SYS amy_ariyo.portfolio · v2026 LOC 51.5074°N, 0.1278°W · LONDON UTC --:--:-- SCROLL 0.00
Amy Ariyo // MSci Get in touch ↗
Earth Sciences × Data

Mapping the planet.
Reading its data.

I'm Amy Ariyo, a final-year MSci Earth Sciences student at UCL. I work with satellite imagery, geospatial tools and environmental data — turning messy datasets into things people can actually use.

View projects ↗ Contact
Discipline Earth Sci. + Data
Based in London, UK
Status Open to roles
01 · About

Trained as a scientist. Wired as an analyst.

// PROFILE

I'm interested in how Earth systems behave — and what the data we collect about them can tell us.

My MSci at UCL has built up a working toolkit across Python, R, QGIS and the wider remote-sensing stack. I've used it on land-cover change detection, emissions storytelling, and lab work at Swansea University's sodium-ion battery group.

I'm comfortable cleaning a messy dataset, writing a technical report, or explaining a finding to someone outside the field. I'm looking for graduate roles that combine environmental subject knowledge with analytical work, environmental consultancy, sustainability analytics, climate-tech, or geospatial data teams.

4 yrs
UCL Earth Sciences
4
Featured projects
5+
Tools in active use
02 · Projects

Selected work.

// 04 PROJECTS
CASE 01 · GEOSPATIAL ML

Comparing K-means and Random Forest for multi-year land-cover classification

I compared K-means clustering with a Random Forest model for mapping urban land cover over five years of Sentinel-2 imagery. The focus wasn't only accuracy, I also looked at how stable each method was over time, and what it cost to run.

Python scikit-learn Sentinel-2 QGIS Google Colab Remote sensing
View project ↗
ROI · 02
Forest Dense canopy Cropland Bare soil Water Built-up
FIG.01 — Stylised land-cover classification preview · Actual maps, notebooks and results on GitHub classified
Problem
How do unsupervised and supervised classifiers compare for tracking land-cover change over several years, not just on accuracy, but on temporal stability and compute cost?
Method
Sentinel-2 imagery (2020–2024) over a fixed area; preprocessing in Python; K-means (unsupervised) against a Random Forest (supervised) trained on labelled samples; per-year confusion matrices and year-on-year change detection.
Tools
Python · scikit-learn · rasterio · NumPy · QGIS · Google Colab. Runtime tracked to estimate carbon footprint.
Results
Random Forest came out on top for per-year accuracy. K-means was more stable across years and far cheaper to run, a real trade-off if you're doing repeat monitoring.
CASE 02 · DATA STORYTELLING

An animated 30-year journey through per-capita CO₂ emissions

An R-based animated visualisation of global CO₂ emissions per capita (1990–2020), built to make long-term trends easier to read.

R ggplot2 gganimate wbstats dplyr Storytelling
View project ↗
20 15 10 5 tCO₂/cap 1990 2000 2010 2020 USA CHN WLD 1990
FIG.02 — Stylised animated chart preview · Full R code and outputs on GitHub animating
Problem
Per-capita emissions vary hugely between countries and decades. How do you make 30 years of data across hundreds of countries readable in seconds?
Method
Pulled World Bank indicators with the wbstats API; cleaned and reshaped in dplyr; rendered year-by-year frames in ggplot2; stitched into a smooth animation with gganimate.
Tools
R · ggplot2 · gganimate · wbstats · dplyr · tidyr. Output as MP4 and GIF.
Results
A short animation that gets the global emissions story across quickly. The same template now works for other indicators — energy intensity, forest cover, GDP per capita.
CASE 03 · CLIMATE DATA STORYTELLING

Climate disaster trends, 1960–2024

An animated data story built from EM-DAT disaster records, showing how reported floods, storms, droughts, heat extremes and wildfires have changed since 1960. The project balances visual impact with caveats around reporting bias and exposure.

Python pandas matplotlib animation climate risk data storytelling
View project ↗
0 100 200 300 400 events/yr 1960 1980 2000 2020 Flood Storm Ext. temp Drought Wildfire
FIG.03 — Stylised disaster-trends animation preview · Full GIF, figures and code on GitHub preview
Problem
Disaster data is often stuck in dense tables, which makes long-term change hard to read. I wanted to show how reported climate-related hazards have shifted over six decades — without overclaiming.
Method
Filtered EM-DAT / Our World in Data records to five weather-related hazard types, grouped by year and decade, then built an animated stacked area chart alongside static summary plots.
Tools
Python · pandas · matplotlib · Pillow · animation · data cleaning · summary tables.
Results
The animation makes the rise in reported disaster events clear, with floods and storms dominating. The README also explains why more reported disasters don't automatically mean more physical events.
CASE 04 · RAINFALL EXTREMES

UK rainfall extremes, 1991–2024

A small climate-data pipeline comparing extreme rainfall days across 12 UK cities. Instead of using one fixed rainfall threshold, each city is compared against its own wet-day distribution.

