Regulatory Language Limits Satellite-Based Enforcement in Marine Protected Areas

Abstract

# Regulatory Language Limits Satellite-Based Enforcement in Marine Protected Areas **Submission package for *Science* (Policy Forum) + Zenodo archive** Authors: Fabio Favoretto, Catalina López-Sagástegui, Paolo Guidetti, Aldo Simone, Jennifer Sletten, Virgil Zetterlind, Octavio Aburto-Oropeza, Keiron Fraser, Enric Sala. Corresponding author: Fabio Favoretto — --- ## Contents ```for_submission/├── manuscript/│ ├── Favoretto_Main_Manuscript.docx Final main text│ ├── Favoretto_Supplementary_Materials.docx Final supplementary│ └── Favoretto_Cover_Letter_Science.docx Cover letter to Science editors├── figures/│ ├── Figure1_regulation_preparedness.{pdf,png}│ ├── Figure2_improvement_potential.{pdf,png}│ ├── Figure3_legislation_enforcement.{pdf,png}│ ├── FigureS1_country_comparison.{pdf,png}│ ├── FigureS2_enforcement_features.{pdf,png}│ ├── FigureS3_legislative_age.{pdf,png}│ └── FigureS4_score_vs_recency.{pdf,png}├── code/│ ├── python/│ │ ├── nlp_contextual_scoring.py Primary scoring (RPI + Legislation Readiness)│ │ ├── build_legislation_index.py Document-year metadata + country index│ │ └── nlp_validation_scoring.py Optional LLM validation cross-check│ └── R/│ ├── 01_policy_unpreparedness_analysis.R Generates Fig 1, 2, S1, S2│ └── 02_legislation_readiness_figures.R Generates Fig 3, S3, S4├── data/│ ├── input/│ │ └── navigator_priority_countries_202507.csv 9,740 MPA regulations│ ├── legislation_texts/ 119 FAOLEX PDFs + 7 EN translations│ ├── faolex_fisheries_legislation.csv FAOLEX document metadata│ └── legislation_extracted_metadata.csv Extracted year/title metadata├── outputs/ Pre-computed analysis outputs│ ├── regulation_scores.csv 9,740 scored MPAs│ ├── legislation_country_scores.csv 15 country legislation scores│ ├── legislation_scores.csv 119 document-level scores│ ├── legislation_readiness_index.csv Country index with year metadata│ ├── manuscript_statistics.json Key stats cited in main text│ ├── manuscript_legislation_statistics.json│ └── derived/{country_summary.csv,fix_impact_summary.csv}├── requirements.txt Python dependencies├── R_dependencies.txt R package list├── LICENSE MIT└── README.md This file``` **One file is intentionally excluded** from this package due to size: `navigator_priority_countries_202507_clean.gpkg` (~581 MB), the GeoPackage of MPA polygon geometries. It is required only to regenerate **Figure 1A** (the map). It can be rebuilt from the ProtectedSeas Navigator shapefile distribution at or obtained on request from the corresponding author. --- ## Reproducing every number in the paper (≤ 5 minutes) The analysis is **fully deterministic**: no machine-learning models, no random seeds, no external APIs in the scoring path. Given the same input data, any machine with Python 3.8+ produces identical scores. ### Environment ```bash# Python (only pdfplumber is strictly required; lxml is used by nlp_validation_scoring.py)pip install -r requirements.txt # R packagesRscript -e 'install.packages(c("tidyverse","sf","viridis","scales","patchwork","ggrepel"))'``` ### Run From the root of this package: ```bash# 1. Score all 9,740 MPAs and all 119 FAOLEX documents (~2 minutes)python3 code/python/nlp_contextual_scoring.py # 2. Build the year-metadata-enriched legislation indexpython3 code/python/build_legislation_index.py # 3. Generate figures 1, 2, S1, S2 (regulation side)Rscript code/R/01_policy_unpreparedness_analysis.R # 4. Generate figures 3, S3, S4 (legislation side)Rscript code/R/02_legislation_readiness_figures.R``` Outputs are written to `outputs/latest/` next to the inputs. Compare against the pre-computed `outputs/` directory in this package to verify reproducibility. > ⚠ **Workflow note:** `build_legislation_index.py` and `nlp_contextual_scoring.py` both write `manuscript_legislation_statistics.json`. To reproduce the manuscript numbers, run `nlp_contextual_scoring.py` **last** so its statistics overwrite the build-script's. ### Optional: LLM validation cross-check `code/python/nlp_validation_scoring.py` reproduces the Claude-API-based independent scoring used to validate the keyword Legislation Readiness Index. Requires `ANTHROPIC_API_KEY` and is NOT needed to reproduce any number in the manuscript — it is only a cross-validation that the keyword scoring captures meaningful legal provisions. --- ## Key reproduced numbers (cross-checked against fresh outputs) | Claim in manuscript | Output value | Source file ||---|---|---|| 9,740 zonations, 15 countries | 9,740 / 15 | `navigator_priority_countries_202507.csv` || Mean RPI = 33 | 33.0 | `manuscript_statistics.json` || Median RPI = 33 | 33.4 | `manuscript_statistics.json` || ~40% MPAs below 30 | 39.2% | `manuscript_statistics.json` || No MPA above 75 | max score 71.3 | `regulation_scores.csv` || 4% mention monitoring | 3.7% | `manuscript_statistics.json` || 68% with prohibition | 68.1% | `manuscript_statistics.json` || 61% with discretionary | 61.3% | `manuscript_statistics.json` || Mean RPI → 76 after fixes (+130%) | 75.9 / 129.7% | `manuscript_statistics.json` || Mean Legislation Readiness = 47 | 47.0 | `manuscript_legislation_statistics.json` || Best: Maldives 67; Worst: Ecuador 19, Gabon 19 | 67 / 19 / 19 | `legislation_country_scores.csv` || Evidence admissibility mean = 11/22 | 10.9 | `manuscript_legislation_statistics.json` || Pearson r = 0.32, p = 0.25 (recency vs readiness) | 0.316 / 0.2506 | computed by `02_legislation_readiness_figures.R` | --- ## Data provenance & licensing - **ProtectedSeas Navigator** regulatory text and zone polygons: © ProtectedSeas / Anthropocene Institute, distributed under the Navigator data-use terms. The CSV extract included here covers the 15 countries used in our sample (release 2025-07).- **FAOLEX** fisheries legislation PDFs: © FAO. Included for reproducibility of the keyword scoring. Original documents are publicly accessible at .- **Analysis code** (`code/`): MIT License (see `LICENSE`).- **Pre-computed outputs and figures**: CC-BY-4.0. --- ## Citing this work ```Favoretto, F., López-Sagástegui, C., Guidetti, P., Simone, A., Sletten, J.,Zetterlind, V., Aburto-Oropeza, O., Fraser, K., & Sala, E. (2026).Regulatory Language Limits Satellite-Based Enforcement in Marine ProtectedAreas. Submitted to Science (Policy Forum). Zenodo. doi:[to be assigned]``` Once the *Science* DOI is issued, please cite the paper directly and use the Zenodo archive only for the data and code. --- ## Acknowledgments AI tools were used as a coding assistant and for proofreading the manuscript. ## Competing interests The authors declare no competing interests.