Global Dryland Desertification Risk (GDDR) Mapping

Global Dryland Desertification Risk (GDDR) Mapping Version 1.0 2026 📜 Credits Mohammad Javad Soltani Hooman Latifi 📃 Overview ◼️ Global Desertification Risk Mapping is a comprehensive remote sensing framework for assessing desertification risk across global drylands (Aridity Index < 0.65) using long-term NDVI trends, climate data, and human footprint pressure. ◼️ The framework integrates satellite-derived vegetation trends (GIMMS 3G + MODIS, 1982–2025) with Mann-Kendall trend analysis, Google Earth Engine cloud computing, and multi-metric fusion to produce continuous, unit-free desertification risk scores. ◼️ Validation is performed against SDG Indicator 15.3.1 (Trends.Earth) and land use/land cover transition analysis, ensuring scientific robustness and policy relevance. Key Innovations Feature Description 🌍 Multi-Decadal Analysis 43+ years of NDVI data (1982–2025) with GIMMS→MODIS quantile-matching harmonization 📈 Server-Side Mann-Kendall Pixel-wise trend analysis via ee.Reducer.kendallsCorrelation() — no client-side loops 🔀 Percentile-Rank Fusion Continuous risk score from geometric mean of AI, tau, and HFI percentile ranks 🧪 Ecological Significance FDR-corrected significance testing with climate-matched null models (latitude bands) ✅ SDG 15.3.1 Alignment Direct comparison with Trends.Earth land degradation indicators 🗺️ Multi-Resolution 5 km analysis resolution, IPCC region-based aggregation, patch-level validation Theoretical Foundation Riskgeomean=(AIr×Taur×HFIr)1/3 Where: AIr = 1 − percentile_rank(Aridity Index) ↳ lower AI → higher risk Taur = 1 − percentile_rank(MK tau) ↳ more negative trend → higher risk HFIr = percentile_rank(Human Footprint Index) ↳ higher pressure → higher risk 🗂️ Repository Structure Desertification-Risk/ │ ├── Base_Maps_Gen/ # Base map generation pipeline │ ├── MApGen_AI.ipynb # Aridity Index map generation │ ├── MApGen_HFP.ipynb # Human Footprint Pressure map generation │ ├── MKTrend_MApGen_MODIS.ipynb # MODIS Mann-Kendall trend maps │ ├── Redundantness.ipynb # Redundancy analysis │ └── Download_HFP_AI.py # Script: download HFP & AI data │ ├── NDVI_Trend/ # Core NDVI trend & desertification analysis │ ├── Desertification_global_1982_2025.ipynb # Full record (GIMMS+MODIS, 1982-2025) │ └── Desertification_global_2000_2025.ipynb # MODIS-era only (2000-2025) │ ├── Desertification_Gen/ # Desertification risk map generation │ ├── Desertification_Map.ipynb # Percentile-rank + geometric-mean fusion │ ├── Eco_Sig_Pathces_Map_Gen_Desert.ipynb # Ecological significance (desert risk) │ └── Eco_Sig_Pathces_Map_Gen_Tau.ipynb # Ecological significance (tau trends) │ ├── LULC_Check/ # Land use/land cover validation │ └── LULC_Check_V01.ipynb # Dryland LULC stability × desertification risk │ ├── Evaluation/ # Validation & evaluation │ └── Fianl_EV_Coor_With_SDG.ipynb # SDG 15.3.1 coordination & final validation │ ├── LICENSE └── README.md 🔬 Methodology Analysis Pipeline (4 Stages) ┌─────────────────────────────────────────────────────────────────────────┐ │ STAGE 1 — Base Map Generation │ │ │ │ GIMMS 3G (1982-1999) MODIS MOD13C2 (2000-2025) │ │ │ │ │ │ └───── QM Harmonisation ──┘ │ │ │ │ │ ┌───────┴───────┐ ┌──────────────────┐ │ │ │ CHIRPS Precip │ │ ERA5-LAND Temp │ │ │ │ (UCSB-CHG) │ │ (2m temperature) │ │ │ └───────────────┘ └──────────────────┘ │ └─────────────────────────────────────────────────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────────────────────┐ │ STAGE 2 — NDVI Trend Analysis (GEE) │ │ │ │ ┌─────────────────────────────────────────────────────────────────┐ │ │ │ ee.Reducer.kendallsCorrelation() — pixel-wise Mann-Kendall │ │ │ │ → τ (Kendall's tau) + p-value for every pixel │ │ │ └─────────────────────────────────────────────────────────────────┘ │ │ │ │ Dryland mask (AI < 0.65) applied before export │ │ Resolution: 5,000 m │ └─────────────────────────────────────────────────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────────────────────┐ │ STAGE 3 — Desertification Risk Fusion │ │ │ │ Pass 1: Stream histograms → empirical CDFs (AI, tau, HFI) │ │ Pass 2: Percentile-rank transform + 3 fusion operators: │ │ • Risk_geomean = (AI_r × Tau_r × HFI_r)^(1/3) ← headline │ │ • Risk_arithmean = (AI_r + Tau_r + HFI_r) / 3 ← sensitivity│ │ • Risk_min = min(AI_r, Tau_r, HFI_r) ← sensitivity│ │ Pass 3: Pairwise agreement & Jaccard between fusion methods │ └─────────────────────────────────────────────────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────────────────────┐ │ STAGE 4 — Validation & Evaluation │ │ │ │ • Ecological significance testing (FDR-corrected, │ │ climate-matched null models) │ │ • LULC transition analysis (degradation pathways) │ │ • SDG 15.3.1 coordination (Trends.Earth comparison) │ └─────────────────────────────────────────────────────────────────────────┘ Analysis Layers Layer Method Inputs Output NDVI Harmonisation Quantile-matching percentile stretch GIMMS 3G + MODIS Harmonised NDVI time series Mann-Kendall Trend ee.Reducer.kendallsCorrelation() NDVI ImageCollection τ + p-value per pixel Risk Map (Boolean) MK p < 0.05 ∩ τ < 0 ∩ AI < 0.65 ∩ HFP > 0.2 MK results + AI + HFP Binary risk classification Risk Map (Continuous) Percentile-rank + geometric-mean fusion AI, τ, HFI Unit-free risk score [0, 1] Ecological Significance FDR Benjamini-Hochberg + latitude-matched null Risk raster + IPCC regions Significance maps + hexbin plots LULC Stability Directed transition analysis GLDAS LULC patches Degradation pathway stats SDG Validation Spatial join + agreement metrics Risk raster + Trends.Earth Accuracy assessment 📊 Performance Metrics Desertification risk is evaluated using multiple complementary metrics, combining trend significance, ecological robustness, and policy alignment. 1. Mann-Kendall τ (Trend Strength) Measures the monotonic trend direction and magnitude of NDVI over time: τ ∈ [-1, 1] τ < 0 → browning / vegetation decline τ > 0 → greening / vegetation increase τ = 0 → no monotonic trend 2. Risk Percentile Fusion Three components are combined via geometric mean to produce a continuous risk surface: Component Variable Rationale AIr 1 − percentile(Aridity Index) Drier areas have higher inherent vulnerability Taur 1 − percentile(MK τ) Stronger negative trends indicate active degradation HFIr percentile(Human Footprint) Higher human pressure increases degradation risk 3. FDR-Corrected Significance (Benjamini-Hochberg) Controls the false discovery rate across spatial hypothesis tests: def benjamini_hochberg(pvals): p = np.asarray(pvals, dtype=float) n = p.size order = np.argsort(p) ranked = p[order] adj_ranked = ranked * n / np.arange(1, n + 1) adj_ranked = np.minimum.accumulate(adj_ranked[::-1])[::-1] adj_ranked = np.clip(adj_ranked, 0.0, 1.0) return adj_ranked[np.argsort(order)] 4. SDG Indicator 15.3.1 Agreement Validation against Trends.Earth land degradation indicators across three time windows: Window Period Source Baseline 2000–2015 SDG 15.3.1 baseline Mid-term 2004–2019 Trends.Earth v2 Recent 2008–2023 Trends.Earth v3 🚀 Quick Start ⚙️ System Requirements Python 3.13 is required. This project has been developed and tested on Python 3.13.x. Using other Python versions (3.9, 3.11+) may cause dependency conflicts, particularly with GDAL, Rasterio, and Google Earth Engine bindings. To verify your Python version: python --version # Expected: Python 3.13.x Google Earth Engine account required. You must sign up at signup.earthengine.google.com and authenticate your account. 