Benchmarking Geometric Similarity Metrics for Building Footprints Using Monte Carlo Simulation

This repository contains the full code and benchmark dataset supporting the study "Benchmarking Geometric Similarity Metrics for Building Footprints Using Monte Carlo Simulation" It provides: a synthetic distortion pipeline applying four classes of geometric degradation (rigid transformation; scale and cartographic generalisation; topological noise; and a fully mixed condition) to reference building footprints; an implementation of 18 geometric similarity metrics spanning positional, areal, boundary, and morphological families; the statistical analysis code used to evaluate metric sensitivity, robustness to building morphology, and cross-metric redundancy; and the full benchmark dataset (N = 100,000 rows) generated by this pipeline. Everything needed to regenerate the benchmark dataset from scratch, or to reproduce every figure and statistic reported in the paper directly from the provided dataset, is included. Contents code/ — distortion engine, 18-metric calculator, statistical analysis library, and two Jupyter notebooks (dataset generation; figures and analysis) data/ — the 10 reference building footprints (Shapefile) used as the basis for all simulated distortions output/ — the generated benchmark dataset, benchmark_dataset_with_metrics.csv README.md — full usage instructions, data dictionary, and code overview LICENSE.md / DATA_LICENSE.md — licensing and attribution terms (see below) Requirements Python 3.9+ with: numpy, pandas, scipy, statsmodels, shapely, geopandas, matplotlib, seaborn, numba, opencv-python, tqdm, and (optional) scikit-posthocs. Quick start To reproduce the paper's figures and statistics directly from the provided dataset, run code/benchmark_analysis.ipynb — no regeneration needed. To regenerate the benchmark dataset from scratch, run code/benchmark_dataset_generation.ipynb. All stochastic distortion sampling uses a fixed NumPy seed (42), so this reproduces the provided CSV exactly. Full instructions are in README.md. License Source code: MIT License.Dataset and reference geometries: Creative Commons Attribution 4.0 International (CC BY 4.0). The 10 reference building geometries were extracted from the Topographic Objects Database (BDOT10k), Head Office of Geodesy and Cartography (GUGiK), Poland, distributed via geoportal.gov.pl, and used here in accordance with its terms of use. See DATA_LICENSE.md for full attribution details.