Historical Climate Reconstruction From Newspaper Phenology

ISEF Category: Earth and Environmental Sciences

Subcategory: Climate Science  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

Old newspapers can act like a hidden weather log. A line about first frost or lilac bloom may tell you what the local climate felt like before thermometers were everywhere. If you can turn those scattered mentions into a timeline, you can compare human memory, written records, and station data. That gives you a real piece of climate history, not just a graph.

What Is It?

This project asks you to reconstruct past local climate from phenology, which means the timing of natural events like flowering, ice break-up, and first frost. Newspapers often mention these events in local news, farm columns, and community notes. Those mentions can act like clues from the past. You can use OCR, which reads scanned text from images, to pull those clues out of old paper archives.

Then you can classify each mention as a real dateable event or a false hit. A text model, such as OpenAI or Llama, can help sort the mentions by topic, location, and event type. After that, you compare the event timeline with modern weather station records. Think of it like building a climate puzzle from newspaper scraps, then checking how close the puzzle comes to the thermometer record.

Why This Is a Good Topic

This is a strong science fair topic because it has real data, clear variables, and a story that matters. You can test whether historical phenology mentions track temperature or seasonal shifts, which connects to climate reconstruction and historical ecology. You also get to practice data cleaning, text classification, and validation against a known reference. That mix makes the project feel original and analytical, not just descriptive.

Research Questions

• How does the number of newspaper phenology mentions change across decades?
• What is the effect of event type, such as first frost, bloom, or ice-out, on agreement with station data?
• Does OCR confidence score predict whether a newspaper mention is usable for climate reconstruction?
• To what extent do newspaper phenology dates match modern station-derived seasonal markers?
• Which text classification method, rule-based labels or Llama-assisted labels, finds more accurate phenology mentions?
• How does newspaper source location affect the match between reported phenology and nearby weather records?

Basic Materials

• Scanned newspaper archives or online newspaper database access.
• A laptop with internet access.
• OCR software such as Tesseract or Adobe Acrobat OCR.
• Spreadsheet software such as Google Sheets or Excel.
• Basic coding environment such as Python in Google Colab.
• Access to local or regional weather station records from NOAA or a state climate office.
• A notes document for manual coding rules.
• A digital map or GIS viewer for checking newspaper source locations.

Advanced Materials

• High-resolution newspaper scans from a library archive.
• Server or cloud account for batch OCR and text processing.
• Python environment with pandas, regex, spaCy, and scikit-learn.
• Access to OpenAI or a local Llama model for text classification experiments.
• Weather station data from NOAA or a university climate archive.
• GIS software such as QGIS for mapping source coverage.
• Statistical software for time-series comparison and error analysis.
• Annotation tool for building a labeled training set.

Software & Tools

• Python: Cleans OCR text, extracts dates, and compares newspaper mentions with weather records.
• Tesseract OCR: Converts scanned newspaper pages into machine-readable text.
• Google Colab: Runs Python notebooks without installing much software on your own computer.
• QGIS: Maps newspaper sources and station locations to check spatial bias.
• PubMed: Finds review papers on phenology, climate reconstruction, and historical ecology.

Experiment Steps

1. Define one local region and a narrow set of phenology events, so your dataset stays manageable.
2. Build a search plan for newspaper archives and decide which date fields and source details you will record.
3. Create a labeling scheme that separates true phenology mentions from unrelated weather chatter.
4. Test OCR and text classification on a small sample, then check where the model misses or mislabels events.
5. Match each confirmed event to modern station data and choose a comparison metric for timing agreement.
6. Analyze bias from newspaper coverage, source location, and event type before you make any climate claim.

Common Pitfalls

• Treating every flower or frost mention as a climate record, which inflates false positives and weakens your timeline.
• Ignoring OCR errors in old scans, which can turn dates, place names, or plant species into nonsense.
• Mixing nearby towns with different microclimates, which makes your climate comparison noisy and misleading.
• Comparing newspaper event dates to the wrong station metric, which creates a mismatch between the story and the data.
• Failing to define clear coding rules, which makes your classification results hard to trust or reproduce.

What Makes This Competitive

A class-level version of this project stops at a few newspaper hits and a simple chart. A stronger version builds a labeled dataset, tests two or more classification approaches, and reports precision, recall, or error patterns. You can also make the project stronger by checking spatial bias, archive coverage, and how far each newspaper source sits from the weather station. That turns the project into a careful reconstruction study instead of a simple text search.

Project Variations

• Use school yearbooks or town history newsletters instead of newspapers to test whether community records give similar phenology signals.
• Focus only on ice-out dates for one lake or river, then compare those mentions with freeze-thaw data from NOAA.
• Compare one OCR pipeline against a manual human coding workflow to measure how much machine reading changes the final climate timeline.

Learn More

• NOAA National Centers for Environmental Information: Search for local station data, climate normals, and historical weather records.
• USGS Phenology Portal: Read about plant and animal seasonal timing and how scientists use it in research.
• USDA National Phenology Network: Find background on phenology methods, event definitions, and observation standards.
• Google Books and local newspaper archives: Search scanned historical newspapers for old phenology mentions and local weather notes.
• MIT OpenCourseWare, Introduction to Environmental Science: Review climate concepts and data interpretation basics from free course materials.
• PubMed: Search review articles on phenology, climate reconstruction, and historical ecology.