| --- |
| license: cc-by-4.0 |
| library_name: onnx |
| tags: |
| - ecology |
| - vegetation |
| - biodiversity |
| - onnx |
| --- |
| |
| # RESOLVE model suites for the browser |
|
|
| The released [RESOLVE](https://github.com/gcol33/resolve) model suites as ONNX |
| graphs, read by the [RESOLVE plot predictor](https://github.com/gcol33/resolve-web). |
| From the species list of a European vegetation plot they predict its EUNIS |
| habitat (level 1), altitude, slope, plot area, survey year and location. |
|
|
| - `V_species_context/`: models that read the species with their cover, the |
| plot's coordinates and its survey year and month (each target reads the |
| inputs its `target.json` names). |
| - `V_minimal/`: models that read the species with their cover only. |
|
|
| Each target directory holds `target.json` (vocabularies, inputs, output heads, |
| the rule that combines the members) and five members, `seed_<k>.onnx`, trained |
| with different seeds. Species are World Flora Online names as matched in the |
| European Vegetation Archive. Every graph takes the assembled species ids, |
| genus and family ids, pool weights, a cover flag and the raw continuous block |
| (NaN where missing), and returns the target in its original units. |
|
|
| The graphs reproduce the engine's `SuitePredictor`: on 2,000 plots of the UK |
| National Plant Monitoring Scheme every EUNIS label agrees, probabilities |
| differ by less than 1e-6 and regressions by less than 5e-4 of their value. |
|
|
