Interpretable machine learning reveals predictability and spatially heterogeneous factors associated with the RSEI in China
Abstract
Large-scale eco-environmental assessment requires not only long-term monitoring but also an understanding of short-term predictability and spatially heterogeneous environmental associations. Here, we evaluated the Remote Sensing Ecological Index (RSEI) across China from 2002 to 2024 and integrated spatiotemporal trend analysis, one-year-ahead prediction, and interpretable machine learning. The national area-weighted mean RSEI increased from 0.6385 in 2002 to 0.6614 in 2024, although the magnitude and timing of change varied markedly among climatic zones. Under a spatially and temporally held-out test design, XGBoost achieved the best predictive performance, with a weighted of 0.9712 and an RMSE of 0.0262, reducing RMSE by 13.67% relative to a persistence baseline. Variance decomposition showed that 97.39% of total RSEI variability was attributable to between-pixel spatial differences, whereas only 2.61% reflected within-pixel interannual variation. Interpretable modelling further revealed contrasting association structures: elevation, warm-season temperature and land cover were most important for spatial-level differences, whereas warm-season precipitation, land cover and potential evapotranspiration dominated interannual variability. These relationships also differed substantially among climatic zones. Our results show that RSEI dynamics are highly predictable at a one-year horizon, but the factors associated with persistent spatial patterns differ from those linked to year-to-year ecological fluctuations, highlighting the need to distinguish spatial baseline conditions from interannual environmental variability in large-scale ecological assessment.
Related
- Atmospheric Predictability beyond 30 Days with Machine Learning
- Scale-sensitive river water quality indicators reveal spatially heterogeneous drivers across China
- Unraveling the Context-Dependent Effects of Microplastics on Soil Enzymatic Activities via Interpretable Machine Learning
- Interpretable machine learning quantifies composition and size influences on aerosol spectral absorption
- Long-Term Monitoring of Greenhouse Gas Emissions and Interpretable Machine Learning Models for Methane Prediction in Dairy Cattle Slurry Storage.
BibTeX
@article{chen2026interpretable,
title = {{Interpretable machine learning reveals predictability and spatially heterogeneous factors associated with the RSEI in China}},
author = {Guoqing Chen and Chao Jian and Yanhui Zhang and Xu Zheng},
journal = {Scientific Reports},
year = {2026},
month = sep,
doi = {10.1038/s41598-026-72438-6},
url = {https://www.nature.com/articles/s41598-026-72438-6},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Interpretable machine learning reveals predictability and spatially heterogeneous factors associated with the RSEI in China
AU - Guoqing Chen
AU - Chao Jian
AU - Yanhui Zhang
AU - Xu Zheng
PY - 2026
DA - 2026/09/30
JO - Scientific Reports
DO - 10.1038/s41598-026-72438-6
UR - https://www.nature.com/articles/s41598-026-72438-6
AB - Large-scale eco-environmental assessment requires not only long-term monitoring but also an understanding of short-term predictability and spatially heterogeneous environmental associations. Here, we evaluated the Remote Sensing Ecological Index (RSEI) across China from 2002 to 2024 and integrated spatiotemporal trend analysis, one-year-ahead prediction, and interpretable machine learning. The national area-weighted mean RSEI increased from 0.6385 in 2002 to 0.6614 in 2024, although the magnitude and timing of change varied markedly among climatic zones. Under a spatially and temporally held-out test design, XGBoost achieved the best predictive performance, with a weighted of 0.9712 and an RMSE of 0.0262, reducing RMSE by 13.67% relative to a persistence baseline. Variance decomposition showed that 97.39% of total RSEI variability was attributable to between-pixel spatial differences, whereas only 2.61% reflected within-pixel interannual variation. Interpretable modelling further revealed contrasting association structures: elevation, warm-season temperature and land cover were most important for spatial-level differences, whereas warm-season precipitation, land cover and potential evapotranspiration dominated interannual variability. These relationships also differed substantially among climatic zones. Our results show that RSEI dynamics are highly predictable at a one-year horizon, but the factors associated with persistent spatial patterns differ from those linked to year-to-year ecological fluctuations, highlighting the need to distinguish spatial baseline conditions from interannual environmental variability in large-scale ecological assessment.
SN - 2045-2322
ER -