This repository contains the code and analysis scripts for the paper: "Upwind terrestrial influences on soil moisture variability across South America" (Huang et al., 2025).
@Feini Huang (feinihuang9554@gmail.com)
We develop an observation-driven framework that integrates atmospheric moisture transport trajectories with deep learning to quantify vegetation's contribution to downwind soil moisture variability across South America.
Soil moisture exhibits coherent spatial patterns that reflect systematic influences from upwind terrestrial conditions. This framework quantifies how vegetation dynamics in the Amazon and other regions modulate soil moisture hundreds to thousands of kilometers downwind via atmospheric moisture transport.
Key findings:
- Upwind information contributes 67% to the model's predictive power
- Vegetation explains 15% of this upwind impact
- Vegetation-mediated influence is significantly enhanced in agricultural zones
- Effects vary nonlinearly across climatic regimes (R1-R4)
- YF-0-data-* Data preprocessing
- YF-1-model-* Model implementations
- YF-2-source_influence.ipynb Source influence analysis
- YF-3-regime_piecewise.ipynb Climatic regime classification
- YF-4-plot-* Main figures
- YF-5-analysis-* Supplementary analyses
- YF-6-plot-* Supplementary figures (S1-S22)
- Python 3.8+
- PyTorch (for deep learning models)
- Jupyter Notebook
- Key packages: numpy, pandas, xarray, matplotlib, scikit-learn, pytorch
git clone https://github.com/hydroshub/Transport-informed-model-for-remote-attribution.git
cd Transport-informed-model-for-remote-attribution
pip install -r requirements.txt # Create this file if needed