Environmental Data Analyses 🌿

This repo is a collection of reproducible environmental and ecological analyses spanning climate, habitat, biodiversity, and human–environment interactions. The emphasis is on clean data workflows, geospatial reasoning, and decision-ready visuals (maps, trend plots, compact writeups).

If you’re here from my profile: this repo is the “environmental counterpart” to my aerospace work — same focus on structure and reproducibility, applied to environmental intelligence and ecological datasets.

---

What you’ll find here

This repo focuses on end-to-end analytical work: ingesting messy public datasets, building tidy analysis-ready tables, running models or statistical summaries, and producing clear outputs.

Typical workflows include geospatial joins, raster/vector preprocessing, timeseries aggregation, and uncertainty-aware evaluation (where labels exist).

---

🌟 Featured Projects (inside this repo)

🐦 Wildlife Image Classification API

Goal: Deploy a lightweight FastAPI service for individual bird identification.

Data: Bird Individual ID dataset (Ferreira et al. 2020).

Methods: Transfer learning (ResNet18), PyTorch dataloaders, FastAPI inference, CI/CD, model versioning.

Outputs: Trained model (.pt), classes.json, API endpoints, benchmark scripts.

🔗 View Project

---

🧊 Sea Ice Trend Analysis

Goal: Quantify long‑term and seasonal changes in Arctic sea ice extent.

Data: NSIDC Sea Ice Index (NOAA), Copernicus Arctic datasets.

Methods: Raster time‑series analysis, anomaly detection, trend fitting, geospatial visualization.

Outputs: Annotated plots, exploratory notebooks, exportable datasets.

🔗 View Project

---

Repo structure

data/ is intentionally not fully checked in if it’s large. This repo favors a structure that makes it easy to reproduce results without guessing.

• notebooks/ exploratory and narrative analyses
• src/ reusable functions (loading, cleaning, geospatial utilities)
• reports/ exported figures and short writeups
• data/ small samples or metadata (large raw data is pulled via scripts)
• scripts/ one-shot CLIs to download/process data

---

Reproducibility

The goal is that a reader can clone the repo and rerun the analysis with minimal friction.

• Environment: environment.yml or pyproject.toml
• Determinism: fixed random seeds where applicable
• Data provenance: dataset sources documented in each project folder
• Outputs: figures/maps exported to reports/ (and kept lightweight)

Quickstart:

# create env (example)
conda env create -f environment.yml
conda activate envdata

# run notebooks (example)
jupyter lab