Methods and limitations
From a 1st-century text to a 2050 habitat map and an action plan in eight steps. Each step names its sources and its main uncertainty.
Pipeline
- STEP 1Read the sourceDe Materia Medica chapters and Greek plant names (Wellmann ed.)
- STEP 2Identify modern taxaLiterature-based match, rated Certain · Probable · Disputed
- STEP 3Gather occurrencesGBIF records, cleaned, thinned to ~1 km, published only as ~10 km cells
- STEP 4Climate layersBaseline and 2041–2060 under SSP2-4.5 and SSP5-8.5, 3 climate models
- STEP 5Model habitatMaxEnt + random-forest ensemble, target-group background
- STEP 6Score riskRange loss, endemism, conservation status, harvest pressure
- STEP 7Find cultivation zonesSuitable in 2050, on farmland, outside Natura 2000, away from wild populations
- STEP 8Turn results into actionDecision rules, grower index per 10 km cell, identification timeline
Historical sources Modern data and models. Steps run left to right; the arrow ▸ marks the flow. Each step is a script in pipeline/ with cached, seeded outputs.
Data sources and licences
| Source | Used for | Licence / terms | Version |
|---|---|---|---|
| De Materia Medica, Wellmann ed. (1906–14) | Greek names, book, chapter, opening words | Public domain (Wellmann d. 1933); First1KGreek digitisation CC BY-SA 4.0 for some excerpts | archive.org scans |
| Wikimedia Commons photographs | One photo per plant | CC0, public domain, CC BY or CC BY-SA; photographer and licence shown under each photo | commons.wikimedia.org |
| Sprengel 1830 commentary; Osbaldeston 2000 | Ancient-to-modern taxon matches | Cited per plant; not redistributed | see each plant |
| GBIF occurrence download | Occurrence cells | Cite the download DOI; per-record CC0 / CC BY / CC BY-NC | DOI 10.15468/dl.4u6k8n |
| GBIF Backbone · Catalogue of Life · WCVP (Kew) | Accepted names | CC BY 4.0 | 2026 |
| WorldClim 2.1 + CMIP6 (2041–2060) | Baseline and 2050 climate | Academic, non-commercial use; no redistribution of the data | worldclim 2.5m · MPI-ESM1-2-HR, MRI-ESM2-0, IPSL-CM6A-LR |
| Copernicus DEM GLO-90 | Elevation, slope | Free licence; mandatory notice below | Copernicus GLO-90 |
| Vascular Plants of Greece checklist | Native status, endemism | Cited as facts | Dimopoulos et al. 2013, 2016 |
| IUCN Red List API v4 | Conservation status | IUCN terms of use (non-commercial) | Red List API v4 (2026) |
| Natura 2000 (EEA / DG ENV) | Protected-area boundaries | CC BY 4.0 | end-2024 |
| CORINE Land Cover 2018 (Copernicus) | Agricultural and transitional land | Full, free and open (Reg. 1159/2013) | CLC2018 100 m |
| OpenStreetMap (Geofabrik) | Roads and settlements for the harvest proxy | ODbL 1.0 | Geofabrik Greece extract |
| EMA herbal monographs (HMPC) | Regulatory status only | © EMA; reuse with acknowledgement | accessed Nov 2026 |
| Natural Earth via world-atlas | Basemap outlines | Public domain; world-atlas ISC © Michael Bostock | 1:50m |
Climate. Fick S.E. & Hijmans R.J. 2017. WorldClim 2: new 1 km spatial resolution climate surfaces for global land areas. Int. J. Climatol. 37: 4302–4315. Future climate is CMIP6 output from MPI-ESM1-2-HR, MRI-ESM2-0, IPSL-CM6A-LR, downscaled by WorldClim. We acknowledge the World Climate Research Programme, which coordinated CMIP6, the modelling groups that produced and made available their output, and the Earth System Grid Federation and its funding agencies. No WorldClim raster is redistributed here.
Elevation. Produced using Copernicus WorldDEM-90 © DLR e.V. 2010-2014 and © Airbus Defence and Space GmbH 2014-2018 provided under COPERNICUS by the European Union and ESA; all rights reserved.
Land cover. Generated using European Union's Copernicus Land Monitoring Service information; doi:10.2909/960998c1. Modified: reclassified and averaged to the climate grid.
