, China Institute of Water Resources and Hydropower Research, Beijing, 100038, China

Short Summary

This study dynamically monitored ecological security patterns in three arid zone oases in China from 2000-2022 using a remote sensing ecological index (RSEI) and interpretable machine learning, identifying precipitation as the primary driver of eco-environmental quality (EEQ) dynamics and establishing critical thresholds for key factors.

Objective

• To assess the spatiotemporal dynamics of eco-environmental quality (EEQ) in three arid zone oases in northwestern China (Hetao, Ningxia, and Minqin) from 2000 to 2022.
• To quantify the contributions of climatic and anthropogenic drivers to EEQ variations at regional and sub-zonal levels, elucidating the roles of irrigation and climate conditions.
• To identify critical factor thresholds that either support or impede EEQ improvement.
• To establish a decision framework for optimizing ecological restoration projects in drylands.

Study Configuration

• Spatial Scale: Three irrigated oases in arid regions of northwestern China: Hetao Irrigated Oasis Ecosystem (HIOE), Ningxia Irrigated Oasis Ecosystem (NIOE), and Minqin Irrigated Oasis Ecosystem (MIOE), covering a total area of 25,000 square kilometers.
• Temporal Scale: 23 years, from 2000 to 2022.

Methodology and Data

• Models used:
  • Remote Sensing Ecological Index (RSEI)
  • Principal Component Analysis (PCA)
  • Theil-Sen Median Trend analysis
  • Mann–Kendall test
  • Hurst index
  • Machine learning regression models: Random Forest (RF), XGBoost (XGB), Decision Tree (DT), LightGBM (LGB), K-Nearest Neighbors (KNN), Gradient Boosting Decision Tree (GBDT). Random Forest (RF) was selected as the optimal model.
  • Explainable machine learning: Shapley Additive Explanations (SHAP) model.
  • InVEST model (for soil salinization (SI) quantification).
• Data sources:
  • Google Earth Engine (GEE) cloud computing platform.
  • MODIS datasets (MOD09A1, MOD11A2 V6, MOD13A1 V6) for surface reflectance, land surface temperature, and NDVI (2000–2022, 500 meters and 1 kilometer resolution).
  • Climatic, ecological, and socioeconomic factors (421,330 grids of 1x1 kilometer): Precipitation (PRE), Potential Evapotranspiration (PET), Soil Moisture Content (SMC), Fractional Vegetation Cover (FVC), Land Use Change (LUCC), Irrigation Volume (IV), Nighttime Light (NL), and Population Density (POP).

Main Results

• The Remote Sensing Ecological Index (RSEI) effectively captured eco-environmental quality (EEQ), with the first principal component (PC1) explaining an average of 65% of the variance.
• Spatially, 99% of RSEI values across the three oases were in the medium-to-poor range. HIOE had 56.5% poor and 42.0% medium quality, NIOE had 19.5% poor and 80.5% medium, and MIOE had 49.4% poor and 48.8% medium.
• Temporally, HIOE showed a fluctuating equilibrium in RSEI (-0.0062 per year), NIOE exhibited a fluctuating rising trend (0.0066 per year), and MIOE displayed a fluctuating downward trend (-0.0003 per year).
• Spatially, HIOE experienced 61.4% degradation, NIOE 89.9% improvement, and MIOE 53.7% degradation.
• The Hurst index indicated that 77.8% of the RSEI trends were unsustainable (Hurst index range: 0.16 to 0.75), suggesting future divergence from current trajectories. Approximately 51.8% of currently improved areas may degrade, while 33.2% of degraded areas could improve.
• Interpretable machine learning (SHAP) identified precipitation (PRE) as the most significant factor influencing RSEI across all three oases, with SHAP values of 0.031, 0.063, and 0.30. Other important factors included irrigation volume (IV), soil moisture content (SMC), potential evapotranspiration (PET), fractional vegetation cover (FVC), and nighttime light (NL), with their relative importance varying by oasis. Land use change (LUCC), population density (POP), and nighttime light (NL) generally had the least impact.
• Critical thresholds were identified: ecosystem suppression shifted to enhancement at approximately 164 millimeters per year for precipitation and 1218 millimeters per year for irrigation volume. Exceeding these thresholds could lead to soil salinization and RSEI deterioration.

Contributions

• Provides a comprehensive, long-term (23 years), and large-scale assessment of EEQ dynamics in arid zone oases using the Google Earth Engine platform, addressing a gap in existing literature.
• Introduces and applies interpretable machine learning (SHAP) to quantitatively disentangle the complex, non-linear contributions and interactions of multiple climatic and anthropogenic drivers on EEQ, moving beyond traditional statistical methods.
• Identifies critical quantitative thresholds for precipitation (164 mm/year) and irrigation volume (1218 mm/year) that mark shifts from ecosystem suppression to enhancement, offering actionable insights for water resource management.
• Establishes a threshold-driven decision framework for optimizing ecological restoration projects, particularly relevant for achieving land degradation neutrality in global drylands.
• Offers theoretical references for tackling ecological issues in dry zone oases and enhancing regional ecosystem resilience under climate change.

Funding

• National Natural Science Foundation Projects (52025093, 51979284)
• Inner Mongolia Autonomous Region Water Conservancy Science and Technology Project (NSK202406)

Citation

@article{Zhang2025Dynamic,
  author = {Zhang, Longlong and Zhai, Jiaqi and He, Fan and Li, Xing and Wang, Tao},
  title = {Dynamic monitoring of ecological security patterns in arid zone oases: a remote sensing-based ecological index evolution analysis},
  journal = {Scientific Reports},
  year = {2025},
  doi = {10.1038/s41598-025-29774-w},
  url = {https://doi.org/10.1038/s41598-025-29774-w}
}