Agent-based modeling and reinforcement learning for equitable waste transitions in Costa Rica: A national policy stress-test

CONTENT

ABMMACHINENEW.py – Agent-based model of household waste and recycling with optional PAYT policy. Outputs province-level and national waste, recycling, CO₂ and biogas indicators (model_summary.csv, agent_final_states.csv).

ECONOMICNEWWASTE.py – Uses ABM and RL outputs (agent_final_states.csv, agent_states_RL.csv) to compute annualized flows, total cost, cost per ton of waste and per kg CO₂, and discounted cost metrics. Outputs economic_comparison_results.csv and cost_breakdown.csv.

WASTESENSIVILITYNEW.py – ±20% one-way sensitivity analysis on key cost parameters for ABM and RL strategies. Outputs sensitivity_analysis_results.csv for tornado plots and robustness checks.

RLFIGURENEW.py – Builds a 2×2 figure with RL results by province (CO₂ avoided, recycling share, landfill mass) and cumulative recycling rate over time using agent_states_RL.csv and model_vars_RL.csv.

FUGREALLPROVINCES.py – Creates heatmaps of household accessibility to existing and potential facilities by province and distance band using ACCESS_FINAL_OUTPUT.csv.

Shapefiles (inputLayers.*, FindCentroidsOutput.*) – GIS layers used to derive accessibility metrics. Provided for transparency; not required to run the Python scripts, which work from CSVs.