Climate_Health_attribution_visuals

Publication-quality diagrams for climate and health attribution methods: how heat exposure is linked to health outcomes, and where statistical and machine learning approaches fit in that chain.

The figures are schematic. Any numbers shown in them are illustrative, chosen to make a method legible, and are not study results. No participant data is used or stored in this repository.

Diagrams

Hand-authored SVGs in diagrams/, edited directly as vectors (Figma, Illustrator, Inkscape).

File	Shows
01_DLNM_framework.svg	Distributed lag non-linear model structure
02_Transformer_DLNM_hybrid.svg	Transformer and DLNM hybrid architecture
03_GNN_spatial_attribution.svg	Graph neural network for spatial attribution
04_causal_ML_framework.svg	Causal machine learning framework
05_end_to_end_pipeline.svg	End to end analysis pipeline
06_explainable_AI_workflow.svg	Explainable AI workflow, SHAP based

Generated figures

A second, separate set of figures is drawn with matplotlib from the scripts in scripts/. These are build outputs, written to figures/ as SVG and PNG at 300 dpi, and are not tracked in git. Edit the script, not the output.

Script	Output
01_conceptual_framework.py	Climate, exposure and health outcome framework
02_exposure_response.py	Exposure response curve and minimum mortality temperature
03_attribution_pathway.py	Causal pathway from anthropogenic forcing to health outcomes, with counterfactual comparison
04_dlnm_methodology.py	DLNM cross-basis construction
ml_01_enhanced_framework.py	ML integration points across the attribution pipeline
ml_02_methods_comparison.py	Traditional, machine learning and hybrid methods compared
ml_03_deep_learning_architecture.py	ConvLSTM and attention architecture
ml_04_explainable_ai_workflow.py	SHAP, LIME and attention interpretability workflow
ml_05_spatial_temporal_pipeline.py	Spatial-temporal processing pipeline

Building

pip install -r requirements.txt
python scripts/generate_all_diagrams.py   # conceptual and methods figures
python scripts/generate_ml_diagrams.py    # machine learning figures

Both commands run from the repository root and write to figures/. Individual figures can be built by running their script directly, for example python scripts/ml_03_deep_learning_architecture.py.

Requirements are matplotlib, numpy and scipy.

Documentation

File	Contents
CONCEPTUAL_FRAMEWORK.md	The attribution framework the diagrams illustrate
DIAGRAM_INDEX.md	Catalogue of the six SVGs in diagrams/, with components and references
DIAGRAMS_GUIDE.md	Notes on the four conceptual scripts
ML_DIAGRAMS_GUIDE.md	Notes on the five machine learning scripts
README_ML_DIAGRAMS.md	Short build guide for the machine learning scripts
ML_DIAGRAMS_SUMMARY.md	Design decisions and file reference for the ML figures
ML_INNOVATIONS.md	Background on the machine learning methods shown

Context

Methods figures for heat and health work at the HE²AT Center, Wits Planetary Health Research. Supported by NIH award U54 TW 012083.

Citing

See CITATION.cff, or use the "Cite this repository" link on the GitHub page.

Licence

MIT, see LICENSE.