Full Written Report

The PDF report is the primary analytical document for this project. It contains the full methods, final figures/tables, statistical testing, benchmark models, and interpretation boundaries.

What the report specifically answers

Main question:

How do NASA EONET observations vary across category, time, seasonality, and space, and how does predictive performance change between two target scopes (full multi-category counts vs wildfire-only counts)?

Supporting questions:

  • Which category dominates and by how much?
  • What is the temporal trend shape, and does it differ by category?
  • How strong is monthly seasonality, and is it category-dependent?
  • Do simple benchmarks improve out-of-sample prediction under strict future-data testing?

Quantitative highlights from the report

  • Dominant class: Wildfires with 24,208 observations (89.1% of all records).
  • Peak annual volume: 2024 with 16,816 observations.
  • Full-scope benchmark: best RMSE from GAM (324.258).
  • Wildfire-only benchmark: best RMSE from OLS (571.294).
  • Best out-of-sample R2 remains weak in both scopes (0.345 full scope; -0.216 wildfire-only).

Structure of the PDF

Introduction

  • Topic motivation, dataset context, and explicit research questions.

Methods

  • Reproducible API ingestion and cleaning pipeline.
  • Descriptive + inferential workflow.
  • Predictive benchmark design with strict temporal split (train <= 2023, test >= 2024).

Results

  • Publication-ready figures and tables for dominance, trend, seasonality, and space.
  • Full-scope and wildfire-only benchmark comparison with MAE/RMSE/R2.

Conclusions and limitations

  • Final claims with numerical support and interpretation boundaries under severe class imbalance.

How to use this page

Use this page as a concise summary of what is in the PDF.
The interactive website pages are for exploration, while final.pdf is the formal report for methods, evidence, and final conclusions.