Glenn Moncrieff
  • Home
  • Papers
  • Blog
  • Projects
  • Contact

Projects

// Selected work

Projects

A few projects I have built or helped lead, across biodiversity monitoring, forecasting, and tools for conservation. Code and live demos are linked where they exist.

Forecasting

Human modification forecasting

The Human Modification metric maps how much people have altered land across the world. I build models that forecast how it will change over the coming decades. The current model is a ConvLSTM that predicts modification at five to twenty year horizons and reports a calibrated range of uncertainty rather than a single number. It is meant to help conservation planning by showing where natural habitat is most likely to be lost.

PyTorchConvLSTMEarth observationUncertainty
Model code Paper code Paper
Tool · LLMs

SciArgus

SciArgus is a free weekly newsletter that keeps up with new research for you. It reads papers published in the last week from OpenAlex, uses a language model to score each one against your interests, and emails a short digest explaining why each paper matters to your work. It runs on free services through GitHub Actions, so it costs nothing to run.

PythonLLMsGeminiOpenAlexGitHub Actions
Code & setup
Remote sensing

Remote sensing of fynbos biodiversity

Fynbos is a small, very diverse, and threatened shrubland in South Africa. I use satellite and hyperspectral imagery to map its plants and track how they recover after fire. This includes deep learning to find invasive trees, spectral unmixing of NASA's EMIT imagery to separate invasive pine from native vegetation, and Bayesian models of post-fire recovery. The work feeds into BioSCape, a NASA-led imaging spectroscopy campaign in South Africa, and EMMA, a shared biodiversity monitoring platform.

HyperspectralNASA EMITDeep learningBayesianBioSCape
hyper-iap aliens-unmix-emit postfire-statespace BioSCape paper
Biodiversity monitoring

Global Renosterveld Watch

Renosterveld is another critically endangered South African shrubland, and only a small fraction of it remains. Most of the loss happens when remnants are ploughed for crops. I trained a neural network to detect this change from Sentinel-2 time series, then turned it into an operational system. A serverless pipeline on Google Cloud and Earth Engine runs predictions every twenty days and updates a public dashboard, so loss can be spotted soon after it happens.

TensorFlowSentinel-2Earth EngineGoogle Cloud
Live dashboard Model code Pipeline Paper

© 2026 Glenn Moncrieff

  • Papers

  • Blog

  • Projects

  • Contact

Cape Town · South Africa