Attributing a deadly landslide disaster in Southeastern Brazil to human-induced climate change

This repository contains the full analytical workflow and dataset supporting the study “Attributing a deadly landslide disaster in Southeastern Brazil to human-induced climate change”. The analysis focuses on land-use and land-cover (LULC) dynamics in landslide-prone areas of Petrópolis, Rio de Janeiro, Brazil, from 1985 to 2024.

The workflow integrates Google Earth Engine (GEE) data processing, Bayesian time-series modeling, and visualization routines to quantify LULC changes and their association with landslide risk. Using MapBiomas Collection 10, a custom landslide inventory, and topographic data (SRTM DEM), the pipeline reclassifies over 60 land-cover categories into 10 consolidated classes, calculates annual class areas, generates transition matrices, and fits hierarchical Bayesian models to detect long-term trends.

The outputs include reproducible datasets, statistical summaries, and graphics illustrating forest loss, urban expansion, and their links to landslide concentration in steep terrain.

Contents

• Python script (codigo_petropolis.py) compatible with Colab/Jupyter.

• Annual LULC area CSVs (1985–2024) for the municipality and mapped landslide points.

• Aggregated transition tables and Sankey diagrams (1985→2012→2022).

• Bayesian model outputs: posterior means, HDI intervals, slope probabilities, and performance metrics.

• Figures used in the article and supplementary material (bar plots, risk charts, HDI trend curves, heatmaps).

Key features

• Reproducible and openly reusable workflow.

• Harmonized land-cover dataset for a critical hotspot of climate-related disasters in Brazil.

• Ready-to-use CSVs for further analysis in R, Python, or GIS.

• Bayesian hierarchical modeling framework to assess long-term LULC trends.

Requirements

• Python ≥ 3.9

• Libraries: earthengine-api, geemap, pandas, numpy, matplotlib, plotly, pymc, arviz, scikit-learn

• Google Earth Engine account with access to specified assets.

This dataset and code provide a valuable resource for understanding the role of land-use change in amplifying landslide risk under climate change in Southeastern Brazil.