🗺️ Installation 1. Clone the repository: git clone https://github.com/MohammadJavadSoltani/Desertification-Risk.git cd Desertification-Risk 2. Create a virtual environment (strongly recommended): python -m venv venv 3. Activate the environment: On Windows: venv\Scripts\activate On macOS/Linux: source venv/bin/activate 4. Upgrade pip, setuptools, and wheel: pip install --upgrade pip setuptools wheel 5. Install dependencies: pip install earthengine-api geemap matplotlib numpy pandas scipy statsmodels scikit-learn rasterio geopandas cartopy seaborn tqdm 6. Authenticate Google Earth Engine: earthengine authenticate 📦 Key Dependencies Package Version Role Python 3.13.x Base interpreter earthengine-api latest Google Earth Engine Python client geemap latest Interactive GEE mapping GDAL ≥3.5 Geospatial data I/O and raster processing Rasterio ≥1.3 Raster read/write built on GDAL GeoPandas ≥1.0 Vector geometry handling and spatial joins NumPy ≥1.24 Numerical computing Pandas ≥2.0 Tabular data management Matplotlib ≥3.8 Visualization Cartopy ≥0.22 Map projection and geographic plotting SciPy ≥1.11 Statistical computations Scikit-learn ≥1.3 Machine learning utilities Statsmodels ≥0.14 Statistical modeling (RESTREND) tqdm latest Progress bars ▶️ Running the Pipeline Each notebook can be run independently, but the recommended workflow order is: Step 1 — Base Maps Open and run notebooks in Base_Maps_Gen/: Download_HFP_AI.py — Download Human Footprint Pressure and Aridity Index data MApGen_AI.ipynb — Generate global Aridity Index base map MApGen_HFP.ipynb — Generate Human Footprint Pressure base map MKTrend_MApGen_MODIS.ipynb — Generate MODIS MK trend base maps Step 2 — NDVI Trend Analysis Choose the appropriate notebook in NDVI_Trend/: Desertification_global_1982_2025.ipynb — Full 43-year record (GIMMS + MODIS) Desertification_global_2000_2025.ipynb — MODIS-era only Update file paths at the top of the notebook: # Example path configuration DATA_DIR = Path("/path/to/your/data") OUT_DIR = Path("/path/to/your/output") Then run all cells. The notebook will: Install dependencies Authenticate GEE Export NDVI ImageCollections by year Run server-side Mann-Kendall trend analysis Generate desertification risk maps Export results as GeoTIFF Step 3 — Desertification Risk Fusion Run Desertification_Gen/Desertification_Map.ipynb with: # Path to your MK trend outputs TAU_RASTER = "/path/to/tau_merged_global.tif" SIG_RASTER = "/path/to/sig_merged_global.tif" AI_RASTER = "/path/to/aridity_index.tif" HFP_RASTER = "/path/to/human_footprint.tif" # Output OUTPUT_DIR = "/path/to/desertification_risk_output" Step 4 — Ecological Significance Run the ecological significance notebooks: Eco_Sig_Pathces_Map_Gen_Desert.ipynb — For desertification risk maps Eco_Sig_Pathces_Map_Gen_Tau.ipynb — For tau trend maps Step 5 — Validation Run LULC_Check/LULC_Check_V01.ipynb for LULC transition analysis Run Evaluation/Fianl_EV_Coor_With_SDG.ipynb for SDG 15.3.1 validation 🌐 Data Sources Dataset Source Period Resolution Usage GIMMS 3G NDVI NASA ARC 1982–1999 8 km → 5 km Historical NDVI MODIS MOD13C2 NASA LP DAAC 2000–2025 0.05° → 5 km Recent NDVI CHIRPS Precipitation UCSB-CHG 1982–2025 0.05° Climate normalization ERA5-LAND Temperature ECMWF/Copernicus 1982–2025 0.1° Climate normalization Aridity Index (AI) CGIAR-CSI / GEE Long-term avg 5 km Dryland delineation Human Footprint (HFP) WCS/NASA SEDAC 2013 1 km Human pressure IPCC Reference Regions IPCC-WGI — Vector Regional aggregation Trends.Earth SDG 15.3.1 Conservation International 2000–2023 Point data Validation 📄 Citation If you use this framework in your research, please cite the following: Soltani, M.J., Latifi, H. (2026). Global Desertification Risk Mapping Using Multi-Decadal NDVI Trends, Satellite Climate Data, and Continuous Risk Fusion. Scientific Data. BibTeX: @article{ title = {Global Desertification Risk Mapping Using Multi-Decadal NDVI Trends, Satellite Climate Data, and Continuous Risk Fusion}, author = {Soltani, M.J. and Latifi, H.}, journal = {Scientific Data}, year = {2026}, note = {Submitted} } 📧 Contact Mohammad Javad Soltani📧 [e-mail]🌐 GitHub Profile