yanziyanny/deep-soil-moisture-african-maize-code

# Deep soil moisture reveals hidden water stress in African rainfed maize systems This repository contains code and analysis-ready processed data for reproducing the submitted main-text figures, supplement figure panels, and Figure 4 machine-learning retraining for the manuscript *Deep soil moisture reveals hidden water stress in African rainfed maize systems*. ## Contents - `figure*/`: main-text figure scripts and packaged figure inputs. - `common/map_layers/`: shared map inputs. - `supplement/`: supplement figure panel reproduction. - `training/`: optional Figure 4 XGBoost retraining workflow. - `DATA_DICTIONARY.md`: packaged input file descriptions. - `MODEL_CARD.md`: Figure 4 model, split, metric, and baseline details. - `COMPUTATIONAL_RESOURCES.md`: recorded benchmark environment. ## Analysis-Ready Data The Figure 4 retraining input is: `training/data/figure4_retraining_input.csv.gz` This file is an 8-day processed panel with 1,000,127 rows. Each row represents an admin2/county-like unit in a given year and 8-day growing-season window. - ID and grouping fields: `admin2_idx`, `year`, `koppen5` - target variable: `sif_anom` - energy predictors: `VPDa_8mean`, `SWa_8mean`, `Tmaxa_8mean` - water predictors: `SMa_L1_8mean`, `SMa_L2_8mean`, `SMa_L3_8mean`, `PPTa_8sum` - raw energy fields: `VPD_8mean_raw`, `SW_8mean_raw`, `Tmax_8mean_raw` The figure scripts use packaged figure-ready inputs under `figure*/data/`, shared map layers under `common/map_layers/`, and supplement inputs under `supplement/data/`. ## Install Python 3.11 is recommended. ```bash pip install -r requirements.txt ``` For the optional ML workflow: ```bash pip install -r training/requirements-ml.txt ``` A conda-style environment file is also provided: ```bash conda env create -f environment.yml conda activate deep-soil-moisture-maize ``` `requirements-lock.txt` records package versions from the development environment. ## Main-Text Figure Reproduction Run all main-text figures: ```b …