Flames of Desire

001326310070

001358750036

001358750072

Fruit Paradise

Night Flower

Research, geospatial work, hardware, or the space in between.

Satellite data, turned into decisions people can act on.

Ricky Winarko

The platform responders and researchers use to find, request and track SAR imagery over a disaster. It searches the Sentinel-1 acquisition plan, dispatches processing jobs to Airflow, and surfaces GDACS events as they happen — Django and Celery behind a CesiumJS globe.

SARFinder

A solar-powered LTE field node that measures water depth in a rice paddy, and the app that turns it into an irrigation decision. Custom PCB, ESP32-S3 firmware, MQTT backend, React dashboard and a Flutter app farmers install themselves by scanning a QR code.

PadiHAT — water in the paddy, measured from orbit and from the mud

Can ESA's P-band BIOMASS mission see flooding in Javanese rice paddies? Scored honestly against ground water-level pipes and a label-shuffle null, mostly no — and the reasons why turned out to be the interesting part.

BIOMASS Demak — flood detection at 435 MHz

An Android app for UHF RFID handhelds that turns a jewellery stock-take into a walk past the shelves. Licensed per device against its own key server, and built to keep counting when the shop's Wi-Fi does not reach.

Jewelsoft RFID Handheld

Satellite remote-sensing tutorials that run in a browser tab. Students open a Codespace and are working with real PACE, SWOT and Sentinel data in minutes, with no environment to install and no way for a course update to overwrite their work.

ES4304 Eyes on Earth

Finding out what is on in a city means scraping a lot of places politely. Ember does it from a fleet of small edge workers that submit signed batches to a Django backend, which stays the single system of record and publishes nothing without review.

Ember — event discovery on a fleet of Raspberry Pis

ESP32 firmware, a Django API and a Flutter app: soil moisture every five minutes, a pump that can be triggered from a phone or left on a schedule, and BLE provisioning so a new device never needs the Wi-Fi password hardcoded.

Plant IoT — a plant that waters itself

A long-distance companion device: messages and photos sent through a Telegram bot appear on a Raspberry Pi e-paper frame, alongside Kkomi — a one-eyed virtual pet whose mood quietly decays until somebody looks after it.

PaperHeart

Getting terabytes of ALOS-2 scenes onto a cluster is nobody's research. This turns it into one command: parallel FTP and SFTP transfers, optional PBS or Slurm submission, extraction into the layout the processors expect, and registration with the scene catalogue.

Ultimate Downloader

A Next.js platform built for Enitio, NTU's freshman orientation program — points tracking, photo galleries, and gratitude messages for participants. A middleware layer syncs and caches points data from Google Sheets; deployed on Vercel.

Enitio Website

A web platform serving the Earth Observatory of Singapore's remote-sensing products, built as part of my work there.

EOS-RS — Earth Observatory of Singapore

## Eyes on Earth

The first hour of a remote-sensing course is usually lost to conda. ES4304's tutorials are built so that it isn't: the repository opens as a pre-built GitHub Codespace with the environment already in place, and the local conda route exists mainly for instructors testing a session beforehand.

### The material

Students set up their environment and register for the data accounts first — NASA Earthdata and JAXA P-Tree, one of which takes days to approve, which is exactly why it is module 0.2 and not a surprise in week six. From there the tutorials work through real scenes: Sentinel-2 over the 2023 Merapi eruption, NDVI and NBR over the Eaton Fire, ocean colour from PACE OCI, sea surface temperature, and sea surface height from SWOT.

### Two details that make it hold up

The setup cell at the top of every notebook checks what is importable and installs only what is missing — so it does nothing at all in a Codespace, and still rescues someone running locally.

And when the course materials change mid-term, `scripts/update.py` merges the changes into a student's copy **without ever leaving conflict markers inside a notebook**, then tells them the few files where their own version was kept instead. A merge conflict in a `.ipynb` is unreadable to a student, so the updater is not allowed to create one.