Roads and settlements. Contains information from OpenStreetMap, available under the Open Database Licence. The derived accessibility layer is not published; the code that makes it is.
Protected areas. Natura 2000 (vector), end-2024 release © European Environment Agency / DG ENV, CC BY 4.0, rasterised to the climate grid.
Full licence notes, with the terms pages we read: DATA_LICENSES.md.
Validation
5-fold spatial block cross-validation (blocks ≈ 25 km). Values are mean ± SD across folds. Every model is also compared with a distance-only baseline on the same folds: a held-out point scores higher the closer it is to a training record. Grade: standard = ≥ 30 records, AUC ≥ 0.75 and above the baseline; data gap = fewer than 10 records; otherwise low. 19 of 32 models are standard grade and 29 of 32 beat the baseline, several only narrowly (Cretan dittany: 0.961 vs 0.959). For narrow endemics, being near known records predicts held-out records almost as well as climate does.
| Species | AUC | Baseline AUC | Boyce | Records | Model |
|---|---|---|---|---|---|
| Origanum dictamnus | 0.96 ± 0.02 | 0.96 | 0.63 | 123 | Standard |
| Hyssopus officinalis | 0.63 ± 0.00 | 0.64 | -0.03 | 20 | Low |
| Althaea officinalis | 0.87 ± 0.07 | 0.77 | 0.20 | 55 | Standard |
| Sideritis syriaca | 0.88 ± 0.10 | 0.79 | -0.04 | 23 | Low |
| Paeonia L. | 0.82 ± 0.06 | 0.66 | 0.49 | 270 | Standard |
| Glycyrrhiza glabra | 0.94 ± 0.04 | 0.86 | 0.09 | 31 | Low |
| Sambucus nigra | 0.73 ± 0.02 | 0.67 | 0.12 | 257 | Low |
| Juniperus deltoides | 0.77 ± 0.03 | 0.66 | 0.23 | 322 | Standard |
| Origanum vulgare | 0.76 ± 0.05 | 0.76 | 0.76 | 365 | Standard |
| Hypericum perforatum | 0.77 ± 0.03 | 0.71 | 0.90 | 849 | Standard |
| Urtica dioica | 0.76 ± 0.02 | 0.73 | 0.88 | 563 | Standard |
| Plantago lanceolata | 0.75 ± 0.03 | 0.71 | 0.96 | 1135 | Standard |
| Teucrium chamaedrys | 0.78 ± 0.01 | 0.76 | 0.89 | 577 | Standard |
| Melissa officinalis | 0.65 ± 0.03 | 0.60 | 0.53 | 452 | Low |
| Teucrium capitatum | 0.75 ± 0.03 | 0.75 | 0.93 | 737 | Low |
| Matricaria chamomilla | 0.78 ± 0.02 | 0.71 | 0.98 | 1296 | Standard |
| Marrubium vulgare | 0.66 ± 0.04 | 0.66 | 0.67 | 261 | Low |
| Artemisia absinthium | 0.78 ± 0.13 | 0.78 | 0.22 | 43 | Standard |
| Malva sylvestris | 0.67 ± 0.03 | 0.62 | 0.87 | 2217 | Low |
| Salvia rosmarinus | 0.69 ± 0.03 | 0.66 | 0.48 | 457 | Low |
| Foeniculum vulgare | 0.73 ± 0.02 | 0.65 | 0.89 | 971 | Low |
| Salvia fruticosa | 0.85 ± 0.02 | 0.78 | 0.93 | 1188 | Standard |
| Lavandula stoechas | 0.86 ± 0.03 | 0.85 | 0.87 | 745 | Standard |
| Thymbra capitata | 0.85 ± 0.02 | 0.78 | 0.95 | 1859 | Standard |
| Ecballium elaterium | 0.75 ± 0.01 | 0.69 | 0.97 | 4900 | Standard |
| Glaucium flavum | 0.87 ± 0.02 | 0.77 | 0.98 | 2967 | Standard |
| Centaurium erythraea | 0.71 ± 0.05 | 0.70 | 0.90 | 532 | Low |
| Mentha pulegium | 0.67 ± 0.05 | 0.61 | 0.40 | 298 | Low |
| Cyclamen graecum | 0.83 ± 0.03 | 0.81 | 0.86 | 918 | Standard |
| Pistacia lentiscus | 0.83 ± 0.01 | 0.76 | 0.97 | 5537 | Standard |
| Capparis spinosa | 0.85 ± 0.02 | 0.76 | 0.95 | 1720 | Standard |
| Mandragora officinarum | 0.90 ± 0.02 | 0.92 | 0.46 | 280 | Low |
Sensitivity of the risk ranking
Weights: range loss 0.40 · harvest pressure 0.20 · conservation status 0.20 · endemism 0.20. Each weight was varied ±50% while the others rescale. Bars show the 5–95% range of rank agreement (Kendall's τ) with the baseline ranking; the dot is the median over 1,000 draws.
