by Borevitz (ideas,team,papers) ClaudeScience-Opus5 (re-prose and python) Nov8, 2026
Tools for reading a landscape at the resolution it actually varies at, and for deciding what to do about it.
I'm Justin Borevitz at the Australian National University. These repositories are the farm-scale end of a larger idea: that the same light-driven programs studied in a 3 cm Arabidopsis rosette operate, through conserved pathways, all the way up to the land surface we can see from orbit.
Plant biology spans two nested jumps of roughly a million length units each, or a trillion in area. A tree holds on the order of a trillion plant cells. The Earth's ice-free surface holds on the order of a trillion "planet cells" — 10 m satellite pixels, each one a patch of crop, grass, shrub or tree responding to its own light, water, nutrients, genetics and management.
The analogy is not decorative. A plant cell type is defined from the bottom up by its transcriptomic signature; a planet cell type can be defined from the bottom up by its spectral-temporal signature — ten Sentinel-2 bands, weekly, for a decade. Both are phenotypes of a developmental program running in a local environment. Annual crops pulse, perennial grasses are dynamic, evergreens persist, deciduous trees switch.
Resolution has made this tractable: 30 m with Landsat-7 in 2000, 10 m with Sentinel-2 in 2017, and now near-daily commercial imagery under 4 m. As pixels approach the size of a single tree crown, Earth observation starts to look like individual-level phenotyping.
Why it matters: humans manage about half the ice-free land surface, so land accounting that averages a woodland into one "ecosystem type" buries both the losses and the gains. Granular, pixel-level accounting can tell the difference between a paddock that should stay in production and a gully that should be trees — and can support honest counterfactuals, additionality tests and origin-based trade measures.
This framing is set out in "Scaling Plant Photomorphogenesis to the Globe" (Borevitz, Brown, Aisbett, Bradley and Burley), written in honour of the late Joanne Chory, whose Harnessing Plants Initiative pursued exactly this scaling — from an Arabidopsis light-response mutant to deeper, tougher roots holding carbon in soil.
Everything here is developed against one 46.3 ha mountain property at 845 Avonside, Snowy Monaro, NSW — 888 to 1197 m, mapped at 3 m from airborne LiDAR. It is deliberately the hardest case: a block as tall as it is wide, 309 m of relief, and rainfall of 537 mm/yr against a reference evapotranspiration of 1043 — a deficit near 506 mm every year.
At that resolution the averages stop being useful. Potential ET spans 318 to 1103 mm/yr across the one property, and on slopes over 20 degrees north-facing ground demands 1071 mm/yr against 599 for south-facing: a 472 mm gap, 88% of the annual rainfall, inside one fence line. The block divides into four parts on its own terrain, and 365 mm/yr separates the thirstiest from the least thirsty.
Six small single-purpose packages, all MIT-licensed, each README leading with what it found and linking its methods. They form a chain, and each link is usable alone.
| package | reads | gives you |
|---|---|---|
| pylidarterrain | classified airborne LiDAR | the ground surface a plant actually stands on: slope, aspect, horizon angles, sky view |
| pyterrainsol | that surface + daily weather | light, PAR, daily light integral and evaporative demand, cell by cell, day by day |
| pyterrainwind | that surface + reanalysis wind | wind exposure by two schemes, and the extra demand it carries |
| pychannelform | that surface | where the water goes: catchments, channel networks, long profiles |
| pylandstore | channels + demand | where to hold water, evaporative loss, weir siting, pumping energy |
| pylandgraze | all of the above + a herd | terrain-driven grazing allocation: which ground, at which hour, in which season (concept stage) |
Light and wind are separate packages on purpose. Aspect-driven radiation is
predictable geometry and matters at any site with relief; wind needs a terrain-to-
speed transfer function that is scheme-dependent and only partly constrained, so it
is an extra position to take rather than a missing feature. The dependency runs one
way — wind needs the light chain, the light chain never needs wind — so a
collaborator modelling light installs pyterrainsol alone and never has to argue
with the wind question.
Still to come: change detection between LiDAR epochs, and a daily water balance aligned with collaborators' models rather than duplicating them.
- DAESim — the dynamic agro-ecosystem simulator this framework uses as its point model (Taghikhah, Borevitz, Costanza & Voinov 2022).
- TerraSim2 — gridded daily soil water on a terrain grid: buckets with an infiltration cap, downslope routing and Darcy subsurface drainage, validated against field logger sites. The per-cell evaporative demand computed here is what drives it.
- PaddockTS — paddock-scale satellite time series.
- shelterbelts — mapping and quantifying windbreaks.
Working notes, open questions and the hard-won traps are kept in NOTES.md, and updated as the work goes.
- 👋 Hi, I'm @jborevitz
- 👀 I'm interested in Precision Landscape Regeneration tools
- 🌱 I'm currently learning (meta)genomics and landscape phenomics
- 💞️ I'm looking to collaborate on bioinformatics, remote sensing, and agro-ecosystem modeling
- 📫 How to reach me justin.borevitz@anu.edu.au