Python requests pandas cartopy ERA5 rainfall extremes climate analysis
View project ↗
Glasgow Edinburgh Belfast Manchester Birmingham Cardiff London Norwich EXTREME DAYS · CITY × DECADE 90s 00s 10s 20s Glasgow Edinburgh Manchester Birmingham Cardiff London Norwich FEWER MORE ERA5 · 0.25° grid
FIG.04 — Stylised rainfall-extremes preview · Full maps, heatmaps and code on GitHub preview
Problem
Annual rainfall totals hide the difference between persistent drizzle and sharp bursts. I wanted to compare where rainfall becomes extreme, and when those days tend to happen.
Method
Pulled daily precipitation from Open-Meteo's ERA5 archive, cached each city locally, defined wet days as ≥ 1 mm, then counted days above each city's own 95th-percentile wet-day threshold.
Tools
Python · requests · pandas · matplotlib · cartopy · ERA5 · Open-Meteo API.
Results
Clear geographic and seasonal contrasts — western and northern cities stand out more strongly than drier eastern locations. The project also notes the limitations of ERA5's coarse grid.
03 · Skills

Stack across the pipeline.

// CAPABILITIES

Data & Analysis

  • Python (pandas, NumPy)
  • R (tidyverse, ggplot2)
  • Statistical reasoning
  • Data visualisation
  • Excel · pivot & modelling

Geospatial

  • QGIS · raster & vector
  • Sentinel-2 imagery
  • Remote sensing workflows
  • Land-cover classification
  • Spatial statistics

Tools & Platforms

  • Google Colab
  • Jupyter notebooks
  • Git basics
  • Markdown & LaTeX
  • Lab data acquisition

Communication

  • Technical reporting
  • Science communication
  • Cross-disciplinary work
  • Strong organisation
  • Presentations & video
04 · Education & Experience

Trained at UCL. Sharpened in the field.

// CV
University College London

MSci Earth Sciences

FINAL YEAR · LONDON, UK

A four-year integrated master's covering geoscience fundamentals alongside quantitative and data-driven approaches to environmental problems.

Research Assistant
Swansea University · Sodium-ion battery lab

Hands-on electrochemistry experiments and data acquisition for a sodium-ion materials group. A useful step into clean-energy research.

Customer Assistant
Marks & Spencer · Retail

Customer-facing work — communication, reliability, and staying composed when things get busy.

// Selected modules include
M.01 GIS & Remote Sensing Geospatial
M.02 AI for Earth Observation ML / EO
M.03 Statistics for Geoscientists Quantitative
M.04 Environmental Geochemistry Environmental
M.05 Independent Research Project MSci
05 · Fieldwork

Time spent on the ground.

// FIELD LOG

Most of the data I work with starts in the field. Here are the places that shaped how I read landscapes and the rocks beneath them.

50.71°N · 2.45°W

Dorset

JURASSIC COAST · UK

Sedimentology, stratigraphy, and coastal processes. Logged sections and reconstructed depositional environments from Jurassic–Cretaceous sequences.

50.27°N · 5.05°W

Cornwall

SOUTH WEST · UK

Petrology and mineral identification in the field, linking rock types to the processes that formed them.

58.57°N · 4.75°W

NW Scotland

DURNESS · ASSYNT

Geological mapping in complex terrain — working with structural data, lithologies, and field interpretation.

53.36°N · 1.81°W

Peak District

2025 · UK

Applied geoscience focused on mining, environmental impact, and resource assessment.

UPCOMING · 2026

Germany

CENTRAL EUROPE

Heading out next — looking forward to comparing field methods and geology with what I've seen in the UK.

06 · Interests

Off the clock.

// PERSONAL

A few of the things I spend time on outside study and work.

Guitar
Mostly self-taught, still learning.
Speedcubing
Solving fast, then trying to go faster.
Physical media
CDs, old hardware, anything tactile. CRT TV included.
Ballet
Long-time hobby. Good for posture, better for patience.
Anime
Always something on the watch list.
07 · Video

Explaining science through video.

// MULTIMEDIA

Short videos where I break down environmental and Earth-science topics for people outside the field.

EP.01 · LAND-COVER CHANGE

Visualising land-cover change with Sentinel-2

A walkthrough of how raw satellite bands turn into a classified map you can actually reason about.

EP.02 · SEA LEVEL RISE

Understanding sea level rise

A short presentation covering the sea level equation, the main contributors (eustatic, isostatic, steric), and how sea level is measured.

08 · Contact
// END OF FEED

Let's build something
that matters.

Open to graduate roles in environmental consultancy, sustainability analytics, climate-tech, and geospatial data. Happy to chat about a project, a placement, or an idea.

ariyoamy@gmail.com ↗ LinkedIn ↗ GitHub ↗ Download CV (PDF) ↗