Jiao et al. (2025) Mapping stability and instability hotspots in Jiangsu’s vegetation: an explainable machine learning approach to climatic and anthropogenic drivers

Identification

• Journal: Frontiers in Plant Science
• Year: 2025
• Date: 2025-12-01
• Authors: Fusheng Jiao, XU Xiao-juan, Haibo Gong, Chuanzhuang Liang, Jing Liu, Kun Zhang, Yue Yang, Dayi Lin, Naifeng Lin, Changxin Zou, Jie Qiu
• DOI: 10.3389/fpls.2025.1678262

Research Groups

• Nanjing Institute of Environmental Sciences, Ministry of Environmental Protection, Nanjing, China
• School of Geography, Nanjing Normal University, Nanjing, China
• Institute of Carbon Neutrality, College of Urban and Environmental Sciences, Peking University, Beijing, China
• Faculty of Geo-Information and Earth Observation (ITC), University of Twente, Enschede, Netherlands

Short Summary

This study investigated the spatiotemporal patterns and climatic drivers of vegetation stability across Jiangsu Province, China, using an explainable machine learning approach. It found that while most areas showed enhanced stability, 15.77% experienced increasing instability, primarily driven by background solar radiation and its temporal variability, followed by vapor pressure deficit.

Objective

• Map the spatial distribution and trends of vegetation stability in Jiangsu Province.
• Identify regional hotspots of increasing vegetation instability.
• Quantify the relative importance of background climate conditions and climate variability in shaping vegetation stability patterns.

Study Configuration

• Spatial Scale: Jiangsu Province, China, covering approximately 1.07 × 10^5 square kilometers. Data resolution was 30 meters.
• Temporal Scale: 1984–2023 (40 years). A 15-year moving window was used for trend analysis of stability metrics.

Methodology and Data

• Models used:
  • Extreme Gradient Boosting (XGBoost)
  • SHapley Additive exPlanations (SHAP)
  • Theil–Sen robust regression method
  • Spatial interpolation methods (thin plate spline (ANUSPLIN) and inverse distance weighting (IDW))
• Data sources:
  • Annual maximum kernel normalized difference vegetation index (kNDVImax) dataset (1984–2023) at 30 m resolution, derived from Landsat surface reflectance imagery (TM, ETM+, OLI sensors).
  • Daily meteorological observations (air temperature, precipitation, relative humidity, shortwave radiation) from ground-based stations (China Meteorological Administration).
  • Annual soil moisture data from the Global Land Evaporation Amsterdam Model (GLEAM).
  • Vegetation stability indicators: Proportional Variability (PV) and Lag-one Autocorrelation (AR).
  • Climatic variables: Precipitation, air temperature, soil moisture, vapor pressure deficit (VPD), solar radiation (background conditions and temporal stability metrics).

Main Results

• 15.77% of Jiangsu Province experienced increases in Proportional Variability (PV) and Lag-one Autocorrelation (AR), indicating growing vegetation instability, particularly in south-central and southeastern regions.
• 84.23% of the area showed declining PV and AR trends, suggesting enhanced stability, mainly in southwestern, northern, and central regions.
• High AR values (low resilience) were observed in approximately 25% of the region, mainly in western and southern Jiangsu, while high PV values (elevated interannual variability) were concentrated along the eastern coast and near Lake Taihu.
• More stable areas (low PV and AR) were primarily located in central and northwestern regions, comprising approximately 32% of vegetated land.
• An interpretable machine learning model identified background solar radiation and its temporal variability as the dominant drivers of vegetation stability, followed by vapor pressure deficit (VPD). Precipitation variability had minimal influence.
• SHAP dependence plots revealed nonlinear responses: moderate radiation and higher soil moisture promoted stability, while elevated VPD and radiation variability reduced it.
• Most regions (44% ideal, 17% acceptable) were in favorable ecological condition, while approximately 20% were classified as poor or abysmal, and another 20% remained uncertain.

Contributions

• First comprehensive mapping of spatiotemporal patterns and trends of vegetation stability (using PV and AR) in Jiangsu Province, China.
• Applied an explainable machine learning framework (XGBoost + SHAP) to identify and quantify the nonlinear climatic drivers of vegetation stability, providing interpretable insights into complex climate-vegetation interactions.
• Identified solar radiation and its variability, along with vapor pressure deficit, as dominant drivers of vegetation stability in intensively managed agroecosystems, contrasting with global studies often highlighting temperature.
• Developed a multi-dimensional ecosystem state classification framework by integrating ecosystem functioning (kNDVI trend) and stability (PV and AR trends) to provide targeted land management recommendations.
• Offers data-driven insights for climate-resilient land management and ecological monitoring in rapidly urbanizing, climate-sensitive agroecosystems.

Funding

• National Natural Science Foundation of Jiangsu Province (BK20240277)
• National Key Research and Development Program of Xizang (XZ202501ZY0038)
• Science and Technology Fundamental sources Investigation Program of China (2023FY100101)
• Special Fund of the Chinese Central Government for Basic Scientific Research Operations in the commonweal Research Institute (GYZX250107)

Citation

@article{Jiao2025Mapping,
  author = {Jiao, Fusheng and Xiao-juan, XU and Gong, Haibo and Liang, Chuanzhuang and Liu, Jing and Zhang, Kun and Yang, Yue and Lin, Dayi and Lin, Naifeng and Zou, Changxin and Qiu, Jie},
  title = {Mapping stability and instability hotspots in Jiangsu’s vegetation: an explainable machine learning approach to climatic and anthropogenic drivers},
  journal = {Frontiers in Plant Science},
  year = {2025},
  doi = {10.3389/fpls.2025.1678262},
  url = {https://doi.org/10.3389/fpls.2025.1678262}
}