Weights move the ranking far less than identification does. Modelling each disputed plant as the rival species its sources name moves several plants in or out of the top 10: see the identification sensitivity.
From results to action (stage 8)
pipeline/08_action_layers.py uses only the Python standard library and reads only what stage 7 exported, so anyone can re-run it without the modelling environment (make action). It adds no new modelling.
Decision rules
- Protect and cultivate (6): Loses habitat in all 3 climate models · has a 2050 cultivation zone.
- Protect: no cultivation fallback (5): Loses habitat in all 3 climate models · no 2050 cultivation zone.
- Cultivation opportunity (10): Gains or keeps habitat · has a 2050 cultivation zone.
- Monitor (11): Gains or keeps habitat · no 2050 cultivation zone.
Plants with a disputed identification or a low-confidence model are marked "resolve data first" in every lane. Open the action plan.
Grower index
For every ~10 km cell, the plants whose 2050 cultivation zone covers it. Zones are unions of whole grid cells, so a cell-centre test is exact. Validated: the per-cell counts match stage 7's independent raster counts in every cell (0 mismatches). SSP2-4.5 only, as stage 7 exports zone outlines for that scenario.
Identification timeline
115 claims parsed from the curated identification table: who proposed, preferred or rejected each candidate, with the authority who reports it ("Matthiolus, rejected by Sprengel"). Dates are each authority's main work, approximate and used only for ordering.
Risk weight lab
Recomputes the stage 5 score in the browser from the exported components. With the published weights it reproduces every published score to within 0.001 and the published ranking exactly; Kendall's τ is computed the same way as above.
Limitations
Identification is interpretation
Matching an ancient name to a modern species is a scholarly judgement. "Disputed" plants may describe the wrong taxon entirely.
Suitability is not presence
Correlative models assume equilibrium with climate and ignore dispersal, soils, grazing and land use. Range change assumes plants can reach every newly suitable cell in Greece, so gains are optimistic; a suitable cell in 2050 may never be colonised.
Sampling bias
Occurrence records cluster near roads and well-known mountains. Target-group background and thinning reduce but do not remove this.
Harvest pressure is a proxy, and it saturates
Estimated from accessibility (range within 2 km of roads or villages) plus a documented wild-harvest flag, not measured collection. In Greece's dense road network almost every range is accessible: the component spans only 0.86–0.99 across species, so it barely separates plants and its weight hardly moves the ranking. Measured collection data (permits, trade volumes) would fix this.
Coarse grid
~10 km cells protect rare plants from collectors but hide microrefugia such as gorges and north slopes.
No health claims
This project maps plants. It makes no statement about the safety or effect of any plant or preparation. Results are hypotheses for conservation planning, not prescriptions.
How to reproduce
git clone https://github.com/SoujanyaChatti/Materia-Medica-2050.git cd Materia-Medica-2050 make env # conda env from environment.yml make data # species list, GBIF, WorldClim, DEM make models # SDMs + spatially blocked CV make risk # risk score, sensitivity, cultivation make web # export + JSON Schema validation + stage 8 make action # stage 8 only (stdlib Python, no GIS stack) make site # builds the Astro site # or simply: make all
Fixed random seeds; GBIF download DOI and all versions recorded in data/manifest.yml. Build 2026-11-08 · GBIF DOI 10.15468/dl.4u6k8n.
Team
- Soujanya ChattiResearch, modelling and web
Cite as: Chatti, S. (2026). Dioscorides' Plants, Then and Now. EthnoHack 2026. github.com/SoujanyaChatti/Materia-Medica-2050. Occurrence data: GBIF.org GBIF Occurrence Download https://doi.org/10.15468/dl.4u6k8n.