Spatiotemporal Evolution and Driving Mechanisms of Eco-Environmental Quality in the Northern Tibetan Plateau Based on an Improved SRSEI Data and code
Description
This repository contains the complete research dataset and executable Python scripts supporting the study on eco-environmental quality (EEQ) dynamics, explainable machine-learning forecasting, and climate change responses in the Northern Tibetan Plateau (NTP). 1. Overview: The package provides historical (2000–2025) and future CMIP6 multi-scenario projected Soil-Adjusted Remote Sensing Ecological Index (SRSEI) rasters, multi-source environmental driving factors, inter-model uncertainty layers, and the full modeling pipeline. 2. Structure and File Contents: (1) Python Scripts: - 'SHAP value calculation.py': Implements TreeExplainer to conduct game-theoretic feature screening and quantify non-linear interaction surfaces between dominant drivers (DEM and TEMP). - 'Multi model Prediction Framework.py' & 'Optimal combined forecasting.py': Systematic benchmarking across machine learning (XGBoost, RF, ANN) and deep learning (ResNet-2D, ConvLSTM, ST-Transformer) architectures under multi-level predictor scenarios with independent temporal holdout validation (2020–2025). - 'Uncertainty, Stability, and Centroid Migration.py': Calculates inter-model standard deviation (SD) across CMIP6 GCMs, computes 26-year historical coefficient of variation (CV) for stability zoning, and tracks ecological gravity-center migration trajectories. (2) Data Archives: - 'SRSEI projection data.zip': Annual historical SRSEI grids and future projected layers under SSP1-2.6, SSP2-4.5, and SSP5-8.5 across 2021–2040, 2041–2060, and 2081–2100. - 'CMIP6 data.zip': Downscaled and bias-corrected climate forcing fields from four reliable GCMs (BCC-CSM2-MR, EC-Earth3-Veg, MPI-ESM1-2-HR, MRI-ESM2-0). - 'Uncertainty.zip': Pixel-wise inter-model prediction standard deviation (SD) maps characterizing projection divergences. - 'Impact factor.zip': Preprocessed multi-source environmental covariates (topography, climate, and human activities). 3. Data Usage: These datasets and scripts are openly shared for academic purposes to support reproducible research in alpine ecological monitoring and climate risk evaluation.
Files
Steps to reproduce
1. Software & Environment Requirements: The analysis scripts are developed in Python (>= 3.8). Required dependencies include: - Data handling & GIS: numpy, pandas, rasterio, scipy - Machine learning & Explainability: xgboost, shap, scikit-learn - Visualization: matplotlib 2. Data Preparation: - Download and extract all compressed archives ('SRSEI projection data.zip', 'CMIP6 data.zip', 'Uncertainty.zip', and 'Impact factor.zip') into your local working directory. - Open the Python scripts and verify/update the file paths at the beginning of each script (e.g., 'root_dir', 'base_dir', 'output_dir') to match your local extraction directory. 3. Reproducing the Analysis Pipeline: - Step 1 (Feature Attribution & Screening): Execute 'SHAP value calculation.py' to run the TreeExplainer on the training dataset, quantify feature importance rankings, and assess the non-linear coupling effects between dominant drivers (DEM and TEMP). - Step 2 (Model Benchmarking & Selection): Run 'Multi model Prediction Framework.py' and 'Optimal combined forecasting.py' to benchmark the multi-level predictor scenarios across machine learning algorithms and evaluate temporal generalization performance on the independent testing period (2020-2025). - Step 3 (Future Projections, Uncertainty & Ecological Dynamics): Run 'Uncertainty, Stability, and Centroid Migration.py' to: (a) Quantify the inter-model prediction standard deviation (uncertainty) across CMIP6 GCMs and generate the 3x3 scenario matrix; (b) Calculate the 26-year historical coefficient of variation (CV) for pixel-level stability classification; (c) Compute the multi-decadal spatial gravity-center migration trajectories and export the coordinates to Excel. 4. Outputs: All generated high-resolution figures (600/300 DPI) and processed spatial raster layers (.tif) will be automatically written to the designated output folders.
Institutions
- Taiyuan University of TechnologyShanxi, Taiyuan